Power load prediction method, related device, model training method and system

Abstract

The invention belongs to a prediction model training method, and provides a power load prediction method, a related device, a model training method and a system for solving the problem that a prediction result generated by adopting a diffusion model at present is poor in prediction effect due to addition of noise. The time sequence data set for training is expanded, and the generalization ability of the power load prediction model is also improved. And then extracting a long-term dependency relationship of historical power load data in a time dimension, aggregating the extracted feature information data, and performing noise reduction processing on the input sequence subjected to coupling diffusion noise addition to obtain a preliminary prediction result. Through the above operation, the power load prediction model can more effectively process the uncertainty in prediction, especially the power load data set influenced by many factors. And field-specific physical constraint information is introduced while diffusion is carried out, and the prediction result is further cleaned to obtain a final prediction result.

Description

Technical Field

[0001] The present application relates to a prediction model training method, and specifically to a power load prediction method and related devices, a model training method, and a system. Background Art

[0002] The power load forecasting task refers to the prediction and estimation of the power load of the power system in the future to assist power companies in formulating more scientific, economical and reasonable power generation plans and dispatching plans. Power load forecasting is the basis of power system planning and operation, and runs through all power operation stages from traditional to intelligent. In the current open power market environment, power load forecasting is not only a key tool to ensure the economic, safe and stable operation of the power system, but also the basis for many important decision-making. Therefore, mastering the dynamic changes of power load data is crucial for the safe and efficient operation of the power system.

[0003] The diffusion model has powerful nonlinear modeling capabilities. By simulating the diffusion process of data from clarity to noise and the reverse process, it can not only flexibly capture the complex dynamics and long-term dependencies in time series data, but also has the ability to efficiently generate diverse and high-fidelity time series data samples from noise, providing strong support for time series prediction. Although the diffusion model has powerful feature learning and data expansion capabilities, the generated prediction results often violate physical behavior due to the addition of noise, affecting the effect of model prediction. Summary of the invention

[0004] The present application aims to solve the technical problem that the prediction results currently generated by the diffusion model have poor prediction effect due to the addition of noise, and provides a power load prediction method and related devices, model training method, and system.

[0005] In order to achieve the above objectives, this application adopts the following technical solutions:

[0006] In a first aspect, the present application proposes a method for training a power load forecasting model, wherein the power load forecasting model is based on a diffusion model; the method comprises:

[0007] Obtaining an input sequence and a target sequence, wherein the input sequence includes a historical power load time series, and the target sequence is a known prediction sequence corresponding to the input sequence;

[0008] The input sequence is used as the input of the power load forecasting model, and the target sequence is used as the true value of the power load forecasting model to train the power load forecasting model. During training:

[0009] Gaussian noise is added to the input sequence and the target sequence synchronously by coupled diffusion;

[0010] Extract features from the input data in the input sequence to obtain temporal and spatial feature information of the input data;

[0011] Aggregate the temporal and spatial feature information of the input data, and denoise the target sequence with added Gaussian noise to obtain the predicted data;

[0012] The training objective is to minimize the difference between the predicted data and the target sequence with Gaussian noise added, and physical constraints are imposed on the power load forecasting model, and the physical constraints are used to minimize the loss between the predicted data and the target sequence.

[0013] Furthermore, the step of synchronously adding Gaussian noise to the input sequence and the target sequence by coupled diffusion includes:

[0014] Two different variance lists are used to control the Gaussian noise adding process of the input sequence and the target sequence respectively, so that the process of adding Gaussian noise to the input sequence and the target sequence can be kept synchronized.

[0015] Furthermore, the extracting features from the input data in the input sequence includes:

[0016] Embed the input data to get the word vector;

[0017] Input the word vector into Encoder to obtain the temporal and spatial feature information of the input data;

[0018] The Encoder includes an attention module, a convolutional layer and a linear layer connected in sequence.

[0019] Furthermore, the method for obtaining the physical constraint includes:

[0020] By training the neural network Minimize the loss between the predicted data and the target sequence and obtain the physical constraint function; where is the predicted data, Y (0) is the target data in the target sequence, and F is the physical constraint function.

