Active response load prediction method and system based on hybrid neural network

Abstract

The application discloses a method and system for active response load prediction based on a hybrid neural network, which adopts the features of real load data and generates time series data as input, adopts a CNN-LSTM neural network model for a load classifier, uses an AD response model to simulate the active response effect of the load, and adopts an RNN neural network model for load prediction, so that the application has the advantages of novel scheme and accurate result, and can perform load feature analysis and preprocessing on a basic data set containing different types of loads; a time series data set is created according to the preprocessed data set as a benchmark data set for load classification research, the application can predict the attribution label of the load and classify the load. The model extracts different types of load features through CNN, introduces a batch normalization layer for standardizing the output of the CNN layer, stabilizes the training process and accelerates convergence, and captures the dependency between time steps through LSTM, so that the output result of the classification model has high accuracy and reliability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of load classification and prediction, in particular to an active response load prediction method and system based on a hybrid neural network. BACKGROUND

[0002] With the increasing intelligence of the power system, the power demand response technology has gradually become an important means for stable operation of the power system. Active response load classification and prediction are key links in demand response. Through accurate classification and prediction of active response loads, reasonable scheduling and optimization of loads can be achieved, and the operation efficiency and stability of the power grid can be improved.

[0003] Traditional load classification methods are divided into unsupervised clustering and supervised classification. The former is based on various clustering methods such as k-means and fuzzy c-means (FCM), and the latter often uses artificial intelligence neural networks as a load classification method. The clustering algorithm is not suitable for dividing data types with many outliers, and often needs to specify the number of categories according to experience, lacking effective basis. Although neural networks have good results, such as BP neural networks, these classification methods only train the mapping relationship between load curves and category labels from a mathematical level, ignoring the time series correlation of load data. In addition, the classification efficiency of neural networks will be severely affected when facing massive data.

[0004] As for load prediction, the composition of the traditional power system is simple, and the prediction scenario is often for system-level or bus-level load. Therefore, the traditional load prediction method is mainly based on statistical analysis methods. However, with the rapid development and continuous evolution of the power system, especially the construction of new power systems, a large number of distributed new energy, electric vehicles and other new elements will be introduced to a great extent, and the load side of the power system will become more flexible and variable. In the face of various complex load influencing factors, traditional load prediction methods are difficult to accurately construct load patterns under the background of new power systems. Therefore, it is urgent to further improve the accuracy of load prediction under the background of new power systems to ensure the safe and stable operation of the power grid and the balance between power supply and demand. SUMMARY

[0005] The purpose of the present application is to provide an active response load prediction method and system based on a hybrid neural network, which solves the problem of inaccurate prediction of future load when multiple active response loads are mixed, and aims to improve the accuracy of load classification and prediction.

[0006] To achieve the above-mentioned purpose, the technical solutions adopted by the present application are as follows:

[0007] The active response load prediction method based on a hybrid neural network comprises the following steps:

[0008] Step S1, obtain the load basic data set and perform load feature analysis to obtain load features, and create a time series data set based on the load features as a benchmark data set for load classification research;

[0009] Step S2, construct a CNN-LSTM neural network model for load classification based on the time series data set;

[0010] Step S3, construct an AD response model for simulating the active response effect of the load;

[0011] Step S4, construct an RNN neural network model for load prediction; classify the to-be-predicted load data based on the CNN-LSTM neural network model, simulate and generate the active response result of the current load based on the AD response model, and predict the active response result data based on the RNN model to obtain the image of each type of load after classification and the curve comparison image of the predicted load and the real load after prediction.

[0012] Preferably, obtaining the load basic data set and performing load feature analysis to obtain load features specifically includes the following steps:

[0013] S11: Collect different types of load data, and add corresponding labels to different types of data;

[0014] S12: Group by label and date, calculate the statistical features of the load, and form a data frame;

[0015] S13: Extract the statistical features and labels in the data frame, standardize the features, and encode the labels into integers;

[0016] S14: Create a time series data set according to the standardized features, encoded labels, and specified time steps, and convert the input load data and encoded labels into time series data suitable for model input.

