Method and system for predicting power load interval in distribution area based on online learning

Abstract

The present invention discloses a method and system for predicting the interval of power load in a substation based on online learning. The method for predicting the interval of power load in a substation based on online learning includes: step S1, in an offline environment, obtaining the recent historical data related to the prediction of the interval of power load in the substation as the input of the model. Step S2, using a double-layer LSTM of a time series neural network, and combining with a load interval prediction loss function, the recent historical data of the input model is trained to obtain a load interval prediction model. Step S3, loading the load interval prediction model. Step S4, at the end of a certain day x, the smart meter collects the power load data of the day. By this, the method for predicting the interval of power load in a substation based on online learning of the present invention can make the actual value of the power load fall within the interval range, and can also make the prediction interval interval as small as possible, meet the real needs of the load interval prediction in the substation, and improve the accuracy of the model prediction.

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and in particular to a method and system for predicting power load intervals in a substation based on online learning. Background Art

[0002] At present, the objects of power load forecasting are either to forecast the power load of a single power user, such as a family or factory, or to forecast the power load of a city or region. There is a lack of power load forecasting for the substation, which is a very important node in the actual power supply environment. The substation is a concept in marketing, and the distribution transformer and its power supply range are called substations.

[0003] Most of the current power load forecasting methods predict power load values, which is a regression problem. However, this prediction method is insufficient in actual needs, because in regression problems, the prediction results can only be as close to the actual value as possible, but cannot truly predict the actual value. Therefore, it is necessary to predict a power load interval and make the actual power load value fall within the interval as much as possible, so as to meet actual business needs.

[0004] However, the existing technology only uses the interval prediction of power load, and does not take into account another important index in interval prediction. That is, it is necessary not only to make the actual value of power load fall within the interval as much as possible, but also to make the prediction interval interval as small as possible, so as to obtain the best interval prediction result.

[0005] The existing technology includes power load data, holiday data, and climate data such as temperature, humidity, and rainfall in the model input feature variables. However, in the actual deployment of the substation environment, there is no way to obtain climate data online, and it is necessary to consider another way to obtain data related to the climate.

[0006] The current implementation method of power load forecasting is to first train and optimize the model offline, and then deploy it to the working environment for online forecasting when the model achieves a relatively good effect. This method has the following disadvantages:

[0007] 1. The cycle of updating a model is long;

[0008] 2. Updating a model consumes manpower and time;

[0009] 3. The power load in the substation usually changes frequently. If the recent power load data is not trained into the model in time, it will easily lead to deviations in the prediction results.

[0010] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgment or any form of suggestion that the information constitutes the prior art already known to a person skilled in the art. Summary of the invention

[0011] The purpose of the present invention is to provide a method and system for predicting the interval of power load in a substation based on online learning, which can make the actual value of the power load fall within the interval range, and can also make the prediction interval interval as small as possible, meet the actual needs of substation load interval prediction, and improve the accuracy of model prediction.

[0012] To achieve the above-mentioned purpose, in the first aspect, the present invention provides a method for predicting the power load interval of a substation based on online learning, the method comprising: step S1, in an offline environment, obtaining the recent historical data related to the prediction of the power load interval of the substation as the input of the model. Step S2, using a double-layer LSTM of a time series neural network, and combining the load interval prediction loss function, the recent historical data of the input model is trained to obtain a load interval prediction model. Step S3, in the substation environment, the load interval prediction model is loaded. Step S4, when a certain day x ends, the smart meter collects the power load data of the day, and pre-processes the data of the most recent m days including the day, where x≥1, and m is the number of days corresponding to the training set data when the load interval prediction model is online learned. Step S5, in the prior learning process, the data obtained after pre-processing in step S4 is used as the training set, and the load interval prediction model loaded in step S3 is online learned, and the training optimization is performed to obtain the latest load interval prediction model. Step S6, the data of the most recent n days including the day is pre-processed, where n is the length of the time series, and n≤m. Step S7, load the latest load interval prediction model, and input the data preprocessed in step S6 to obtain the model's load interval prediction result after the xth day.

