TW201915838A - Particle swarm optimization (PSO) fuzzy logic control (FLC) charging method applicable to smart grid in which a current-state-of-charge input membership function and a state-of-charge-variation input membership function are used to provide fuzzy results through a first and a second fuzzy operations - Google Patents

Abstract

A particle swarm optimization (PSO) fuzzy logic control (FLC) charging method which comprises: carrying out a first fuzzy operation on a current-state-of-charge according to a current-state-of-charge input membership function, and carrying out a second fuzzy operation on a state-of-charge-variation according to a state-of-charge-variation input membership function, wherein the current-state-of-charge input membership function and the state-of-charge-variation input membership function are both determined in advance according to particle swarm optimization (PSO); mapping a rule base according to a first fuzzy result of the first fuzzy operation and a second fuzzy result of the second fuzzy operation to generate a third fuzzy result; carrying out a defuzzification operation on the third fuzzy result according to an output membership function to obtain a charging ratio; and multiplying the charging ratio with a battery rating power to determine a charging power instruction.

Description

   Particle Swarm Optimization Fuzzy Logic Control Charging Method Applied to Smart Grid   

The invention relates to the optimization of smart grid power dispatching, and in particular to a method for optimizing and charging a battery energy storage module applied to a smart grid.

Today, petrochemical fuel and nuclear energy are still the main sources of global energy. However, due to the growing shortage of petrochemical energy and the large amount of greenhouse gases emitted by the use of petrochemical fuels will cause the deterioration of the global environment, climate and ecology. The 21st United Nations Climate Change Conference (COP21) in Paris, France adopted a carbon reduction agreement-the Paris Agreement to control greenhouse gas emissions, and set a target for the global average temperature to rise no more than 2 ° C by 2100. Electricity is the most important issue for human beings to continue to become civilized. As environmental protection concepts and sustainable development have become a global consensus, how to use existing energy sources more efficiently and actively develop new alternative energy sources is currently the top priority in the engineering science and technology community. Business.

In many emerging energy developments, renewable energy with low carbon emissions and high energy security has become the most promising new generation energy generation technology. Many investments and efforts have been made globally to deploy renewable energy power generation systems, and governments have also provided many incentives to promote this energy transition. And with the input of many renewable energy sources into the existing grid system, it will inevitably impact the safety and reliability of the grid's operation. Due to the different positions, output characteristics and capacity of various renewable energy sources, various renewable energy sources may have different impacts when they are connected to the power grid. In addition, renewable energy sources such as wind energy and solar energy are intermittent sources of energy. The amount of output is greatly affected by local climate factors, so power supply with renewable energy generally cannot meet the instantaneous load increase and decrease requirements. Therefore, if you can store excess electrical energy through a battery storage system (BSS), you can not only supplement peak demand, but also sell electricity to power companies, which can solve the problem of high energy costs and insufficient power, but also Can ease the pressure of electric field construction and reduce environmental pollution.

In order to integrate decentralized renewable energy generation into the grid, the current grid must be redesigned and upgraded as a whole. The smart grid is the power system architecture generated under this wave. Different renewable energy sources such as wind, solar power and energy storage systems form a diverse and decentralized power network. Unlike traditional grids, which only allow electricity to flow from generators to consumers, smart grids focus on the integration of renewable energy, two-way transmission and dispatch of electrical energy, and analysis and sharing of data to optimize the use of power resources.

With the reduction of investment cost of solar power generation and the innovation of power generation methods, the installation of solar power generation systems has become more attractive. However, when the resale price of electricity drops and retail energy prices rise, local households can generate more profits by generating electricity for their own use. Therefore, general residential houses will be equipped with a robust and efficient battery energy storage system to compensate for the often fluctuating solar power generation. Using a battery storage system, residential customers can save solar energy for use in the absence of sunlight, and power losses can be eliminated through appropriate power dispatching strategies in the system.

However, in terms of design and implementation, there are also complex and highly optimized optimization issues in smart grids involving power system engineering or telecommunications technology. The literature on the optimization of smart grids mainly includes the following three aspects:

