CN107506932A - Power grid risk scenes in parallel computational methods and system - Google Patents

Abstract

The present invention relates to a method and system for parallel computing of power grid risk scenarios, and constructs a parallel computing pool. The parallel computing pool includes multiple computing nodes, and the computing nodes include multiple computing engines; multiple power grid risk scenario data are obtained for each The computing engine allocates corresponding grid risk scenario data; transmits each grid risk scenario data to the corresponding computing engine, and performs parallel computing on each grid risk scenario data through each computing engine; obtains parallel computing results. In this solution, the power grid risk scenario data includes large-scale branch disconnection data. Through the distribution of the scenario data, each grid scenario data is transmitted to the corresponding computing engine for parallel calculation, which can Efficiently complete the calculation tasks of power grid risk scenarios with a large amount of calculation, a large number of iterations, and high real-time requirements.

Description

Parallel computing method and system for power grid risk scenarios

technical field

The invention relates to the technical field of power grids, in particular to a method and system for parallel computing of power grid risk scenarios.

Background technique

With the growth of socio-economic level and the increase of population, the scale of power grid is getting bigger and bigger, and the society has higher and higher requirements for the safety and reliability of power grid. In order to ensure the normal operation of the power grid, researchers have proposed risk assessment indicators and calculation methods for power grid operation. Due to the large number of expected fault scenarios to be considered, and calculations such as power flow analysis and load reduction for each fault scenario, the calculation amount is very large.

In order to improve the calculation efficiency of power grid risk assessment, it is generally completed by purchasing high-performance servers or workstations, but due to the high price and high maintenance costs, it is generally difficult to own; there are also attempts to optimize algorithms, but due to the wide range and depth of technology involved, the optimization algorithm is not suitable for The gains in computational efficiency are negligible, so grid risk assessment is computationally inefficient.

Contents of the invention

Based on this, it is necessary to provide a parallel calculation method and system for power grid risk scenarios to solve the problem of low calculation efficiency in traditional power grid risk assessment.

A method for parallel computing of power grid risk scenarios, comprising the following steps:

Build a parallel computing pool, which includes multiple computing nodes, and the computing nodes include multiple computing engines;

Obtain multiple grid risk scenario data, and assign corresponding grid risk scenario data to each computing engine;

Transmit the risk scenario data of each power grid to the corresponding computing engine, and perform parallel calculations on the risk scenario data of each power grid through each computing engine;

Get parallel computing results.

A power grid risk scenario parallel computing system includes the following modules:

The parallel computing pool building block is used to build a parallel computing pool. The parallel computing pool includes multiple computing nodes, and the computing nodes include multiple computing engines;

The scenario data allocation module is used to obtain multiple grid risk scenario data, assign corresponding grid risk scenario data to each computing engine, and transmit each grid risk scenario data to the corresponding computing engine;

The parallel computing module is used to perform parallel computing on the risk scene data of each power grid through each computing engine;

The result collection module is used to obtain parallel computing results.

According to the grid risk scenario parallel computing method and system of the present invention, multiple computing nodes are used to construct a parallel computing pool, multiple grid scenario data are distributed to each computing engine in each computing node, and the grid risk scenario data is processed by each computing engine Perform parallel calculations and obtain the calculation results of each calculation engine. In this solution, through the allocation of scene data, each power grid scene data is transmitted to the corresponding computing engine for parallel computing, which can efficiently complete the power grid risk scene computing tasks with large amount of calculation, many iterations, and high real-time requirements. In this way, the calculation efficiency of power grid risk scenarios is improved.

A readable storage medium, on which an executable program is stored, and when the program is executed by a processor, the steps of the above-mentioned parallel computing method for power grid risk scenarios are realized.

A computing device includes a memory, a processor, and an executable program stored on the memory and operable on the processor. When the processor executes the program, the steps of the above-mentioned parallel computing method for power grid risk scenarios are realized.

According to the above-mentioned parallel computing method for power grid risk scenarios of the present invention, the present invention also provides a readable storage medium and a computing device for realizing the above-mentioned parallel computing method for power grid risk scenarios through programs.