[0013] In one embodiment of the present invention, the method for predicting the power load interval of the substation based on online learning also includes: step S8, when the x+1th day ends, repeating steps S3 to S6 to obtain the model's load interval prediction result after the x+1th day.

[0014] In one embodiment of the present invention, the recent historical data is in days, and the recent historical data includes district power load data, holiday data, lunar calendar data and 24 solar terms data, and the district power load data, holiday data, lunar calendar data and 24 solar terms data are pre-processed and combined as input to the model.

[0015] In one embodiment of the present invention, step S2 includes: for the M-day historical power load data of the substation, a vector representation of each day is obtained through step S1. A time series of length n is constructed, and the vector representations of n consecutive days are combined to obtain a 24*n matrix, and there are a total of M-n+1 matrices. The obtained M-n+1 time series are used as input and passed into the time series neural network double-layer LSTM for training. In the time series neural network double-layer LSTM training, the load interval prediction loss function is used as the target.

[0016] In the second aspect, the embodiment of the present invention also provides a system for predicting the power load interval of the substation based on online learning, which includes: an acquisition module, a training module, a loading module, a first preprocessing module, a learning module, a second preprocessing module and a prediction module. The acquisition module is used to obtain the recent historical data related to the prediction of the power load interval of the substation in an offline environment as the input of the model. The training module is used to use the double-layer LSTM of the time series neural network and combine the load interval prediction loss function to train the recent historical data of the input model to obtain the load interval prediction model. The loading module is used to load the load interval prediction model in the substation environment. The first preprocessing module is used when a certain day x ends, the smart meter collects the power load data of the day, and preprocesses the data of the recent m days including the day, where x≥1, and m is the number of days corresponding to the training set data when the load interval prediction model is online learning. The learning module is used to use the data obtained after processing by the first preprocessing module as the training set in the previous learning process, and to perform online learning on the load interval prediction model loaded by the loading module, and to train and optimize to obtain the latest load interval prediction model. The second preprocessing module is used to preprocess the data of the latest n days including the current day, where n is the length of the time series and n≤m. The prediction module is used to load the latest load interval prediction model and input the data processed by the second preprocessing module to obtain the model's load interval prediction results after the xth day.

[0017] In one embodiment of the present invention, the recent historical data is in days, and the recent historical data includes district power load data, holiday data, lunar calendar data and 24 solar terms data, and the district power load data, holiday data, lunar calendar data and 24 solar terms data are pre-processed and combined as input to the model.

[0018] In one embodiment of the present invention, a time series neural network double-layer LSTM is used, and combined with a load interval prediction loss function, the recent historical data of the input model is trained to obtain a load interval prediction model including: for the M-day historical electricity load data of the substation, a vector representation of each day is obtained through step S1. A time series of length n is constructed, and the vector representations of n consecutive days are combined to obtain a 24*n matrix, and there are a total of M-n+1 matrices. The obtained M-n+1 time series are used as input and passed into the time series neural network double-layer LSTM for training. In the time series neural network double-layer LSTM training, the load interval prediction loss function is used as the target.

[0019] Compared with the prior art, the method and system for predicting the interval of power load in the substation based on online learning according to the present invention use a specially designed loss function when predicting the interval of the substation load, so as to make the actual value of the power load fall within the interval as much as possible, and also make the prediction interval interval as small as possible, so as to meet the real needs of the prediction of the interval of the substation load. At the same time, in view of the situation that there is no way to obtain climate data in the substation environment, this method creatively uses lunar calendar data and 24 solar terms data to replace climate data as the input of the model to improve the accuracy of the model prediction. In addition, because the power load law of the substation itself changes frequently, this method adopts the online learning method to be applied in the substation environment, and timely learns the changing law of the power load of the substation, so as to ensure that the accuracy of the prediction of the power load interval of the substation is not reduced after the model is deployed for a long time. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart of a method for predicting the interval of power load in a substation based on online learning according to an embodiment of the present invention;

[0021] Figure 2 It is a logical schematic diagram of a method for predicting power load intervals in an area based on online learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings, but it should be understood that the protection scope of the present invention is not limited by the specific implementation modes.

