Power load prediction method and device, electronic equipment and storage medium

Abstract

One or more embodiments of the invention provide a power load prediction method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring first power loads corresponding to a preset number of continuous historical moments; obtaining a power load characteristic corresponding to the first power load based on the first power load; predicting a second power load at the current moment by using a pre-trained power load prediction model based on the power load characteristics; wherein the power load prediction model is constructed based on a selective state space model, the selective state space model is used for describing the dynamic process of the power load changing along with time so as to obtain the state of the current moment, and the state represents accumulated dynamic information of the power load in the time dimension.

Description

Technical Field

[0001] This disclosure relates to the field of power technology, and more particularly to a power load forecasting method, apparatus, electronic device, and storage medium. Background Technology

[0002] It should be noted that the above description of the technical background is only for the purpose of providing a clear and complete explanation of the technical solutions of the present invention and facilitating understanding by those skilled in the art. It should not be assumed that the above technical solutions are known to those skilled in the art simply because they have been described in the background section of this invention.

[0003] The power industry is a key foundational industry for national development, closely linked to the development of various sectors and playing a vital supporting role in the stable operation of the social economy. Accurate short-term power load forecasting is a core element in ensuring the stable operation of the power system and maintaining the orderly functioning of society. In recent years, with the continuous development of artificial intelligence technology, methods such as machine learning and deep learning have been gradually introduced into the field of power load forecasting. These technologies can identify key factors affecting power load through in-depth analysis of historical data, providing a technical foundation for improving forecast accuracy.

[0004] The mainstream technical solutions in the field of current power load forecasting are mainly divided into two categories. One category is a model with recurrent neural networks as its core, which predicts the current power load based on the power load data of the previous historical moment; the other category is a neural network model based on the Transformer structure, which uses power load data from multiple historical moments to capture global temporal correlations to achieve prediction.

[0005] However, both technical solutions have obvious limitations: the former relies on the recursive logic of historical data at a single moment, making it difficult to fully extract the deep dependency features in the time series, resulting in poor prediction accuracy; the latter, although it can capture global correlations, has a large number of model parameters and high computational complexity, which easily generates high computing power and storage overhead, thus limiting its adaptability. Summary of the Invention

[0006] In view of the above, the purpose of one or more embodiments of this disclosure is to provide a power load forecasting method, apparatus, electronic device and storage medium to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the first aspect of this disclosure provides a method for predicting electricity load, comprising: Obtain the first power load corresponding to a preset number of consecutive historical moments; Based on the first power load, the power load characteristics corresponding to the first power load are obtained; Based on the power load characteristics, the second power load at the current moment is predicted using a pre-trained power load prediction model; The power load prediction model is constructed based on a selective state-space model, which is used to describe the dynamic process of power load changing over time to obtain the state at the current moment. The state represents the cumulative dynamic information of the power load in the time dimension.

[0008] Optionally, based on the power load characteristics, a pre-trained power load prediction model is used to predict the second power load at the current moment, including: The power load characteristics are input into the selective state space model to output a first intermediate feature corresponding to the first power load; the first intermediate feature carries the correlation between the first power load at a historical time and the second power load at the current time; A gating mechanism is used to filter the second intermediate feature from the first intermediate feature, where the second intermediate feature represents the effective feature from the first intermediate feature. The second intermediate feature is linearly mapped to obtain the second power load.

[0009] Optionally, the power load characteristics are input into the selective state-space model to output a first intermediate feature corresponding to the first power load, including: Multiplying the power load characteristics by the control matrix yields the effect of the power load characteristics on the state. Multiply the current state by the state matrix to obtain the evolution term of the state; Integrating the sum of the action term and the evolution term yields the updated state; Multiply the updated state by the output matrix to obtain the contribution of the state to the output result; Multiply the power load characteristics by the instruction matrix to obtain the main item of the output result; The contribution item and the main item are added together to obtain the first intermediate feature.

[0010] Optionally, it also includes: Configure the updated state as the state corresponding to the next round of input data.

[0011] Optionally, based on the power load data at the historical time, power load characteristics are obtained, including: The historical power load data is divided into multiple data segments along the time-series dimension; The power load characteristics are obtained by performing convolution processing on the multiple data segments.

[0012] Optionally, it also includes: The following steps are executed iteratively until the first parameter meets the preset condition: Obtain the first parameters of the power load prediction model sent by the central server, and initialize the power load prediction model based on the first parameters; In response to determining that the amount of data for the first power load is greater than a preset threshold, the power load prediction model is trained based on the first power load to obtain the second parameter of the power load prediction model; the second parameter is used by the central server to integrate the data to obtain the first parameter.

