Mass transformer area weight and overload early warning method and system based on transfer learning

Abstract

The invention relates to a mass transformer area weight and overload early warning method and system based on transfer learning, and the method comprises the following steps: obtaining the real-time load data of each transformer area distribution transformer, and carrying out the preprocessing; the preprocessed real-time load data of the distribution transformers of all the transformer areas are input into a pre-trained load prediction model of all the transformer areas for prediction, predicted load data of the distribution transformers of all the transformer areas are obtained, and the load prediction model is obtained through structure and parameter migration of a network prediction model based on LSTM; and calculating the correlation between the predicted load data of the distribution transformer in each transformer area and the temperature to obtain a correlation coefficient, and combining a heavy overload early warning judgment standard to obtain a mass heavy overload early warning result of the distribution transformer in each transformer area. Compared with the prior art, the method has the advantages of improving the prediction efficiency of massive transformer area distribution transformers and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of power load forecasting, and in particular to a method and system for early warning of heavy overload in a large number of substations based on transfer learning. Background Art

[0002] As one of the important infrastructures of modern society, the power system plays a vital role in various fields such as economy, industry, and residents' lives. With the acceleration of urbanization and the continuous growth of the economy, the power system is facing increasing load pressure and challenges in energy supply. Load forecasting plays a vital role in energy management and power system operation. Especially in urban centers with developed economies, dense populations, and prosperous industries and commerce, the power load is complex, the load characteristics are increasingly prominent, and the load rate of distribution stations fluctuates violently, and the power system bears huge load pressure. Effective load forecasting can help power grid companies and operators to reasonably arrange power generation plans, optimize energy utilization, improve power supply reliability, and help to allocate power resources economically and rationally. Load forecasting refers to the process of predicting power demand in the future based on historical data and related factors. Its main purpose is to provide a reliable basis for the planning and operation of the power system. Through load forecasting, power grid companies can prepare for power generation and transmission in advance to ensure that user needs can be met during peak power consumption periods and avoid power shortages or power waste.

[0003] Load forecasting involves the analysis and modeling of a variety of complex factors, including climate change, economic development level, population growth, industrial structure adjustment, etc. Through the comprehensive analysis of these factors, future electricity demand can be predicted more accurately, thus providing a reliable basis for the planning and operation of the power system. The accuracy of load forecasting is crucial to the stable operation of the power system. Accurate load forecasting can help power grid companies to reasonably arrange power generation plans and avoid economic losses and waste of resources caused by insufficient or excessive power supply. At the same time, by optimizing the allocation of power resources, environmental pollution can be effectively reduced and the sustainable use of energy can be promoted. Load forecasting not only provides a reliable guarantee for the operation of the power system, but also plays an important role in energy management. Through accurate prediction of power demand, energy optimization can be achieved, energy utilization efficiency can be improved, energy waste can be reduced, and energy costs can be reduced. Especially today when the proportion of new energy power generation is gradually increasing, load forecasting is of great significance for achieving efficient use of energy and sustainable development.

[0004] Load forecasting has important practical significance and broad application prospects in modern power systems. With the continuous advancement of technology, load forecasting methods will become more accurate and intelligent, providing a solid guarantee for the stable operation of power systems and the efficient use of energy. In future development, load forecasting technology for power systems will continue to play a key role and safeguard the sustainable and healthy development of the economy and society.

[0005] At present, there are many existing technologies for the research of power load forecasting. Patent application CN103617467A discloses a short-term combined load forecasting method, which adopts a combined model forecasting method, and uses weights to construct a combined model to improve the prediction accuracy of a single model; Patent application CN113537571A discloses a building energy consumption load forecasting method based on a CNN-LSTM hybrid network model, which constructs a CNN-LSTM hybrid network model to predict building power consumption load and verify its accuracy and effectiveness. However, there are still many technical challenges for regional load forecasting research. First, with the development of intelligent power systems, more and more electrical equipment are connected to the power system, resulting in the load size of the power system being affected by meteorological conditions, electricity prices, electricity habits, social time, holidays and other factors. How to pick out high correlation and low redundancy features from a large number of mutually restrictive influencing factors is the first major challenge facing load forecasting. Secondly, there are different high correlation characteristics between different distribution substations, and the specific data between each substation is different, and the power load has obvious seasonality, periodicity and uncertainty. These characteristics bring many problems to the accuracy of power load forecasting that need to be solved. Finally, in a large area, there are a large number of distribution transformers, and the amount of historical load data is also huge. Faced with the influx of massive data, conventional load forecasting models often lead to reduced prediction efficiency due to the excessive amount of data, which greatly weakens the timeliness of load forecasting. Therefore, it is urgent to build an efficient load forecasting model to process massive data and retain the timeliness of load forecasting to the greatest extent. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for heavy overload warning in a large number of substations based on transfer learning, which can improve the efficiency of load prediction and the timeliness of heavy overload warning.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] A massive overload warning method based on transfer learning includes the following steps:

[0009] Obtain the real-time load data of each distribution transformer and perform preprocessing;

[0010] The pre-processed real-time load data of each distribution transformer in each substation are respectively input into the pre-trained load prediction model of each substation for prediction, so as to obtain the predicted load data of each distribution transformer in each substation, wherein the load prediction model is obtained by performing structure and parameter migration on the LSTM-based network prediction model;

[0011] The correlation between the predicted load data and the temperature of the distribution transformers in each substation is calculated to obtain a correlation coefficient, and then combined with the heavy overload warning evaluation criteria to obtain the heavy overload warning results of the distribution transformers in a large number of substations.

[0012] Furthermore, the pre-processing step includes:

[0013] Based on the real-time load data of the distribution transformers in each substation, abnormal values ​​and outliers are identified and mean values ​​are corrected, and null values ​​are filled with polynomial interpolation to form complete real-time load data.

[0014] Furthermore, the expression of the polynomial interpolation method is:

[0015] L n (x) = a 0 +a 1 x+a 2 x 2 +...+a n x n

[0016] Where, L n (x) represents n+1 different sample points (x i ,y i ) There exists a unique polynomial, a 0 、a 1 ,……,a n is the coefficient.

[0017] Furthermore, the training steps of the load forecasting model of each substation include:

[0018] Obtain massive historical load data of distribution transformers in substations and perform preprocessing, including removing abnormal values ​​and outliers in the load data and filling in empty values ​​using polynomial interpolation method;

[0019] Obtaining a meteorological factor data set and a humanistic factor feature data set for a date corresponding to the massive historical load data of the distribution transformer in the substation, performing preliminary processing and merging them into a feature data set, wherein the humanistic factor feature data set includes electricity price factor features, electricity usage habit factor features, social time factor features, holiday factor features, and economic and social development statistical factor features;

[0020] Performing feature analysis on the massive historical load data of distribution transformers in the substations to obtain historical load features, and using the maximum mutual information coefficient method to calculate the maximum mutual information coefficient matrix between the massive historical load data of distribution transformers in the substations and various meteorological factors and human factors;

[0021] According to the maximum mutual information coefficient matrix, considering the correlation and redundancy of features, selecting from the historical load features as an input feature set, and merging the historical load data set with the feature data set to construct a sample data set;

[0022] Inputting the massive distribution transformer sample data set into the LSTM-based network prediction model for training until the training end condition is met, thereby obtaining a trained LSTM-based network prediction model;

[0023] The structure and parameters of the trained LSTM-based network prediction model are migrated to obtain the initial load prediction model of the massive distribution transformers in each substation;

[0024] The initial load forecasting models of the massive distribution transformers in each substation are fine-tuned, and finally each distribution transformer in the substation obtains its own load forecasting model.

[0025] Furthermore, the step of using the maximum mutual information coefficient method to calculate includes:

[0026] Given m sets of observations of a two-dimensional random variable (ξ,η) Then the mutual information of ξ and η is:

[0027]

[0028] Among them, p(x,y) is the joint distribution of ξ and η, p(x) and p(y) are the marginal distributions of ξ and η, and the variables ξ and η represent the massive historical load data of distribution transformers in the distribution area, various meteorological factors or human factors respectively;

[0029] Discretize the relationship between variables ξ and η into a two-dimensional space of a×b, and calculate the maximum mutual information value in the two-dimensional space:

[0030]

[0031] Among them, MI(A,a,b) is the maximum mutual information value, I c (ξ,η) is the mutual information value in two-dimensional space, c is the two-dimensional grid space of a×b;

[0032] The maximum mutual information value is normalized by using a normalization method to obtain the maximum mutual information coefficient matrix, wherein the normalized expression is:

[0033]

[0034] Where B(m) is the upper limit of the number of grid divisions, and MIC(A) is the normalized maximum mutual information value.

[0035] Furthermore, the step of obtaining the correlation coefficient includes:

[0036] The Pearson correlation coefficient method is used to calculate the correlation between the predicted load data of the distribution transformers in each substation and the maximum temperature or the minimum temperature to obtain the correlation coefficient.