[0023] Unless explicitly stated otherwise, throughout the specification and claims, the term “comprise” or variations such as “include” or “comprising”, etc., will be understood to include the stated elements or components but not to exclude other elements or components.

[0024] Figure 1 It is a flowchart of a method for predicting power load intervals in an area based on online learning according to an embodiment of the present invention. Figure 2It is a logical schematic diagram of a method for predicting power load intervals in an area based on online learning according to an embodiment of the present invention.

[0025] like Figure 1 to Figure 2 As shown, in the first aspect, a method for predicting the power load interval of a substation based on online learning according to a preferred embodiment of the present invention is provided. The method comprises: step S1, in an offline environment, obtaining the recent historical data related to the prediction of the power load interval of the substation as the input of the model. Step S2, using a double-layer LSTM of a time series neural network and combining the load interval prediction loss function, the recent historical data of the input model is trained to obtain a load interval prediction model. Step S3, in the substation environment, the load interval prediction model is loaded. Step S4, when a certain day x ends, the smart meter collects the power load data of the day, and pre-processes the data of the most recent m days including the day, where x≥1, and m is the number of days corresponding to the training set data when the load interval prediction model is online learned. Step S5, in the prior learning process, the data obtained after pre-processing in step S4 is used as the training set, and the load interval prediction model loaded in step S3 is online learned, and the training optimization is performed to obtain the latest load interval prediction model. Step S6, pre-processing the data of the most recent n days including the day, where n is the length of the time series, and n≤m. Step S7, load the latest load interval prediction model, and input the data preprocessed in step S6 to obtain the model's load interval prediction result after the xth day.

[0026] In one embodiment of the present invention, the method for predicting the power load interval of the substation based on online learning also includes: step S8, when the x+1th day ends, repeating steps S3 to S6 to obtain the model's load interval prediction result after the x+1th day.

[0027] In one embodiment of the present invention, the recent historical data is in days, and the recent historical data includes district power load data, holiday data, lunar calendar data and 24 solar terms data, and the district power load data, holiday data, lunar calendar data and 24 solar terms data are pre-processed and combined as input to the model.

[0028] In one embodiment of the present invention, step S2 includes: for the M-day historical power load data of the substation, a vector representation of each day is obtained through step S1. A time series of length n is constructed, and the vector representations of n consecutive days are combined to obtain a 24*n matrix, and there are a total of M-n+1 matrices. The obtained M-n+1 time series are used as input and passed into the time series neural network double-layer LSTM for training. In the time series neural network double-layer LSTM training, the load interval prediction loss function is used as the target.

[0029] In the second aspect, the embodiment of the present invention also provides a system for predicting the power load interval of the substation based on online learning. The system includes: an acquisition module, a training module, a loading module, a first preprocessing module, a learning module, a second preprocessing module and a prediction module. The acquisition module is used to obtain the recent historical data related to the prediction of the power load interval of the substation in an offline environment as the input of the model. The training module is used to use the double-layer LSTM of the time series neural network and combine the load interval prediction loss function to train the recent historical data of the input model to obtain the load interval prediction model. The loading module is used to load the load interval prediction model in the substation environment. The first preprocessing module is used when a certain day x ends, the smart meter collects the power load data of the day, and preprocesses the data of the recent m days including the day, where x≥1, and m is the number of days corresponding to the training set data when the load interval prediction model is online learning. The learning module is used to use the data obtained after processing by the first preprocessing module as the training set in the previous learning process, and to perform online learning on the load interval prediction model loaded by the loading module, and the training optimization obtains the latest load interval prediction model. The second preprocessing module is used to preprocess the data of the latest n days including the current day, where n is the length of the time series and n≤m. The prediction module is used to load the latest load interval prediction model and input the data processed by the second preprocessing module to obtain the model's load interval prediction results after the xth day.

[0030] In one embodiment of the present invention, the recent historical data is in days, and the recent historical data includes district power load data, holiday data, lunar calendar data and 24 solar terms data, and the district power load data, holiday data, lunar calendar data and 24 solar terms data are pre-processed and combined as input to the model.