[0021] Furthermore, the training objective is to minimize the difference between the predicted data and the target sequence with added Gaussian noise, including:

[0022] The KL divergence and MSE loss between the predicted data and the target sequence with Gaussian noise added are used as the difference measurement indicators between the predicted data and the target sequence.

[0023] In a second aspect, the present application proposes a power load forecasting model training system, wherein the power load forecasting model is based on a diffusion model; comprising:

[0024] A first acquisition module is used to acquire an input sequence and a target sequence, wherein the input sequence includes a historical power load time series, and the target sequence is a known prediction sequence corresponding to the input sequence;

[0025] The training module is used to train the power load forecasting model using the input sequence as the input of the power load forecasting model and the target sequence as the true value of the power load forecasting model; during training:

[0026] Gaussian noise is added to the input sequence and the target sequence synchronously by coupled diffusion;

[0027] Extract features from the input data in the input sequence to obtain temporal and spatial feature information of the input data;

[0028] Aggregate the temporal and spatial feature information of the input data, and denoise the target sequence with added Gaussian noise to obtain the predicted data;

[0029] The training objective is to minimize the difference between the predicted data and the target sequence, and physical constraints are imposed on the power load forecasting model, and the physical constraints are used to minimize the loss between the predicted data and the target sequence.

[0030] In a third aspect, the present application proposes a method for predicting power load, comprising:

[0031] Obtain historical power load time series;

[0032] Input the acquired historical power load time series into the trained power load forecasting model to obtain the corresponding forecast data;

[0033] Through multiple back-propagation iterations, the power load forecasting results are obtained.

[0034] In a fourth aspect, the present application proposes a power load forecasting system, comprising:

[0035] The second acquisition module is used to obtain the historical power load time series;

[0036] The model prediction module is used to input the acquired historical power load time series into the trained power load prediction model to obtain the corresponding prediction data;

[0037] The reverse module is used to obtain the power load forecasting result through multiple back-propagation iterations.

[0038] In a fifth aspect, the present application proposes an electronic device, comprising:

[0039] Memory for storing computer programs;

[0040] A processor is used to implement the steps of the above-mentioned power load forecasting method when executing the computer program.

[0041] In a sixth aspect, the present application proposes a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned power load forecasting method are implemented.

[0042] Compared with the prior art, this application has the following beneficial effects:

[0043] The present application proposes a method for training an electric load forecasting model. First, the data set is diffused through coupled diffusion, which not only expands the time series data set used for training, but also improves the generalization ability of the electric load forecasting model. Then, the long-term dependencies in the time dimension of the historical electric load data are extracted, and the extracted feature information data is aggregated, and the input sequence after coupled diffusion and noise is reduced to obtain a preliminary prediction result. Through the above operations, the electric load forecasting model can more effectively handle the uncertainty in the prediction, especially for data sets such as electric load that are affected by many factors. At the same time, domain-specific physical constraint information is introduced during diffusion. Through this physical constraint, the electric load forecasting model can better generate more reasonable prediction results that meet the physical rules of the electric load. When making predictions, even if the sequence input to the electric load forecasting model has noise added, accurate output prediction results can be guaranteed.

[0044] The present application also proposes a power load forecasting model training system, a power load forecasting method, an electronic device and a computer storage medium, which have all the advantages of the above-mentioned power load forecasting model training method. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0046] Figure 1 This is a schematic diagram of the first process of the power load forecasting model training method of this application;

[0047] Figure 2 A second flow chart of the power load forecasting model training method of this application;

[0048] Figure 3 Schematic diagram of the principle of the coupled diffusion method in the embodiment of the present application;

[0049] Figure 4 A schematic diagram of a power load forecasting model training system for this application;

[0050] Figure 5 A schematic diagram of a flow chart of the power load forecasting method of the present application;

[0051] Figure 6 A schematic diagram of the power load forecasting system of the present application;

[0052] Figure 7 A schematic diagram of the loss rate of the power load prediction model on the training set in an embodiment of the present application;

[0053] Figure 8 A schematic diagram of the loss rate of the power load prediction model on the validation set in an embodiment of the present application;