[0037] Furthermore, the calculation expression of the Pearson correlation coefficient method is:

[0038]

[0039] In the formula, r, ρ X,Y are Pearson correlation coefficients. X and Y represent two continuous variables to be measured, which are the predicted load data and the maximum temperature or the minimum temperature. cov represents the covariance between the variables. σ X , σ Y represents the standard deviation of X and Y, E represents the expected value, μ X , μ Y Represent the average values ​​of X and Y respectively.

[0040] Furthermore, the step of obtaining the heavy overload warning results of the massive distribution transformers in each substation area includes:

[0041] Determine the relationship between the correlation coefficient and a set threshold value to obtain the correlation strength between the predicted load data and the temperature;

[0042] According to the correlation strength and combined with the heavy overload warning evaluation criteria, the heavy overload warning results of each substation are obtained.

[0043] Furthermore, the relationship between the correlation coefficient and the set threshold is:

[0044] When 1≥|r i When |≥0.6, it is considered that there is a strong correlation between the predicted load data and the temperature, where |r i | represents the absolute value of the correlation coefficient;

[0045] When 0.6≥|r i When |≥0.2, it is considered that there is a moderate correlation between the predicted load data and the temperature;

[0046] When 0.2≥|r i When |≥0, it is considered that there is a weak correlation between the predicted load data and temperature.

[0047] The present invention also provides a massive overload warning system based on transfer learning, comprising:

[0048] Acquisition module: used to obtain the real-time load data of each distribution transformer in each area and perform preprocessing;

[0049] Prediction module: used to input the pre-processed real-time load data of each distribution transformer into the pre-trained load prediction models for prediction, and obtain the predicted load data of each distribution transformer, wherein the load prediction models are obtained by performing structure and parameter migration on the LSTM-based network prediction model;

[0050] Heavy overload warning module: used to calculate the correlation between the predicted load data and the temperature of the distribution transformers in each substation, obtain the correlation coefficient, and then combine it with the heavy overload warning evaluation criteria to obtain the heavy overload warning results of a large number of distribution transformers in each substation.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] (1) Considering the problem that conventional load forecasting models often suffer from reduced prediction efficiency due to the large amount of data flowing into them when processing massive load data, the present invention trains a network prediction model based on LSTM, and then combines the model with the transfer learning method to migrate parameters, so as to obtain a load prediction model for each distribution transformer in each substation, thereby enabling the massive load data of each distribution transformer in each substation to be used for load prediction according to their respective load prediction models. While maintaining the original load prediction accuracy, the load processing capability of the big data can be greatly improved, the prediction efficiency can be improved, and the timeliness of the heavy overload warning can be improved.

[0053] (2) The present invention uses the maximum mutual information coefficient method to study the influence of external factors on the distribution transformer load in the substation area, and can simultaneously take into account the correlation and redundancy between the meteorological factor data and the human factor data. This method helps to accurately select load characteristics, reduce feature redundancy, speed up calculations, and avoid dimensionality problems.

[0054] (3) The present invention uses the Pearson correlation coefficient to calculate the correlation between the load data of all distribution transformers in the substation and the temperature characteristics, and combines it with the heavy overload warning evaluation criteria to improve the accuracy of the heavy overload warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0056] Figure 2 It is a data preprocessing flow chart of the present invention;

[0057] Figure 3 It is the maximum mutual information coefficient heat map of the present invention;

[0058] Figure 4 It is the transfer learning logic diagram of the present invention;

[0059] Figure 5 It is the LSTM neural network diagram of the present invention;

[0060] Figure 6 The predicted and actual load rates of the distribution transformer 1 of the present invention;

[0061] Figure 7 The predicted and actual load rates of the distribution transformer 2 of the present invention;

[0062] Figure 8 The predicted and actual load rates of the distribution transformer 3 of the present invention;

[0063] Fig. 9 It is a scatter diagram of the early warning distribution transformer of the present invention. DETAILED DESCRIPTION

[0064] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0065] This embodiment provides a massive overload warning method based on transfer learning, such as Figure 1 As shown, the method comprises the following steps:

[0066] S1. Obtain massive historical load data of distribution transformers in substations and perform preprocessing.

[0067] according to Figure 2 The specific implementation process of this step is as follows:

[0068] Obtain massive historical load data of distribution transformers in substations and preprocess the data. The preprocessing process includes mean correction for abnormal values ​​and outliers in the load data and filling in empty values ​​using polynomial interpolation method.