[0013] Optionally, it also includes: In response to determining that the amount of data of the first power load is less than or equal to the preset threshold, the first parameter sent by the central server is obtained; The power load prediction model is deployed based on the first parameter.

[0014] A second aspect of this disclosure provides an electrical load forecasting device, comprising: The acquisition module is configured to acquire the first power load corresponding to a preset number of consecutive historical moments; The extraction module is configured to obtain the power load characteristics corresponding to the first power load based on the first power load; The prediction module is configured to predict the second power load at the current moment based on the power load characteristics using a pre-trained power load prediction model; The power load prediction model is constructed based on a selective state-space model, which is used to describe the dynamic process of power load changing over time to obtain the state at the current moment. The state represents the cumulative dynamic information of the power load in the time dimension.

[0015] A third aspect of this disclosure provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in the first aspect.

[0016] In a fourth aspect, this disclosure provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method as described in the first aspect.

[0017] As described above, the power load forecasting method, apparatus, electronic device, and storage medium disclosed herein first acquire a first power load corresponding to a preset number of consecutive historical time points, then extract the power load characteristics corresponding to the first power load, and finally, based on the power load characteristics, use a pre-trained power load forecasting model to predict the second power load at the current time. The power load forecasting model is constructed based on a selective state-space model. Through the power load forecasting model, the temporal dependence characteristics and dynamic change patterns of power load can be fully explored, effectively improving the accuracy and reliability of power load forecasting, thereby providing technical support for the stable operation and scheduling optimization of the power system. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in one or more embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only one or more embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating one or more embodiments of the power load forecasting method disclosed herein; Figure 2 This is a load data distribution feature diagram of one or more embodiments of this disclosure; Figure 3 This is a comparison chart of load data prediction results for one or more embodiments of this disclosure; Figure 4 This is a schematic diagram of the structure of a power load forecasting device according to one or more embodiments of the present disclosure; Figure 5 This is a schematic diagram of the hardware structure of an electronic device according to one or more embodiments of this disclosure. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0021] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of this disclosure should have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms "first," "second," and similar words used in one or more embodiments of this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0022] As described in the background section, the mainstream technical solutions in the current field of power load forecasting are mainly divided into two categories. One category is models with recurrent neural networks as the core, including recurrent neural networks (RNN), long short-term memory networks (LSTM), and gated recurrent units (GRU); the other category is the Transformer structure model, which retains the positional information of time-series data through positional encoding and relies on self-attention mechanisms to capture global dependencies, thereby optimizing load forecasting performance.

[0023] However, traditional RNN models suffer from problems such as missing gradients, gradient explosion, and loss of historical data over long distances.

[0024] When using LSTM neural networks to predict short-term power load data, the introduction of gating and long-short-term modules on top of RNN networks requires more parameters and computational resources. It also necessitates extensive training with historical data to prevent overfitting. Furthermore, the complexity of LSTM makes the iterative process less intuitive, leading to difficulties in interpreting decision results in certain scenarios.

[0025] Furthermore, in the prediction process based on the recurrent neural network structure, when the sequence length is very long, the state closer to the initial time has less impact on the prediction result at the current time, which means it faces the problem of gradient vanishing or exploding, resulting in the need to improve the prediction accuracy.

[0026] The Transformer preserves temporal positional information through positional encoding and captures global dependencies through a self-attention mechanism, thus alleviating the long-distance dependency problem to some extent. However, the Transformer has a large number of parameters and a complex structure, which can impose enormous computational and storage pressure on edge devices with limited computing resources.

[0027] Therefore, this disclosure proposes a method for electricity load forecasting. (Reference) Figure 1 The power load forecasting method of one or more embodiments disclosed herein includes the following steps: Step S101: Obtain the first power load corresponding to a preset number of consecutive historical moments.

[0028] Step S102: Based on the first power load, obtain the power load characteristics corresponding to the first power load.

[0029] Step S103: Based on the power load characteristics, predict the second power load at the current moment using a pre-trained power load prediction model.

[0030] The power load prediction model is constructed based on a selective state-space model, which is used to describe the dynamic process of power load changing over time to obtain the state at the current moment. The state represents the cumulative dynamic information of the power load in the time dimension.

[0031] The power load refers to the total power consumption demand of power users (such as industrial enterprises, residential buildings, public facilities, etc.) on the power system at a specific moment or time period.