1. Optimization of capacity planning for smart grids: Hybrid power generation systems are expected to have appropriate capacity or to continuously meet load demand capabilities, while at the same time looking for the best configuration of power systems, or even the best location, with the lowest possible investment and cost. Before the installation, type, and installation, one or more specific criteria need to be determined in advance. The reliability of electricity is usually measured by the probability of loss of electrical energy and the loss of load probability. The net present value cost, average energy cost, and life cycle cost are used. As an economic indicator. The capacity planning optimization methods can basically be divided into probabilistic methods, analytical methods, iterative methods, and computational intelligence methods. Some literatures use statistical methods to describe the capacity distribution of batteries and supercapacitors. Such methods are probabilistic capacity planning optimization methods. In addition, in order to study the impact of different capacity planning and photovoltaic cell system prices on their own consumption and self-sufficiency, there are The literature conducts technical and economic analysis and sensitivity analysis. This type of analysis method is an analysis method that belongs to the optimization of capacity planning. Some literatures have proposed simulation tools for hybrid optimization models for renewable energy, which are widely used in analysis and model verification. Such methods are based on iterative methods. Some literatures have proposed an integrated renewable energy optimization model. When field conditions and seasonal load distributions are known, different renewable energy system capacity options are optimized based on energy costs and reliability indicators. For the application of computational intelligence methods, there are literatures that use genetic algorithms to optimize the components required for smart grid systems. In addition, based on considerations of cost, reliability, and carbon emission standards, some literatures have proposed particle swarm optimization methods. To address this multi-objective planning problem. In summary, the choice of these capacity planning optimization techniques depends on the available information, goals, and simplicity.

2. Optimization of smart grid power dispatching: Some literature first established a mathematical model of components in a hybrid system, and proposed the hourly energy management of a monitoring system based on fuzzy logic control (FLC) to achieve power dispatching And energy management; there are also literatures that use mathematical methods to propose power dispatching strategies based on mixed-integer nonlinear programming to obtain the cheapest price and maximum utilization of decentralized energy; or there are literatures that summarize the constraints to mixed-integer linear Planning models and integrating power reduction strategies into load and power management. In addition, methods based on computational intelligence are widely used in hybrid power generation systems. Genetic algorithms have been used in literature to optimize independent hybrid power generation systems consisting of wind turbines, microturbines, solar cell arrays, and batteries. On the other hand, some scholars proposed the cuckoo search method to optimize the fuzzy logic control system for the operation of battery packs, photovoltaic systems, and diesel generators, thereby achieving multi-target minimization, which includes the loss of power supply probability, excess energy, and Uniform energy costs can be achieved. In order to minimize the energy provided by diesel generators and battery packs, the literature has accurately predicted wind speed and load power. There is also a literature that proposes a hierarchical control strategy that consists of master-slave control for an independent hybrid power generation system. The master control strategy is responsible for determining reference power, while the slave control strategy modifies these references to meet dynamic limits. In addition, some scholars have proposed a method based on Liquor Group Optimization to solve the multi-objective optimization problem, whose purpose is to minimize the total operating cost, fuel emissions, and loss of load probability. In another similar study, a differential evolution algorithm accompanied by fuzzy technology was proposed. For the grid-connected architecture, there is a literature design of a monitoring strategy for the microgrid power system. The proposed strategy can switch the operation mode to minimize the operation cost. There are also literatures that consider solar power levels and control modes of fuel cells to minimize the number of changes in operating modes, and control solar and fuel cells to operate at maximum power points and high efficiency, respectively.

3. Optimization of smart grid control design: In order to make full use of the subsystems in the hybrid power generation system, control strategies play an important role in further improving efficiency and cooperation with other subsystems. For battery storage systems in smart grids, There are many literatures suggesting different operating strategies. For example, when the photovoltaic cell system is considered in the literature, the needs of distribution system operators and regional power companies are clearly stated. Among them, the optimization of battery charging power is determined by the dynamic programming method, and various charging targets are included first. The objective function is then multi-objective optimized. The proposed dynamic programming method discretizes the remaining power value every 15 minutes to evaluate all possible charging trajectories from the initial remaining power at the beginning of the day to the end of the day. An optimal battery charging curve is obtained with the remaining power sequence corresponding to the minimum objective function value. This method does not set targets like the previous literature, but uses the self-consumption ratio, self-supply ratio, peak voltage reduction ratio, and loss ratio to determine the characteristics between different methods, and uses these indicators to analyze residential photovoltaic energy storage Different control strategies of the system. On the other hand, there are literatures that combine computational intelligence and advanced control technologies to promote wider battery energy storage system applications. For example, in order to reduce operating costs, there are literatures that use evolutionary algorithms to optimize the charge-discharge control in closed-loop controllers. Parameter, in this way, if the feed-in compensation tax rate is variable, energy costs can be reduced. In addition, there are references in the literature that use Newton-Raphson linear regularization iterative algorithms for energy management in microgrids, and genetic algorithms for capacity planning of distributed renewable energy power generation systems with distributed energy storage systems. In addition, fuzzy logic control-based methods are often used in battery energy storage systems to perform control techniques for different purposes. Some literatures have proposed a day-to-day scheduling algorithm to reduce the operating cost of microgrids, and established a fuzzy Expert system to control the output power of the energy storage system. In order to minimize the effective power exchange at the common coupling point of the microgrid, the literature considers the remaining power of the battery, and designs a control strategy based on fuzzy logic control for the battery energy storage system in the microgrid to adjust the battery according to different operating modes. Reference value of the effective power of the energy storage system. In order to maintain the battery cycle life, some literatures have designed a fuzzy logic control mechanism to make the battery energy storage system operate within the required remaining power range.

