A demand response potential prediction method and device based on intelligent model migration

Abstract

The present invention discloses a method and device for predicting demand response potential based on intelligent model migration, wherein the method comprises: obtaining important feature data of a time period to be predicted, and inputting the obtained important feature data into a demand response potential prediction model trained by transfer learning and online learning methods, obtaining a demand response potential value for the time period to be predicted, and performing regulation according to the prediction results. The demand response potential prediction model proposed by the method of the present invention combines the advantages of transfer learning and online learning, and compared with the existing technology, it can dynamically track changes in the response behavior of new customers. It can be seen that the demand response potential prediction accuracy of newly signed customers is improved, the prediction results are more accurate, and it is conducive to the formulation of regulatory strategies.

Description

Technical Field

[0001] The present invention relates to the technical field of demand response potential prediction, and in particular to a demand response potential prediction method and device based on intelligent model migration. Background Art

[0002] Predicting the demand response potential of load aggregators in power systems is crucial, especially given the high penetration of renewable energy, whose randomness, intermittency, and volatility pose challenges to stable grid operation. Demand response guides load-side adjustments through pricing or incentive mechanisms, enhancing the flexibility of the power system and its ability to accommodate renewable energy. Residential loads, a crucial resource, are difficult to directly participate in the market due to their fragmented and uncertain nature. Load aggregators, acting as intermediaries, integrate these resources and enable them to participate in market transactions. Accurately predicting their demand response potential is crucial.

[0003] Existing forecasting methods fall into two categories: model-driven and data-driven. Data-driven approaches rely on historical data, such as price, temperature, and preferences, using models like recurrent neural networks to learn the relationship between features and potential. This allows for accurate forecasts for customers with sufficient data. However, directly applying existing models to newly registered customers with limited data can easily lead to significant errors. Existing models often rely on offline learning, assuming complete data. This is ineffective for forecasting new customers, as data accumulates over time and customer behavior changes dynamically with incentives and the environment. Offline learning struggles to capture these changes, resulting in reduced forecast accuracy. Summary of the Invention

[0004] The present invention provides a method and device for predicting demand response potential based on intelligent model migration to solve the technical problem that the existing technology cannot accurately predict the demand response potential of new customers.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides a demand response potential prediction method based on intelligent model migration, comprising:

[0006] Obtaining important feature data for the forecast period; wherein the important features are selected from basic features through feature selection; the basic features include: demand response incentive price, baseline load, temperature, humidity, rainfall, and comfort index;

[0007] Inputting the important feature data into a trained demand response potential prediction model so that the demand response potential prediction model predicts the time period to be predicted to obtain a predicted demand response potential value;

[0008] According to the demand response potential value, the power system is regulated so that the amount of electricity supplied by the power system meets the demand response potential value;

[0009] The training process of the demand response potential prediction model includes:

[0010] Constructing a first initial model, a second initial model, and a third initial model having the same structure;

[0011] Obtaining a number of first sample important feature data and corresponding first demand response values; obtaining a number of second sample important feature data and corresponding second demand response values; wherein the first sample important feature data is the historical important feature data of the original contracted user; and the second sample important feature data is the historical important feature data of the newly added contracted user;

[0012] Training the first initial model according to each of the first sample important feature data and the corresponding first demand response value to obtain a first model;

[0013] Performing online training on the second initial model according to each of the second sample important feature data and the corresponding second demand response value to obtain a second model;

[0014] Migrating the parameters of the first model to the third initial model to obtain a third model;

[0015] The second model and the third model are adaptively integrated to obtain a trained demand response potential prediction model.