[0032] Electricity load characteristics not only include regular cyclical features such as daily fluctuations and seasonal changes, but also complex features such as dynamic changes and long-term dependencies. Traditional techniques for load feature extraction focus on capturing relatively stable regularities in historical data to build load forecasting models. However, these techniques lack depth in mining the complex features of dynamic changes, long-term dependencies, and non-stationarity in load data, failing to fully utilize the key forecasting information carried by these features. Furthermore, they ignore the dynamic evolution of the probability distribution of load data, building models based on the assumption of "consistent distribution," which makes it difficult to adapt to the actual changing patterns of load data in real-world scenarios. This limits the comprehensiveness and accuracy of feature extraction, thus affecting the performance of the forecasting model.

[0033] In the embodiments of this disclosure, the first power load corresponding to a preset number of consecutive historical moments is used as the data basis for the predicted power load at the current moment, so as to fully consider the temporal characteristics of the historical power load.

[0034] In some embodiments, the first power load at N consecutive historical moments within a preset time window can be collected and used as the data basis for model prediction, so that the model can better capture the local dynamic information of the sequence. It should be noted that each input to the power load prediction model is the first power load at N consecutive historical moments.

[0035] In some embodiments, in order to eliminate the impact of scale differences between different load data and prevent large-scale data from dominating the training process, the power load data must first be instance-normalized.

[0036] In some embodiments, features of the first electrical load can be extracted using a convolutional neural network.

[0037] Then, the power load characteristics are input into the pre-trained power load prediction model to obtain the predicted second power load at the current moment.

[0038] In the embodiments of this disclosure, the power load forecasting model is constructed based on a selective state-space model. A selective state-space model focuses on dynamic modeling of time-series data. Its core principle is to construct state evolution equations using parameters such as state matrices and control matrices, adaptively selecting key features while preserving accumulated dynamic information over time.

[0039] In some embodiments, predicting the second power load at the current moment using a pre-trained power load prediction model based on the power load characteristics may include: inputting the power load characteristics into the selective state-space model to output a first intermediate feature corresponding to the first power load; the first intermediate feature carries the correlation between the first power load at a historical moment and the second power load at the current moment; using a gating mechanism to filter a second intermediate feature from the first intermediate features, the second intermediate feature representing a valid feature in the first intermediate features; and performing a linear mapping on the second intermediate feature to obtain the second power load.

[0040] In some embodiments, the power load forecasting model can be trained based on a linear time series model with a selective state space (Mamba). The core advantages of Mamba are: through recursive state updates, it compresses all historical time-series information into a hidden state, fully preserving long-range dependencies while avoiding redundant computation; and it dynamically filters key information and noise to adapt to time-series scenarios such as power load forecasting. In other words, Mamba can achieve higher forecasting accuracy with less computation.

[0041] Specifically, in some embodiments, inputting the power load characteristics into the Selective State Space Model to output a first intermediate feature corresponding to the first power load may include: multiplying the power load characteristics by a control matrix to obtain an action term of the power load characteristics on the state; multiplying the current state by a state matrix to obtain an evolution term of the state; integrating the sum of the action term and the evolution term to obtain an updated state; multiplying the updated state by an output matrix to obtain a contribution term of the state to the output result; multiplying the power load characteristics by an instruction matrix to obtain a main term of the output result; and adding the contribution term and the main term to obtain the first intermediate feature.

[0042] In some embodiments, the updated state is configured as the state corresponding to the next round of input data to ensure that the state is a superposition of dynamic information from historical moments.

[0043] The process of integrating the sum of the action term and the evolution term to obtain the updated state can be regarded as a state update to capture the temporal dependence and dynamic information of historical data; the process of adding the contribution term and the subject term to obtain the first intermediate feature can be regarded as an output mapping to obtain the first intermediate feature at the current moment.

[0044] It is important to note that the first intermediate feature data obtained at this point represents both structural and temporal information.

[0045] In the process of realizing this disclosure, the inventors discovered that the single prediction model used in the related technology has limitations when facing power load prediction in different application scenarios, is prone to large deviations when processing specific types of data, and is difficult to effectively cope with the highly nonlinear and time-varying characteristics of the data.

[0046] Therefore, improving the generalization ability of the model is crucial.

[0047] A gated perceptron is a neural network module with a dynamic gating mechanism. It selectively filters and fuses input features through learnable gating vectors to retain key and effective information, suppress redundant noise, and adapt to the dynamic dependencies of time-series data.