[0017] Preferably, the statistical features include mean, standard deviation, maximum value, and minimum value.

[0018] Preferably, constructing a CNN-LSTM neural network model for load classification based on the time series data set specifically includes the following steps:

[0019] S21: Dimension conversion, convert the load time series data into the input shape expected by the convolution layer;

[0020] S22: Set CNN layer 1 and batch normalization layer 1; CNN layer 1 is used to extract local features of the input sequence data and generate feature maps; batch normalization layer 1 is used to standardize the output of the convolution layer; select a ReLU activation function to introduce a nonlinear activation;

[0021] S23: Set the maximum pooling layer, down-sample the feature map, reduce the data size, and retain the main features;

[0022] S24: Set CNN layer 2 and batch normalization layer 2; further extract the output features of CNN layer 1 to generate higher-level feature maps; standardize the output of CNN layer 2, also select the ReLU activation function, and improve the model expression ability;

[0023] S25: Set the LSTM layer to process time series data and capture the dependency between time steps; take the output of the last time step of the LSTM as the summary of the entire sequence;

[0024] S26: Set the fully connected layer to map the output of the LSTM to the class score for the final classification decision.

[0025] Preferably, the AD response model for simulating the active response effect of the load comprises the following steps:

[0026] If the response model of the consumer is linear and time-invariant, a simple finite impulse response model is suitable:

[0027]

[0028] where k is the discrete time index. ad(k) is the standard AD response load, is a zero-mean independent and identically distributed error term for modeling random components, is the actual AD response load. is the response degree of the user at a certain time, which is a polynomial of the backward shift operator and is specifically expressed as:

[0029]

[0030] where, the term represents the estimation of the actual AD response according to the verified AD profile and the response degree of the user.

[0031] Preferably, the RNN neural network model for load prediction comprises the following steps:

[0032] S41: Select and preprocess the active response load data;

[0033] S42: Set RNN layer 1 and Dropout layer 1, RNN layer 1 is used to extract time series features and return the output of the entire sequence; Dropout layer 1 prevents overfitting;

[0034] ​S43: Set RNN layer 2 and Dropout layer 2, RNN layer 2 is used to further extract time series features, and the output of the last time step is returned; Dropout layer 2 prevents overfitting;

[0035] S44: Set a fully connected layer to map the output of the RNN to the target space to generate a predicted value;

[0036] S45: Train the RNN model and monitor the loss on the validation set; use the trained model to predict the test set and denormalize the results; finally generate the load prediction value and compare it with the true value.

[0037] Preferably, the filtering and preprocessing of the active response load data includes normalization, generation of load time series data set, and splitting of training set and test set.

[0038] The active response load prediction system based on a hybrid neural network includes a data collection and arrangement module, a load classification module, a load active response module, and a load prediction module.

[0039] The data collection and arrangement module obtains a load basic data set and performs load feature analysis to obtain load features, and creates a time series data set based on the load features as a benchmark data set for load classification research.

[0040] The load classification module constructs a CNN-LSTM neural network model for load classification based on the time series data set.

[0041] The load active response module constructs an AD response model for simulating the active response effect of the load.

[0042] The load prediction module constructs an RNN neural network model for load prediction; classifies the to-be-predicted load data based on the CNN-LSTM neural network model, simulates the active response result of the current load based on the AD response model, and predicts the active response result data based on the RNN model to obtain the image of each classified load and the curve comparison image of the predicted load and the true load after prediction.

[0043] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above-mentioned active response load prediction method based on a hybrid neural network when executing the computer program.

[0044] A computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the above-mentioned active response load prediction method based on a hybrid neural network.

[0045] Compared with the prior art, the application has the following beneficial technical effects:

[0046] The active response load prediction method based on the hybrid neural network adopts the features of real load data and generates time series data as input, the load classifier adopts a CNN-LSTM neural network model, the AD response model is used to simulate the active response effect of the load, and the RNN neural network model is used for load prediction, so that the application has the advantages of novel scheme and accurate result.