[0016] As a preferred solution, the feature selection process includes:

[0017] Obtaining some basic characteristic data and corresponding demand response values ​​as the original data set;

[0018] Repeat the following steps until the number of remaining basic features equals the preset number, and then use the remaining basic features as important features:

[0019] Performing multiple sampling on the original data set by resampling to obtain multiple training sets;

[0020] For each of the training sets, construct a decision tree model;

[0021] For each currently remaining basic feature, performing a basic feature importance calculation operation to obtain an importance index value of the basic feature;

[0022] From the currently remaining basic features, remove the basic feature with the smallest importance index value;

[0023] The basic feature importance calculation operation includes:

[0024] For each decision tree model, calculate the first out-of-bag data error corresponding to the decision tree model, randomly add noise interference to the current basic features in the out-of-bag data corresponding to the decision tree model, and calculate the second out-of-bag data error corresponding to the decision tree model;

[0025] According to the first out-of-bag data error and the second out-of-bag data error of all decision tree models, the importance index value of the current basic feature is calculated.

[0026] As a preferred solution, the calculation formula of the importance index value is:

[0027]

[0028] Where, P x Represents the importance index value of the basic feature x; Represents the first out-of-bag data error corresponding to the i-th decision tree model; represents the second out-of-bag data error corresponding to the i-th decision tree model; N is the total number of decision trees.

[0029] As a preferred solution, the adaptive integration operation of the second model and the third model includes:

[0030] Inputting each of the important feature data of the second samples into the second model to obtain a second model prediction value;

[0031] Inputting each of the important feature data of the second samples into the third model to obtain a third model prediction value;

[0032] Calculating a second model penalty factor and a third model penalty factor according to the second demand response value, the second model prediction value, and the third model prediction value corresponding to each of the second sample important feature data;

[0033] Calculating a second model weight value and a third model weight value according to the second model penalty factor and the third model penalty factor;

[0034] The second model and the third model are integrated according to the second model weight value and the third model weight value.

[0035] As a preferred solution, the calculation formula of the second model penalty factor is:

[0036]

[0037] Where, ε a represents the second model penalty factor; represents the second model prediction value corresponding to the important feature data of the i-th second sample; represents the predicted value of the third model corresponding to the important feature data of the i-th second sample; y i represents the second demand response value corresponding to the important characteristic data of the i-th second sample; M is the number of newly added contracted users;

[0038] The calculation formula of the third model penalty factor is:

[0039]

[0040] Where, ε b Represents the third model penalty factor

[0041] The solution formula for the second model weight value and the third model weight value is:

[0042]

[0043] Where, ω a Represents the weight value of the second model; ω a Represents the weight value of the third model.

[0044] As a preferred solution, the first initial model, the second initial model and the third initial model are all improved 1D-CNN networks, each including a convolutional layer, a batch normalization layer, an activation function layer, a SCAM attention mechanism, a maximum pooling layer and a fully connected layer.

[0045] As a preferred solution, the SCAM attention mechanism includes: a channel attention mechanism and a temporal attention mechanism.

[0046] Based on the above embodiment, another embodiment of the present invention provides a demand response potential prediction device based on intelligent model migration, comprising: a data acquisition module, a prediction module and a control module;

[0047] The data acquisition module is used to obtain important feature data of the forecast period; wherein the important features are selected from basic features through feature selection; the basic features include: demand response incentive price, baseline load, temperature, humidity, rainfall and comfort index;

[0048] The prediction module is used to input the important feature data into a trained demand response potential prediction model, so that the demand response potential prediction model predicts the time period to be predicted and obtains a predicted demand response potential value; wherein, the training process of the demand response potential prediction model includes: constructing a first initial model, a second initial model and a third initial model with the same structure; obtaining a number of first sample important feature data and corresponding first demand response values; obtaining a number of second sample important feature data and corresponding second demand response values; wherein, the first sample important feature data is the historical important feature data of the original contracted user; the second sample important feature data is the historical important feature data of the newly added contracted user; according to each of the first sample important feature data and the corresponding first demand response value, the first initial model is trained to obtain a first model; according to each of the second sample important feature data and the corresponding second demand response value, the second initial model is trained online to obtain a second model; the parameters of the first model are transferred to the third initial model to obtain a third model; the second model and the third model are adaptively integrated to obtain a trained demand response potential prediction model;

[0049] The control module is used to control the power system according to the demand response potential value so that the power supplied by the power system meets the demand response potential value.