[0048] Therefore, in some embodiments, after obtaining the first intermediate feature, a gating mechanism can be used to filter the second intermediate feature among the first intermediate features, where the second intermediate feature represents a valid feature among the first intermediate features.

[0049] In the process of developing this disclosure, the inventors also discovered that in the actual regional power system user load forecasting work, edge users need to upload data, and this data includes sensitive electricity consumption data such as unit production energy consumption and production load characteristics. When edge user electricity consumption data is uploaded to the cloud-based power grid dispatch load management center for centralized load forecasting, there is a risk of leakage of sensitive user data. Therefore, while achieving accurate user load forecasting, protecting the privacy of electricity consumption data has become an urgent problem to be solved.

[0050] Therefore, in some embodiments, the power load forecasting model is trained through edge-cloud collaboration. Specifically, after edge users train a local model based on their own power load data, they upload the parameters of the local model to the cloud, and the cloud integrates the model parameters uploaded by multiple edge users to obtain the globally optimal model parameters.

[0051] In some embodiments, considering that not all edge users have enough training data, in each iteration, only edge users with more than a preset threshold are selected to participate in training, and the final model parameters are sent to edge users who have not participated in training, so as to ensure that all edge users follow the iteration progress.

[0052] In other words, in some embodiments, a training round may include: obtaining the first parameters of the power load prediction model sent by the central server (cloud), and initializing the power load prediction model based on the first parameters; in response to determining that the amount of data of the first power load is greater than a preset threshold, training the power load prediction model based on the first power load to obtain the second parameters of the power load prediction model; the second parameters are used by the central server to integrate them to obtain the first parameters.

[0053] For edge users who cannot participate in training, a training round may include: in response to determining that the amount of data for the first power load is less than or equal to the preset threshold, obtaining the first parameter sent by the central server; and deploying the power load prediction model based on the first parameter.

[0054] Taking one embodiment of this disclosure as an example, the model training process may include: A subset of users is selected from all users, and each subset is provided with the current global model. The total number of samples is m, the proportion of users participating in this process is η, and the sample size is kj. The global model network initialization parameters include initial values ​​W0 and B0 for the weights and biases. These parameters can be randomly initialized or pre-trained using data. During each round of training, the cloud distributes the global model parameters to user subsets with sufficient data records and sufficient required features to enhance the features of the training data.

[0055] Selected edge users update their parameters using their locally stored historical data and send the updated parameters back to the cloud. The cloud then aggregates and integrates these updated model parameters. During this integration process, multiple model parameters can be cleaned and preprocessed.

[0056] The cloud determines whether the integrated model parameters meet the preset convergence conditions. If they do, the iteration ends and the integrated model parameters are synchronized to all edge users. If they do not meet the conditions, the iteration continues.

[0057] Through the above iterative calculations, the average gradient method can ensure global accuracy and stability, and effectively overcome the problem of data imbalance between devices.

[0058] To verify the technical effectiveness of the disclosed technical solution, the inventors also conducted a series of experiments.

[0059] The inventors used centralized electrolytic aluminum plant load data, distributed electric vehicle load data, and temperature-controlled load data as research subjects. The time span for all types of load data used in the experiment was one year, with a data collection granularity set at 15 minutes (i.e., 96 data points collected daily). Specifically, the centralized load data included electricity consumption data from four electrolytic aluminum plants; the distributed load data encompassed five electric vehicle datasets and five temperature-controlled load datasets, each containing electricity consumption records from 100 users.

[0060] Figure 2 The daily load distribution map shown illustrates the specific distribution characteristics of the three types of datasets. Table 1 presents the statistical characteristics of each type of load data.

[0061] Table 1

[0062] Among them, the loads of electric vehicles and temperature control systems fluctuate significantly, while the loads of electrolytic aluminum fluctuate less, indicating that centralized industrial users maintain a stable state over long periods of time.

[0063] The power load forecasting model disclosed herein (hereinafter referred to as HFL+Mamba) and four other forecasting models from related technologies, namely Transformer, RNN, etc., were selected. The root mean square error (RMSE) was used as a quantitative evaluation index to compare and analyze the power load forecasting performance of multiple users.

[0064] The comparison results are as follows Figure 3As shown in Table 2, for electrolytic aluminum load, compared with the Mamba prediction method, the Transformer prediction method, and the RNN prediction method, the root mean square error of the HFL+Mamba-based prediction scheme proposed in this disclosure is reduced by 18.89%, 54.40%, and 64.14%, respectively. Similarly, for electric vehicle load and temperature control load, the root mean square error of the proposed HFL+Mamba-based prediction scheme is reduced by more than 45.85% and 52.83%, respectively, compared with other methods, indicating that the proposed method has a significant improvement in prediction accuracy.