Description of drawings

Fig. 1 is a schematic flow chart of a method for parallel computing of power grid risk scenarios in an embodiment of the present invention;

Fig. 2 is a schematic flow chart of the steps of constructing a parallel computing pool in another embodiment of the present invention;

Fig. 3 is a schematic flow chart of the step of allocating scene data in another embodiment of the present invention;

4 is a schematic structural diagram of a parallel computing system for power grid risk scenarios in another embodiment of the present invention;

FIG. 5 is a schematic flow diagram of a method for parallel computing of power grid risk scenarios in another embodiment of the present invention;

FIG. 6 is a schematic flow diagram of a method for parallel computing of power grid risk scenarios in another embodiment of the present invention;

FIG. 7 is a parallel computing architecture diagram of a parallel computing method for power grid risk scenarios in another embodiment of the present invention.

detailed description

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, and do not limit the protection scope of the present invention.

Referring to FIG. 1 , it is a schematic flowchart of a method for parallel computing of power grid risk scenarios according to an embodiment of the present invention. The grid risk scenario parallel computing method in this embodiment includes the following steps:

Step S110: building a parallel computing pool, the parallel computing pool includes multiple computing nodes, and the computing nodes include multiple computing engines;

In this step, the computing node may be a computer device, including a multi-core processor and a storage device, and a parallel computing pool may be constructed by connecting processors and storage devices of multiple computing nodes.

Step S120: Obtain multiple power grid risk scenario data, and assign corresponding power grid risk scenario data to each computing engine;

In this step, by allocating grid risk scenario data to each computing engine, the computing pressure of each computing engine can be adjusted.

Step S130: Transmit the risk scenario data of each power grid to the corresponding calculation engine, and perform parallel calculation on the risk scenario data of each power grid through each calculation engine;

In this step, after the allocation of grid risk scenario data is completed, each scenario data is transmitted to each computing engine of each computing node for parallel computing according to the distribution result.

Step S140: Acquiring parallel computing results.

In this step, since each calculation result is still stored in each calculation engine after the parallel calculation is completed, it is necessary to return the calculation result through each calculation engine.

In this embodiment, multiple computing nodes are used to build a parallel computing pool, multiple power grid scenario data are distributed to each computing engine in each computing node, and each computing engine performs parallel computing on the power grid risk scenario data to obtain the data of each computing engine calculation results. In this solution, through the allocation of scene data, each power grid scene data is transmitted to the corresponding computing engine for parallel computing, which can efficiently complete the computing tasks of power grid risk scenarios with a large amount of calculation, a large number of iterations, and high real-time requirements.

Optionally, the power grid risk scenario data may include branch disconnection data of a large-scale power grid. By sorting out the disconnection data of each branch, multiple power grid risk scenario data are formed;

Referring to Figure 2, it is a schematic flow diagram of building a parallel computing pool in one of the embodiments, and the steps of building a parallel computing pool in this embodiment include the following steps:

Step S111: Obtain multiple computing nodes, add each computing node to the cluster, and create task management for the cluster;

Step S112: Configure the number of computing engines of each computing node, add the cluster to the initial cluster configuration file, obtain the target cluster configuration file, and create a parallel computing pool according to the target cluster configuration file.

In this embodiment, obtain computing nodes that will be used to participate in parallel computing, add computing nodes to the same cluster, create task management and configure the number of computing engines, add the cluster to the initial cluster configuration file, and obtain the target cluster Configuration files that can be used to create parallel computing pools. In practical applications, for different computing tasks, corresponding clusters need to be created in a targeted manner. Due to the multi-node characteristics of the cluster system, the cluster configuration is intricate. Especially in large-scale systems, manual configuration of cluster nodes is inefficient. By using cluster configuration files, the efficiency of creating and managing parallel pools can be improved.

Optionally, the above operations can be done on one of the computing nodes in the parallel pool. Through the host IP address or host name, find the computing nodes that can be used to participate in parallel computing, and add the found computing nodes to a cluster. Through the cluster Create tasks for unified management.

In one of the embodiments, the step of configuring the number of computing engines of each computing node includes the following steps:

Configure the number of computing engines in the current computing node as the number of CPU cores of the current computing node.

In this embodiment, since the calculation of power grid risk scenario data is data-intensive calculation, there is basically no IO operation during the calculation process, and IO blocking rarely occurs; on the other hand, when a single CPU core processes multiple processes, it is For queuing processing, since one computing engine corresponds to one process, setting the number of computing engines of each computing node to be consistent with the number of CPU cores of the computing engine can reduce the performance loss caused by system scheduling and make full use of the resources of each computing node. computing resources. In one embodiment, each computing node is in the same local area network, and data transmission is performed between computing nodes through the local area network.