[0047] In the application, load feature analysis and preprocessing are performed on the basic data set containing different types of loads; and a time series data set is created according to the preprocessed data set as a benchmark data set for load classification research.

[0048] The application can predict the attribution label of the load and classify the load based on the CNN-LSTM network model. The model extracts different types of load features through CNN, introduces a batch normalization layer to standardize the output of the CNN layer and stabilize the training process and accelerate convergence, and captures the dependency relationship between time steps through LSTM, so that the output result of the classification model has high accuracy and reliability.

[0049] The application introduces an AD response model to simulate the active response effect of the load. The generation mode of AD information is determined; the AD profile is generated; the AD effect is simulated; the AD effect is added to the original data set; and according to the date and load information of the data set, the simulated AD effect is added to the corresponding time point, so as to obtain a data set containing AD information.

[0050] The application predicts the active response load after classification based on the RNN neural network model. The RNN model is trained, and the loss is monitored on the validation set; and MAPE is used as the standard for verifying the prediction accuracy of the model.

[0051] The application will help to solve the problem that it is difficult to accurately predict future load when multiple active response loads are mixed, improve the accuracy of load prediction under the background of demand response, and ensure the safe and stable operation of the power grid and the balance between power supply and demand. These aspects or other aspects of the application will be more clear and easy to understand in the following description of the embodiments. It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the application. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0053] Figure 1 This is a flowchart of the active response load classification and prediction method based on a hybrid neural network according to an embodiment of the present invention.

[0054] Figure 2 This is a model diagram of the CNN-LSTM hybrid neural network according to an embodiment of the present invention.

[0055] Figure 3 This is a CNN feature extractor according to an embodiment of the present invention.

[0056] Figure 4 This is an LSTM cell structure according to an embodiment of the present invention.

[0057] Figure 5 The figure shows the experimental results of active response load classification based on hybrid neural networks in an embodiment of the present invention.

[0058] Figure 6 This is a comparison chart of experimental results for active response load prediction based on hybrid neural networks according to an embodiment of the present invention. Detailed Implementation

[0059] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0060] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0061] The purpose of this invention is to address the shortcomings mentioned above by proposing an active response load classification and prediction method based on a hybrid neural network. To make the above-mentioned objectives, features, and advantages of this invention more apparent and understandable, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0062] The application provides a proactive response load prediction method based on a hybrid neural network, which comprises the following steps:

[0063] Step S1, constructing a basic data set for load classification prediction research.

[0064] This step can specifically include:

[0065] Step S11, obtaining basic load data from the OSF public database;

[0066] The load data set is obtained from the OSF public database, and the data set contains one-year time resolution of 30 minutes of power load data of different types of buildings in an industrial park. These data come from smart meters. It contains four load types: Public, Office, Residential, and Commercial. Each load type has 365 days of data. Each day's data contains 49 data points with a 30-minute time resolution. Each data point has two features: time and corresponding load.

[0067] Step S12, preprocessing the load data obtained in step S11 to complete the construction of the benchmark data set for load classification research;

[0068] Different types of data are labeled; group by label and date, and calculate the statistical characteristics (mean, standard deviation, maximum, minimum) of the load, form a data frame; extract the statistical characteristics and labels in the data frame, standardize the features and encode the labels into integers. Feature standardization will transform the feature data as follows:

[0069]

[0070] wherein, is the mean of the feature data, is the standard deviation of the feature data. After this processing, the distribution of the data is more concentrated in the range of mean 0 and standard deviation 1.

[0071] Create a time series data set according to the standardized features, encoded labels, and specified time steps. The specific process is as follows:

[0072]

[0073] wherein X is the feature data, Y is the corresponding label, load is the input load data, labels_encoded is the encoded label data, and time_steps is the specified time step. The application specifies the time step as 10.