[0054] Fig. 9 This is a schematic diagram of the prediction results of power load data using the power load prediction model in an embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0056] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0057] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0058] In the description of the embodiments of the present application, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. appear, the orientation or position relationship indicated is based on the orientation or position relationship shown in the drawings, or the orientation or position relationship in which the invented product is usually placed when used. It is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0059] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", which does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0060] In the description of the embodiments of the present application, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0061] The diffusion model is a generative model based on probability theory. It gradually adds noise to the data and then learns how to reverse this process to restore the original data distribution from the noisy data, and finally generates new data samples. The working mechanism of the diffusion model can be divided into two stages: the forward diffusion process and the reverse diffusion process. Among them, the forward diffusion process is to gradually add noise to the data, making the data gradually more complex until it is finally converted into pure Gaussian noise. This process is similar to gradually adding a blur effect to an image until the image is completely unrecognizable. The reverse diffusion process is also called the generative process. It starts with pure Gaussian noise, gradually removes the noise, and restores the original high-quality data. This process is similar to playing a video in reverse, gradually revealing the details hidden under the noise.

[0062] The diffusion model demonstrates a strong nonlinear modeling capability. By simulating the diffusion process of data from a clear state to a noisy state and its inverse process, it can not only flexibly capture the complex dynamics and long-term dependencies in time series data, but also has the ability to efficiently generate diverse and high-quality time series data samples from noisy data. However, the prediction results generated by the diffusion model often deviate from physical reality due to the introduction of noise, thus affecting the prediction effect of the diffusion model.

[0063] Based on the above situation, the present application proposes a power load forecasting method and related devices, a model training method, and a system. The present application is described in detail below in conjunction with embodiments and drawings.

[0064] like Figure 1 As shown, it is a first flow chart of the power load forecasting model training method of the present application, which may include:

[0065] S101, obtaining an input sequence and a target sequence, wherein the input sequence includes a historical power load time series, and the target sequence is a known prediction sequence corresponding to the input sequence.

[0066] It should be noted that when training the power load forecasting model, it is necessary to use the historical power load time series and the corresponding target series. When actually using the power load forecasting model for forecasting, it is necessary to input the historical power load time series, and then the corresponding target series can be obtained, that is, the desired forecast result, and during training, the forecast result is known.

[0067] S102, using the input sequence as the input of the power load prediction model and the target sequence as the true value of the power load prediction model to train the power load prediction model.

[0068] It should be noted that the true value of the power load forecasting model usually refers to the output result of the power load forecasting model under ideal conditions.

[0069] During training:

[0070] (1) Gaussian noise is added to the input sequence and the target sequence synchronously by coupled diffusion.

[0071] In practical applications, the number of diffusion steps can be set, which refers to the number of intermediate states to be experienced in the process from clear data to complete noise. During the diffusion process, it is necessary to ensure that the input sequence and the target sequence are affected by the same degree and type of Gaussian noise. After each diffusion step, the input sequence and the target sequence will gradually become more blurred until they are completely noised.

[0072] (2) Extract features from the input data in the input sequence to obtain temporal and spatial feature information of the input data.

[0073] It should be noted that when obtaining temporal feature information, since the input sequence is a time series, the input data has a clear timestamp, and the periodic features in the input data can be identified and extracted, such as seasonal changes, daily cycles, weekly cycles, etc., and then the changing trend of the input data over time is analyzed to obtain temporal feature information, and the changing trend can be a linear trend or a nonlinear trend, etc. When obtaining spatial feature information, the location features can be extracted, and the distribution features of the location features in space can be analyzed, such as clustering and dispersion, and then the spatial relationship features between different location data, such as distance and direction, can be extracted to obtain spatial feature information.

[0074] (3) Aggregate the temporal and spatial feature information of the input data, and denoise the target sequence with added Gaussian noise to obtain the predicted data.

[0075] In practical applications, the temporal and spatial feature information of the input data is aggregated to form a feature vector. For example, a statistical method can be used to aggregate the feature information.

[0076] (4) Minimizing the difference between the predicted data and the target sequence with added Gaussian noise is used as the training goal, and physical constraints are imposed in the power load forecasting model, and the physical constraints are used to minimize the loss between the predicted data and the target sequence.

[0077] like Figure 2 As shown, it is a second flow chart of the power load forecasting model training method of the present application, which may include:

[0078] S201, establishing a power load data set.