However, the above method has high computational complexity and cannot meet the diverse energy usage behaviors of power consumers to achieve the lowest operating cost. Therefore, a novel optimization fuzzy logic control charging method is urgently needed in this field.

It is an object of the present invention to disclose a particle swarm optimization fuzzy logic control charging method, which uses the current state of charge (SOC) and the change amount of battery power ΔSOC as fuzzy logic control input variables to effectively Perform fuzzy logic control operations and prevent battery damage caused by improper charging and discharging.

Another object of the present invention is to disclose a particle swarm optimization fuzzy logic control charging method, which uses a particle swarm optimization algorithm to optimize the set values of the SOC and △ SOC input attribution functions to satisfy various energy uses. To achieve the lowest operating costs.

Another object of the present invention is to disclose a particle swarm optimization fuzzy logic control charging method, which can achieve the lowest operating cost of 98.6% of houses, and can also minimize the average annual energy cost of a single house.

In order to achieve the foregoing object, a particle swarm optimization fuzzy logic control charging method is proposed, which is implemented using a control circuit, including the following steps: inputting a current battery level and comparing the current battery level with a previous battery level To obtain a battery power fluctuation amount; perform a first fuzzification operation on the current battery power according to a current battery power input attribution function to obtain a first fuzzification result; and input a battery function power attribution function to the battery A second fuzzification operation is performed to obtain a second fuzzification result for the amount of power variation, wherein the current battery power input attribution function and the battery power input variation attribution function are determined in advance according to a particle group algorithm; Mapping a first fuzzy result and the second fuzzy result to a rule base to generate a third fuzzy result; performing a deblurring operation on the third fuzzy result according to an output assignment function to obtain a charging rate; And multiplying the charging rate by a battery rated power to determine a charging power command.

In an embodiment, the current battery power input attribution function corresponds to five semantic parameters, where S represents small, MS represents small, M represents medium, ML represents large, and L represents large; the amount of battery power variation The input assignment function corresponds to five semantic parameters, where NL represents negative large, NS represents negative small, Z represents zero, PS represents positive small, and PL represents positive large; the output assignment function corresponds to five semantic parameters, Among them, S is small, MS is small, M is medium, ML is large, and L is large. In an embodiment, the current battery power input attribution function and the battery power variation input attribution function are determined in advance by the particle swarm algorithm, and three fuzzy subsets remain.

In one embodiment, the particle swarm optimization algorithm uses an fitness function to minimize the operation cost of multi-residential energy management.

In one embodiment, the fitness function is: Among them, Cost (t) is the total operating cost of a system, P _grid-buy is the electricity purchased by a grid, C _buy is the tax rate for purchase, P _grid-sell is the electricity fed into the grid, and C _sell is the feed Tax rate, C _battery is the cost of a battery module, and SOH _remain represents the remaining SOH (state of health) of the battery module.

In order to enable your reviewers to further understand the structure, characteristics, and purpose of the present invention, drawings and detailed descriptions of the preferred embodiments are attached below.

Step a‧‧‧Enter a current battery level and compare the current battery level with a previous battery level to obtain a battery level change

Step b‧‧‧ performs a first fuzzification operation on the current battery power according to a current battery power input attribution function to obtain a first fuzzification result, and performs a battery power change on the battery power variation based on a battery power variation input A second fuzzification operation to obtain a second fuzzification result, wherein the current battery power input attribution function and the battery power variation input attribution function are determined in advance according to a particle group algorithm

Step c‧‧‧ maps a rule base according to the first fuzzy result and the second fuzzy result to generate a third fuzzy result

Step d‧‧‧ performs a deblurring operation on the third fuzzy result according to an output attribution function to obtain a charging rate

Step e‧‧‧ multiplies the charging rate by a battery rated power to determine a charging power command

FIG. 1 is a flowchart of steps in an embodiment of a particle swarm optimization fuzzy logic control charging method according to the present invention.

Fig. 2 is a schematic diagram showing the application of a residential photovoltaic battery energy storage system.

Figure 3a shows the average daily excess power of the user.

Figure 3b shows the maximum excess power increase of the user within one year.

FIG. 4 shows the cycle life versus discharge depth curve of the lithium iron phosphate battery obtained from the Rosenkranz model.

FIG. 5 illustrates the assignment functions of the input and output variables adopted by the present invention.

FIG. 6 illustrates a system architecture of fuzzy logic control adopted by the present invention.

FIG. 7 illustrates a rule base derivation situation of fuzzy logic control.

FIG. 8 is a sectional view of a power simulation result on a certain day.