[0031] In one embodiment of the present invention, a time series neural network double-layer LSTM is used, and combined with a load interval prediction loss function, the recent historical data of the input model is trained to obtain a load interval prediction model including: for the M-day historical electricity load data of the substation, a vector representation of each day is obtained through step S1. A time series of length n is constructed, and the vector representations of n consecutive days are combined to obtain a 24*n matrix, and there are a total of M-n+1 matrices. The obtained M-n+1 time series are used as input and passed into the time series neural network double-layer LSTM for training. In the time series neural network double-layer LSTM training, the load interval prediction loss function is used as the target.

[0032] In practical applications, the specific implementation steps of the online learning-based power load interval prediction method of the present invention are as follows:

[0033] Step 1: Obtain the recent historical data related to the power load interval forecast of a certain area. The data is in days, including the power load data of the area, holiday data, lunar calendar data and 24 solar terms data. After preprocessing these three types of data, they are combined as the input of the model; (Lunar calendar data and 24 solar terms data are related to climate, which ensures that climate data can be roughly used in the area environment)

[0034] For one day's data, the preprocessing process includes three parts:

[0035] (1) The power load data of the substation area, that is, the power load value of the day is standardized, and the mean and variance of all historical power load data are used for standardization.

[0036] (2) Holiday data: a day is divided into four categories: weekend, working day, statutory holiday and compensatory holiday. They are represented by one-hot. For example, working day is represented as [0,1,0,0] and compensatory holiday is represented as [0,0,0,1].

[0037] (3) 24 solar terms data. There are 24 solar terms in a year, corresponding to a vector of length 24. The first vector element represents the first solar term, the second vector element represents the second solar term, and so on. The initial values ​​of the vector elements are all zero. The time length between the day and the two solar terms is used to calculate and update the vector element values. For example, a certain day is between the second and third solar terms, with a time interval of a day from the second solar term and b days from the third solar term. Therefore, the vector element value corresponding to the second solar term is expressed as b / (a+b), and the vector element value corresponding to the third solar term is expressed as a / (a+b). The values ​​of the remaining elements are all 0, so the 24 solar terms data is expressed as L = [0, b / (a+b), a / (a+b), 0, ..., 0, 0].

[0038] Concatenate the vector representations of the three parts of the above preprocessing to obtain a vector of size 29 (1+4+24) representing the information of this day.

[0039] Step 2, using a double-layer LSTM time series neural network and combining the loss function designed for the prediction interval in the present invention, training the preprocessed data to obtain a load interval prediction model;

[0040] The specific process is as follows:

[0041] (1) For the M-day historical electricity load data of the substation, the vector representation of each day is obtained through step 1.

[0042] (2) Construct a time series of length n, that is, combine the vector representations of n consecutive days to obtain a 24*n matrix. There are a total of M-n+1 such matrices.

[0043] (3) Use the obtained M - n + 1 time series as input and feed them into the model for training. Any time series model can be used. In this invention, a double - layer LSTM is used as an example.

[0044] (4) In the training of the time series model, use the loss function defined in this invention as the objective.

[0045] Among them, the formula of the load interval prediction loss function is as follows:

[0046] y_u, y_l, and y are all vector values, and the lengths of the three are the same. The length is n, which means there are n groups of true values, n groups of predicted upper bounds, and n groups of predicted lower bounds.

[0047] For example: y = [1.0, 2.0, 3.0, 4.0, 5.0], y_u = [1.3, 2.4, 2.9, 4.1, 4.5], y_l = [0.8, 2.1, 2.4, 3.6, 4.0].

[0048] 1. K_u = max(0, sign(y_u - y))

[0049] In y_u, find the values greater than y as 1 and the values less than or equal to y as 0.

[0050] sign(y_u - y) = [1, 1, -1, 1, -1]

[0051] K_u = max(0, sign(y_u - y)) = [1, 1, 0, 1, 0];

[0052] 2. K_l = max(0, sign(y - y_l))

[0053] Similarly

[0054] sign(y - y_l) = [1, -1, 1, 1, 1]

[0055] K_l = max(0, sign(y - y_l)) = [1, 0, 1, 1, 1];

[0056] 3. K_h = multiply(K_u, K_l)

[0057] The dot - product of K_u and K_l means that in a set of data, if y_l(i) < y(i) < y_u(i), then K_h(i) = 1; otherwise K_h(i) = 0.