[0050] Based on the above embodiments, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the demand response potential prediction method described in the above embodiment of the invention.

[0051] Based on the above embodiment, another embodiment of the present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute the demand response potential prediction method described in the above embodiment of the invention.

[0052] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0053] The present invention obtains important feature data for the time period to be predicted and inputs the obtained important feature data into a demand response potential prediction model trained through transfer learning and online learning methods to obtain the demand response potential value for the time period to be predicted, and then regulates and controls according to the prediction results. The demand response potential prediction model proposed by the present invention method combines the advantages of transfer learning and online learning. Compared with existing technologies, it can dynamically track changes in the response behavior of new customers. This shows that the accuracy of demand response potential prediction for newly signed customers is improved, making the prediction results more accurate and facilitating the formulation of regulatory strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a flow chart of a demand response potential prediction method based on intelligent model migration provided by one embodiment of the present invention;

[0055] Figure 2 This is a structural diagram of a demand response potential prediction device based on intelligent model migration provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0056] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions in this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned figure descriptions are intended to cover non-exclusive inclusions.

[0058] In the description of the embodiments of this application, the technical terms "first" and "second" are used only to distinguish different objects and should not be understood to indicate or imply relative importance or implicitly specify the quantity, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, the meaning of "several" is two or more, unless otherwise clearly and specifically defined.

[0059] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0061] Example 1

[0062] Please refer to Figure 1 , which is a flow chart of a demand response potential prediction method based on intelligent model migration provided by one embodiment of the present invention, including:

[0063] S1. Obtain important feature data for the time period to be predicted; wherein the important features are selected from basic features through feature selection; the basic features include: demand response incentive price, baseline load, temperature, humidity, rainfall and comfort index.

[0064] In step S1, the time when the demand response potential value needs to be predicted is first determined as the time period to be predicted, and then important characteristic data of the time period to be predicted is obtained.

[0065] It should be noted that the time period to be predicted is generally the current time period, and the important feature data obtained is for each time node in the time period to be predicted. The time period to be predicted can be a specific time node, in which case only the important feature data of that time node need to be obtained. The important features are obtained from the basic features through feature selection. For example, if the important features obtained through feature selection are outdoor temperature θ and baseline load l, the data obtained is x = [θ1, θ2, ..., θ n ,l1,l2,...,l n ], where n represents the number of time nodes in the period to be predicted.

[0066] It should also be noted that, among the basic characteristics, the demand response incentive price refers to the incentive price used to encourage users to participate in demand response during the demand response period, the baseline load refers to the load that users should consume when they do not participate in demand response, the temperature refers to the local outdoor temperature, the humidity refers to the local outdoor humidity, the rainfall refers to the local rainfall, and the comfort index is an index used to evaluate the impact of meteorological conditions on human comfort.

[0067] In a preferred embodiment, the feature selection process includes:

[0068] Obtaining some basic characteristic data and corresponding demand response values ​​as the original data set;

[0069] Repeat the following steps until the number of remaining basic features equals the preset number, and then use the remaining basic features as important features:

[0070] Performing multiple sampling on the original data set by resampling to obtain multiple training sets;

[0071] For each of the training sets, construct a decision tree model;

[0072] For each currently remaining basic feature, performing a basic feature importance calculation operation to obtain an importance index value of the basic feature;

[0073] From the currently remaining basic features, remove the basic feature with the smallest importance index value;

[0074] The basic feature importance calculation operation includes:

[0075] For each decision tree model, calculate the first out-of-bag data error corresponding to the decision tree model, randomly add noise interference to the current basic features in the out-of-bag data corresponding to the decision tree model, and calculate the second out-of-bag data error corresponding to the decision tree model;

[0076] According to the first out-of-bag data error and the second out-of-bag data error of all decision tree models, the importance index value of the current basic feature is calculated.

[0077] It should be noted that basic characteristic data includes the demand response incentive price, baseline load, temperature, humidity, rainfall, and comfort index at a specific time point. The demand response value corresponding to the basic characteristic data is the actual demand response power generated at that time point in reality.