FIG. 9 shows the universe setting values of the optimized input MF of SOC and ΔSOC.

FIG. 10a shows the simulation results of the percentage of the minimum power management operation cost achieved by the greedy method, the FID method, and the FuzzyN method.

Figure 10b shows the difference between the minimum cost of the total operation of the computing system and the cost of the FuzzyN method.

FIG. 11 illustrates the g _best value recorded in each iteration of the PSO.

Figure 12 shows the annual energy management operating costs of 74 homes in four ways.

Please refer to FIG. 1, which illustrates a flowchart of steps in an embodiment of the particle swarm optimization fuzzy logic control charging method of the present invention.

As shown in FIG. 1, the particle swarm optimization fuzzy logic control charging method of the present invention includes the following steps: inputting a current battery level and comparing the current battery level with a previous battery level to obtain a battery level change amount ( Step a); Perform a first fuzzification operation on the current battery power according to a current battery power input attribution function to obtain a first fuzzification result and perform a The second fuzzification operation obtains a second fuzzification result, wherein the current battery power input attribution function and the battery power variation input attribution function are determined in advance according to a particle group algorithm (step b); according to the first A fuzzing result and the second fuzzing result are mapped to a rule base to generate a third fuzzing result (step c); a deblurring operation is performed on the third fuzzing result according to an output attribution function to obtain a The charging rate (step d); and multiplying the charging rate by a battery rated power to determine a charging power command (step e).

Among them, the current battery power input attribution function corresponds to five semantic parameters, where S stands for small, MS stands for small, M stands for ML, large stands for L, and L stands for large. To five semantic parameters, where NL represents negative large, NS represents negative small, Z represents zero, PS represents positive small, and PL represents positive large; the output attribution function corresponds to five semantic parameters, where S represents small , MS stands for small, M stands for medium, ML stands for large, and L stands for large.

The current battery power input attribution function and the battery power change input attribution function are determined by the particle swarm algorithm in advance, and there are three fuzzy subsets left, and the particle swarm algorithm uses an fitness function to minimize Cost of multiple residential energy management operations.

The following describes the system architecture of the present invention:

Different from the previous research on the design and verification of power dispatching strategies for a single house, this case proposes an adaptive control algorithm to achieve the lowest power generation cost without weather or load forecasting conditions.

Please refer to FIG. 2, which shows a schematic diagram of the application of a residential photovoltaic energy storage system.

As shown in the figure, in addition to the power generated by the photovoltaic power generation system, it can be used to charge the battery energy storage system if there is surplus power. The battery energy storage system, in addition to being charged by the photovoltaic power generation system, can also discharge to the outside world, and can decide to sell or purchase electricity from the public power grid depending on the demand for electricity.

Please refer to FIG. 3a to FIG. 3b together, wherein FIG. 3a shows the average daily excess power of the user; and FIG. 3b shows the maximum excess power increase of the user within one year.

As shown in the figure, the community studied by the present invention has a total of 74 households, and each house has a photovoltaic power generation system with the same roof area. Since each household has different energy usage behaviors, the individual power consumption curves are different from each other.

According to statistics, the average annual power consumption of a family of 4-6 people is between 4.3MWh and 4.75MWh. In order to make a fair comparison of the costs required for various control methods, the present invention sets the annual household energy consumption data measured to 4.5MWh.

Although the power demand of each house is the same, the load power consumption patterns are still different, so the characteristics of different power consumption patterns can be clearly seen, but the present invention does not classify various load power consumption patterns.

Among them, photovoltaic power generation data is calculated by a solar cell installed on the roof with a resolution of 1 second. The same photovoltaic data is applicable to 74 residential houses, but the power loss of analog / digital conversion and the solar cell model are ignored. The degradation rate of the group and the peak power of photovoltaic power generation are set to 4.4kWp according to the annual power generation of 4017kWh.

In terms of battery energy storage systems, because lithium-ion batteries have the characteristics of high efficiency, high scalability, and long cycle life in various electrochemical technologies, the research of the present invention uses the most advanced lithium iron phosphate at present. (LiFePO ₄ ) battery with a capacity of 4.4 kWh.

For batteries in fixed storage systems, the aging effect has a crucial impact on capacity decline, and even dominates the battery's economic performance. Battery aging will increase battery replacement costs. The aging effect of a battery is a complex electrochemical reaction process, involving many variables, and there are situations where capacity degradation occurs such as timed aging and cyclic aging. In the present invention, the individual effects are added up and the total aging effect is considered.

In order to calculate the capacity degradation of cyclic aging, Rosenkranz and other scholars proposed a model based on the Wöhler curve to evaluate the cyclic aging of the battery under the assumption of equivalent full cycle. At a given depth of discharge (DoD), the remaining cycle life is shown in equation (1).