[0058] K_h = multiply(K_u, K_l) = [1, 0, 0, 1, 0];

[0059] 4. Loss_h = reduce_sum(multiply(y_u - y_l, K_h)) / reduce_sum(K_h)

[0060] Loss_h represents the interval target between y_l and y_u. The smaller the interval, the smaller Loss_h. The meanings of the numerator and denominator of the formula are as follows:

[0061] (1) reduce_sum(K_h) = reduce_sum([1, 0, 0, 1, 0]) = 3

[0062] To reduce Loss_h, the denominator needs to be increased. Therefore, the goal is to ensure that y_l(i) < y(i) < y_u(i) for each set of data, that is, K_h = [1, 1, 1, 1, 1].

[0063] (2) reduce_sum(multiply(y_u - y_l, K_h)) = reduce_sum(multiply([0.5, 0.3, 0.5, 0.5, 0.5], [1, 0, 0, 1, 0])) = reduce_sum([0.5, 0, 0, 0.5, 0]) = 1

[0064] To reduce Loss_h, the numerator needs to be decreased. Therefore, the goal is to make y_u(i) - y_l(i) as small as possible under the condition of y_l(i) < y(i) < y_u(i). Only when y_u - y_l is smaller, the numerator will be smaller.

[0065] 5. Loss_u and Loss_l are quantile loss functions;

[0066] 6. The final loss function combines the three, that is

[0067] Loss = λ_u * Loss_u + λ_l * Loss_l + λ_h * Loss_h.

[0068] Among them, y is the true value, y_u is the predicted upper bound, y_l is the predicted lower bound, r_u is the upper bound quantile coefficient, r_l is the predicted lower bound quantile coefficient, and r_u > 0.5 > r_l, λ_u is the predicted upper bound loss function coefficient, λ_l is the predicted lower bound loss function coefficient, and λ_h is the interval loss function coefficient.

[0069] Enter the substation area environment

[0070] Step 3: Load the load interval prediction model obtained in Step 2 in the substation area environment.

[0071] Step 4: At the end of the xth day on a certain day x (x>=1), the smart meter collects the power load data of the day and preprocesses the data of the most recent m days including the day, where m is the number of days corresponding to the training set data when the model is learning online.

[0072] Step 5, the online learning process, uses the data obtained after preprocessing in step 4 as the training set, and performs online learning on the load interval prediction model loaded in step 3, and obtains the latest load interval prediction model through training optimization.

[0073] Step 6: preprocess the data of the most recent n days including the current day, where n is the length of the time series, that is, at least n days of data need to be input for model prediction, and in general, n<=m.

[0074] Step 7, load the latest load interval prediction model, input the preprocessed data in step 6, and obtain the model's load interval prediction results after the xth day.

[0075] Step 8: At the end of the x+1 day, repeat steps 3 to 6 to obtain the model's prediction results for the load interval after the x+1 day.

[0076] In summary, the method and system for predicting the interval of power load in the substation based on online learning of the present invention use a specially designed loss function when predicting the interval of the substation load, so as to make the actual value of the power load fall within the interval as much as possible, and also make the prediction interval interval as small as possible, so as to meet the real needs of the prediction of the interval of the substation load. At the same time, in view of the situation that there is no way to obtain climate data in the substation environment, this method creatively uses lunar calendar data and 24 solar terms data to replace climate data as the input of the model, thereby improving the accuracy of the model prediction. In addition, because the law of power load in the substation itself changes frequently, this method adopts the method of online learning to be applied in the substation environment, and timely learns the changing law of power load in the substation, so as to ensure that the accuracy of the prediction of power load interval in the substation does not decrease after the model is deployed for a long time.

