CN110601193A - A power load forecasting system and method based on environmental variable perception and lag - Google Patents

Abstract

A power load forecasting system and method based on environmental variable perception and hysteresis, including a sensor module, a storage module, a variable hysteresis module, a Res-LSTM prediction module, an error redundancy calculation module, an electric energy storage calculation module, and a power dispatching calculation module; the sensor There are multiple modules; the sensor module is connected to the storage module by communication; the storage module is connected to the variable lag module by communication; the variable lag module is connected to the Res-LSTM prediction module, and the Res-LSTM prediction module is connected to the error redundancy calculation module. The remaining calculation module is connected with the power dispatching calculation module and the electric energy storage calculation module through communication. After pre-distributing the power loads in each area, the present invention will compare the actual power loads in the future, so as to serve as a reference for power dispatching and power storage, and reduce power waste.

Description

A power load forecasting system and method based on environmental variable perception and lag

technical field

The present invention relates to the technical field of electric power internet of things, in particular to an electric load forecasting system and method based on environmental variable perception and hysteresis.

Background technique

The power load refers to the value of the power consumed by various electrical equipment installed in state agencies, enterprises, residents and other users. Power load forecasting refers to the process of predicting the future power consumption or the time and space distribution of power consumption with the power load as the research object. Power load forecasting is the decision-making basis for power dispatching in the power grid, an important part of power system planning, and the basis for the economic operation of the power system. It is extremely important for the planning and smooth operation of the entire power grid.

At present, the power load in the power grid is mainly monitored in terms of practicability, and the forecast of power load is mainly based on the historical data of power load to roughly estimate the future short-term power demand.

In the prior art, the prediction of power load is mainly based on the historical data of power load, and its defects or deficiencies are mainly as follows:

1. Large prediction deviation: the actual power load is largely affected by various complex factors such as weather (temperature, humidity, precipitation, etc.), holidays and special conditions, emergencies of large industrial users, economic operation conditions, and national policies and regulations. Therefore, the traditional forecasting method purely based on the historical data of power load has a large deviation.

2. The prediction area is small: the existing power load forecasting model usually does not consider environmental factors, so it can only predict the power load in a small area in space. Once it involves multiple areas with large differences in environmental factors, the existing Forecasting methods based on historical load data are often no longer reliable.

In order to solve the above problems, this application proposes a power load forecasting system and method based on environmental variable perception and hysteresis.

Contents of the invention

(1) Purpose of the invention

In order to solve the technical problems of the prior art in the background technology that the prediction deviation of electric power load is large and the prediction area is small, the present invention proposes a power load prediction system and method based on environmental variable perception and hysteresis. The present invention has high prediction accuracy, Multiple prediction areas, pre-allocated redundancy and simultaneous scheduling can be referred to to reduce power waste.

(2) Technical solution

In order to solve the above problems, the present invention provides a power load forecasting system based on environmental variable perception and hysteresis, including a sensor module, a storage module, a variable hysteresis module, a Res-LSTM prediction module, an error redundancy calculation module, and an electric energy storage calculation module and power dispatching calculation module;

There are multiple sensor modules; the sensor module is connected to the storage module by communication; the storage module is connected to the variable lag module by communication; the variable lag module is connected to the Res-LSTM prediction module, and the Res-LSTM prediction module is connected to the error redundancy calculation module. The redundant calculation module is connected with the power dispatching calculation module and the electric energy storage calculation module through communication.

Preferably, the sensor module includes a rain sensor, a temperature sensor and a humidity sensor, and the sensor modules are distributed in area 1, area 2, ..., area N; the rain sensor is used to measure regional rainfall; meanwhile, the rainfall information is converted into a digital signal and carried out Transmission; the temperature sensor monitors the current temperature in the area, converts it into an available output signal and transmits it; the humidity sensor can measure the current humidity in the area, converts it into an available output signal and transmits it.

Preferably, at least one storage module is set in each area; it is used to store the historical information and current information of rainfall, temperature and humidity uploaded by the sensor module.

Preferably, the input of the variable hysteresis module is the sensing data of each regional sensor module and the historical data of electric load, and the output is a data set used for training or prediction of the Res-LSTM prediction module.

Preferably, the Res-LSTM prediction module is used to train the historical data about the environmental variables and power loads in each area formed by the variable lag module, and determine the parameters of the model so that it can be based on the subsequent input of environmental variables and historical power loads. Future electricity loads are forecasted.

Preferably, the power dispatch calculation module calculates the actual power load values of each region in the future.

Preferably, the electric energy storage calculation module calculates the electric energy storage power according to the actual electric load of each area and the pre-allocated power of each area.