[0078] In this embodiment, a number of basic feature data and the corresponding demand response values ​​for each basic feature data are obtained, the so-called original dataset. Using a resampling method, m samples are randomly sampled with replacement from the original dataset to form the training set. Each sample represents a basic feature data item and its corresponding demand response value. Data not sampled in the original dataset is called out-of-bag data. Sampling is performed N times to obtain N training sets. Using these N sampled training sets, N decision tree models are trained to generate a random forest consisting of N decision trees.

[0079] In order to select important features from basic features, multiple rounds of elimination of basic features are required until the number of remaining basic features is equal to a preset number, which is generally 3 to 5.

[0080] During each round of basic feature elimination, a basic feature importance calculation operation is performed on each remaining basic feature to calculate the importance index value of that feature. When calculating the importance index value of a specific basic feature A, for each decision tree model, the out-of-bag data error of that decision tree is first calculated as the first out-of-bag data error. After calculating the first out-of-bag data error, noise interference is added to the basic feature A of all out-of-bag data, and then the out-of-bag data error of the decision tree is calculated again as the second out-of-bag data error. Finally, the importance index value of the basic feature is calculated based on the first and second out-of-bag errors. After calculating the importance index values ​​of all remaining basic features, the basic feature with the smallest importance index value is eliminated.

[0081] It should be noted that each decision tree model corresponds to a training set sampled from the original data. The out-of-bag data for this decision tree refers to data that was not sampled from the original data. The out-of-bag data error is calculated by inputting the basic feature data of each out-of-bag data into the decision tree model to obtain the corresponding output of the decision tree model. The error of the out-of-bag data is then calculated based on the output of the decision tree model and the demand response value corresponding to the out-of-bag data. Finally, the average of the errors of all out-of-bag data is the out-of-bag data error of the decision tree.

[0082] In a preferred embodiment, the calculation formula of the importance index value is:

[0083]

[0084] Where, P x Represents the importance index value of the basic feature x; Represents the first out-of-bag data error corresponding to the i-th decision tree model; represents the second out-of-bag data error corresponding to the i-th decision tree model; N is the total number of decision trees.

[0085] S2. Inputting the important feature data into a trained demand response potential prediction model, so that the demand response potential prediction model predicts the time period to be predicted to obtain a predicted demand response potential value;

[0086] The training process of the demand response potential prediction model includes:

[0087] Constructing a first initial model, a second initial model, and a third initial model having the same structure;

[0088] Obtaining a number of first sample important feature data and corresponding first demand response values; obtaining a number of second sample important feature data and corresponding second demand response values; wherein the first sample important feature data is the historical important feature data of the original contracted user; and the second sample important feature data is the historical important feature data of the newly added contracted user;

[0089] Training the first initial model according to each of the first sample important feature data and the corresponding first demand response value to obtain a first model;

[0090] Performing online training on the second initial model according to each of the second sample important feature data and the corresponding second demand response value to obtain a second model;

[0091] Migrating the parameters of the first model to the third initial model to obtain a third model;

[0092] The second model and the third model are adaptively integrated to obtain a trained demand response potential prediction model.

[0093] In step S2, the important data features obtained in step S1 are input into the trained demand response potential prediction model so that the model predicts the demand response potential of the prediction period and obtains the predicted demand response potential value.

[0094] It should be noted that if the forecast period is a certain time period, the output demand response potential value is the demand response potential value of all time nodes within the period. If the forecast period is a certain time node, the output demand response potential value is the demand response potential value of the time node.

[0095] The training process of the demand response potential prediction model is as follows:

[0096] Models with the same structure are constructed, namely: Model A, Model B and Model C; Model A is first trained using the historical important feature data of the original contracted customers and the corresponding demand response values; Model B is then trained online using the historical important feature data of the newly added contracted users and the corresponding demand response values; the parameters of the trained Model A are then migrated and frozen into Model C; finally, the trained Model B and the model C with completed parameter migration are integrated together to obtain a trained demand response potential prediction model.