[0069] For n+1 different sample points (x i ,y i ), there exists a unique polynomial:

[0070] L n (x) = a 0 +a 1 x+a 2 x 2 +...+a n x n

[0071] Make L n (x j)=y j (j=0,1,2,…,n)

[0072] Therefore, the null value load ratio under any date can be filled according to the unique polynomial.

[0073] S2. Obtain meteorological characteristic data sets and human factor characteristic data sets, perform characteristic analysis on massive distribution and transformation historical load data, and calculate the maximum mutual information coefficient matrix.

[0074] The meteorological characteristic data set and the human factor characteristic data set of the corresponding dates of the massive historical load data of the distribution and transformation areas are obtained and merged into a characteristic data set after preliminary processing. The human factor characteristics include electricity price factor characteristics, electricity usage habit factor characteristics, social time factor characteristics, holiday factor characteristics, and economic and social development statistical factor characteristics.

[0075] The characteristics of the distribution transformer in the substation area are analyzed. The maximum mutual information coefficient method (MIC) is used to calculate the maximum mutual information coefficient matrix between the distribution transformer load and various meteorological and human factors. Considering the correlation and redundancy of the characteristics, appropriate load characteristics are selected. The maximum mutual information coefficient thermal matrix diagram is shown in Figure 3 shown.

[0076] The maximum mutual information coefficient is calculated as follows:

[0077] For a given set of m observations of a two-dimensional random variable (ξ,η) Then the mutual information of ξ and η is:

[0078]

[0079] Among them, p(x,y) is the joint distribution of (ξ,η), p(x) and p(y) are marginal distributions of ξ and η, and the variables ξ and η represent the massive historical load data of distribution transformers and various meteorological factors or human factors, respectively.

[0080] The relationship between the two variables is discretized into the two-dimensional space c of a×b. The maximum mutual information value in this two-dimensional space is:

[0081]

[0082] Among them, I c (ξ,η) is the mutual information value in two-dimensional space.

[0083] Using the normalization method, we can normalize the above formula and get:

[0084]

[0085] Where B(m) is the upper limit of the number of grid divisions, and B(m) is generally taken as m.0.6 .

[0086] S3. Establish a load forecasting model based on transfer learning and long short-term memory neural network, and perform load forecasting to obtain massive load forecast data for distribution transformers in various substations.

[0087] First, a network prediction model based on long short-term memory neural network (LSTM) is constructed. The data set and feature set of massive distribution transformers in the substation area are input into the model for training. When the training of the selected benchmark distribution transformer is completed and good prediction results can be achieved, a trained network prediction model is obtained. The network structure diagram of LSTM is shown in the figure below: Figure 5 shown.

[0088] Secondly, combined with Figure 4 The transfer learning method logic diagram shown in FIG1 migrates the network prediction model structure and parameters to each substation to perform load prediction of the distribution transformer. Therefore, an initial load prediction model based on transfer learning and LSTM is first obtained for each substation, and then fine-tuned to obtain the final load prediction model.

[0089] The real-time load data of massive distribution transformers in each substation are predicted according to the final load prediction model, and the predicted load data of massive distribution transformers in each substation can be obtained.

[0090] The prediction results are expressed in terms of the daily maximum load factor of the distribution transformer. The prediction results of the maximum load factor of a distribution transformer in the next 7 days are as follows: Figure 6-8 As shown, Figure 6 is the maximum load factor prediction result of distribution transformer 1, Figure 7 is the maximum load factor prediction result of distribution transformer 2, Figure 8 The maximum load factor prediction result of distribution transformer 3 is Figure 6-8 It can be seen that the change trend of the actual maximum load rate curve of the distribution transformer in the next 7 days is basically consistent with the predicted maximum load rate curve, and the predicted and actual load rate curves are also similar in value. By calculating the average prediction errors of the three distribution transformers in the next week, they are 0.0958, 0.0277, and 0.0985 respectively.

[0091] S4. Calculate the correlation between the predicted load data and the temperature, and combine it with the heavy overload judgment standard to obtain the heavy overload warning result.

[0092] Calculate the Pearson correlation coefficient between the distribution transformer in the substation area and the highest or lowest temperature, and classify the distribution transformer into strong temperature correlation, medium temperature correlation, and weak temperature correlation based on the correlation, and assign weights to different correlations. The calculation method of the Pearson correlation coefficient is as follows:

[0093]

[0094] In the formula, r, ρ X,Y are Pearson correlation coefficients. X and Y represent two continuous variables to be measured, which are the predicted load data and the maximum temperature or the minimum temperature. cov represents the covariance between the variables. σ X , σ Y represents the standard deviation of X and Y, E represents the expected value, μ X , μ Y Represent the average values ​​of X and Y respectively.