Preferably, the error redundancy calculation module calculates a pre-allocated power for each area based on the predicted power load of each area and the variance of the historical power load of the area.

A method for electric load forecasting based on environmental variable perception and lag, comprising the following specific steps:

S10, the sensor modules installed in each area periodically monitor information such as rainfall, temperature, humidity, etc., and after collecting the information, upload and store it to the storage module in each area;

S20, collect the historical data of electric load in each region, and match it with the sensor data in S10 by using the environmental variable hysteresis module by setting a reasonable time interval to become an environmental information-electrical load data set;

S30, divide the environmental information-electric load data set generated in S20 into a training set, a verification set and a test set according to a certain ratio, use the training set to train the Res-LSTM model, and use the verification set to determine the hyperparameters in the Res-LSTM , using the test set to test the accuracy of the model.

S40, inputting the current environmental monitoring data such as rainfall, temperature, and humidity in S10 and the historical power load data in S20 into the trained Res-LSTM model in S30, to obtain the predicted power load values in each region;

S50, combine the power load value predicted by each region in S40 with the variance of the respective historical data of the power load in each region, and obtain the pre-allocated power of each region through the error redundancy calculation module;

S60, input the pre-allocated power of each region in S50 and the actual power load calculated by the power load dispatching module of each region into the power storage calculation module to obtain the calculated value of power storage, and provide an operation basis for the actual power dispatching process.

Preferably, the process of training, verifying and testing the Res-LSTM model in step S30 is as follows:

S31, initialize the parameters in the Res-LSTM model, including the matrix U _f , U _u , U _c , U _o and the matrix G _f , G _u , G _o and the matrix W _f , Wu _u , W in each neural network unit _c , W _o ;

S32, inputting the training set of the data set formulated in S20, using the mainstream BPTT algorithm to repeatedly iterate the parameters in the Res-LSTM model until the algorithm converges or the error is lower than the preset threshold;

S33, inputting the verification set of the data set formulated in S20, using methods such as grid search or random search, to verify the hyperparameters such as the number of neural network units, and selecting the hyperparameter combination with the smallest generalization error;

S34. Test the trained Res-LSTM. If the error is higher than the preset threshold, retrain and verify the model until the model accuracy meets the requirements.

The above-mentioned technical scheme of the present invention has the following beneficial technical effects: 1. High prediction accuracy: the present invention combines the historical data of the power load in the region due to consideration of environmental factors including temperature, humidity and precipitation, that is, simultaneously takes into account Compared with the existing power load forecasting methods, the prediction accuracy is higher due to the universality of the historical law and the particularity of the current situation of the regional power load.

2. There are many prediction areas: the present invention inputs the environmental data such as temperature, humidity and precipitation in multiple areas and the historical data of electric power load in each area into the Res-LSTM model for prediction. The relevance of regional power loads is considered, and the prediction of power loads in multiple regions is realized at the same time.

3. Pre-allocation redundancy: After the present invention predicts the power load of each region, it does not simply use the predicted power load to the power system for power dispatching as a reference, but refers to the variance of the historical power load prediction of the region, It makes the power pre-distribution in some areas with large load fluctuations very flexible.

4. Scheduling can be used as a reference: After the present invention pre-allocates the power loads in each area, it will compare the actual power loads in the future, so as to serve as a reference for power dispatching and power storage, and reduce power waste.

Description of drawings

FIG. 1 is a structural block diagram of an electric power Internet of Things load forecasting system based on environmental variable perception and hysteresis electric load forecasting system and method proposed by the present invention.

Fig. 2 is a structural block diagram of the variable hysteresis module in the environmental variable perception and hysteresis based power load forecasting system and method proposed by the present invention.

Fig. 3 is a Res-LSTM forecasting module in the electric load forecasting system and method based on environmental variable perception and hysteresis proposed by the present invention.

FIG. 4 is a structural block diagram of an error redundancy calculation module in the environmental variable perception and lag-based power load forecasting system and method proposed by the present invention.

Fig. 5 is a structural block diagram of the electric energy storage calculation module in the electric load forecasting system and method based on environmental variable perception and hysteresis proposed by the present invention.

Detailed ways

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 combination with specific embodiments and with reference to the accompanying drawings. It should be understood that these descriptions are exemplary only, and are not intended to limit the scope of the present invention. Also, in the following description, descriptions of well-known structures and techniques are omitted to avoid unnecessarily obscuring the concept of the present invention.

