Method, device, storage medium and product for load prediction of a local network

Abstract

The application discloses a kind of regional network supply load prediction method, equipment, storage medium and product, the prediction method includes obtaining the historical data of target area before prediction;History data is analyzed and sorted, constructs time derivative feature;According to historical network supply load and historical meteorological data, select the meteorological parameter with high correlation degree with historical network supply load, and then according to historical meteorological data and selected meteorological parameter constructs meteorological derivative feature;According to the daily historical peak power consumption and historical valley power consumption of each user extracted, user derivative feature is constructed;Load and distributed energy output prediction model are constructed, and historical data and time derivative feature, meteorological derivative feature and user derivative feature are used to train and verify load and distributed energy output prediction model, obtain target prediction model;Regional network supply load prediction is carried out using target prediction model.The application improves the prediction accuracy of network supply load.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of new energy grid connection, and particularly relates to a regional grid supply load prediction method and device, a storage medium and a product. BACKGROUND

[0002] The installed capacity of distributed new energy in an industrial park is increasingly high. However, the output of new energy (such as photovoltaic and wind turbine) is significantly affected by meteorological conditions and has intermittency and uncertainty, which changes the original load curve trend and poses new challenges to the prediction of the grid supply load of the industrial park. The grid supply load of the industrial park is the difference between the actual load in the industrial park and the output of the distributed energy. At present, the prediction of the grid supply load of the industrial park is increasingly difficult, and accurate prediction of the grid supply load of the industrial park is of great significance to maintaining the stability and safety of power in the park.

[0003] At present, most of the grid supply load prediction methods of the industrial park only consider the data of equivalent load and only rely on historical load data as a data source to establish a prediction model. This prediction method is intuitive and has a certain accuracy, but as the scale of distributed energy increases, the sensitivity of the grid supply load of the industrial park to new features such as meteorological conditions increases, and only relying on historical load data cannot meet the prediction accuracy requirements.

[0004] The patent document with the publication number CN117713049A discloses a regional grid supply load prediction method, which predicts the total grid supply load of the region by predicting the historical grid supply load of residential electricity, commercial electricity, industrial electricity and agricultural electricity in the region. Only historical load data is considered, and the influence of new features is not considered, resulting in poor prediction accuracy of the grid supply load. SUMMARY

[0005] The purpose of the present application is to provide a regional grid supply load prediction method, device, storage medium and product to solve the problem that the traditional prediction method only considers historical load data, resulting in prediction accuracy that cannot meet the requirements.

[0006] The present application solves the above technical problems through the following technical solutions: a regional grid supply load prediction method, comprising:

[0007] obtaining historical data of the target region before prediction; wherein the historical data includes historical grid supply load, historical total electricity load, historical distributed energy output, historical electricity load of each user and historical meteorological data of the target region;

[0008] analyzing and classifying the historical data to construct time-derived features; wherein the time-derived features include a weekday data set, a non-weekday data set, an intra-day load peak value data set, and an intra-day load valley value data set; the weekday data set includes historical grid supply load, historical total electricity load, historical distributed energy output, and historical electricity load of each user of each working day in chronological order; the non-weekday data set includes historical grid supply load, historical total electricity load, historical distributed energy output, and historical electricity load of each user of each non-working day in chronological order; the intra-day load peak value data set includes daily historical load peak values in chronological order; and the intra-day load valley value data set includes daily historical load valley values in chronological order;

[0009] According to the historical grid supply load and historical meteorological data, a meteorological parameter with high correlation with the historical grid supply load is selected, and then a meteorological-derived feature is constructed according to the historical meteorological data and the selected meteorological parameter;

[0010] From the historical electricity load of each user, the daily historical peak-time electricity consumption and the daily historical valley-time electricity consumption of each user are extracted, and a user-derived feature is constructed according to the extracted daily historical peak-time electricity consumption and the daily historical valley-time electricity consumption of each user;

[0011] A load and distributed energy output prediction model is constructed, and the historical data and the time-derived features, the meteorological-derived features, and the user-derived features are used to train and verify the load and distributed energy output prediction model to obtain a target prediction model;

[0012] In the prediction stage, the total electricity load prediction value and the distributed energy output prediction value are obtained by using the target prediction model, and the regional grid supply load prediction value is calculated according to the total electricity load prediction value and the distributed energy output prediction value.

