JPH08211161A - Prediction system of weather disaster - Google Patents

Abstract

PURPOSE: To realize a prediction method in which both a miss and an overlook can be reduced by a method wherein the existence of a weather disaster in every region is predicted by a neural network on the basis of weather data and a physical quantity as a cause of the weather disaster is computed for every region. CONSTITUTION: A prediction-range comparison device 4 uses a prediction by a snow-damage prediction neural network device 1 and a prediction by a threshold judgment device 3, and it controls the operation of a warning output device 5 on the basis of the overlap of a mesh predicted as snow damage with a mesh in which a snow accumulated amount exceeds a threshold. The device 5 outputs snow damage information and a snow damage warning on the basis of a judgment by the device 1 and of a judgment by the device 3. Then, in a warning system, a snow damage prediction is performed for every region. As a result, e.g. a mesh which is partitioned by a mesh of a net at intervals of 6km is used as an object. By using weather data in every mesh for 6km around, the snow damage prediction of every mesh is performed, and a snow damage warning or the like is issued to individual meshes. As the weather data, a rainfall amount, a temperature and a wind velocity are used.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a meteorological disaster forecasting system for issuing a meteorological disaster warning by adding forecasts of meteorological disasters by neural networks to forecasts of meteorological disasters by calculation results of physical quantities that cause meteorological disasters. In particular, the present invention relates to a meteorological disaster prediction method that can reduce missed shots and misses.

[0002]

2. Description of the Related Art A natural disaster (hereinafter referred to as a meteorological disaster) in which a land or a structure is destroyed or invaded by rain, wind, snow, or the like causes a great loss in a damaged area. It is well known that meteorological disasters are strongly related to meteorological data obtained by meteorological observation. It is also well known that a structure receives a deforming force due to a physical quantity brought to the ground by the weather. The present invention is intended to predict a weather disaster from such weather data and physical quantities. The conventional technique will be described below.

In today's highly electrified world, the supply of electricity should never be interrupted. If the supply of electricity is cut off, it will cause a great deal of damage over a wide range from ordinary households to industries. Therefore, in the event of an accident (disaster) in the power transmission line, a maintenance management system that immediately detects the damaged point and promptly recovers from the disaster has been developed and put into practical use.

Further, in recent years, a predictive maintenance system has been developed to prevent accidents from occurring by predicting the accidents by using weather data etc. rather than knowing the accidents as a result. There is.

[0005] Among these systems, there is a snow damage prediction system that predicts the occurrence of a snow damage accident due to snow accretion on a power transmission line which is widely damaged in winter. This snow damage prediction system uses neural networks to determine whether snow damage will occur 3 hours ahead, based on weather data supplied online by meteorological agencies and private weather service businesses (example Iida et al., Development of transmission line snow damage warning system using neural network, 1994, The Institute of Electrical Engineers of Japan, 1
164).

As shown in FIG. 2, a conventional transmission line snow damage warning system includes a snow damage prediction neural network device 1, a snow accretion amount estimation device 2, a threshold value judgment device 3, and an alarm output device 5.
Consists of The snow damage prediction neural network device 1 is configured to learn weather data patterns from past accidents and to determine the presence or absence of snow damage by inputting the weather data patterns. The snow accretion amount estimation device 2 calculates an estimated snow accretion amount on a power transmission line based on weather data input. The threshold determination device 3 determines the threshold value of the snow accretion amount with a high probability of accident occurrence by the examination using the weather data at the time of the past accident, and gives an alarm when the estimated snow accretion amount exceeds the snow accretion amount threshold value. It is judged as a level, and a threshold value of snow accretion that is lower than this probability but is highly likely to cause an accident is set, and when the estimated snow accretion amount reaches this snow accretion amount threshold, a warning level Is determined. The alarm output device 5 outputs a snow damage warning and a snow damage warning based on the judgment of the snow damage prediction neural net device 1 and the judgment of the threshold value judgment device 3.

In the transmission line snow damage warning system of FIG. 2, in order to predict the snow damage for each area, an area (mesh) divided by meshes of 6 km intervals is targeted. Snow damage prediction for each mesh is performed using the meteorological data for each 6 km square, and a snow damage warning or the like is issued for each mesh. Precipitation, temperature, and wind speed are used as meteorological data.

The meteorological data input to the snow damage prediction neural network device 1 is numerical data for a total of 6 hours including past 3 hours data including the current time and forecast data 3 hours ahead. By inputting this numerical data to the snow damage prediction neural network device 1, it is possible to predict the presence or absence of snow damage 3 hours ahead.