[0095] Define the weight selection of Pearson correlation coefficient:

[0096] α max ≥|r i |≥α min

[0097] where |r i | represents the absolute value of the Pearson correlation coefficient, α max , α min Represent the upper and lower bounds of the threshold respectively.

[0098] When defining the temperature-load correlation, according to the Pearson correlation coefficient, when 1≥|r i |≥0.6, the distribution transformer load in this area has a high correlation with temperature and is considered to be strongly correlated; when 0.6≥|r i |≥0.2, the distribution transformer load in the area is moderately correlated with the temperature and is considered to be moderately correlated; when 0.2≥|r i When |≥0, the correlation between the distribution transformer load and temperature in this area is relatively low and is considered to be weakly correlated.

[0099] Set the heavy overload warning line for distribution transformers. In engineering, distribution transformers with a maximum load rate exceeding 80% are generally considered to be heavily overloaded distribution transformers. In order to leave a margin, this application sets two warning lines, namely a 60% general warning line and an 80% emergency warning line. Different warning results are given based on whether the distribution transformer predicts that the maximum load rate exceeds the warning line. If it exceeds 80%, the warning is emergency, and if it exceeds 60%, the warning is general. Based on the correlation between the load rate change and the temperature, the distribution transformers with strong correlation are marked. The combination of the two gives the final heavy overload warning results for the distribution transformers in a large number of substations, such as Fig. 9 shown.

[0100] Example 2

[0101] This embodiment provides a massive overload warning system based on transfer learning, including:

[0102] Acquisition module: used to obtain the real-time load data of each distribution transformer in each area and perform preprocessing;

[0103] Prediction module: used to input the pre-processed real-time load data of each distribution transformer into the pre-trained load prediction models for prediction, and obtain the predicted load data of each distribution transformer, wherein the load prediction models are obtained by performing structure and parameter migration on the LSTM-based network prediction model;

[0104] Heavy overload warning module: used to calculate the correlation between the predicted load data and the temperature of the distribution transformers in each substation, obtain the correlation coefficient, and then combine it with the heavy overload warning evaluation criteria to obtain the heavy overload warning results of a large number of distribution transformers in each substation.

[0105] The rest is the same as in Example 1.

[0106] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0107] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0108] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks 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 produce 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.

[0109] 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.

[0110] 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. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0111] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0112] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A massive overload warning method based on transfer learning, It is characterized in that The following steps are involved: Obtain the real-time load data of each distribution transformer and perform preprocessing; The pre-processed real-time load data of each distribution transformer in each substation are respectively input into the pre-trained load prediction model of each substation for prediction, so as to obtain the predicted load data of each distribution transformer in each substation, wherein the load prediction model is obtained by performing structure and parameter migration on the LSTM-based network prediction model; The correlation between the predicted load data and the temperature of the distribution transformers in each substation is calculated to obtain a correlation coefficient, and then combined with the heavy overload warning evaluation criteria to obtain the heavy overload warning results of the distribution transformers in a large number of substations.

2. According to the method of heavy overload warning for massive substations based on transfer learning in claim 1, It is characterized in that The pre-processing steps include: Based on the real-time load data of the distribution transformers in each substation, abnormal values ​​and outliers are identified and mean values ​​are corrected, and null values ​​are filled with polynomial interpolation to form complete real-time load data.

3. According to the method of heavy overload warning for massive substations based on transfer learning in claim 2, It is characterized in that The expression of the polynomial interpolation method is: L n (x)=a 0 +a 1 x+a 2 x 2 +...+a n x n Where, L n (x) represents n+1 different sample points (x i ,u i ) There exists a unique polynomial, a 0 、a 1 ,……,a n is the coefficient.