[0013] Further, when constructing the time-derived features, binary indexes are added to the weekday data set, the non-weekday data set, the intra-day load peak value data set, and the intra-day load valley value data set.

[0014] Further, the maximum mutual information coefficient is used to select the meteorological parameter with high correlation with the historical grid supply load, which specifically includes:

[0015] According to the historical grid supply load and historical meteorological data, the mutual information coefficient of each two-dimensional sequence under different rows and different columns is calculated, and the specific calculation formula is:

[0016]

[0017] wherein, represents the mutual information coefficient of the nth two-dimensional sequence D n , and the two-dimensional sequence Dn It is a two-dimensional sequence D composed of all specific values ​​of the nth meteorological parameter in historical meteorological data and the specific value Y of the historical network load corresponding to each specific value X of the nth meteorological parameter. n The sequence D is distributed in a two-dimensional space, which is divided into a rows and b columns; p(i,j) represents the two-dimensional sequence D. n The probability of falling in the i-th row and j-th column; p(i) represents the two-dimensional sequence D. n The probability of falling in the i-th row; p(j) represents the two-dimensional sequence D. n The probability of falling in the j-th column;

[0018] The maximum mutual information coefficient of each two-dimensional sequence and its corresponding number of rows and columns are determined based on the mutual information coefficient of each two-dimensional sequence under different rows and columns;

[0019] The maximum mutual information coefficient of each two-dimensional sequence is normalized based on the number of rows and columns corresponding to the maximum mutual information coefficient. The specific formula is as follows:

[0020]

[0021] in, D represents the nth two-dimensional sequence. n The normalized value of the maximum mutual information coefficient; D represents the nth two-dimensional sequence. n Maximum mutual information coefficient; a m D represents the nth two-dimensional sequence. n The number of rows corresponding to the maximum mutual information coefficient; b m D represents the nth two-dimensional sequence. n The number of columns corresponding to the maximum mutual information coefficient; N represents the nth two-dimensional sequence D. n The number of data sets in the dataset, each data set includes a specific value of the nth meteorological parameter and its corresponding historical network load value; α represents the hyperparameter;

[0022] Based on the normalized value of the maximum mutual information coefficient of each two-dimensional sequence, meteorological parameters with high correlation to the historical network load are selected.

[0023] Furthermore, the historical meteorological data involves meteorological parameters including weather type, humidity, temperature, air pressure, precipitation, wind speed, and wind direction. The meteorological parameters selected that are highly correlated with the historical grid power supply load include weather type, temperature, and precipitation.

[0024] Furthermore, the load and distributed energy output prediction model includes a load prediction model and a distributed energy output model;

[0025] Alternatively, the load and distributed energy output prediction model is a multi-task model, the multi-task model adopts a customized gating network architecture, and a task tower in the customized gating network architecture is replaced by a fully connected layer.

[0026] Further, the prediction method further comprises correcting the calculated grid load prediction value, and the specific correction process is:

[0027] An error prediction model is constructed in advance, and the grid load prediction value is subjected to error prediction by using the error prediction model to obtain an error prediction value.

[0028] The error prediction value and the grid load prediction value are added to obtain a final grid load prediction value.