In this embodiment, since the power grid risk scenario data is designed with a very large amount of data in actual operation, and the scenario data needs to be transmitted to each computing engine for parallel computing, there is a high requirement on the data transmission rate. Each computing node in the parallel computing pool uses a specially laid transmission medium for networking, which has a high transmission rate and low delay, which improves the transmission of scene data in the parallel computing pool, thereby improving the efficiency of parallel computing.

In one of the embodiments, after building the parallel computing pool, select any computing node in the parallel computing pool as the client, and transmit the data of each power grid risk scenario to the client; wherein, the client assigns the corresponding power grid to each computing engine risk scenario data;

The step of transmitting each data to the corresponding calculation engine includes the following steps:

Each data is transmitted to the corresponding computing engine through the client.

In this embodiment, the calculation of power grid risk scenario data involves a large number of calculations. By using the client to allocate calculation tasks, the calculation tasks can be better divided and the efficiency of parallel calculation can be improved.

Referring to FIG. 3 , it is a schematic flow diagram of assigning corresponding grid risk scenario data to each computing engine in one embodiment, and the step of assigning corresponding grid risk scenario data to each computing engine in this embodiment includes the following steps:

Step S121: Obtain the total number of computing engines in the parallel computing pool and the engine serial number of each computing engine;

Step S122: number the grid risk scenario data sequentially to obtain the scenario number;

Step S123: Add 1 to the remainder of the current scenario number and the total number of computing engines to obtain the target engine serial number corresponding to the current scenario number, and distribute the power grid risk scenario data to which the current scenario number belongs to the computing engine to which the target engine serial number belongs.

In this embodiment, by adopting the allocation method of scene data, each scene data can be assigned a calculation engine responsible for its calculation, and at the same time, the amount of scene data allocated to each calculation engine is basically the same, so parallel computing The pool achieves the effect of load balancing, which is conducive to improving CPU utilization, minimizing task idle time, and improving parallel computing efficiency.

In one of the embodiments, the platform for parallel computing is MATLAB, and the method of single program and multiple data streams of the platform is used to perform parallel computing on the grid risk scenario data.

In this embodiment, the power grid risk scenario data has a large amount of data, but the calculation method of each scenario data is similar, and only the data is different. SPMD (Single Program Multiple Data) refers to the parallel computing method of single program and multiple data streams. By using MATLAB to run parallel computing programs, the same program can be run on multiple computing engines. Corresponding codes are written for different data in the program. Each calculation engine uses the same program to calculate different power grid risk scenario data. At the same time, the program contains the necessary logic. Each calculation engine can only execute some statements in the program instead of the entire program. The return value of the calculation result is composite type of object storage, thus improving the efficiency of parallel computing.

In one of the embodiments, the step of parallel computing includes the following steps:

Transmit the corresponding power grid risk scenario data to the GPU corresponding to the current computing engine through the current computing engine;

Among them, the GPU judges the type of the received grid risk scenario data, and if the received grid risk scenario data is dense matrix data, use the gpuArray() function to transfer the dense matrix data to the GPU memory for parallel calculation;

If the received power grid risk scenario data is sparse matrix data, use the MEX function and CUDA computing library to perform parallel calculations on the sparse matrix data;

If the received power grid risk scenario data is a plurality of data and is scalar, then generate a custom function through the arrayfun function, and use the custom function to perform parallel calculation on the received power grid risk scenario data.

In this embodiment, after receiving the allocated scene data, the calculation engine transmits the data to a GPU (Graphics Processing Unit, graphics processing unit) for parallel calculation. GPU is the core component of a computer display device. Each GPU contains multiple stream processors. For image processing, these stream processors are designed to work in parallel. CUDA (Compute Unified Device Architecture) is a general-purpose parallel computing architecture of the GPU, which enables the GPU to be used for floating-point operations. Utilizing the floating-point computing performance, high memory bandwidth, and high cost performance of the GPU, which far exceeds that of the CPU, the CUDA architecture is applied to parallel computing tasks in large-scale power grid risk scenarios, which is conducive to improving computing efficiency.

The calculation of the matrix is the basic component of the grid risk scenario calculation. In the matrix, if the number of elements with a value of 0 is far more than the number of non-zero elements, and the distribution of non-zero elements is irregular, the matrix is called a sparse matrix; On the contrary, if the number of non-zero elements accounts for the majority, the matrix is called a dense matrix.