[0065] In addition, through Figure 3 It is evident that the two Mamba-based prediction schemes are more accurate than those based on Transformer or RNN. This is because: 1) The Mamba-based prediction scheme uses a state-space equation to describe the mapping relationship between input data and output prediction results, and can dynamically adjust the model's parameter matrix based on the input values, thus enabling the prediction model to respond more quickly to sudden changes; 2) The Transformer or RNN-based prediction schemes iteratively update the predicted values ​​by obtaining the correlation between input sequences, therefore they cannot accurately characterize the random characteristics of user electricity consumption behavior.

[0066] Furthermore, considering that the HFL+Mamba proposed in this disclosure can make more targeted predictions for users, by updating all model parameters through local training by the user and uploading and aggregating some parameters (Mamba neural network weight parameters and bias parameters), each local model contains both overall features and retains local user-specific features (sampling step size). HFL+Mamba achieves the protection of local dataset privacy while jointly training the charging load prediction model with multiple parties. The richer and more diverse training sample set significantly improves the accuracy of load prediction.

[0067] Table 2

[0068] It is understandable that this method can be executed by any device, equipment, platform, or cluster of devices with computing and processing capabilities.

[0069] It should be noted that the methods of one or more embodiments of this disclosure can be executed by a single device, such as a computer or server. The methods of this embodiment can also be applied in a distributed scenario, where multiple devices cooperate to complete the process. In such a distributed scenario, one of these devices may execute only one or more steps of the methods of one or more embodiments of this disclosure, and the multiple devices will interact with each other to complete the method described.

[0070] It should be noted that the above description pertains to specific embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than those shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0071] Based on the same inventive concept, corresponding to any of the methods in the above embodiments, this disclosure also provides a power load forecasting device. For example... Figure 4 As shown, the device includes: The acquisition module 41 is configured to acquire the first power load corresponding to a preset number of consecutive historical moments; Extraction module 42 is configured to obtain the power load characteristics corresponding to the first power load based on the first power load; Prediction module 43 is configured to predict the second power load at the current moment based on the power load characteristics using a pre-trained power load prediction model; The power load prediction model is constructed based on a selective state-space model, which is used to describe the dynamic process of power load changing over time to obtain the state at the current moment. The state represents the cumulative dynamic information of the power load in the time dimension.

[0072] Optionally, the prediction module 43 is specifically configured as follows: The power load characteristics are input into the selective state space model to output a first intermediate feature corresponding to the first power load; the first intermediate feature carries the correlation between the first power load at a historical time and the second power load at the current time; A gating mechanism is used to filter the second intermediate feature from the first intermediate feature, where the second intermediate feature represents the effective feature from the first intermediate feature. The second intermediate feature is linearly mapped to obtain the second power load.

[0073] Optionally, the prediction module 43 is specifically configured as follows: Multiplying the power load characteristics by the control matrix yields the effect of the power load characteristics on the state. Multiply the current state by the state matrix to obtain the evolution term of the state; Integrating the sum of the action term and the evolution term yields the updated state; Multiply the updated state by the output matrix to obtain the contribution of the state to the output result; Multiply the power load characteristics by the instruction matrix to obtain the main item of the output result; The contribution item and the main item are added together to obtain the first intermediate feature.

[0074] Optionally, it is also configured as follows: Configure the updated state as the state corresponding to the next round of input data.

[0075] Optionally, it is also configured as follows: The historical power load data is divided into multiple data segments along the time-series dimension; The power load characteristics are obtained by performing convolution processing on the multiple data segments.

[0076] Optionally, it is also configured as follows: The following steps are executed iteratively until the first parameter meets the preset condition: Obtain the first parameters of the power load prediction model sent by the central server, and initialize the power load prediction model based on the first parameters; In response to determining that the amount of data for the first power load is greater than a preset threshold, the power load prediction model is trained based on the first power load to obtain the second parameter of the power load prediction model; the second parameter is used by the central server to integrate the data to obtain the first parameter.

[0077] Optionally, it is also configured as follows: In response to determining that the amount of data of the first power load is less than or equal to the preset threshold, the first parameter sent by the central server is obtained; The power load prediction model is deployed based on the first parameter.

[0078] For ease of description, the above apparatus is described in terms of its functions, divided into various modules. Of course, when implementing one or more embodiments of this disclosure, the functions of each module can be implemented in one or more software and / or hardware.