[0097] In a preferred embodiment, the adaptive integration operation of the second model and the third model includes:

[0098] Inputting each of the important feature data of the second samples into the second model to obtain a second model prediction value;

[0099] Inputting each of the important feature data of the second samples into the third model to obtain a third model prediction value;

[0100] Calculating a second model penalty factor and a third model penalty factor according to the second demand response value, the second model prediction value, and the third model prediction value corresponding to each of the second sample important feature data;

[0101] Calculating a second model weight value and a third model weight value according to the second model penalty factor and the third model penalty factor;

[0102] The second model and the third model are integrated according to the second model weight value and the third model weight value.

[0103] In a preferred embodiment, the calculation formula of the second model penalty factor is:

[0104]

[0105] Where, ε a represents the second model penalty factor; represents the second model prediction value corresponding to the important feature data of the i-th second sample; represents the predicted value of the third model corresponding to the important feature data of the i-th second sample; y i represents the second demand response value corresponding to the important characteristic data of the i-th second sample; M is the number of newly added contracted users;

[0106] The calculation formula of the third model penalty factor is:

[0107]

[0108] Where, ε b Represents the third model penalty factor

[0109] The solution formula for the second model weight value and the third model weight value is:

[0110]

[0111] Where, ω a Represents the weight value of the second model; ω a Represents the weight value of the third model.

[0112] In a preferred embodiment, the first initial model, the second initial model and the third initial model are all improved 1D-CNN networks, each including a convolutional layer, a batch normalization layer, an activation function layer, a SCAM attention mechanism, a maximum pooling layer and a fully connected layer.

[0113] In a preferred embodiment, the SCAM attention mechanism includes: a channel attention mechanism and a temporal attention mechanism.

[0114] It should be noted that the process of the channel attention mechanism includes: the input feature sequence is dimensionally compressed by two parallel maximum pooling layers and average pooling layers; it is sent to the shared fully connected layer to learn the attention weight of each channel; the two output results are added together and then activated by the sigmoid function to obtain the channel attention matrix.

[0115] The process of the temporal attention mechanism includes: splicing the data of different channels of the input feature sequence according to the time dimension, extracting features through a one-dimensional convolution layer, transforming the dimension of the feature sequence to the original dimension, and finally obtaining the temporal attention matrix through a sigmoid activation function.

[0116] S3. According to the demand response potential value, the power system is regulated so that the amount of electricity supplied by the power system meets the demand response potential value.

[0117] It's important to note that the demand response potential assessment results for user clusters can provide reference information for load aggregators selecting bidding strategies and for power companies formulating demand response strategies, reducing decision-making risks. This invention predicts the demand response potential of load aggregators primarily to maximize their profits. Demand response behavior itself assists in power system regulation, and the predicted demand response potential should also assist in power system regulation.

[0118] By predicting demand response potential, power systems can more accurately allocate resources, rationally plan renewable energy generation and grid connection, and promote the absorption and utilization of renewable energy. Furthermore, demand response potential predictions can help power systems balance supply and demand by guiding users to adjust their electricity usage when faced with supply and demand imbalances, thereby improving system stability and reliability.

[0119] It should also be noted that the demand response potential prediction model proposed by the method of the present invention combines the advantages of transfer learning and online learning, can dynamically track changes in the response behavior of new customers, and improve the demand response potential prediction accuracy of newly signed customers.