[0074] The active response load classification and prediction method based on hybrid neural networks described in this invention further includes the following steps:

[0075] Step S2: Construct a CNN-LSTM neural network model for load classification.

[0076] Figure 2 This is a system framework diagram of the load classification based on the CNN-LSTM model used in this invention.

[0077] like Figure 2 As shown, the CNN-LSTM model consists of five parts: CNN layer 1, max pooling layer, CNN layer 2, LSTM layer, and fully connected layer. Figure 3 The diagram shows the convolutional neural network (CNN) structure used by the feature extractor.

[0078] Step S2 may specifically include:

[0079] Step S21: Dimension transformation, converting the load time series data obtained in step S1 into the input shape expected by the convolutional layer;

[0080] Step S22: Set up CNN layer 1 and batch normalization layer 1; Convolutional layer 1 is used to extract local features of the input sequence data and generate feature maps; Batch normalization layer 1 is used to normalize the output of the convolutional layer, stabilize the training process, and accelerate convergence; Select the ReLU activation function to introduce non-linear activation and improve the model's expressive power;

[0081] Batch Normalization (BN) algorithm standardizes intermediate computation data, accelerating network convergence and reducing the network's sensitivity to weight initialization. The algorithm is described below:

[0082]

[0083] In the above algorithm, batch-size is the size of each batch of sample data; For a minimum value (e.g.) )

[0084] avoid The denominator of the expression is 0; and These are the mean and variance of the data sample, respectively. For standard normalization intermediate variables; The parameters that the network needs to learn are the scaling factor and the translation factor, respectively.

[0085] To avoid gradient vanishing or exploding, the ReLU activation function is used to process the batch-normalized data. The ReLU activation function can be expressed as:

[0086]

[0087] Step S23: Set up a max pooling layer to downsample the feature map, reduce the data size, and retain the main features;

[0088] Step S24: Set up CNN layer 2 and batch normalization layer 2; further extract the output features of convolutional layer 1 to generate higher-level feature maps; normalize the output of convolutional layer 2 to further stabilize the training process and accelerate convergence; also select the ReLU activation function to improve the model's expressive power;

[0089] Step S25: Set up an LSTM layer to process the sequence data, capture the dependencies between time steps, and only take the output of the last time step of the LSTM as a summary of the entire sequence.

[0090] Ordinary recurrent neural networks (RNNs) cannot handle long-range dependencies when processing sequential data, and the vanishing gradient problem makes the model difficult to train. LSTM networks were proposed to address this issue. LSTM neural networks can solve the problem of RNNs handling long-range dependencies, fully utilize historical information, and have stronger adaptability in time series data analysis. The unit structure of LSTM is as follows: Figure 4 As shown.

[0091] Figure 4 middle: and These represent the input vector and hidden layer state value at time t, respectively. The memory unit is the memory of the neuron's state, used to record the current state. The input gate and output gate are used to read, output, and correct parameters. The forget gate is used to selectively forget and correct parameters based on the unit state from the previous time step. The calculation formula is as follows:

[0092]

[0093] In the formula: f, i, g, C, o represent the forget gate, input gate, alternative cell states to be updated, updated cell state, and output gate, respectively; W and b are the corresponding weight coefficient matrix and bias term, respectively. σ and tanh represent the sigmoid activation function and the hyperbolic tangent activation function, respectively.

[0094] Step S26: Set up a fully connected layer to map the output of the LSTM to the category score for the final classification decision.