In order to improve the parallel computing efficiency of scene data, the scene data of dense matrix type is copied to the GPU memory through the gpuArray() function, and the matrix and vector with gpuArray attribute are generated, which can be automatically calculated by GPU. For scene data of sparse matrix type, if the same calculation method as the dense matrix type is used above, a large amount of GPU memory space will be consumed, resulting in a limited calculation scale. Therefore, MEX function and CUDA calculation library are used to perform parallel calculations on sparse matrix data. , where the MEX function can call various CUDA computing libraries for parallel computing of mixed GPUs. Specifically, two sparse matrix numerical computing libraries, CUSPARSE and CUSOLVER provided by CUDA, can be used to perform parallel computing on GPU. If the received power grid risk scenario data is a plurality of data and is a scalar quantity, it may be a situation where there is more than one input or output variable.

In one of the embodiments, the step of obtaining the parallel calculation result includes the following steps:

Receive calculation results returned by each computing node; wherein, each computing node collects the calculation results of each computing engine included in each through a gather function.

In this embodiment, since the power grid risk scenario data is distributed to each calculation engine for parallel calculation, it is necessary to obtain the calculation results returned by each calculation engine. First, each computing node recycles the calculation result from the GPU memory to the physical memory through the gather function, and then each computing node returns the calculation result. By calling the gather function, the speed of obtaining the calculation results of each calculation engine is improved.

According to the above-mentioned parallel computing method for power grid risk scenarios, the present invention also provides a parallel computing system for power grid risk scenarios, and an embodiment of the parallel computing system for power grid risk scenarios of the present invention will be described in detail below.

Referring to FIG. 4, it is a schematic structural diagram of a power grid risk scenario parallel computing system according to an embodiment of the present invention. The power grid risk scenario parallel computing system in this embodiment includes:

A parallel computing pool building module 210, configured to build a parallel computing pool, the parallel computing pool includes a plurality of computing nodes, and the computing nodes include a plurality of computing engines;

The scene data distribution module 220 is used to obtain multiple power grid risk scene data, assign corresponding power grid risk scene data to each calculation engine, and transmit each power grid risk scene data to the corresponding calculation engine;

Parallel calculation module 230, used for performing parallel calculation on each power grid risk scenario data through each calculation engine;

The result collection module 240 is used for obtaining parallel computing results.

In one of the embodiments, the parallel computing pool construction module 210 acquires multiple computing nodes, adds each computing node to the cluster, creates task management for the cluster; configures the number of computing engines of each computing node, and adds the cluster to the initial cluster configuration file, obtain the target cluster configuration file, and create a parallel computing pool based on the target cluster configuration file.

In one of the embodiments, the parallel computing pool construction module 210 configures the number of computing engines in the current computing node as the number of CPU cores of the current computing node.

In one of the embodiments, the parallel computing pool construction module 210 configures each computing node in the same local area network, so that data transmission between each computing node is performed through the local area network.

In one of the embodiments, the scenario data distribution module 220 selects any computing node in the parallel computing pool as a client, and transmits the scenario data of each power grid risk to the client; wherein, the client assigns the corresponding power grid risk to each computing engine Scene data, each data is transmitted to the corresponding computing engine through the client.

In one of the embodiments, the scenario data allocation module 220 obtains the total number of computing engines in the parallel computing pool and the engine serial numbers of each computing engine; numbers the grid risk scenario data in order to obtain the scenario number; assigns the current scenario number to the computing engine Add 1 to the remainder of the total to obtain the target engine serial number corresponding to the current scenario number, and distribute the power grid risk scenario data to which the current scenario number belongs to the calculation engine to which the target engine serial number belongs.

In one of the embodiments, the parallel computing pool construction module 210 uses the MATLAB platform to build a parallel computing pool, and the parallel computing module 230 uses the single program multiple data stream method of the MATLAB platform to perform parallel computing on the power grid risk scenario data.