[0077] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0078] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0079] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0080] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0081] The foregoing description of specific exemplary embodiments of the present invention is for the purpose of illustration and demonstration. These descriptions are not intended to limit the present invention to the precise form disclosed, and it is clear that many changes and variations can be made based on the above teachings. The purpose of selecting and describing the exemplary embodiments is to explain the specific principles of the present invention and its practical application, so that those skilled in the art can realize and utilize various different exemplary embodiments of the present invention and various different selections and changes. The scope of the present invention is intended to be limited by the claims and their equivalents.

Claims

1. A method for predicting power load intervals in a substation based on online learning, characterized in that: include: Step S1, in an offline environment, obtaining recent historical data related to the power load interval prediction of the substation as input of the model; Step S2, using a time series neural network double-layer LSTM and combining it with a load interval prediction loss function, training the recent historical data of the input model to obtain a load interval prediction model; Step S3, loading a load interval prediction model in a substation environment; Step S4, when a certain day x ends, the smart meter collects the power load data of that day, and pre-processes the data of the most recent m days including that day, where x≥1, and m is the number of days corresponding to the training set data when the load interval prediction model is online learning; Step S5, in the prior learning process, using the data obtained after preprocessing in step S4 as a training set, and performing online learning on the load interval prediction model loaded in step S3, training and optimizing to obtain the latest load interval prediction model; Step S6, preprocessing the data of the latest n days including the current day, where n is the length of the time series and n≤m; Step S7, loading the latest load interval prediction model, and inputting the data preprocessed in step S6, to obtain the model's load interval prediction result after the xth day; The recent historical data is in days, and includes the power load data of the area, holiday data, lunar calendar data and 24 solar terms data, and the power load data of the area, the holiday data, the lunar calendar data and the 24 solar terms data are pre-processed and combined as the input of the model; Wherein, the step S2 comprises: For the M-day historical power load data of the substation, a vector representation of each day is obtained through step S1; Construct a time series of length n, combine the vector representations of n consecutive days to obtain a 24*n matrix, and there are a total of M-n+1 matrices; The obtained M-n+1 time series are used as input and passed into the time series neural network double-layer LSTM for training; In the time series neural network two-layer LSTM training, the load interval prediction loss function is used as the target.

2. The method for predicting the power load interval of a substation based on online learning according to claim 1, characterized in that: Also includes: Step S8, when the x+1th day ends, repeat steps S3 to S6 to obtain the load interval prediction result of the model after the x+1th day.

3. A power load interval prediction system for a substation based on online learning, characterized in that: include: An acquisition module is used to obtain the recent historical data related to the power load interval prediction of the substation in an offline environment as the input of the model; A training module, for training the recent historical data of the input model using a time series neural network double-layer LSTM and combining it with a load interval prediction loss function to obtain a load interval prediction model; A loading module is used to load the load interval prediction model in the substation environment; The first preprocessing module is used for collecting the power load data of the day when the smart meter ends on a certain day x, and preprocessing the data of the latest m days including the day, where x≥1, and m is the number of days corresponding to the training set data when the load interval prediction model is online learning; A learning module is used to use the data obtained after being processed by the first preprocessing module as a training set in a prior learning process, and to perform online learning on the load interval prediction model loaded by the loading module, and to obtain the latest load interval prediction model through training and optimization; The second preprocessing module is used to preprocess the data of the latest n days including the current day, where n is the length of the time series and n≤m; as well as The prediction module is used to load the latest load interval prediction model and input the data processed by the second preprocessing module to obtain the model's load interval prediction result after the xth day; The recent historical data is in days, and includes the power load data of the area, holiday data, lunar calendar data and 24 solar terms data, and the power load data of the area, the holiday data, the lunar calendar data and the 24 solar terms data are pre-processed and combined as the input of the model; Among them, the time series neural network double-layer LSTM is used, and the load interval prediction loss function is combined to train the recent historical data of the input model, and the load interval prediction model is obtained, including: For the M-day historical power load data of the substation, a vector representation of each day is obtained through step S1; Construct a time series of length n, combine the vector representations of n consecutive days to obtain a 24*n matrix, and there are a total of M-n+1 matrices; The obtained M-n+1 time series are used as input and passed into the time series neural network double-layer LSTM for training; In the time series neural network two-layer LSTM training, the load interval prediction loss function is used as the target.

Images