[0095] The active response load classification and prediction method based on hybrid neural networks described in this invention further includes the following steps:

[0096] Step S3 involves simulating the active response effect of the load data based on the constructed AD response model, specifically including:

[0097] Step S31: Assuming the consumer response model is linear and time-invariant, it can be described using a simple finite impulse response model. The relationship with ad, that is ;

[0098] Step S32: Generate a specified number of AD information items, including (AD volume) (AD duration) and (The start time interval between two continuous AD products); randomly select AD volume in the interval [30, 100] kW and duration. In the interval [1, 4]h, Randomly generated within different ranges, taking values ​​of [1, 24]h and [12, 72]h respectively. An AD product is formed when a consumer may reduce or increase their power consumption based on a signal within a certain time period. Each load data point corresponds to a set of information;

[0099] Step S33: Simulate the effect of the AD response model and calculate the impact of AD on the load curve. Generate the basic part of the AD effect, in [ , Uniformly distributed within the range; the calculated energy recovery amplitude is equal to 30%, its duration is assumed to be (1 / 2). An energy recovery sequence is generated based on the magnitude and duration of energy recovery to simulate the energy recovery process after AD in practice. The energy recovery process occurs after a demand modification (i.e., AD effect), and may involve a modification with the opposite sign to the previous demand modification. The energy recovery sequence is then merged with the basic part of the AD effect; considerations are made regarding... The impact of this is addressed by padding the current AD effect sequence with a padding length of (Tint * 2). When the padded AD effect sequence is convolved with the polynomial B(q), the time interval between AD products can be more accurately considered; in the AD model... During the experiment, take This part is computed using convolution; a random noise part is generated, which is a simulated sequence of independent and identically distributed random numbers with zero mean. The noise is also added to obtain the final AD effect.

[0100] Step S34: For each load data point, add the original load data to the AD effect to obtain the load data after considering the AD effect, and add it to the dataset.

[0101] The active response load classification and prediction method based on hybrid neural networks described in this invention further includes the following steps:

[0102] Step S4, constructing an RNN neural network model for load forecasting, specifically including:

[0103] Step S41: Filter and preprocess the active response load data, including normalization, generating a load time series dataset, and splitting the training and test sets;

[0104] Step S42: Set up two RNN layers and a Dropout layer. The RNN layer is used to extract time series features and return the output of the entire sequence; the Dropout layer prevents overfitting; the first RNN layer has 128 output units and the second RNN layer has 64 output units, both using ReLU as the activation function; the Dropout layer has a dropout rate of 0.2.

[0105] Step S43: Set up a fully connected layer to map the RNN output to the target space and generate predicted values;

[0106] Step S44: Train the RNN model with 30 training epochs and a batch size of 8. Improve the model's performance and generalization ability by adjusting the learning rate and using dropout, and monitor the loss on the validation set. Use the trained model to predict the test set and denormalize the results. Finally, generate the load prediction values ​​and compare them with the true values.

[0107] A Recurrent Neural Network (RNN) is a type of neural network used to process sequential data. The basic structure of an RNN includes an input layer, hidden layers, and an output layer. In the hidden layers, recurrent connections exist between neurons, allowing information to be passed over time. Specifically, for the input at time t... Output of hidden layer It depends not only on the current input It also depends on the output of the hidden layer at the previous time step. This recurrent connection allows RNNs to capture temporal dependencies in sequential data. The mathematical expression for an RNN is as follows:

[0108]

[0109] in, , , These are the weight matrices from input to hidden layer, from hidden layer to hidden layer, and from hidden layer to output layer, respectively. , are the bias vectors of the hidden layer and the output layer, respectively, and f and g are the activation functions.

[0110] The active response load classification and prediction method based on hybrid neural networks described in this invention further includes the following steps:

[0111] Step S5, combining the CNN-LSTM classification model and the RNN prediction model, involves testing and verification, and also includes the following steps:

[0112] Step S51: Select the verification test dataset. In this experiment, we select the electricity load data of a residential building in an industrial park in Suzhou, China, with a time resolution of 30 minutes for one year (January 1, 2021 to December 31, 2021).

[0113] Step S52: Preprocess the data, resample the test data, and calculate the statistical characteristics of the load, such as the mean, standard deviation, maximum and minimum values, by day; fill missing values ​​with the mean; standardize the features of the test data using the standardizer of the previously trained data; create a time series dataset using the method used for training data and specify a time step of 10.