As shown in Figures 1-5, the present invention proposes a power load forecasting system based on environmental variable perception and hysteresis, including a sensor module, a storage module, a variable hysteresis module, a Res-LSTM prediction module, an error redundancy calculation module, and an electric energy Storage computing module and power scheduling computing module;

There are multiple sensor modules; the sensor module is connected to the storage module by communication; the storage module is connected to the variable lag module by communication; the variable lag module is connected to the Res-LSTM prediction module, and the Res-LSTM prediction module is connected to the error redundancy calculation module. The redundant calculation module is connected with the power dispatching calculation module and the electric energy storage calculation module through communication.

In an optional embodiment, the sensor module includes a rain sensor, a temperature sensor and a humidity sensor, and the sensor modules are distributed in area 1, area 2, ..., area N; the rain sensor is used to measure regional rainfall; simultaneously the rainfall information It is converted into a digital signal and transmitted; the temperature sensor monitors the current temperature of the area, converts it into an available output signal and transmits it; the humidity sensor can measure the current humidity of the area, converts it into an available output signal and transmits it.

In an optional embodiment, at least one storage module is set in each area; it is used to store the historical information and current information of rainfall, temperature and humidity uploaded by the sensor module.

In an optional embodiment, the input of the variable hysteresis module is the sensing data of each regional sensor module and the historical power load data, and the output is a data set used for training or prediction of the Res-LSTM prediction module. As shown in Figure 2, the main function of the variable lag module is to shift the environmental historical data and historical power load of each region according to the required forecast span, so as to facilitate the formation of data sets for the subsequent load forecasting model training. The corresponding relationship between the original environmental variable data and the electric load data is:

in, Indicates the rainfall in a certain area at time t, Indicates the temperature in a certain area at time t, Indicates the humidity in a certain area at time t, and y ^(t) indicates the load data in a certain area at time t.

Assuming that the forecast lag is Δ, that is, it is necessary to predict the future power load after the Δ time period based on the current environmental data and historical load data. After the variable lag module is processed, the corresponding relationship between the environmental variable data and the power load data is:

In an optional embodiment, the Res-LSTM prediction module is used to train the historical data about the environmental variables and power loads in each area formed by the variable lag module, and determine the parameters of the model so that it can be based on the subsequent input environment variables and historical power load to predict the future power load; as shown in Figure 3, compared with the traditional LSTM model, the network structure of the Res-LSTM model adds a Res residual operation, which is mainly to make the power load The selection of prediction intervals is more flexible, avoiding the problem of gradient disappearance due to too many neural network layers in model training. In the Res-LSTM model, multiple neural networks are set according to the time interval between environmental variables and historical power load data. Network unit, each neural network unit takes environmental variables as input and predicted power load as output, denoted as A ⁽¹⁾ , A ⁽²⁾ ,...,A ^(t) , and neural network unit A ^{(t )} as an example, the input-output relationship between the environmental variable input x ^(t) and the power load forecast value y( ^t ) is:

The connection output of the classic LSTM model in the neural network unit A ^(t-1) and Directly as the connection input c ^(t-1) and a ^(t-1) of the neural network unit A ⁽ t), where the Res residual operation is to output the connection output c ^(t -2) of the neural network unit A ^{(t-2) )} and a ^(t-2) , and the connection output of the neural network unit A ^(t-1) and Together, it is also used as the connection input c ^(t-1) and a ^(t-1) of the neural network unit A ^(t) . The specific method of the above operation is:

and, there are:

Similarly, for the connection output c ^(t-1) and a ^(t-1) of the neural network unit A ⁽ t-1), and the connection output of the neural network unit A ^(t) and Together, it is also used as the connection input c ^(t) and a ^(t) of the neural network unit A ^(t+1) . The specific method of this operation is:

and, there are:

For the parts of the Res-LSTM prediction module that are the same as those of the traditional LSTM model, details will not be repeated here.

In an optional embodiment, the error redundancy calculation module calculates a pre-allocated power for each area according to the predicted power load of each area and in combination with the variance of the historical power load of the area. As shown in Figure 4, assume that the predicted load of region n at time t is The variance of historical load data in this area is var _{area n} , and the redundancy ratio is θ, then the pre-allocated power of area n at time t for:

In an optional embodiment, the power dispatch calculation module calculates the actual power load value of each area in the future, and the actual load of area n at time t is recorded as

In an optional embodiment, the electric energy storage calculation module calculates the electric energy storage power according to the actual electric load of each area and the pre-allocated power of each area. As shown in Figure 5, firstly, the actual power load of each area at time t is calculated according to the power dispatch calculation module Calculate the sum Load ^(t) of the actual power load of each area at time t, which can be expressed as:

At the same time, according to the pre-allocated power of each area at time t obtained by the error redundancy calculation module The sum P ^(t) of the pre-allocated power of all areas at time t is obtained, which can be expressed as:

Then the electric energy storage calculation module can obtain the calculation result of the electric energy storage value, which can be expressed as:

Ch ^(t) = P ^(t) -Load ^(t) (11)

Wherein, if Ch ^(t) >0, it indicates that the energy storage device should be in a charging state; otherwise, if Ch ^(t) <0, it indicates that the energy storage device should be in a discharging state.