With the snow damage prediction neural network device 1 alone,
It is desirable to be able to predict the probability of an accident and the extent of the accident, but at present it is difficult to obtain good prediction results due to lack of accident data. Therefore, in order to classify the degree of possibility of an accident, the snow accretion amount estimation device 2 which calculates the estimated snow accretion amount by the snow accretion amount estimation formula derived based on the experiment is also used. It was decided to classify it into a warning and a warning. For this reason, the threshold determination device 3 outputs an alarm level when the estimated snow accretion amount is 10 g / cm or more, and outputs a warning level when the estimated snow accretion amount is 10 g / cm or more, based on an examination using the weather data in the past accident. Output. Then, the alarm output device 5 combines the prediction result of the snow damage prediction neural net device 1 and the judgment result of the threshold value judgment device 3 to issue a snow damage warning when the estimated snow accretion amount is 10 g / cm or more and an accident is predicted. , Estimated snow accretion amount is 5g
/ Cm or more and if there is an accident, a snow damage warning will be issued.

[0010]

When the conventional transmission line snow damage warning system of FIG. 2 is actually operated, an accident occurs in an area where the warning is not issued (this is called overlooking), or conversely, the warning is given. There were frequent occurrences of accidents that did not occur in the area where they came out (this is called a miss).

In order to investigate the cause, what is the extent to which the mesh where the accident occurs is predicted for each of the two basic prediction methods, the neural network and the snow accretion estimation? I checked if there was something.
As a result, although the neural network predicted the accident, there were many cases where the snowfall was missed because the snow accretion amount was below the threshold value. Moreover, when the prediction by the neural network and the prediction by the snow accretion amount estimation are in good agreement, there was a tendency for missed shots to occur frequently but missed frequently.

Both the neural network and the estimation of the amount of snowfall are learned or threshold values are set by using the weather data at the same accident,
With regard to past accidents, it is possible to judge the existence of accidents to almost the same extent. Therefore, if snow damage prediction is performed from the same meteorological data, the predictions made by the respective prediction methods should essentially match. Actually, when the weather actual condition data consisting of the past actual condition values was input to the transmission line snow damage warning system of FIG. 2, there were few overlooked cases when there were many overlooked cases when the weather forecast data was input, and there were many missed events. As for the example, there are fewer misses. That is,
It was found that when the meteorological data is input, the prediction accuracy is high, and when the weather forecast data is input, the prediction accuracy is lower than that.

The cause of this is that there is a discrepancy between the value of the weather forecast data and the actual value. In particular, regarding the temperature, the sustainability forecast value is used in the weather forecast data.
The sustainability forecast value is a substitute of the forecast value after 3 hours by the current live performance value, and in the field of meteorology, it is generally considered that the current live performance value lasts for a short time of about 3 hours, so there is no problem. It is said that. However, the temperature is an important parameter in estimating the snow accretion amount, and the estimated snow accretion amount greatly differs when the temperature is different. Therefore, it was found that the use of the sustainability prediction value for the temperature of the weather forecast data has a great influence on the estimation of the amount of snow accretion, which affects the prediction accuracy.

As described above, the snow damage predictions by the neural network and the snow accretion amount estimation are erroneous due to the weather forecast error, but it is unavoidable that there is an error in the weather forecast. We aimed to make the snow damage warning system as a whole highly predictive.

Therefore, an object of the present invention is to solve the above-mentioned problems and to provide a meteorological disaster prediction method capable of reducing both missed shots and missed shots.

[0016]

In order to achieve the above object, the present invention predicts the presence / absence of a weather disaster for each region from meteorological data by a neural network, and from the meteorological data, determines the physical quantity that causes the weather disaster. It is calculated for each area to determine whether or not this physical quantity exceeds a threshold value that causes a meteorological disaster, and the area that issues a meteorological hazard warning is determined based on how the area predicted to have a meteorological disaster and the area that exceeds the threshold value overlap. It is a thing.

When the common range between the area predicted to have a meteorological disaster and the area exceeding the threshold value is wider than the predetermined range, an alarm is issued only to the area adjacent to the area within the common range that is predicted to have a meteorological disaster. If the common range is narrower than the predetermined range, a warning may be issued to an area within the common range and a warning may be issued to an area predicted to have a weather disaster.

[0018]

As described above, as a result of investigating the spread range of the mesh in which the prediction that an accident will occur is made by each prediction method, the snow accretion amount is below the threshold value even though the neural network predicts the accident. Because there were many cases that I missed,
Moreover, when the prediction by the neural network and the prediction by the snow accretion amount estimation are in good agreement, there was a tendency for missed shots to occur frequently but missed frequently. Therefore, it is possible to determine whether the mesh is often overlooked or the missed case is frequently detected depending on how the meshes overlap.

That is, by determining the mesh for issuing the weather disaster warning from the overlapping manner of the meshes predicted to cause an accident by each prediction method, both miss and miss can be reduced.

Preferably, when the common range of the mesh in which the accident is predicted is wide, the number of meshes is reduced to reduce the idling, and when the common range is narrow, the prediction by the neural network alone is used as a warning. By doing so, the oversight is reduced.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described in detail below with reference to the accompanying drawings.

As shown in FIG. 1, the transmission line snow damage warning system of the present invention is a snow damage prediction neural network device 1,
It comprises a snow accretion amount estimation device 2, a threshold value determination device 3, a prediction range comparison device 4, and an alarm output device 5. The snow damage prediction neural network device 1, the snow accretion amount estimation device 2, and the threshold value determination device 3 are shown in FIG.
The description is omitted because it is the same as the above.