[0029] Further, the construction process of the error prediction model is:

[0030] Step A1: input the historical data and the time-derived features, the weather-derived features and the user-derived features into the target prediction model to obtain a total power load prediction value and a distributed energy output prediction value;

[0031] Step A2: calculate a grid load initial prediction value according to the total power load prediction value and the distributed energy output prediction value;

[0032] Step A3: obtain an error initial prediction value according to the grid load initial prediction value and the historical grid load;

[0033] Step A4: add the error initial prediction value and the grid load initial prediction value to obtain a new grid load prediction value;

[0034] Step A5: obtain a new error prediction value according to the new grid load prediction value obtained in step A4 and the historical grid load;

[0035] Step A6: compare the error initial prediction value obtained in step A3 and the new error prediction value obtained in step A5, if the error initial prediction value is greater than the new error prediction value, repeat steps A1 to A6; if the error initial prediction value is less than or equal to the new error prediction value, stop training to obtain a trained error prediction model.

[0036] Based on the same concept, the application further provides an electronic device comprising a memory, a processor and a computer program / instruction stored in the memory, wherein the processor executes the computer program / instruction to implement the regional grid load prediction method as described above.

[0037] Based on the same concept, the application also provides a computer readable storage medium, which stores computer programs / instructions, and the computer programs / instructions are executed by a processor to realize the area network supply load prediction method as described above.

[0038] Based on the same concept, the application also provides a computer program product, which comprises computer programs / instructions, and the computer programs / instructions are executed by a processor to realize the area network supply load prediction method as described above.

[0039] Compared with the prior art, the application has the following advantages:

[0040] The application constructs derived features from three aspects of time, weather and user behavior, mines high-sensitive features of the network supply load, and uses historical data and the derived features to predict the load and distributed energy output, so that the prediction accuracy of the load and distributed energy output is greatly improved, and the prediction accuracy of the network supply load is improved.

[0041] The application constructs a load and distributed energy output prediction model based on a customized gate control network architecture, realizes the prediction of the total power load and distributed energy output through one model, dynamically fuses the shared expert network and specific expert network in the customized gate control network architecture and the gate control mechanism, so that the model can find a balance between the load prediction task and the distributed energy output prediction task, improves the prediction accuracy, and better handles the conflict between the tasks and the sample correlation problem.

[0042] The application also uses an error prediction model to correct the error of the network supply load prediction value, and further improves the prediction accuracy of the network supply load. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only one embodiment of the application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0044] Figure 1 is the flow chart of the area network supply load prediction method in the embodiment of the application;

[0045] Figure 2 is the multi-task model based on CGC in the embodiment of the application. DETAILED DESCRIPTION

[0046] With reference to the accompanying drawings, the technical solutions in the present application will be described clearly and completely in embodiments of the present application. Obviously, the described embodiments are only a part but not all of the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work fall into the protection scope of the present application.

[0047] The technical solutions of the present application will be described in detail with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described in some embodiments.

[0048] Embodiment one

[0049] Figure 1 The flow chart of the regional grid supply load prediction method provided by the present application is shown. As shown in the figure, the regional grid supply load prediction method comprises the following steps: Figure 1

[0050] Step S1: obtaining the historical data of the target region before prediction.

[0051] The historical data includes the historical grid supply load of the target region, the historical total power consumption load, the historical distributed energy output, the historical power consumption load of each user and the historical meteorological data. The meteorological parameters involved in the historical meteorological data include weather type, humidity, temperature, air pressure, precipitation, wind speed and wind direction.

[0052] In this embodiment, the industrial park is taken as the target region, and the historical data of the industrial park grid supply load prediction for 30 days before the day is obtained. The data is collected every 15 minutes, and the collected data each time is the grid supply load, the total power consumption load, the distributed energy output, the power consumption load of each user and the meteorological data.

[0053] In order to avoid the influence of data noise on the prediction result, the historical data is also preprocessed in the present application. The specific preprocessing includes missing value, abnormal value and normalization processing. In this embodiment, the missing value processing adopts the method of piecewise linear interpolation, the abnormal value processing first eliminates and then interpolates, and the normalization specifically adopts Log normalization processing.

[0054] Step S2: analyzing and classifying the historical grid supply load, the historical total power consumption load, the historical distributed energy output and the historical power consumption load of each user, and constructing time derivative features.