[0079] The power load data set used in this embodiment is the actual power load data of a certain place from January 1, 2018 to September 1, 2020, a total of 3 years, with a sampling period of 15 minutes, 96 samples per day, and the unit of power load data is KW. In order to observe the correlation between the daily maximum temperature, daily minimum temperature, average temperature, relative humidity, rainfall and power load data in weather factors, this embodiment collects and exports the meteorological data set of the area from the meteorological station according to the corresponding sampling time and period, and merges them into a table. Some data are shown in Table 1:

[0080] Table 1 Power load data set combined with some meteorological data sets in the same region

[0081]

[0082] S202, by means of a coupled diffusion method, Gaussian noise is synchronously added to the input sequence and the target sequence for data enhancement.

[0083] The input sequence and target sequence are obtained from the above power load dataset.

[0084] In the diffusion model, the forward diffusion process can be fixed on a Markov chain, and Gaussian noise is gradually added to the data. Compared with the conventional Gaussian noise addition method, this application proposes a coupled diffusion method, such as Figure 3 , which is a schematic diagram of the principle of the coupled diffusion method in this embodiment. By adding Gaussian noise to the input sequence and the target sequence simultaneously to enhance the data, the accidental uncertainty of the time series data can be effectively reduced and the generalization ability of the generated model can be improved.

[0085] Specifically, two different variance lists β and β' can be used to control the noise addition process of the input sequence and the target sequence respectively, so that the input sequence and the target sequence are synchronized during the diffusion process, thereby reducing the difference between the generated noise and the original data noise. Where β'=wβ,w∈(0,1). By coupling the diffusion process, we can get the input data X in the input sequence (0) and the target data Y in the target sequence (0) The corresponding data after diffusion are the input data X in the input sequence (0) Noise data after diffusion X (T) and the target data Y in the target sequence (0) Noise data Y after diffusion (T) , corresponding to Figure 3 , X (1) To X (T-1) is the data after each diffusion in the diffusion process of the input sequence, Y (1) To Y (T-1) is the data after each diffusion in the diffusion process of the target sequence, T is the total number of diffusions, β1 to β T is the variance list used in each diffusion process of the input sequence, β1′ to β T ′ is the variance list used in each diffusion process of the target sequence. Figure 3 In, δ Y is the Gaussian noise added to the target sequence, δ X It is the Gaussian noise added to the input sequence. × represents multiplication with the corresponding weight coefficient, and the difference between 1 and the variance list is used as the weight coefficient.

[0086] S203, extracting features through an Encoder to capture the time dependency of the input sequence in the time dimension.

[0087] The historical power load data without noise, that is, the input data X in the input sequence (0) After data embedding, the word vector is obtained and input into the Encoder for feature learning.

[0088] Encoder usually includes attention modules, convolutional layers and linear layers connected in sequence. In the power load forecasting model, multiple encoders are also included. The attention modules, convolutional layers and linear layers of each encoder are superimposed and mainly used to capture the long-term dependency of the time dimension of historical power load data and the spatial coupling relationship of variables of different dimensions. After feature extraction by encoder, the feature information of time and space of historical power load data can be obtained. enc . This process can be expressed as:

[0089] X enc =Encoder(X (0) )

[0090] S204, performing denoising processing through a Decoder.

[0091] The input of the decoder is the input data after noise diffusion, that is, the input data X in the input sequence (0) Noise data after diffusion X (T) , and the high-dimensional feature X obtained by Encoder feature extraction enc .

[0092] The Decoder mainly aggregates the feature information extracted by the Encoder through the cross-attention mechanism , and for noisy data Perform denoising and obtain predicted data after multiple iterations . This process can be expressed as:

[0093] S205, calculating prediction data With noisy data KL divergence and MSE loss .

[0095] It should be noted that KL divergence (Kullback-Leibler Divergence) is an asymmetric measure of the difference between two probability distributions P and Q. In machine learning, it is often used to measure the difference between predicted values ​​and true values. MSE (Mean Squared Error) loss measures the average of the sum of the squares of the differences between predicted values ​​and true values. KL divergence and MSE loss are both indicators used to measure model performance in machine learning, but they focus on different angles. KL divergence is more suitable for evaluating the difference between probability distributions, while MSE loss is more suitable for evaluating the accuracy of numerical predictions. Combining the two indicators can improve the prediction effect of the power load forecasting model.