A method for electric load forecasting based on environmental variable perception and lag, comprising the following specific steps:

S10, the sensor modules installed in each area periodically monitor information such as rainfall, temperature, humidity, etc., and after collecting the information, upload and store it to the storage module in each area;

S20, collect the historical data of electric load in each region, and match it with the sensor data in S10 by using the environmental variable hysteresis module by setting a reasonable time interval to become an environmental information-electrical load data set;

S30, divide the environmental information-electric load data set generated in S20 into a training set, a verification set and a test set according to a certain ratio, use the training set to train the Res-LSTM model, and use the verification set to determine the hyperparameters in the Res-LSTM , using the test set to test the accuracy of the model.

S40, inputting the current environmental monitoring data such as rainfall, temperature, and humidity in S10 and the historical power load data in S20 into the trained Res-LSTM model in S30, to obtain the predicted power load values in each region;

S50, combine the power load value predicted by each region in S40 with the variance of the respective historical data of the power load in each region, and obtain the pre-allocated power of each region through the error redundancy calculation module;

S60, input the pre-allocated power of each region in S50 and the actual power load calculated by the power load dispatching module of each region into the power storage calculation module to obtain the calculated value of power storage, and provide an operation basis for the actual power dispatching process.

In an optional embodiment, the process of training, verifying and testing the Res-LSTM model in step S30 is as follows:

S31, initialize the parameters in the Res-LSTM model, including the matrix U _f , U _u , U _c , U _o and the matrix G _f , G _u , G _o and the matrix W _f , Wu _u , W in each neural network unit _c , W _o ;

S32, inputting the training set of the data set formulated in S20, using the mainstream BPTT algorithm to repeatedly iterate the parameters in the Res-LSTM model until the algorithm converges or the error is lower than the preset threshold;

S33, inputting the verification set of the data set formulated in S20, using methods such as grid search or random search, to verify the hyperparameters such as the number of neural network units, and selecting the hyperparameter combination with the smallest generalization error;

S34. Test the trained Res-LSTM. If the error is higher than the preset threshold, retrain and verify the model until the model accuracy meets the requirements.

The present invention considers the environmental factors including temperature, humidity and precipitation, and combines the historical data of the regional power load, that is, simultaneously considers the universality of the historical law of the regional power load and the particularity of the current situation, so compared with the present Some power load forecasting methods have higher forecasting accuracy.

In the present invention, environmental data such as temperature, humidity and precipitation in multiple regions and historical data of electric load in each region are simultaneously input into the Res-LSTM model for prediction. Responsibility is considered, and the forecasting of power loads in multiple regions is realized at the same time.

After the present invention predicts the power load in each region, it does not simply use the predicted power load to the power system for power dispatching as a reference, but refers to the variance of the historical power load forecast in the region, so that some loads with large fluctuations Regional power pre-distribution has good flexibility.

After pre-distributing the power loads in each area, the present invention will compare the actual power loads in the future, so as to serve as a reference for power dispatching and power storage, and reduce power waste.

Due to the existence of two types of defects of "large prediction deviation" and "small prediction area" in the existing electric load monitoring or forecasting technology, the present invention has a certain degree of irreplaceability in the following two practical application scenarios:

1. Power load changes: In practical applications, if the power load changes greatly between seasons or years in some areas due to climate factors or economic operation factors, it is difficult for the existing load forecasting methods to take this into account. Due to the particularity of stage power loads relative to historical power loads in the same period, the present invention can play an irreplaceable role to a certain extent in this scenario.

2. Changes in power dispatching: that is, in practical applications, if there is a large difference between the actual load and the pre-allocated power in some areas due to some unknown factors, at this time, the existing method simply considers the power load in this area And research, it is impossible to take into account the relevance of power dispatching in each area. At this time, the present invention has a certain irreplaceability in this scenario.

It should be understood that the above specific embodiments of the present invention are only used to illustrate or explain the principles of the present invention, and not to limit the present invention. Therefore, any modification, equivalent replacement, improvement, etc. made without departing from the spirit and scope of the present invention shall fall within the protection scope of the present invention. Furthermore, it is intended that the appended claims of the present invention cover all changes and modifications that come within the scope and metespan of the appended claims, or equivalents of such scope and metesight.