[0055] ​The historical grid supply load, the historical total power consumption load, the historical distributed energy output, and the historical power consumption load of each user are analyzed, and it is found that the data corresponding to the working day, non-working day, daily load peak, and daily load valley are quite different. Therefore, the historical grid supply load, the historical total power consumption load, the historical distributed energy output, and the historical power consumption load of each user are classified and sorted to construct time-derived features.

[0056] The time-derived features include a working day data set, a non-working day data set, a daily load peak data set, and a daily load valley data set. The working day data set includes historical grid supply load, historical total power consumption load, historical distributed energy output, and historical power consumption load of each user of each working day sorted by time. The non-working day data set includes historical grid supply load, historical total power consumption load, historical distributed energy output, and historical power consumption load of each non-working day sorted by time. The daily load peak data set includes daily historical load peaks sorted by time. The daily load valley data set includes daily historical load valleys sorted by time.

[0057] The time-derived features involve a large amount of data. In order to facilitate the calling of the time-derived features, index items are added to the working day data set, the non-working day data set, the daily load peak data set, and the daily load valley data set. In order to avoid the sequence difference caused by the index, the indexes of the working day data set, the non-working day data set, the daily load peak data set, and the daily load valley data set are binary indexes. For example, the indexes of the working day data set, the non-working day data set, the daily load peak data set, and the daily load valley data set are (0, 0, 1), (0, 1, 0), (0, 1, 1), and (1, 0, 0) respectively, that is, the grid supply load, the total power consumption load, the distributed energy output, and the power consumption load of each user of the historical working day are found in (0, 0, 1).

[0058] Step S3: According to the historical grid supply load and the historical meteorological data, a meteorological parameter with high correlation with the historical grid supply load is selected, and then a meteorological-derived feature is constructed according to the historical meteorological data and the selected meteorological parameter.

[0059] In the specific embodiments of the present application, the maximal information coefficient (MIC) is used to select the meteorological parameter with high correlation with the historical grid supply load, which specifically includes:

[0060] Step S3.1: According to the historical grid supply load and the historical meteorological data, the mutual information coefficient of each two-dimensional sequence under different rows and columns is calculated.

[0061] The meteorological parameters involved in the historical meteorological data include weather type, humidity, temperature, air pressure, precipitation, wind speed and wind direction. One meteorological parameter is selected from the 7 meteorological parameters, all specific values of the meteorological parameter and specific values of the grid power load corresponding to each specific value are extracted from the historical meteorological data, and a two-dimensional sequence D is constructed n ∈(X,Y), the two-dimensional sequence D corresponding to the nth meteorological parameter n Each group of data (or each data point) includes a specific value X of the meteorological parameter and a specific value Y of the corresponding grid power load. The two-dimensional sequence D n is distributed in a two-dimensional space, the two-dimensional space is divided into a grid of a rows and b columns, and the mutual information coefficient of the two-dimensional sequence D n is calculated by calculating the probability of falling in different rows and different columns, and the specific calculation formula is:

[0062]

[0063] Wherein, represents the mutual information coefficient of the nth two-dimensional sequence D n ; p(i,j) represents the probability of the two-dimensional sequence D n falling in the ith row and the jth column, that is, the ratio of the number of data points falling in the ith row and the jth column in the two-dimensional sequence D n to the total number of data points; p(i) represents the probability of the two-dimensional sequence D n falling in the ith row, that is, the ratio of the number of data points falling in the ith row in the two-dimensional sequence D n to the total number of data points; p(j) represents the probability of the two-dimensional sequence D n falling in the jth column, that is, the ratio of the number of data points falling in the jth column in the two-dimensional sequence D n to the total number of data points.

[0064] Different rows and different columns are taken, and the mutual information coefficients of the two-dimensional sequences under different rows and different columns are obtained according to formula (1).