[0096] S206, implementing physical constraint assisted denoising by training a neural network.

[0097] Although the advantages of data diffusion are obvious, the samples after diffusion may be destroyed, causing the generated power load forecasting model to move towards a noisy target. To solve this problem, this application introduces physical constraints, incorporates auxiliary prior knowledge into the diffusion generation process, and improves the prediction accuracy of the power load forecasting model. The physical constraint is based on domain expertise and imposes a physical constraint function F on the predicted value. By training a neural network To achieve physical constraint assisted denoising.

[0098] Neural Networks The input is the target sequence and prediction data , find a suitable physical constraint function through model training , so that the predicted data The loss between the target sequence Minimum.

[0099] The final loss function of the power load forecasting model in this application is the sum of the losses at different stages, and the KL divergence D KL , MSE loss L MSE And the loss superposition of the physical constraint function F.

[0100] S207, through multiple back-propagation iterations, the final prediction result can be obtained.

[0101] Although transformer models can capture the long-term dependencies of power load data very well, when the power load data records are short and include noise, the generalization of transformer models is poor, the prediction effect is average, and the prediction process of transformer models is not interpretable. In response to these problems, the power load prediction model of the present application is based on a diffusion model, which is used to predict the time series of power load. It can not only expand the time series data, but also enhance the generalization ability and interpretability of the diffusion model. In addition, by introducing physical constraints, the accuracy of power load data prediction can be further improved.

[0102] In summary, the power load forecasting model in this application mainly performs four parts of work: (1) coupled diffusion stage: synchronously enhance the input sequence and target sequence to enhance the generalization ability of the power load forecasting model; (2) Encoder feature extraction process: capture the time dependency of the input sequence time dimension; (3) Decoder data denoising process: aggregate the features extracted by the encoder, denoise the noisy input sequence, and obtain the prediction result; (4) physical constraint assisted denoising process: further physical feature constraints are imposed on the prediction result obtained by the decoder.

[0103] In the above-mentioned embodiment, the coupled diffusion process mainly uses Markov chains to perform noise processing on the input sequence and the target sequence synchronously to expand the time series data set. The Encoder mainly extracts the features of the input data in the input sequence, and then aggregates the data features extracted by the Encoder through the Decoder, and performs noise reduction processing on the noisy input sequence to obtain the predicted data. Among them, the use of physical constraints for noise reduction is mainly to further clean the prediction results according to the physical characteristics of the data to improve the prediction accuracy. Therefore, the power load forecasting model training method proposed in this application can enable the power load forecasting model to accurately and efficiently predict the power load data, and provide strong data support for the supply and demand balance, carbon saving and emission reduction, and green environmental protection of the power system.

[0104] The power load forecasting model is trained using the power load forecasting model training method proposed above, and then the trained power load forecasting model is used to forecast the power load data to obtain the power load data forecast results for a period of time in the future. In this example, the final forecast result is obtained through multiple back propagation iterations 10 times. The loss rates of the power load forecasting model on the training set and the validation set are as follows: Figure 7 He Ru Figure 8 As shown, Figure 7 is the loss rate on the training set, which shows an overall downward trend as the number of iterations increases. Figure 8 The loss rate on the validation set decreases as the number of iterations increases.

[0105] In this example, the prediction results of the power load data using the power load prediction model are as follows: Fig. 9 As shown, through analysis, it can be seen that the power load forecasting model trained in this application can better predict the future trend changes of ground current. It should be noted that Fig. 9 In , true represents the true value of the power load forecasting model, and predict represents the prediction result of the power load forecasting model.

[0106] In order to prove the prediction effect of the power load forecasting model of the present application, the obtained prediction results are analyzed. The mean relative error (MRE) and mean error (ME) are selected as evaluation criteria, and the performance of the power load forecasting model of the present application and other mainstream time series forecasting methods are compared in power load forecasting tasks of different lengths. The specific comparison results are shown in Table 2. It can be found from Table 2 that the power load forecasting model of the present application is significantly better than other comparison methods, with an average reduction of 7.9% in the average error and 23.9% in the average relative error.