The prediction range comparison device 4 uses the prediction of the snow damage prediction neural net device 1 and the prediction of the threshold value judgment device 3 to overlap the mesh predicted to have snow damage with the mesh whose snow accretion amount exceeds the threshold value. To control the operation of the alarm output device 5. The alarm output device 5 outputs a snow damage warning and a snow damage warning based on the judgment of the snow damage prediction neural network device 1 and the judgment of the threshold value judgment device 3, but unlike the conventional case, it is controlled by the prediction range comparison device 4. To be done. Specifically, when the common range between a mesh predicted to have snow damage and a mesh that exceeds the threshold is wider than a predetermined range, only the meshes that are predicted to have snow damage will be alarmed even if the adjacent meshes within the common range have an alarm. When the common range is narrower than the predetermined range, an alarm is issued to the meshes within the common range and a warning is issued to the meshes predicted to have snow damage.

Next, the operation of the embodiment will be described.

In the transmission line snow damage warning system of FIG. 1, in order to predict snow damage for each region, a mesh divided by meshes of 6 Km intervals is targeted. Snow damage prediction for each mesh is performed using the meteorological data for each 6 km square, and a snow damage warning or the like is issued for each mesh. Precipitation, temperature, and wind speed are used as meteorological data.

The meteorological data input to the snow damage prediction neural network device 1 is numerical data for a total of 6 hours including past 3 hours data including the current time and forecast data 3 hours ahead. By inputting this numerical data to the snow damage prediction neural network device 1, it is possible to predict the presence or absence of snow damage 3 hours ahead.

The snow accretion amount estimation device 2 uses the same meteorological data to calculate the estimated snow accretion amount by a snow accretion amount estimation formula derived based on experiments.

In the threshold determination device 3, the estimated snow accretion amount is 10 g / cm, as a result of the examination using the meteorological data at the time of the past accident.
In the above cases, the alarm level is output, and in the case of 5 g / cm or more, the warning level is output.

The prediction range comparison device 4 obtains a common range between the range in which the mesh predicted to have snow damage spreads and the range in which the mesh exceeds the threshold spreads. When this common range is wider than the predetermined range, the accuracy of the weather forecast data is expected to be high, and there is a tendency that snow damage will not be overlooked. On the other hand,
The number of misses tends to increase, and the weather forecast data is 6 km.
Considering that it is the average value on all sides, snow damage is often predicted not only in the mesh where snow damage was predicted, but also in the surrounding meshes. Therefore, when snow damage is predicted with a single mesh, its accuracy is rarely high, and even if an alarm is issued, it is often missed. Therefore, an alarm is issued only for the meshes where snow damage is predicted even in adjacent meshes.

On the other hand, when the common range is narrower than the predetermined range, the number of misses tends to increase, but the number of misses tends to increase. The accuracy of the weather forecast data is expected to be low. Here, it is necessary to prevent a decrease in the accuracy of the transmission line snow damage warning system. Due to its generalization ability, the neural network is unlikely to make a large error even if the input changes a little. On the other hand, the snow accretion amount estimation calculation formula is experimentally obtained based on the snow accretion condition on the power transmission line, and is therefore particularly sensitive to changes in temperature as in the actual snow accretion phenomenon. Therefore, although the temperature actually changes and snow accretion increases, the estimated snow accretion amount does not change due to a temperature forecast error, and the possibility of snow damage may be overlooked. Therefore, when the common range is narrower than the predetermined range, the snow damage prediction neural net device 1 only issues a warning to all the meshes that are predicting snow damage to prevent overlooking.

[0031]

The present invention exhibits the following excellent effects.

(1) Even if there is an error in the weather forecast
Both misses can be reduced, and the forecast accuracy of the weather warning system as a whole will be high.

[Brief description of drawings]

FIG. 1 is a block diagram of a power line snow damage warning system according to an embodiment of the present invention.

FIG. 2 is a block diagram of a conventional transmission line snow damage warning system.

[Explanation of symbols]

 1 Snow Damage Prediction Neural Network Device 2 Snow Accumulation Amount Estimation Device 3 Threshold Judgment Device 4 Prediction Range Comparison Device 5 Alarm Output Device

Claims (2)

[Claims] 1. A neural network is used to predict the presence / absence of a weather disaster in each region from meteorological data, and a physical quantity that causes a meteorological disaster is calculated for each area from the meteorological data. A meteorological disaster prediction method characterized by determining whether or not the meteorological disaster has occurred, and determining an area for issuing a meteorological hazard warning based on how the area predicted to have a meteorological disaster and the area exceeding the threshold value overlap. 2. When the common range between an area predicted to have a meteorological disaster and an area exceeding a threshold is wider than a predetermined range,
Only adjacent areas within the common range that are predicted to have meteorological disasters will be issued a warning, and if the common range is narrower than the specified range, an alarm will be issued to the areas within the common range and a meteorological disaster will be predicted. The meteorological disaster prediction method according to claim 1, wherein a warning is issued to the area.