[0065] Step S3.2: Determine the maximum mutual information coefficient of each two-dimensional sequence and the corresponding row number and column number according to the mutual information coefficient of each two-dimensional sequence under different rows and different columns.

[0066] The maximum mutual information coefficient of each two-dimensional sequence is selected from the mutual information coefficients of each two-dimensional sequence under different rows and different columns, that is, the maximum mutual information coefficient of each two-dimensional sequence is obtained, and then the row number and column number corresponding to the maximum mutual information coefficient are obtained.

[0067] Step S3.3: Normalize the maximum mutual information coefficient of each two-dimensional sequence according to the row number and column number corresponding to the maximum mutual information coefficient, and the specific formula is:

[0068]

[0069] wherein, denotes the normalized value of the maximum mutual information coefficient of the nth two-dimensional sequence D n ; n denotes the number of the two-dimensional sequence; and denotes the maximum mutual information coefficient of the nth two-dimensional sequence D n ; n denotes the number of the two-dimensional sequence; and m denotes the row number corresponding to the maximum mutual information coefficient of the nth two-dimensional sequence D n ; n denotes the number of the two-dimensional sequence; and m denotes the column number corresponding to the maximum mutual information coefficient of the nth two-dimensional sequence D n ; n denotes the number of the two-dimensional sequence; and N denotes the number of data groups or data points in the nth two-dimensional sequence D n ; and a denotes a hyperparameter. In the present embodiment, the value of the hyperparameter is 0.6.

[0070] Step S3.4: selecting, according to the normalized value of the maximum mutual information coefficient of each two-dimensional sequence, the meteorological parameters having high correlation with the historical grid supply load.

[0071] For example, the normalized values of the maximum mutual information coefficients calculated according to the historical grid supply load and the historical meteorological data are as shown in Table 1:

[0072] Features Humidity Barometric pressure Temperature Wind direction Wind speed Weather type Precipitation Symbols MIC D1 ]]> MIC D2 ]]> MIC D3 ]]> MIC D4 ]]> MIC D5 ]]> MIC D6 ]]> MIC D7 ]]> Specific values 0.192 0.017 0.417 0.033 0.058 0.240 0.312

[0073] The normalized values of the maximum mutual information coefficients of all the two-dimensional sequences are sorted, and the meteorological parameters corresponding to the top 3 two-dimensional sequences are selected. The specific values of the 3 meteorological parameters from the historical meteorological data are extracted, thereby constructing the meteorological derived features. The meteorological parameters selected in the present embodiment are the weather type, the temperature and the precipitation.

[0074] Step S4: extracting the daily historical peak power consumption and the daily historical valley power consumption of each user from the historical power consumption of each user, and constructing the user derived features according to the extracted daily historical peak power consumption and the daily historical valley power consumption of each user.

[0075] The behavior of each user is characterized by the daily historical peak power consumption and the daily historical valley power consumption of each user.

[0076] Step S5: constructing the load and distributed energy output prediction model, and training and verifying the load and distributed energy output prediction model by using the historical data and the time derived features, the meteorological derived features and the user derived features, to obtain the target prediction model.

[0077] ​​The load and distributed energy output prediction model of the present application has two implementation modes: one is to construct two independent models, train and verify the two independent models respectively, obtain a load prediction model and a distributed energy output prediction model independent of each other, predict the total power load by using the load prediction model, and predict the distributed energy output by using the distributed energy output prediction model; the other is to construct a multi-task model, train and verify the multi-task model, obtain a load and distributed energy output prediction model, and simultaneously predict the total power load and the distributed energy output by using a load and distributed energy output prediction model.

[0078] In this embodiment, as shown in Figure 2 The multi-task model adopts a customized gate control network (CGC) architecture, and replaces the task tower in the customized gate control network architecture with a fully connected layer. Based on the multi-task model of the customized gate control network architecture, the dynamic fusion of the shared expert network and the specific expert network in the customized gate control network architecture and the gating mechanism enables the multi-task model to find a balance between the load prediction task and the distributed energy output prediction task, improves the prediction accuracy, and better handles the conflict between tasks and the sample correlation problem.