[0107] Table 2 Comparison results of power load prediction effects

[0108]

[0109] Electricity load is affected by many unpredictable factors, such as weather changes, emergencies, etc. Traditional electricity load forecasting methods can usually only provide a single best estimate, lacking the expression of uncertainty, which in turn affects the accuracy of the electricity load forecasting model. In addition, the shortage of high-quality data often affects the feature extraction capabilities of traditional time series forecasting models. In contrast, the advantages of this application are mainly reflected in:

[0110] (1) High prediction accuracy: It can effectively handle the uncertainty in the prediction and generate multiple possible future scenarios, which is crucial for processing data such as power load that is affected by multiple unpredictable factors.

[0111] (2) Explicit physical constraints: By introducing physical constraints to guide the diffusion generation process of the power load forecasting model, it is ensured that the generated forecast results are consistent with the physical laws of the specific field and have better forecasting accuracy.

[0112] (3) Strong robustness: The physical constraint-assisted denoising used in this application can be used as a plug-in in the power load forecasting model without affecting the overall training of the power load forecasting model. This modular approach allows lightweight networks to be trained to impose different constraints without having to train a complete model, and can quickly build constrainers with different domain knowledge.

[0113] (4) Multi-time scale prediction: This application can meet the short-term, medium-term and long-term power load forecasting needs, which can well assist power grid scheduling and planning, thereby improving the overall stability and security of the power grid.

[0114] Based on the above power load forecasting model training method, Figure 4 As shown, the present application also proposes a power load forecasting model training system, which may include:

[0115] A first acquisition module is used to acquire an input sequence and a target sequence, wherein the input sequence includes a historical power load time series, and the target sequence is a known prediction sequence corresponding to the input sequence;

[0116] The training module is used to train the power load forecasting model using the input sequence as the input of the power load forecasting model and the target sequence as the true value of the power load forecasting model; during training:

[0117] Gaussian noise is added to the input sequence and the target sequence synchronously by coupled diffusion;

[0118] Extract features from the input data in the input sequence to obtain temporal and spatial feature information of the input data;

[0119] Aggregate the temporal and spatial feature information of the input data, and denoise the target sequence with added Gaussian noise to obtain the predicted data;

[0120] The training objective is to minimize the difference between the predicted data and the target sequence, and physical constraints are imposed on the power load forecasting model, and the physical constraints are used to minimize the loss between the predicted data and the target sequence.

[0121] like Figure 5 As shown, the present application also proposes a power load forecasting method, which may include:

[0122] S301, obtaining a historical power load time series;

[0123] S302, inputting the acquired historical power load time series into the trained power load forecasting model to obtain corresponding forecast data;

[0124] S303, obtaining the power load prediction result through multiple back propagation iterations.

[0125] Correspondingly, such as Figure 6 As shown, the present application also proposes a power load forecasting system, which may include:

[0126] The second acquisition module is used to obtain the historical power load time series;

[0127] The model prediction module is used to input the acquired historical power load time series into the trained power load prediction model to obtain the corresponding prediction data;

[0128] The reverse module is used to obtain the power load forecasting result through multiple back-propagation iterations.

[0129] It should be noted that in the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of each module is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules can be combined or integrated into another device, or some features can be ignored or not executed. The module described as a separate component may or may not be physically separated. The component displayed as a module may be a physical unit or multiple physical units, that is, it may be located in one place, or it may be distributed in multiple different places. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0130] In addition, each module in each embodiment of the present invention may be integrated into a processing unit, each module may exist physically separately, or two or more modules may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0131] An electronic device provided in an embodiment of the present application includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps of the power load forecasting method described in any of the above embodiments are implemented.

[0132] Another electronic device provided in the embodiment of the present application may also include: an input port connected to the processor, used to transmit multimodal data collected by an external acquisition device to the processor; and a display unit connected to the processor, used to display the processing results of the processor to the outside world; a communication module connected to the processor, used to realize communication between the electronic device and the outside world. The display unit can be a display panel, a laser scanning display, etc.; the communication mode adopted by the communication module includes but is not limited to mobile high-definition link technology (HML), universal serial bus (USB), high-definition multimedia interface (HDMI), wireless connection (including wireless fidelity technology (WiFi), Bluetooth communication technology, low-power Bluetooth communication technology, and communication technology based on IEEE802.11s).