[0079] The apparatus described above is used to implement the corresponding methods in the foregoing embodiments and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0080] Figure 5This embodiment illustrates a more specific hardware structure of an electronic device. The device may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0081] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this disclosure.

[0082] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this disclosure are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0083] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0084] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0085] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0086] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this disclosure, and not necessarily all the components shown in the figures.

[0087] The electronic devices described above are used to implement the corresponding methods in the foregoing embodiments and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0088] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0089] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this disclosure (including the claims) is limited to these examples; within the framework of this disclosure, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this disclosure as described above, which are not provided in detail for the sake of brevity.

[0090] Additionally, to simplify the description and discussion, and to avoid obscuring one or more embodiments of this disclosure, the provided drawings may or may not show well-known power / ground connections to integrated circuit (IC) chips and other components. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring one or more embodiments of this disclosure, and this also takes into account the fact that the details of implementation of these block diagram apparatuses are highly dependent on the platform on which one or more embodiments of this disclosure will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuitry) are set forth to describe exemplary embodiments of this disclosure, it will be apparent to those skilled in the art that one or more embodiments of this disclosure may be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0091] Although this disclosure has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0092] This disclosure includes one or more embodiments intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for predicting electricity load, characterized in that, include: Obtain the first power load corresponding to a preset number of consecutive historical moments; Based on the first power load, the power load characteristics corresponding to the first power load are obtained; Based on the power load characteristics, the second power load at the current moment is predicted using a pre-trained power load prediction model; The power load prediction model is constructed based on a selective state-space model, which is used to describe the dynamic process of power load changing over time to obtain the state at the current moment. The state represents the cumulative dynamic information of the power load in the time dimension.

2. The method according to claim 1, characterized in that, Based on the aforementioned power load characteristics, a pre-trained power load prediction model is used to predict the second power load at the current moment, including: The power load characteristics are input into the selective state space model to output a first intermediate feature corresponding to the first power load; the first intermediate feature carries the correlation between the first power load at a historical time and the second power load at the current time; A gating mechanism is used to filter the second intermediate feature from the first intermediate feature, where the second intermediate feature represents the effective feature from the first intermediate feature. The second intermediate feature is linearly mapped to obtain the second power load.

3. The method according to claim 2, characterized in that, The power load characteristics are input into the selective state-space model to output the first intermediate characteristics corresponding to the first power load, including: Multiplying the power load characteristics by the control matrix yields the effect of the power load characteristics on the state. Multiply the current state by the state matrix to obtain the evolution term of the state; Integrating the sum of the action term and the evolution term yields the updated state; Multiply the updated state by the output matrix to obtain the contribution of the state to the output result; Multiply the power load characteristics by the instruction matrix to obtain the main item of the output result; The contribution item and the main item are added together to obtain the first intermediate feature.

4. The method according to claim 3, characterized in that, Also includes: Configure the updated state as the state corresponding to the next round of input data.

5. The method according to claim 1, characterized in that, Based on the power load data at the aforementioned historical time points, power load characteristics are obtained, including: The historical power load data is divided into multiple data segments along the time-series dimension; The power load characteristics are obtained by performing convolution processing on the multiple data segments.

6. The method according to claim 1, characterized in that, Also includes: The following steps are executed iteratively until the first parameter meets the preset condition: Obtain the first parameters of the power load prediction model sent by the central server, and initialize the power load prediction model based on the first parameters; In response to determining that the amount of data of the first power load is greater than a preset threshold, the power load prediction model is trained based on the first power load to obtain the second parameters of the power load prediction model; The second parameter is used by the central server to integrate and obtain the first parameter.

7. The method according to claim 6, characterized in that, Also includes: In response to determining that the amount of data of the first power load is less than or equal to the preset threshold, the first parameter sent by the central server is obtained; The power load prediction model is deployed based on the first parameter.

8. A power load forecasting device, characterized in that, include: The acquisition module is configured to acquire the first power load corresponding to a preset number of consecutive historical moments; The extraction module is configured to obtain the power load characteristics corresponding to the first power load based on the first power load; The prediction module is configured to predict the second power load at the current moment based on the power load characteristics using a pre-trained power load prediction model; The power load prediction model is constructed based on a selective state-space model, which is used to describe the dynamic process of power load changing over time to obtain the state at the current moment. The state represents the cumulative dynamic information of the power load in the time dimension.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executed by the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions for causing the computer to perform the method of any one of claims 1 to 7.

Images