In one of the embodiments, the scenario data distribution module 220 transmits the corresponding power grid risk scenario data to the GPU corresponding to the current computing engine through the current computing engine; wherein, the GPU judges the type of the received power grid risk scenario data, and in the received When the power grid risk scene data is dense matrix data, use the gpuArray() function to transfer the dense matrix data to the GPU memory for parallel calculation;

When the received power grid risk scenario data is sparse matrix data, use the MEX function and CUDA computing library to perform parallel calculations on the sparse matrix data;

When the received power grid risk scenario data is multiple data and scalar, a custom function is generated through the arrayfun function, and the custom function is used to perform parallel calculation on the received power grid risk scenario data.

In one of the embodiments, the result collection module 240 receives the calculation results returned by each computing node; wherein, each computing node collects the calculation results of the computing engines included in each through a gather function.

The grid risk scenario parallel computing system of the present invention corresponds to the grid risk scenario parallel computing method of the present invention, and the technical features and beneficial effects described in the above-mentioned embodiment of the grid risk scenario parallel computing method are applicable to the grid risk scenario parallel computing In the embodiment of the system.

According to the above-mentioned parallel computing method for power grid risk scenarios, an embodiment of the present invention further provides a readable storage medium and a computing device. An executable program is stored on the readable storage medium, and when the program is executed by the processor, the steps of the above-mentioned parallel computing method for power grid risk scenarios are implemented; the computing device includes a memory, a processor, and an executable program stored on the memory and operable on the processor. Executing the program, when the processor executes the program, the steps of implementing the above-mentioned parallel computing method for power grid risk scenarios.

In a specific embodiment, the grid risk scenario parallel calculation method includes the following steps:

Utilize the mathematical software MATLAB to build a parallel computing pool, the parallel computing pool includes a plurality of computing nodes, and the computing node includes a plurality of computing engines; wherein, the steps of constructing the parallel pool include the following steps:

Obtain multiple computing nodes, add MATLAB to the system firewall of the computing nodes, install MATLAB distributed parallel computing server for each node, and start the service, each computing node is in the same local area network, and data transmission is performed between computing nodes through the local area network;

Use an interface Admin Center provided by MATLAB to add each computing node to the cluster, and the MATLAB Job Scheduler interface in the interface manages the cluster creation task;

Configure the number of computing engines in the current computing node as the number of CPU cores in the current computing node, add the cluster to the initial cluster configuration file, obtain the target cluster configuration file, and create Monitor Jobs for the configuration to monitor the cluster task status; according to the target cluster configuration file to create parallel computing pools.

Obtain multiple grid risk scenario data, and assign corresponding grid risk scenario data to each computing engine;

Among them, after the parallel computing pool is built, any computing node in the parallel computing pool is selected as the client, and the data of each power grid risk scenario is transmitted to the client; each data is transmitted to the corresponding computing engine through the client.

Wherein, the step of assigning corresponding power grid risk scenario data to each calculation engine includes the following steps:

Obtain the total number of computing engines in the parallel computing pool, and use the labindex() function to return the engine serial number of the computing engine;

Number the grid risk scenario data sequentially to obtain the scenario number;

Add 1 to the remainder of the current scenario number and the total number of computing engines to obtain the target engine serial number corresponding to the current scenario number, and assign the grid risk scenario data to which the current scenario number belongs to the computing engine to which the target engine serial number belongs.

Use the parallel computing toolbox provided by MATLAB to perform parallel computing, including:

Transmit the risk scenario data of each power grid to the corresponding computing engine, and perform parallel computing on the risk scenario data of each power grid through each computing engine, and the parallel computing is realized by using the single program multiple data stream method provided by the MATLAB platform; the parallel computing steps include the following steps :

Transmit the corresponding power grid risk scenario data to the GPU corresponding to the current computing engine through the current computing engine; if the received power grid risk scenario data is dense matrix data, use the gpuArray() function to transfer the dense matrix data to the GPU memory for parallelism Calculation; if the received power grid risk scenario data is sparse matrix data, use the MEX function and CUDA computing library to perform parallel calculations on the sparse matrix data; if the received power grid risk scenario data is multiple data and is a scalar value, generate it through the arrayfun function User-defined functions, using user-defined functions to perform parallel calculations on the received power grid risk scenario data.

Obtain the results of parallel computing; among them, each computing node gathers the computing results of each computing engine through the gather function, and returns the result to a variable of composite object type, and then receives the computing results returned by each computing node on the client side, and each computing node's The calculation result is obtained through the engine serial number index of the calculation engine.