[0133] An embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of the power load forecasting method described in any of the above embodiments are implemented.

[0134] The computer-readable storage medium involved in the present application includes random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field.

[0135] For the description of the relevant parts of the power load forecasting system, electronic device and computer-readable storage medium provided in the embodiments of the present application, please refer to the detailed description of the corresponding parts in the power load forecasting method provided in the embodiments of the present application, which will not be repeated here. In addition, the parts of the above technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive elaboration.

[0136] The above are only preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for training a power load forecasting model, wherein the power load forecasting model is based on a diffusion model; characterized in that: include: Obtaining an input sequence and a target sequence, wherein the input sequence includes a historical power load time series, and the target sequence is a known prediction sequence corresponding to the input sequence; The input sequence is used as the input of the power load forecasting model, and the target sequence is used as the true value of the power load forecasting model to train the power load forecasting model. During training: Gaussian noise is added to the input sequence and the target sequence synchronously by coupled diffusion; Extract features from the input data in the input sequence to obtain temporal and spatial feature information of the input data; Aggregate the temporal and spatial feature information of the input data, and denoise the target sequence with added Gaussian noise to obtain the predicted data; The training objective is to minimize the difference between the predicted data and the target sequence with Gaussian noise added, and physical constraints are imposed on the power load forecasting model, and the physical constraints are used to minimize the loss between the predicted data and the target sequence.

2. The method for training a power load prediction model according to claim 1, characterized in that: The method of synchronously adding Gaussian noise to the input sequence and the target sequence by coupled diffusion includes: Two different variance lists are used to control the Gaussian noise adding process of the input sequence and the target sequence respectively, so that the process of adding Gaussian noise to the input sequence and the target sequence can be kept synchronized.

3. The method for training a power load prediction model according to claim 2, characterized in that: The feature extraction of the input data in the input sequence includes: Embed the input data to get the word vector; Input the word vector into Encoder to obtain the temporal and spatial feature information of the input data; The Encoder includes an attention module, a convolutional layer and a linear layer connected in sequence.

4. The method for training a power load prediction model according to claim 3, characterized in that: The method for obtaining the physical constraint includes: By training the neural network Minimize the loss between the predicted data and the target sequence and obtain the physical constraint function; where is the predicted data, Y (0) is the target data in the target sequence, and F is the physical constraint function.

5. The method for training a power load prediction model according to claim 4, characterized in that: The training objective is to minimize the difference between the predicted data and the target sequence with added Gaussian noise, including: The KL divergence and MSE loss between the predicted data and the target sequence with Gaussian noise added are used as the difference measurement indicators between the predicted data and the target sequence.

6. A power load forecasting model training system, wherein the power load forecasting model is based on a diffusion model; characterized in that: include: A first acquisition module is used to acquire an input sequence and a target sequence, wherein the input sequence includes a historical power load time series, and the target sequence is a known prediction sequence corresponding to the input sequence; The training module is used to train the power load forecasting model using the input sequence as the input of the power load forecasting model and the target sequence as the true value of the power load forecasting model; during training: Gaussian noise is added to the input sequence and the target sequence synchronously by coupled diffusion; Extract features from the input data in the input sequence to obtain temporal and spatial feature information of the input data; Aggregate the temporal and spatial feature information of the input data, and denoise the target sequence with added Gaussian noise to obtain the predicted data; The training objective is to minimize the difference between the predicted data and the target sequence, and physical constraints are imposed on the power load forecasting model, and the physical constraints are used to minimize the loss between the predicted data and the target sequence.

7. A method for predicting power load, characterized in that: include: Obtain historical power load time series; Input the acquired historical power load time series into the trained power load forecasting model to obtain the corresponding forecast data; Through multiple back-propagation iterations, the power load forecasting results are obtained.

8. A power load forecasting system, characterized in that: include: The second acquisition module is used to obtain the historical power load time series; The model prediction module is used to input the acquired historical power load time series into the trained power load prediction model to obtain the corresponding prediction data; The reverse module is used to obtain the power load forecasting result through multiple back-propagation iterations.

9. An electronic device, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the power load forecasting method as claimed in claim 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the power load forecasting method according to claim 7 are implemented.

Images