[0079] The specific training process of the multi-task model is as follows: input the historical data and the time-derived features, the meteorological-derived features and the user-derived features into the multi-task model, output the total power load prediction value and the distributed energy output prediction value; calculate the difference between the total power load prediction value and the distributed energy output prediction value, and the difference is the grid supply load prediction value; calculate the loss function according to the grid supply load prediction value and the label, adjust the weight parameters of the multi-task model according to the loss function, and realize the training of the multi-task model. In this embodiment, the specific calculation formula of the loss function is as follows:

[0080]

[0081] Wherein, Γ represents the total loss; represents the grid supply load prediction value; y ture represents the historical grid supply load true value; MSE represents the mean square error; R g represents the gating regularization term, which adopts L1 regularization; R e represents the expert network regularization term, which adopts L2 regularization; γ1, γ2, γ3 all represent the weight of each term.

[0082] For example, the historical data is the data of the previous 30 days, and in the training stage, the historical data and derived features of the first day to the seventh day are used as input, and the historical grid supply load of the eighth day is used as label to train the multi-task model.

[0083] Step S6: In the prediction stage, the target prediction model is used to obtain the total electricity load prediction value and the distributed energy output prediction value, and the grid supply load prediction value is calculated according to the total electricity load prediction value and the distributed energy output prediction value.

[0084] In the prediction stage, the target prediction model constructed through steps S1 to S5 is called, and the historical data and derived features of the prediction day are input into the target prediction model, so that the total electricity load prediction value and the distributed energy output prediction value are obtained, and then the grid supply load prediction value can be obtained.

[0085] Step S7: The grid supply load prediction value obtained in step S6 is corrected to obtain the final grid supply load prediction value.

[0086] In the specific embodiment of the present application, the grid supply load prediction value is corrected, and the specific correction process is as follows:

[0087] Step S7.1: The error prediction model is used to predict the error of the grid supply load prediction value, and the error prediction value is obtained;

[0088] Step S7.2: The error prediction value and the grid supply load prediction value are added to obtain the final grid supply load prediction value.

[0089] The error prediction model is constructed before the prediction application. The error prediction model of the present embodiment adopts a recurrent neural network (RNN) which can capture the dependency relationship in time series data, thereby further improving the prediction accuracy of the regional grid supply load.

[0090] In the specific embodiment of the present application, the construction process of the error prediction model is as follows:

[0091] Step A1: The historical data and the time-derived features, the weather-derived features and the user-derived features are input into the target prediction model to obtain the total electricity load prediction value and the distributed energy output prediction value;

[0092] Step A2: The grid supply load initial prediction value is calculated according to the total electricity load prediction value and the distributed energy output prediction value obtained in step A1;

[0093] Step A3: The error initial prediction value is obtained according to the grid supply load initial prediction value obtained in step A2 and the historical grid supply load;

[0094] Step A4: The error initial prediction value obtained in step A3 and the grid supply load initial prediction value obtained in step A2 are added to obtain a new grid supply load prediction value;

[0095] Step A5: The new error prediction value is obtained according to the new grid supply load prediction value obtained in step A4 and the historical grid supply load;

[0096] Step A6: comparing the error initial prediction value obtained in step A3 with the new error prediction value obtained in step A5, if the error initial prediction value is greater than the new error prediction value, repeating steps A1 to A6 to continue training, if the error initial prediction value is less than or equal to the new error prediction value, stopping training to obtain the trained error prediction model.

[0097] Embodiment two

[0098] The embodiments of the present application also provide an electronic device, which comprises a memory, a processor and computer programs / instructions stored in the memory, and the processor executes the computer programs / instructions to implement the area network load prediction method in the embodiments of the present application.