4. According to the method of heavy overload warning for massive substations based on transfer learning in claim 1, It is characterized in that The training steps of the load forecasting model of each substation include: Obtain massive historical load data of distribution transformers in substations and perform preprocessing, including removing abnormal values ​​and outliers in the load data and filling in empty values ​​using polynomial interpolation method; Obtaining a meteorological factor data set and a humanistic factor feature data set for a date corresponding to the massive historical load data of the distribution transformer in the substation, performing preliminary processing and merging them into a feature data set, wherein the humanistic factor feature data set includes electricity price factor features, electricity usage habit factor features, social time factor features, holiday factor features, and economic and social development statistical factor features; Performing feature analysis on the massive historical load data of distribution transformers in the substations to obtain historical load features, and using the maximum mutual information coefficient method to calculate the maximum mutual information coefficient matrix between the massive historical load data of distribution transformers in the substations and various meteorological factors and human factors; According to the maximum mutual information coefficient matrix, considering the correlation and redundancy of features, selecting from the historical load features as an input feature set, and merging the historical load data set with the feature data set to construct a sample data set; Inputting the massive distribution transformer sample data set into the LSTM-based network prediction model for training until the training end condition is met, thereby obtaining a trained LSTM-based network prediction model; The structure and parameters of the trained LSTM-based network prediction model are migrated to obtain the initial load prediction model of the massive distribution transformers in each substation; The initial load forecasting models of the massive distribution transformers in each substation are fine-tuned, and finally each distribution transformer in the substation obtains its own load forecasting model.

5. According to the method of heavy overload early warning for massive substations based on transfer learning in claim 4, It is characterized in that The steps of using the maximum mutual information coefficient method for calculation include: Given m sets of observations of a two-dimensional random variable (ξ,η) Then the mutual information of ξ and η is: Among them, p(x,y) is the joint distribution of ξ and η, p(x) and p(y) are the marginal distributions of ξ and η, and the variables ξ and η represent the massive historical load data of distribution transformers in the distribution area, various meteorological factors or human factors respectively; Discretize the relationship between variables ξ and η into a two-dimensional space of a×b, and calculate the maximum mutual information value in the two-dimensional space: Among them, MI(A,a,b) is the maximum mutual information value, I c (ξ,η) is the mutual information value in two-dimensional space, c represents the two-dimensional grid space of a×b; The maximum mutual information value is normalized by using a normalization method to obtain the maximum mutual information coefficient matrix, wherein the normalized expression is: Where B(m) is the upper limit of the number of grid divisions, and MIC(A) is the normalized maximum mutual information value.

6. According to the method of claim 1, a massive overload early warning method based on transfer learning, It is characterized in that The step of obtaining the correlation coefficient comprises: The Pearson correlation coefficient method is used to calculate the correlation between the predicted load data of the distribution transformers in each substation and the maximum temperature or the minimum temperature to obtain the correlation coefficient.

7. According to claim 6, a method for early warning of heavy overload in a large number of substations based on transfer learning, It is characterized in that The calculation expression of the Pearson correlation coefficient method is: In the formula, r, ρ X,Y are Pearson correlation coefficients. X and Y represent two continuous variables to be measured, which are the predicted load data and the maximum temperature or the minimum temperature. cov represents the covariance between the variables. σ X , σ Y represents the standard deviation of X and Y, E represents the expected value, μ X , μ Y Represent the average values ​​of X and Y respectively.

8. According to the method of claim 1, a massive overload early warning method based on transfer learning, It is characterized in that The step of obtaining the heavy overload warning results of a large number of distribution transformers in each substation area includes: Determine the relationship between the correlation coefficient and a set threshold value to obtain the correlation strength between the predicted load data and the temperature; According to the correlation strength and combined with the heavy overload warning evaluation criteria, the heavy overload warning results of each substation are obtained.

9. According to claim 8, a massive overload warning method based on transfer learning, It is characterized in that The relationship between the correlation coefficient and the set threshold is: When 1≥|r i When |≥0.6, it is considered that there is a strong correlation between the predicted load data and the temperature, where |r i | represents the absolute value of the correlation coefficient; When 0.6≥|r i When |≥0.2, it is considered that there is a moderate correlation between the predicted load data and the temperature; When 0.2≥|r i When |≥0, it is considered that there is a weak correlation between the predicted load data and temperature.

10. A massive overload warning system based on transfer learning, It is characterized in that include: Acquisition module: used to obtain the real-time load data of each distribution transformer in each area and perform preprocessing; Prediction module: used to input the pre-processed real-time load data of each distribution transformer into the pre-trained load prediction models for prediction, and obtain the predicted load data of each distribution transformer, wherein the load prediction models are obtained by performing structure and parameter migration on the LSTM-based network prediction model; Heavy overload warning module: used to calculate the correlation between the predicted load data and the temperature of the distribution transformers in each substation, obtain the correlation coefficient, and then combine it with the heavy overload warning evaluation criteria to obtain the heavy overload warning results of a large number of distribution transformers in each substation.

Images