Among them, cycle lifetime refers to the situation in which the battery capacity gradually decreases under repeated charging and discharging. Generally, the rated capacity of the secondary battery is used as the standard. DoD is the discharge depth refers to the rated capacity of the secondary battery. By comparison, the ratio of discharged electricity.

Please refer to FIG. 4, which illustrates the cycle life versus discharge depth curve of a lithium iron phosphate battery obtained from the Rosenkranz model.

As shown in the figure, the cycle life versus discharge depth curve is called the cyclic aging situation in Rosenkranz model. Assuming that the lithium iron phosphate (LiFePO ₄ ) battery used in the present invention has a life span of 20 years, the periodic aging can be calculated from each sampling time. The standard for replacing the battery means that when the remaining capacity of the battery is reduced to 80% of the rated capacity , Which is 80% battery health (SOH).

The following describes the calculation of funds for various operating situations. Previous literature studies have analyzed the feasibility and necessity of home battery energy storage systems. Although energy incentive policies and regulations vary from country to country, rising residential electricity prices and falling battery costs have been disclosed in the literature. The operating situation of the present invention assumes that the investment of installing a battery energy storage system is more favorable than the system without installing, and the more favorable operating situation is highly dependent on the system's composition, specifications, and related regulations.

Among them, the electricity retail tax is estimated from the historical price of Germany from 2004 to 2014, and the cost of the battery is estimated from the future value estimate. Based on a 20-year depreciation period, according to the analysis of previous literature, when the retail electricity price is higher than 33 euros / kWh and the battery cost is less than 430 Euros / kWh, the device battery energy storage system will be profitable, including the annual 4% interest and 2% inflation rate, and set the feed-in price of the rooftop photovoltaic power generation system to 12.31 euros / kW in the next 20 years, and the maximum feed-in limit for a fully subsidized domestic photovoltaic energy storage system is 50% of the peak photovoltaic power . For fair comparison, all control strategies are calculated and compared for one year under the same funding calculation conditions.

In a single-family household photovoltaic battery energy storage system, battery power and grid power can be considered as controllable variables to meet power balance and load requirements, as shown in equation (2).

P _load + P _PV + P _grid + P _battery = 0 (2)

Among them, P _load is the load demand, P _PV is the photovoltaic power generation, P _grid is the power system power, and P _battery is the battery charge and discharge power.

However, rising electricity prices and low feed-in electricity prices have made consumers reluctant to use electricity from the grid. Therefore, increasing self-consumption is a more favorable method. Therefore, the more photovoltaic power generation can meet the household electricity demand, the less electricity needs to be purchased from the grid. In order to achieve this purpose, it is important to make full use of battery power. When the photovoltaic power generation is insufficient, the battery can store the remaining power and output the stored power, and the net power is shown in equation (3).

P _net = P _load - P _PV (3)

The excess power is shown in equation (4), where P _net is the net power after the amount of photovoltaic power generation deducted from the battery energy storage system, and P _surplus is the excess power.

P _surplus = -P _net (4)

However, due to the fact that the battery capacity decreases with time and number of cycles, and the power reduction caused by the feed-in restriction, the charge / discharge control of the battery energy storage system becomes difficult. The lowest operating cost requires a comprehensive control strategy.

Considering the non-linear and dynamic characteristics of residential photovoltaic cell energy storage systems, the conventional control method for modeling actual physical systems is no longer advantageous, and there are many variables and uncertainties in photovoltaic cell energy storage systems. As a result, the known system is neither suitable for practical experiments nor for mathematical modeling.

In order to solve the power dispatching problem of residential photovoltaic battery energy storage systems, the present invention adopts a fuzzy logic control (FLC) design. Because FLC has the advantage of not needing weather forecasting, it can also properly charge electrical energy into the battery. The battery charge (SOC) and the remaining power are used as input variables for the FLC charging method to achieve lower operating costs.

The present invention uses the current SOC and △ SOC as the input variables of the FLC:

The diversification of the types of electricity used by residents in the community makes the traditional FLC charging method no longer advantageous in a single household, and the set value of the membership function (MF) of the remaining power is fixed, which cannot meet the various input surpluses. The state of power, so the FLC charging method with the remaining power as the input variable may not be the best. In order to effectively perform the FLC operation, the present invention uses the current SOC and ΔSOC as the input variables, where ΔSOC is defined as equation (5 ).

△ SOC = SOC ( t ) -SOC ( t -1) (5)

The current SOC can provide necessary information for the battery energy storage system, and the present invention uses ΔSOC as one of the input variables, which can prevent battery damage caused by improper charging and discharging.

Please refer to FIG. 5, which illustrates the assignment functions of the input and output variables adopted by the present invention. As shown in the figure, the output variable of the present invention is a charging ratio (CR), which is obtained by defuzzifying the range from 0 to 1, and then multiplying CR by the rated power of the battery as the charging power command.