[0099] Although not shown, the electronic device comprises a processor, which can perform various appropriate operations and processes according to programs and / or data stored in a read-only memory (ROM) or loaded from a storage part into a random access memory (RAM). The processor can be a multi-core processor or can comprise a plurality of processors. In some embodiments, the processor can comprise a general-purpose main processor and one or more special-purpose coprocessors, such as a central processing unit, a graphics processing unit (GPU), a neural network processing unit (NPU), a digital signal processor (DSP), etc. In the RAM, various programs and data required for device operation are also stored. The processor, the ROM and the RAM are connected to each other through a bus. An input / output (I / O) interface is also connected to the bus.

[0100] The processor and the memory are used together to execute programs / instructions stored in the memory, and the programs / instructions are executed by a computer to implement the methods, steps or functions described in the above embodiments.

[0101] Although not shown, the embodiments of the present application also provide a computer readable storage medium having computer programs / instructions stored thereon, and the computer programs / instructions are executed by a processor to implement the area network load prediction method in the embodiments of the present application.

[0102] Read only memory (ROM), electrically programmable read only memory (EPROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital versatile discs (DVD) or other optical storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information for access by a computing device. According to the definition provided herein, computer readable medium does not include transitory computer readable medium, such as a modulated data signal and a carrier wave.

[0103] Although not shown, the embodiments of the present application further provide a computer program product, comprising: computer programs / instructions, which, when executed by a processor, implement the area network load prediction method in the embodiments of the present application.

[0104] The above disclosure is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or modifications within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for predicting load of a local area network, characterized by, The prediction method comprises: obtaining historical data of a target area before prediction; wherein the historical data comprises historical grid supply load, historical total electricity consumption load, historical distributed energy output, historical electricity consumption load of each user and historical meteorological data of the target area; analyzing and classifying the historical data to construct time-derived features; wherein the time-derived features comprise weekday data sets, non-weekday data sets, daily load peak value data sets and daily load valley value data sets; the weekday data sets comprise historical grid supply load, historical total electricity consumption load, historical distributed energy output and historical electricity consumption load of each user of each weekday in chronological order; the non-weekday data sets comprise historical grid supply load, historical total electricity consumption load, historical distributed energy output and historical electricity consumption load of each non-weekday in chronological order; the daily load peak value data sets comprise daily historical load peak values in chronological order; and the daily load valley value data sets comprise daily historical load valley values in chronological order; selecting meteorological parameters with high correlation with the historical grid supply load according to the historical grid supply load and the historical meteorological data, and then constructing meteorological-derived features according to the historical meteorological data and the selected meteorological parameters; extracting daily historical peak-time electricity consumption and daily historical valley-time electricity consumption of each user from the historical electricity consumption load of each user, and constructing user-derived features according to the extracted daily historical peak-time electricity consumption and daily historical valley-time electricity consumption of each user; constructing a load and distributed energy output prediction model, and training and verifying the load and distributed energy output prediction model by using the historical data and the time-derived features, the meteorological-derived features and the user-derived features to obtain a target prediction model; in the prediction stage, obtaining total electricity consumption load prediction values and distributed energy output prediction values by using the target prediction model, and calculating regional grid supply load prediction values according to the total electricity consumption load prediction values and the distributed energy output prediction values.

2. The method of claim 1, wherein, When constructing the time-derived features, binary indexes are added to the weekday data sets, the non-weekday data sets, the daily load peak value data sets and the daily load valley value data sets.