[0114] Step S53: Based on the CNN-LSTM model, predict the load data to be predicted, obtain the load label, and visualize the average load curves corresponding to different labels to more intuitively observe and analyze the classification results. The classified image is as follows: Figure 5 As shown;

[0115] Step S54: Simulate the active response effect of the classified load according to the method in step S3 to obtain load data with active response information.

[0116] Step S55: Predict the active response load data based on the RNN model. Obtain a curve comparison image of the predicted load and the actual load, as well as the prediction error. The curve comparison image after prediction is shown below. Figure 6 As shown.

[0117] A hybrid neural network-based active response load forecasting system includes a data acquisition and processing module, a load classification module, an active load response module, and a load forecasting module.

[0118] The data acquisition and processing module obtains the basic load dataset and performs load characteristic analysis to obtain load characteristics. Based on the load characteristics, a time series dataset is created as the benchmark dataset for load classification research.

[0119] The load classification module constructs a CNN-LSTM neural network model for load classification based on time series datasets.

[0120] The active load response module constructs an AD response model to simulate the effects of active load response.

[0121] The load forecasting module constructs an RNN neural network model for load forecasting; it classifies the load data to be predicted based on a CNN-LSTM neural network model, simulates and generates the active response results of the current load based on an AD response model, and predicts the active response results data based on the RNN model, obtaining images of various types of loads after classification and curve comparison images of predicted load and actual load.

[0122] In another embodiment of the present invention, a terminal device is provided, comprising a processor and a memory. The memory stores a computer program, the computer program including program instructions, and the processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), or field-programmable gate arrays (FPGAs). Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., are the computing and control core of the terminal. They are suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to realize the corresponding method flow or corresponding function. The processor described in this embodiment of the invention can be used to operate an active response load forecasting method based on a hybrid neural network, including the following steps: obtaining a load base dataset and performing load feature analysis to obtain load features; creating a time series dataset based on the load features as a benchmark dataset for load classification research; constructing a CNN-LSTM neural network model for load classification based on the time series dataset; constructing an AD response model for simulating the effect of active load response; constructing an RNN neural network model for load forecasting; classifying the load data to be predicted based on the CNN-LSTM neural network model; simulating and generating the active response result of the current load based on the AD response model; predicting the active response result data based on the RNN model; obtaining images of various types of loads after classification and curve comparison images of predicted load and actual load after prediction.

[0123] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory). This computer-readable storage medium is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0124] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the active response load forecasting method based on a hybrid neural network in the above embodiments. One or more instructions in the computer-readable storage medium are loaded and executed by the processor as follows: obtaining a load base dataset and performing load feature analysis to obtain load features; creating a time series dataset based on the load features as a benchmark dataset for load classification research; constructing a CNN-LSTM neural network model for load classification based on the time series dataset; constructing an AD response model for simulating the effect of active load response; constructing an RNN neural network model for load forecasting; classifying the load data to be predicted based on the CNN-LSTM neural network model; simulating and generating the active response results of the current load based on the AD response model; predicting the active response result data based on the RNN model; obtaining images of various types of loads after classification and a curve comparison image of the predicted load and the actual load after prediction.