Among them, each variable contains 5 fuzzy subsets corresponding to the degree of semantics. The semantic parameter S represents small, MS represents small, M represents medium, ML represents large, L represents large, and NL represents negative large. , NS represents negative small, Z represents zero, PS represents positive small, and PL represents positive large.

Please refer to FIG. 6, which illustrates a system architecture of fuzzy logic control adopted by the present invention.

As shown in the figure, the system architecture consists of fuzzy input / output attribution function, fuzzy inference engine, rule base, and defuzzification.

Since the input attribution functions SOC and △ SOC each correspond to 5 semantic parameters, such as SOC corresponds to S, MS, M, ML, L; △ SOC corresponds to NL, NS, Z, PS, PL, so the fuzzy logic controller will have 25 rules. Based on the empirical results of the simulation, the rule base shown in Table 1 can be derived, where a negative ΔSOC indicates discharge; a positive ΔSOC indicates charging.

The operating principle is that the battery is charged as much as possible during the day, while maintaining the battery capacity appropriately for the power reduction demand that is likely to occur at noon, so it is expected to benefit from the good use of renewable energy and the reduction of power loss from energy storage systems.

Please refer to FIG. 7, which illustrates a rule base derivation situation of fuzzy logic control.

As shown in the figure, its inference rule is: if the battery SOC is low (SOC is S) and the SOC is rapidly decreasing (△ SOC is NL), the CR should be large (CR is L), as shown in Rule1 (rule in the figure) 1); However, if the SOC increases very quickly (ΔSOC is PL), the CR should decrease slightly (CR is MS), ie Rule5 (rule 5) in the figure.

On the other hand, if the battery SOC is high (SOC is L), and the SOC is decreasing rapidly (ΔSOC is NL), CR should be slightly reduced (CR is MS), as shown in Rule21 (rule 21) in the figure; If the SOC increases very quickly (ΔSOC is PL), CR should be small (CR is S), which is Rule25 (rule 25) in the figure.

In short, the charging rate decreases as the SOC increases. Compared to positive ΔSOC, charging with negative ΔSOC is always preferred. At negative ΔSOC, the battery can store solar energy as much as possible. However, when ΔSOC is positive, the charging will be more conservative to prevent the battery from being fully charged too early and reducing power loss.

Please refer to FIG. 8, which is a cross-sectional view of a power simulation result on a certain day.

As shown in the figure, in the period A, when there is no solar power generation at night, the battery will discharge and meet the load demand; in the period B at noon, when the remaining power is greater than the feed-in limitation, it will The curtailment losses are charged into the battery instead of being controlled by the FLC. In addition to these two periods, the FLC is responsible for determining the charging power. Therefore, the power charged to the battery changes dynamically. The battery will be fully charged before night. When the solar power is insufficient, the battery energy storage system will always discharge. When the body limit is exceeded, all the power that cuts the power will be used. Charging, in other words, the FLC will only operate if there is remaining power under the feed limit.

The invention uses a particle swarm optimization algorithm to optimize the set values of the SOC and △ SOC input attribution functions:

According to the attribution function (MF) of the input variables shown in Figure 5, it can be seen that the values of the fuzzy subset are evenly distributed in the corresponding range. Because the setting of the MF's domain value governs which rules should be triggered, the uniformly distributed fuzzy The input variable MF of the subset cannot meet the diverse power usage behavior of each user and achieve the lowest operating cost. At the same time, the aging effect of the battery and the influence of different residual power distributions must be considered. Therefore, the design of the set value of the MF for optimizing the input variable has become a key issue.

Because the particle swarm optimization algorithm (PSO) can simply implement exploration in a high-dimensional space to obtain the best solution, that is, PSO is good at solving high-complexity problems that contain many variables. The present invention proposes a PSO to optimize the set value of the input domain MF to further improve the problem that the input variable MF of the uniformly distributed fuzzy subset of the conventional technology cannot reach the minimum running cost in the community.

PSO was proposed by two scholars, Kennedy and Eberhart in 1995. Kennedy and Eberhart were inspired by observing the fish and bird foraging process. When a fish or a bird finds the location of food, it will share the information with other peers. , And finally everyone will concentrate in the direction of food. If each particle is regarded as an individual in a flock of birds or fish, all particles will be randomly scattered in the solution space at first, and the optimal solution position of the whole world will be determined by comparing the fitness of each particle. These particles basically move according to the following two criteria:

(1) Follow the best performing particles.

(2) Each particle will move towards its best position.

In this way, each particle will eventually approach the optimal solution or approach the optimal solution.