3. The method of claim 1, wherein, The meteorological parameters with high correlation with the historical grid supply load are selected by using the maximum mutual information coefficient, which specifically comprises: calculating the mutual information coefficient of each two-dimensional sequence under different rows and different columns according to the historical grid supply load and the historical meteorological data, and the specific calculation formula is: ； wherein, denotes the mutual information coefficient of the n th two-dimensional sequence D n , the two-dimensional sequence D n is composed of all specific values of the n th meteorological parameter and the specific values Y of the historical network load corresponding to each specific value X of the n th meteorological parameter in the historical meteorological data, the two-dimensional sequence D n is distributed in a two-dimensional space and the two-dimensional space is divided into rows and b columns; denotes the probability of the two-dimensional sequence D n falling in the i th row and the j th column; denotes the probability of the two-dimensional sequence D n falling in the i th row; denotes the probability of the two-dimensional sequence D n falling in the j th column; determining the maximum mutual information coefficient of each two-dimensional sequence and the corresponding row number and column number according to the mutual information coefficient of each two-dimensional sequence under different rows and different columns; normalizing the maximum mutual information coefficient of each two-dimensional sequence according to the corresponding row number and column number of the maximum mutual information coefficient, and the specific formula is: ； wherein, represents the normalized value of the maximum mutual information coefficient of the n th two-dimensional sequence D n ; represents the maximum mutual information coefficient of the n th two-dimensional sequence D n ; represents the row number corresponding to the maximum mutual information coefficient of the n th two-dimensional sequence D n ; represents the column number corresponding to the maximum mutual information coefficient of the n th two-dimensional sequence D n ; N represents the number of data groups in the n th two-dimensional sequence D n , each group of data comprising a specific value of the n th meteorological parameter and a specific value of the corresponding historical grid supply load; represents a hyperparameter; selecting the meteorological parameters with high correlation with the historical grid supply load according to the normalized value of the maximum mutual information coefficient of each two-dimensional sequence.

4. The method of claim 3, wherein, The hyperparameters a value of 0.

6.

5. The method of claim 1, wherein, The meteorological parameters involved in the historical meteorological data include weather type, humidity, temperature, air pressure, precipitation, wind speed and wind direction, and the selected meteorological parameters with high correlation with the historical grid supply load include weather type, temperature and precipitation.

6. The method of claim 1, wherein, The load and distributed energy output prediction model comprises a load prediction model and a distributed energy output model. Alternatively, the load and distributed energy output prediction model is a multi-task model, the multi-task model adopts a customized gating network architecture, and a task tower in the customized gating network architecture is replaced by a fully connected layer.

7. The method of claim 1-6, wherein, The prediction method further comprises correcting the calculated grid load prediction value, and the specific correction process is as follows: An error prediction model is constructed in advance, and the grid load prediction value is subjected to error prediction by using the error prediction model to obtain an error prediction value. The error prediction value and the grid load prediction value are added to obtain a final grid load prediction value.

8. The method of claim 7, wherein, The construction process of the error prediction model is as follows: Step A1: inputting the historical data and time-derived features, weather-derived features, and user-derived features into the target prediction model to obtain a total power consumption load prediction value and a distributed energy output prediction value; Step A2: calculating a grid load initial prediction value according to the total power consumption load prediction value and the distributed energy output prediction value; Step A3: obtaining an error initial prediction value according to the grid load initial prediction value and the historical grid load; Step A4: adding the error initial prediction value and the grid load initial prediction value to obtain a new grid load prediction value; Step A5: obtaining a new error prediction value according to the new grid load prediction value obtained in step A4 and the historical grid load; Step A6: comparing the error initial prediction value obtained in step A3 and the new error prediction value obtained in step A5, if the error initial prediction value is greater than the new error prediction value, repeating steps A1 to A6, if the error initial prediction value is less than or equal to the new error prediction value, stopping training to obtain a trained error prediction model.

9. An electronic device comprising a memory, a processor, and a computer program / instructions stored on the memory, wherein, The processor executes the computer program / instruction to realize the regional grid load prediction method according to any one of claims 1-8.

10. A computer readable storage medium having stored thereon computer programs / instructions, characterized in that, The computer program / instruction is executed by the processor to realize the regional grid load prediction method according to any one of claims 1-8.

11. A computer program product comprising computer programs / instructions, characterized in that, The computer program / instruction is executed by the processor to realize the regional grid load prediction method according to any one of claims 1-8.

Images