[0120] Example 2

[0121] Please refer to Figure 2 , is a schematic structural diagram of a demand response potential prediction device based on intelligent model migration provided by an embodiment of the present invention, comprising: a data acquisition module, a prediction module and a control module;

[0122] The data acquisition module is used to obtain important feature data of the forecast period; wherein the important features are selected from basic features through feature selection; the basic features include: demand response incentive price, baseline load, temperature, humidity, rainfall and comfort index;

[0123] The prediction module is used to input the important feature data into a trained demand response potential prediction model, so that the demand response potential prediction model predicts the time period to be predicted and obtains a predicted demand response potential value; wherein, the training process of the demand response potential prediction model includes: constructing a first initial model, a second initial model and a third initial model with the same structure; obtaining a number of first sample important feature data and corresponding first demand response values; obtaining a number of second sample important feature data and corresponding second demand response values; wherein, the first sample important feature data is the historical important feature data of the original contracted user; the second sample important feature data is the historical important feature data of the newly added contracted user; according to each of the first sample important feature data and the corresponding first demand response value, the first initial model is trained to obtain a first model; according to each of the second sample important feature data and the corresponding second demand response value, the second initial model is trained online to obtain a second model; the parameters of the first model are transferred to the third initial model to obtain a third model; the second model and the third model are adaptively integrated to obtain a trained demand response potential prediction model;

[0124] The control module is used to control the power system according to the demand response potential value so that the power supplied by the power system meets the demand response potential value.

[0125] Example 3

[0126] Accordingly, an embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the demand response potential prediction method described in the above-mentioned embodiment of the invention.

[0127] Example 4

[0128] Accordingly, an embodiment of the present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute the demand response potential prediction method described in the above-mentioned embodiment of the invention.

[0129] It should be noted that the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement the present invention without inventive work.

[0130] Those skilled in the art will clearly understand that for the sake of convenience and brevity, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0131] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0132] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the device, connecting various parts of the entire device using various interfaces and lines.

[0133] The memory can be used to store the computer program, and the processor realizes various functions of the device by running or executing the computer program stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0134] The storage medium is a storage medium, and the computer program is stored in the storage medium. When the computer program is executed by the processor, it can implement the steps of the above-mentioned method embodiments. The computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0135] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A demand response potential prediction method based on intelligent model migration, characterized in that: include: Obtaining important feature data for the forecast period; wherein the important features are selected from basic features through feature selection; the basic features include: demand response incentive price, baseline load, temperature, humidity, rainfall, and comfort index; Inputting the important feature data into a trained demand response potential prediction model so that the demand response potential prediction model predicts the time period to be predicted to obtain a predicted demand response potential value; According to the demand response potential value, the power system is regulated so that the amount of electricity supplied by the power system meets the demand response potential value; The training process of the demand response potential prediction model includes: Constructing a first initial model, a second initial model, and a third initial model having the same structure; Obtaining a number of first sample important feature data and corresponding first demand response values; obtaining a number of second sample important feature data and corresponding second demand response values; wherein the first sample important feature data is the historical important feature data of the original contracted user; and the second sample important feature data is the historical important feature data of the newly added contracted user; Training the first initial model according to each of the first sample important feature data and the corresponding first demand response value to obtain a first model; Performing online training on the second initial model according to each of the second sample important feature data and the corresponding second demand response value to obtain a second model; Migrating the parameters of the first model to the third initial model to obtain a third model; Inputting each of the important feature data of the second samples into the second model to obtain a second model prediction value; Inputting each of the important feature data of the second samples into the third model to obtain a third model prediction value; Calculating a second model penalty factor and a third model penalty factor according to the second demand response value, the second model prediction value, and the third model prediction value corresponding to each of the second sample important feature data; Calculating a second model weight value and a third model weight value according to the second model penalty factor and the third model penalty factor; Integrating the second model and the third model according to the second model weight value and the third model weight value to obtain a trained demand response potential prediction model; The calculation formula of the second model penalty factor is: ； Where, represents the second model penalty factor; Indicates the The second model prediction value corresponding to the second sample important feature data; Indicates the The third model prediction value corresponding to the important feature data of the second sample; Indicates the a second demand response value corresponding to the second sample important characteristic data; is the number of newly signed-up users; The calculation formula of the third model penalty factor is: ； Where, Represents the third model penalty factor.