Referring to FIG. 5 , it is a schematic flowchart of a method for parallel computing of power grid risk scenarios according to another embodiment of the present invention. The grid risk scenario parallel computing method in this embodiment includes the following steps:

Step S310: building a parallel computing pool, including the following steps:

Install the same version of MATLAB software to each computing node participating in parallel computing, and add MATLAB to the firewall; installing the same version of MATLAB software can avoid compatibility problems caused by different versions; Computing nodes perform data interaction through the LAN. In order to avoid certain operations being affected by the firewall during the interaction process, it is necessary to set up the firewall in advance.

Install the MATLAB distributed parallel computing server for each computing node as an administrator, and start the service;

Configure each computing node in the same local area network, and then use Admin Center to add each computing node to a cluster; create task management in MATLAB Job Scheduler; configure the number of computing engines for each computing node, preferably, the calculation of each computing node The number of engines is the same as the number of CPU cores of the computing node;

Add the cluster to the cluster configuration file, and create Monitor Jobs for the configuration file to monitor the task status of the cluster.

Use a cluster configuration file to create a parallel computing pool.

Step S320: Acquiring multiple grid risk scenario data, and assigning corresponding grid risk scenario data to each calculation engine, including the following steps:

Obtain the grid risk scenario data that needs to be calculated, and number the scenarios in sequence;

Calculate the remainder of the number of computing engines for the number of the scene, and obtain the remainder + 1. The result obtained is the engine serial number of the computing engine responsible for executing the scene;

Step S330: Transmitting the risk scenario data of each power grid to the corresponding computing engine, and performing parallel computing on the risk scenario data of each power grid through each computing engine; including the following steps:

Use MATLAB's parallel computing toolbox for parallel computing, and use spmd-end to start the parallel computing program; among them, spmd-end is used to mark statements that need to use SPMD single program multiple data streams for parallel computing. Among them, MATLAB's parallel computing toolbox can use multi-core processors, GPUs, and computer clusters to solve computing problems and data-intensive problems. By using high-level structures such as parallel for loops, special array types, and parallel numerical algorithms, MATLAB applications can be parallelized without CUDA or MPI programming.

Use the labindex() function to return the engine serial number of the current computing engine;

According to the calculation result of the number of the scene above, each scene is transmitted to the calculation engine corresponding to the engine serial number for parallel calculation.

Among them, one computing engine corresponds to one GPU, and each computing engine uses the gpuDevice() function to select the corresponding GPU for computing, including:

Dense matrix calculation: MATLAB provides built-in functions that can be directly parallelized by GPU to improve the calculation efficiency of dense matrices. Copy the scene data to the GPU memory through the gpuArray() function to automatically perform parallel computing on the GPU;

Sparse matrix calculation: Since MATLAB's built-in functions do not support direct GPU parallel calculation of sparse matrices, two sparse matrix numerical calculation libraries, CUSPARSE and CUSOLVER, provided by CUDA for free, are used in combination with the MEX function provided by MATLAB that includes a GPU interface for hybrid GPU parallel computing;

Custom function: Through the arrayfun function provided by MATLAB that supports custom functions, parallel calculations can be performed on more than one input and output data and the input data is scalar scene data;

In this step, the GPU acts as a coprocessor to assist the CPU to complete operations with high parallelism, data-intensive, and simple logic. The ability of GPU parallel computing is even more powerful. It has a fast storage system inside. In addition, the hardware design of GPU can manage thousands of parallel threads. These thousands of threads are all created and managed by GPU without any programming by developers. and management.

Step S340: Acquiring parallel computing results, including the following steps:

Each calculation engine collects the calculation results and returns them to the CPU through the gather function, and resets the memory of the GPU at the same time;

Since the return value of each calculation engine is stored in the type of composite when parallel computing is performed through the SPMD method, the result needs to be returned to an object of the composite type. Optionally, composite objects can be pre-created and initialized before parallel computing.

Obtain the calculation results of each computing node through the index value of the engine serial number.

In this embodiment, the overall basic process is shown in Figure 6. The parallel computing environment shown in Figure 7 is constructed through the distributed parallel computing server of MATLAB, and the CUDA computing library is called by the MATLAB parallel computing toolbox to efficiently complete the calculation amount. For risk scenario calculation tasks with large size, many iterations and high real-time requirements, the risk scenario calculation tasks can be calculated in parallel on multiple computers, which has good fault tolerance. GPU is suitable for complex calculations, which helps to improve the efficiency of power grid risk scenario calculations and realize the parallelization of large-scale calculations.