Claims (10)

1. A power load prediction system based on environment variable perception and hysteresis is characterized by comprising a sensor module, a storage module, a variable hysteresis module, a Res-LSTM prediction module, an error redundancy calculation module, an electric energy storage calculation module and an electric power scheduling calculation module;

the sensor modules are arranged in plurality; the sensor module is in communication connection with the storage module; the storage module is in communication connection with the variable hysteresis module; the variable hysteresis module is in communication connection with the Res-LSTM prediction module, the Res-LSTM prediction module is in communication connection with the error redundancy calculation module, and the error redundancy calculation module is in communication connection with the power scheduling calculation module and the electric energy storage calculation module.

2. The system according to claim 1, wherein the sensor modules comprise rain sensors, temperature sensors and humidity sensors, and are distributed in zone 1, zone 2, …, zone N; the rainfall sensor is used for measuring the rainfall of the region; simultaneously converting rainfall information into digital signals and transmitting the digital signals; the temperature sensor monitors the current temperature of the area, converts the current temperature into an available output signal and transmits the available output signal; the humidity sensor can measure the current humidity of the area, convert the current humidity into a usable output signal and transmit the usable output signal.

3. The system according to claim 2, wherein at least one memory module is disposed in each region; the system is used for storing the historical information and the current information of rainfall, temperature and humidity uploaded by the sensor module.

4. The system of claim 1, wherein the inputs to the variable hysteresis module are the sensed data and the power load history data of each regional sensor module, and the outputs are the data set for Res-LSTM prediction module training or prediction.

5. The system according to claim 1, wherein the Res-LSTM prediction module is configured to train historical data about environmental variables and power loads of each region formed by the variable lag module, and determine parameters of the model to enable prediction of future power loads based on subsequently inputted environmental variables and historical power loads.

6. The system according to claim 1, wherein the power scheduling calculation module calculates the actual power load value of each future region.

7. The system according to claim 1, wherein the electrical energy storage calculation module calculates the electrical energy storage power according to the actual electrical load of each region and the pre-allocated power of each region.

8. The system of claim 1, wherein the error redundancy calculation module calculates a pre-allocated power for each of the regions based on the predicted power load for each region in combination with the variance of the historical power load for the region.

9. The system according to any one of claims 1-8, further comprising a method for predicting an electrical load based on the perception and the hysteresis of the environmental variables, comprising the steps of:

s10, the sensor modules arranged in each area periodically monitor information such as rainfall, temperature, humidity and the like, and after the information is collected, the information is uploaded and stored to the storage modules of each area;

s20, collecting power load historical data of each region, and matching the historical data with the sensor data in S10 by using an environment variable hysteresis module through setting a reasonable time interval to form an environment information-power load data set;

and S30, dividing the environmental information-power load data set generated in S20 into a training set, a verification set and a test set according to a certain proportion, training the Res-LSTM model by using the training set, determining the hyper-parameters in the Res-LSTM by using the verification set, and testing the accuracy of the model by using the test set.

S40, inputting the current environmental monitoring data such as rainfall, temperature and humidity in S10 and the historical data of the power load in S20 into the Res-LSTM model trained in S30 to obtain the predicted power load value of each area;

s50, combining the power load values obtained by prediction of each region in S40 with the variance of the power load historical data of each region, and obtaining the pre-distributed power of each region through an error redundancy calculation module;

and S60, inputting the pre-distributed power of each region in the S50 and the actual power load calculated by the power load scheduling module of each region into the electric energy storage calculation module to obtain an electric energy storage calculation value, and providing an operation basis for the actual power scheduling process.

10. The method of claim 9, wherein the method comprises the steps of: the training, verifying and testing process of the Res-LSTM model in step S30 is as follows:

s31, initializing Res-LSTM modelIncluding the matrix U in each neural network unit_f，U_u，U_c，U_oAnd matrix G_f，G_u，G_oAnd a matrix W_f，W_u，W_c，W_o；

S32, inputting a training set of the data set formulated in S20, and repeatedly iterating parameters in the Res-LSTM model by using a mainstream BPTT algorithm until the algorithm converges or the error is lower than a preset threshold value;

s33, inputting a verification set of the data set formulated in S20, verifying the hyperparameters such as the number of neural network units by using methods such as grid search or random search, and selecting the hyperparameter combination with the minimum generalization error;

and S34, testing the trained Res-LSTM, and if the error is higher than a preset threshold value, re-training and verifying the model until the model precision meets the requirement.