Among them, the speed calculation method is shown in equation (6).

v _i ( k +1) = wv _i ( k ) + c ₁ r ₁ ( p _{best , i} - x _i ( k )) + c ₂ r ₂ ( g _best - x _i ( k )) (6)

The position update calculation method is shown in equation (7).

x _i ( k +1) = x _i ( k ) + v _i ( k +1) (7)

Where x _i and v _i represent the position and velocity of the i- th particle, k is the number of iterations, w is the inertia, r ₁ and r ₂ are random values between [0,1], and c ₁ and c ₂ represents the learning coefficient, usually between 0 ~ 2, the variable p _{best, i} stores the best position that the i- th particle walks through, and the variable g _best stores the best position among all particles.

The present invention uses PSO to minimize the operation cost of multi-residential energy management. The implementation procedure is as follows:

Step 1: Select parameters

Due to operational limitations, the maximum and minimum values of the input MF are fixed. The input range of the SOC is 0% to 100%. According to the rated power of the converter connected to the battery system, the input range of △ SOC is -10. % ~ 10%.

Please refer to FIG. 9, which illustrates the set values of the optimal input MF of the SOC and ΔSOC.

Because changing the universe value of the semantic parameter can change the shape of the MF and the slope value that is triggered. As shown in the figure, there are 3 fuzzy subsets left in each input MF, SOC is OP1, OP2, OP3; △ SOC is OP4, OP5, OP6. That is, each particle in the PSO is considered as a possible solution in the six-dimensional search space. The position of individual particles in the solution space can be expressed as x _ij and its velocity as v _ij , where i is the number of particles and j is the number of elements in the particles. Each particle represents a solution and corresponds to an adaptive value.

Step 2: Initialize the particle swarm algorithm

In the initialization phase of the particle swarm algorithm, the particle swarm is allocated to a fixed position or placed in a searched solution space by random numbers. In order to deal with the solution space with unknown characteristics fairly, this case uses the most commonly used uniform random number allocation method for initialization.

Step 3: Calculate the fitness value

The purpose of the present invention is to minimize the operation cost of multi-dwelling power management, in which the total operation cost of a single dwelling is composed of battery replacement costs and electricity costs (including the purchase price and resale revenue). It is worth noting that when the SOH is 80%, the battery must be replaced. Because the one-year simulation range is too short to observe the replacement, the cost of the replacement is based on the aging effect to dissipate the SOH (rated capacity of 20%), and the operation cost operation, that is, the fitness function, is shown in equation (8).

Cost (t) is the total operating cost of the system, P _grid-buy is the electricity purchased by the grid, C _buy is the tax rate for electricity purchase, P _grid-sell is the electricity fed into the grid, and C _sell is the tax rate fed C _battery is the cost of the battery, and SOH _remain represents the remaining SOH.

Step 4: Update regional and global best fit values

In each iteration calculation, calculating the fitness of each particle, each particle of the current adaptation value of p _best individual history optimum value comparison, with a larger value of the adaptation and updating the current p _best current for each individual particle The fitness value p _{best is} compared with the global best value g _best and the current g _best is updated with a larger fitness value.

Step 5: Update the speed and position of each particle

After the PSO has calculated all the particles, each particle needs to update the next speed and position.

Step 6: Termination conditions

If the calculation is terminated when the termination condition is met, the current global best value g _best is the best adaptive value; if the termination condition is not met, then return to step 4.

Simulation results:

The following will be directed to three methods in the previous literature, the Greedy method, the Feed-in damping (FID) method, the Normal fuzzy (FuzzyN) method, and the particle swarm proposed by the present invention. The optimized fuzzy logic control charging method (Optimized Fuzzy (FuzzyOP)) was compared to verify the feasibility and performance improvement of the proposed method.

Please refer to FIGS. 10a to 10b together, wherein FIG. 10a shows the simulation results of the percentage of the minimum power management operation cost achieved by the greedy method, the FID method and the FuzzyN method; and FIG. 10b shows the minimum cost of the total operation of the computing system and the fuzzyN method. Cost difference.

As shown in the figure, before optimizing the input attribution function (MF), a one-year sample of the greedy method, the FID method, and the FuzzyN method was sampled every 10 minutes. Compared with the greedy method and the FID method, FuzzyN method The method achieves the lowest energy management operation costs in 59% of homes.

Compared with the minimum cost of the total system operation, there is still room for improvement in the additional cost of the FuzzyN method in some homes. In order to improve the FuzzyN method to reduce the operating cost of most houses, in the present invention, PSO is used to optimize the input MF. Considering the worst 5 results obtained using the FuzzyN method for optimization, the goal is to use the equation ( 8) To minimize the total cost of 5 houses, the PSO related parameters and configuration are shown in Table 2.

Please refer to FIG. 11, which shows the g _best value recorded in each iteration operation of the PSO.