The technical features of the above-mentioned embodiments can be combined arbitrarily. To make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, should be considered as within the scope of this specification.

The above-mentioned embodiments only express several implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but should not be construed as limiting the patent scope of the invention. It should be pointed out that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention should be based on the appended claims.

Claims (10)

1. A grid risk scenario parallel computing method, characterized in that, comprising the following steps: Build a parallel computing pool, the parallel computing pool includes a plurality of computing nodes, and the computing nodes include a plurality of computing engines; Acquiring a plurality of grid risk scenario data, and assigning corresponding grid risk scenario data to each of the calculation engines; Transmitting each of the power grid risk scenario data to a corresponding computing engine, and performing parallel computing on each of the power grid risk scenario data through each of the computing engines; Get parallel computing results. 2. The grid risk scenario parallel computing method according to claim 1, wherein the step of building a parallel computing pool comprises the following steps: Obtaining multiple computing nodes, adding each of the computing nodes to the cluster, and creating task management for the cluster; Configuring the number of computing engines of each computing node, adding the cluster to the initial cluster configuration file, obtaining a target cluster configuration file, and creating a parallel computing pool according to the target cluster configuration file. 3. The grid risk scenario parallel computing method according to claim 2, wherein the step of configuring the number of computing engines of each computing node comprises the following steps: Configure the number of computing engines in the current computing node as the number of CPU cores of the current computing node. 4. The method for parallel computing of power grid risk scenarios according to claim 2, wherein each of the computing nodes is in the same local area network, and data transmission is performed between computing nodes through the local area network. 5. The grid risk scenario parallel calculation method according to claim 2, further comprising the following steps: selecting any computing node in the parallel computing pool as a client, and transmitting the data of each power grid risk scenario to the client; Wherein, the client assigns corresponding power grid risk scenario data to each of the computing engines; The step of transmitting each of the data to the corresponding computing engine includes the following steps: Each of the data is transmitted to the corresponding computing engine through the client. 6. The grid risk scenario parallel computing method according to claim 1, wherein the step of distributing corresponding grid risk scenario data for each computing engine comprises the following steps: Obtain the total number of computing engines in the parallel computing pool and the engine serial numbers of each computing engine; Sequentially numbering the grid risk scenario data to obtain the scenario number; Adding 1 to the remainder of the total number of computing engines from the current scenario number to obtain the target engine serial number corresponding to the current scenario number, and assigning the power grid risk scenario data to which the current scenario number belongs to the computing engine to which the target engine serial number belongs. 7. The method for parallel computing of power grid risk scenarios according to claim 1, wherein the platform of the parallel computing is MATLAB, and the method of single program and multiple data streams of the platform is used to parallelize the power grid risk scenario data calculate. 8. The grid risk scenario parallel computing method according to claim 7, wherein the step of parallel computing comprises the following steps: Transmit the corresponding power grid risk scenario data to the GPU corresponding to the current computing engine through the current computing engine; Wherein, the GPU judges the type of the received power grid risk scene data, and if the received power grid risk scene data is dense matrix data, then use the gpuArray() function to transfer the dense matrix data to the GPU video memory for parallel calculation; If the received power grid risk scenario data is sparse matrix data, then use the MEX function and CUDA calculation library to perform parallel calculations on the sparse matrix data; If the received power grid risk scenario data is a plurality of data and is a scalar value, generate a custom function through the arrayfun function, and use the custom function to perform parallel calculation on the received power grid risk scenario data. 9. The method for parallel computing of power grid risk scenarios according to claim 8, wherein the step of obtaining parallel computing results comprises the following steps: Receive calculation results returned by each computing node; wherein, each computing node collects the calculation results of each computing engine included in each through a gather function. 10. A grid risk scenario parallel computing system, characterized in that it comprises the following modules: A parallel computing pool construction module, used to build a parallel computing pool, the parallel computing pool includes a plurality of computing nodes, and the computing nodes include a plurality of computing engines; The scene data distribution module is used to obtain multiple power grid risk scene data, allocate corresponding power grid risk scene data to each of the calculation engines, and transmit each of the power grid risk scene data to the corresponding calculation engine; A parallel computing module, configured to perform parallel computing on each of the power grid risk scenario data through each of the computing engines; The result collection module is used to obtain parallel computing results.