2. The demand response potential prediction method according to claim 1, characterized in that: The feature selection process includes: Obtaining some basic characteristic data and corresponding demand response values ​​as the original data set; Repeat the following steps until the number of remaining basic features equals the preset number, and then use the remaining basic features as important features: Performing multiple sampling on the original data set by resampling to obtain multiple training sets; For each of the training sets, construct a decision tree model; For each currently remaining basic feature, performing a basic feature importance calculation operation to obtain an importance index value of the basic feature; From the currently remaining basic features, remove the basic feature with the smallest importance index value; The basic feature importance calculation operation includes: For each decision tree model, calculate the first out-of-bag data error corresponding to the decision tree model, randomly add noise interference to the current basic features in the out-of-bag data corresponding to the decision tree model, and calculate the second out-of-bag data error corresponding to the decision tree model; According to the first out-of-bag data error and the second out-of-bag data error of all decision tree models, the importance index value of the current basic feature is calculated.

3. The demand response potential prediction method according to claim 2, characterized in that: The calculation formula of the importance index value is: ； Where, Represents basic features Importance index value; Indicates the The first out-of-bag data error corresponding to the decision tree model; Indicates the The second out-of-bag data error corresponding to the decision tree model; is the total number of decision trees.

4. The demand response potential prediction method according to claim 1, wherein: The first initial model, the second initial model and the third initial model are all improved 1D-CNN networks, each including a convolutional layer, a batch normalization layer, an activation function layer, a SCAM attention mechanism, a maximum pooling layer and a fully connected layer.

5. The demand response potential prediction method according to claim 4, characterized in that: The SCAM attention mechanism includes: channel attention mechanism and temporal attention mechanism.

6. A demand response potential prediction device based on intelligent model migration, characterized in that: include: Data acquisition module, prediction module and control module; The data acquisition module is used to obtain important feature data of the forecast period; wherein the important features are selected from basic features through feature selection; the basic features include: demand response incentive price, baseline load, temperature, humidity, rainfall and comfort index; The prediction module is used to input the important feature data into the trained demand response potential prediction model, so that the demand response potential prediction model predicts the time period to be predicted and obtains the predicted demand response potential value; wherein, the training process of the demand response potential prediction model includes: constructing a first initial model, a second initial model and a third initial model with the same structure; obtaining a number of first sample important feature data and corresponding first demand response values; obtaining a number of second sample important feature data and corresponding second demand response values; wherein, the first sample important feature data is the historical important feature data of the original contracted user; the second sample important feature data is the historical important feature data of the newly added contracted user; according to each of the first sample important feature data and the corresponding first demand response value, the first initial model is trained to obtain the first model; according to each of the first sample important feature data and the corresponding first demand response value, the first initial model is trained to obtain the first model; according to each of the second sample important feature data and the corresponding first demand response value, the first initial model is trained to obtain the first model. The second initial model is trained online based on the two sample important feature data and the corresponding second demand response value to obtain a second model; the parameters of the first model are transferred to the third initial model to obtain a third model; each of the second sample important feature data is input into the second model to obtain a second model prediction value; each of the second sample important feature data is input into the third model to obtain a third model prediction value; the second model penalty factor and the third model penalty factor are calculated according to the second demand response value, the second model prediction value and the third model prediction value corresponding to each of the second sample important feature data; the second model weight value and the third model weight value are calculated according to the second model penalty factor and the third model penalty factor; the second model and the third model weight value are integrated according to the second model weight value and the third model weight value to obtain a trained demand response potential prediction model; The control module is configured to control the power system according to the demand response potential value so that the amount of electricity supplied by the power system meets the demand response potential value; The calculation formula of the second model penalty factor is: ； Where, represents the second model penalty factor; Indicates the The second model prediction value corresponding to the second sample important feature data; Indicates the The third model prediction value corresponding to the important feature data of the second sample; Indicates the a second demand response value corresponding to the second sample important characteristic data; is the number of newly signed-up users; The calculation formula of the third model penalty factor is: ； Where, Represents the third model penalty factor.

7. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for predicting demand response potential as claimed in any one of claims 1 to 5 is implemented.

8. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is executed, the device where the storage medium is located is controlled to execute the demand response potential prediction method according to any one of claims 1 to 5.

Images