As shown in the figure, g _best starts to decline from the 13th iteration, and finally converges to a total annual expenditure of 5376.66 euros for 5 houses, and finally obtains the best set value of input MF.

Please refer to FIG. 12, which shows the annual energy management operation cost of 74 homes in four methods.

As shown in the figure, the four methods are compared with the ideal prediction method first, and then the actual cost is calculated after the extra cost is calculated. The results obtained using the ideal prediction method provide a reference for other methods to improve.

Table 3 shows the comparison of the proportion of community users who can reach the minimum cost among the four methods. Before the optimization, 59.4% (44/74) of houses in the FuzzyN method reached the lowest cost, and the next best method was the greedy method. 25.7% (19/74) of houses in the community reached the lowest cost. It can be inferred that The excess power of each house makes the uniformly distributed input MF not a proper design, so PSO is used to determine the optimal input variable MF theory and achieve better performance. However, the present invention (FuzzyOP method) uses PSO optimization, so there is no relevant data before optimization. After the optimization, 98.6% (73/74) houses of the present invention achieved the lowest cost operation.

Table 4 lists the average cost of the four methods and the ideal prediction method. The present invention (FuzzyOP method) has the advantage of the lowest operating cost compared to the other three methods.

It can be seen from Fig. 12, Table 3 and Table 4 that the FuzzyOP method proposed by the present invention can achieve the lowest cost among 73 houses, that is, the community performance is 98.6%, and it can also reach the goal of the lowest average annual energy expenditure of a single house.

With the design disclosed above, the present invention has the following advantages:

1. The present invention discloses a particle swarm optimization fuzzy logic control charging method, which uses the current SOC (battery power) and ΔSOC (battery power variation) as FLC input variables to effectively perform FLC operations and prevent improper The battery is damaged by charging and discharging.

2. The present invention discloses a particle swarm optimization fuzzy logic control charging method, which uses a particle swarm optimization algorithm to optimize the set values of the SOC and △ SOC input attribution functions to meet a variety of power usage behaviors. Lowest operating costs.

3. The invention discloses a particle swarm optimization fuzzy logic control charging method, which can enable 98.6% of the houses to achieve the lowest operating cost, and can also minimize the average annual electrical energy expenditure of a single house.

What is disclosed in this case is a preferred embodiment. For example, those who have partial changes or modifications that are derived from the technical ideas of this case and are easily inferred by those skilled in the art, do not depart from the scope of patent rights in this case.

To sum up, regardless of the purpose, method and effect, this case is showing its technical characteristics that are quite different from the conventional ones, and its first invention is practical, and it is also in line with the patent requirements of the invention. Granting patents at an early date will benefit society and feel good.

Claims (5)

    A particle swarm optimization fuzzy logic control charging method includes the following steps: inputting a current battery level and comparing the current battery level with a previous battery level to obtain a battery The amount of power change; a first fuzzification operation is performed on the current battery power according to a current battery power input attribution function to obtain a first fuzzification result and the battery power change is performed according to a battery power change input attribution function. A second fuzzing operation to obtain a second fuzzing result, wherein the current battery power input attribution function and the battery power variation input attribution function are determined in advance according to a particle group algorithm; according to the first fuzzification result And the second fuzzing result is mapped to a rule base to generate a third fuzzing result; a deblurring operation is performed on the third fuzzing result according to an output attribution function to obtain a charging rate; and the charging rate Multiply by a battery rated power to determine a charge power command.         The particle swarm optimization fuzzy logic control charging method described in item 1 of the scope of patent application, wherein the current battery power input attribution function corresponds to five semantic parameters, where S represents small, MS represents small, and M represents medium , ML stands for large, L stands for large; the input function of the battery power variation corresponds to five semantic parameters, of which NL represents negative large, NS represents negative small, Z represents zero, PS represents positive small, PL Represents positive large; the output attribution function corresponds to five semantic parameters, where S represents small, MS represents small, M represents medium, ML represents large, and L represents large.         The particle swarm optimization fuzzy logic control charging method described in item 1 of the scope of the patent application, wherein the current battery power input attribution function and the battery power variation input attribution function are determined in advance by the particle swarm algorithm with three fuzzy ambiguities. set.         The particle swarm optimization fuzzy logic control charging method described in item 1 of the scope of the patent application, wherein the particle swarm optimization algorithm uses an fitness function to minimize the operation cost of multi-residential energy management.     The particle swarm optimization fuzzy logic control charging method described in item 4 of the scope of patent application, wherein the fitness function is: Among them, Cost (t) is the total operating cost of a system, P _grid-buy is the electricity purchased by a grid, C _buy is the tax rate for electricity purchase, P _grid-sell is the electricity fed into the grid, and C _sell is the feed C _battery is the cost of a battery module, and SOH _remain represents the remaining SOH (state of health) of the battery module.