[0125] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0126] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0127] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0128] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0129] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. An active response load forecasting method based on hybrid neural networks, characterized in that, Includes the following steps: S1. Obtain the basic load dataset and perform load feature analysis to obtain load features. Based on the load features, create a time series dataset as the benchmark dataset for load classification research. S2, Construct a CNN-LSTM neural network model for load classification based on time series datasets; S3, Construct an AD response model to simulate the effect of active load response; S4. Construct an RNN neural network model for load forecasting; classify the load data to be predicted based on the CNN-LSTM neural network model, simulate and generate the active response results of the current load based on the AD response model, and predict the active response results data based on the RNN model to obtain images of various types of loads after classification and curve comparison images of predicted load and actual load after prediction. Obtaining the basic load dataset and performing load characteristic analysis to obtain load characteristics specifically includes the following steps: S11: Collect different types of load data and add corresponding labels to different types of data; S12: Group by tag and date, calculate the statistical characteristics of load, and form a data frame; S13: Extract statistical features and labels from the data frame, standardize the features and encode the labels into integers; S14: Create a time series dataset based on the standardized features, encoded labels, and specified time steps, converting the input load data and encoded labels into time series data suitable for model input; the statistical features include mean, standard deviation, maximum value, and minimum value; Building a CNN-LSTM neural network model for load classification based on time series datasets includes the following steps: S21: Dimension transformation, converting the load time series data into the input shape expected by the convolutional layer; S22: Set up CNN layer 1 and batch normalization layer 1; CNN layer 1 is used to extract local features of the input sequence data and generate feature maps; batch normalization layer 1 is used to normalize the output of the convolutional layer; select the ReLU activation function to introduce non-linear activation; S23: Set a max pooling layer to downsample the feature map, reduce the data size, and retain the main features; S24: Set up CNN layer 2 and batch normalization layer 2; further extract the output features of CNN layer 1 to generate higher-level feature maps; normalize the output of CNN layer 2, and also select the ReLU activation function to improve the model's expressive power; S25: Set up an LSTM layer to process time series data, capture the dependencies between time steps, and take the output of the last time step of the LSTM as a summary of the entire sequence. S26: Set up a fully connected layer to map the output of the LSTM to the category score for the final classification decision.

2. The active response load forecasting method based on hybrid neural networks according to claim 1, characterized in that, The specific steps involved in constructing an Active Response (AD) model to simulate the effects of active load response are as follows: Assuming the consumer's response model is linear and time-invariant, then a simple finite impulse response model is appropriate: Where: k is the discrete-time index; ad(k) is the standard AD response load. They are zero-mean, independent and identically distributed Error term, used to model random components. This is the actual AD response load; It refers to the user's responsiveness at a certain moment, and is a backward shift operator. The polynomial of is specifically represented as: in, The item represents the response rate based on the validated AD profile and the user's response. To estimate the actual AD response.

3. The active response load forecasting method based on hybrid neural networks according to claim 1, characterized in that, The specific steps involved in constructing an RNN neural network model for load forecasting are as follows: S41: Filter and preprocess the active response load data; S42: Set up RNN layer 1 and Dropout layer 1. RNN layer 1 is used to extract time series features and return the output of the entire sequence; Dropout layer 1 prevents overfitting. S43: Set up RNN layer 2 and Dropout layer 2. RNN layer 2 is used to further extract time series features and return the output of the last time step; Dropout layer 2 prevents overfitting. S44: Set up a fully connected layer to map the output of the RNN to the target space and generate the predicted value; S45: Train the RNN model and monitor the loss on the validation set; use the trained model to make predictions on the test set and denormalize the results; finally, generate the load predictions and compare them with the actual values.

4. The active response load forecasting method based on a hybrid neural network according to claim 3, characterized in that, The filtering and preprocessing of active response load data includes normalization, generating load time series datasets, and splitting the training and test sets.

5. An active response load forecasting system based on a hybrid neural network for use in the method of claim 1, characterized in that, It includes a data acquisition and processing module, a load classification module, a proactive load response module, and a load forecasting module. The data acquisition and processing module obtains the basic load dataset and performs load characteristic analysis to obtain load characteristics. Based on the load characteristics, a time series dataset is created as the benchmark dataset for load classification research. The load classification module constructs a CNN-LSTM neural network model for load classification based on time series datasets. The active load response module constructs an AD response model to simulate the effects of active load response. The load forecasting module constructs an RNN neural network model for load forecasting; it classifies the load data to be predicted based on a CNN-LSTM neural network model, simulates and generates the active response results of the current load based on an AD response model, and predicts the active response results data based on the RNN model, obtaining images of various types of loads after classification and curve comparison images of predicted load and actual load.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the active response load prediction method based on a hybrid neural network as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the active response load prediction method based on a hybrid neural network as described in any one of claims 1 to 4.

Images