Adjustable load capacity evaluation method and system

Abstract

The invention discloses an adjustable load capacity evaluation method and system, and belongs to the field of adjustable load capacity evaluation. The adjustable load capability evaluation method comprises the following steps: acquiring load and environment data of a power grid, and preprocessing the load and environment data; constructing an MS-Inception-LSTM model, wherein the MS-Inception-LSTM model comprises an Inception module, an MHA module, an SE module, an AFF feature fusion module, an LSTM module and a full connection layer; the Inception module carries out feature extraction on load data, then the load data are respectively input into the MHA module and the SE module, output features are fused by the AFF feature fusion module, the LSTM module carries out modeling on the fused features, environmental data and load data time sequence features generated by the LSTM are fused in a full connection layer, and the environmental data and the load data time sequence features jointly act on a final load regulation capability prediction result; carrying out load capacity prediction after the model is trained; the method not only can capture local and global features in the load regulation capability, but also can effectively process the sequential relationship, and is suitable for complex load regulation capability evaluation tasks.

Description

Technical Field

[0001] The present invention belongs to the field of adjustable load capacity evaluation, and in particular relates to an adjustable load capacity evaluation method and system. Background Art

[0002] With the continuous development of the power system and the rapid access to new energy, the evaluation of load regulation capability has become an important task to ensure the safe and stable operation of the power grid. Load regulation capability reflects the potential of the power grid to cope with fluctuations in power demand and maintain a balance between supply and demand, and is of great significance for formulating reasonable operation plans, optimizing power resource allocation, and improving system flexibility.

[0003] At present, load capacity assessment methods are mainly concentrated in two major areas: statistical analysis and machine learning models. Traditional statistical analysis methods perform simple feature extraction and prediction based on historical load data. Although they can provide certain references, they perform poorly when dealing with complex load change patterns. Machine learning methods have made significant progress in mining deep features in load data in a data-driven manner. However, there are still shortcomings in practical applications. First, many models are difficult to take into account both local change characteristics and global change trends at the same time, resulting in limited prediction accuracy; second, load data has significant time correlation, and existing methods are insufficient in modeling complex time series relationships, making it difficult to fully characterize the load change law; finally, the correlation between different features has not been effectively utilized, resulting in limited comprehensive modeling capabilities of the model for complex load characteristics. To this end, an adjustable load capacity assessment method and system is proposed. Summary of the invention

[0004] In view of the deficiencies in the prior art, the purpose of the present invention is to provide an adjustable load capacity evaluation method and system to solve the problems in the prior art.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A method for evaluating adjustable load capacity comprises the following steps:

[0007] Collect power grid load data and environmental data, and pre-process the load data and environmental data;

[0008] Build the MS-Inception-LSTM model and train it using the preprocessed load data and environmental data;

[0009] After preprocessing, the collected load data and environmental data are input into the trained MS-Inception-LSTM model to output the predicted value of load regulation capacity;

[0010] The MS-Inception-LSTM model includes: an Inception module, an MHA module, an SE module, an AFF feature fusion module, an LSTM module and a fully connected layer; wherein the Inception module extracts features of the preprocessed load data and then inputs them into the MHA module and the SE module respectively; the AFF feature fusion module fuses the features output by the MHA module and the SE module; the LSTM module models the fused features to capture the long-term dependent relationship in the time series; the preprocessed environmental data and the load data time series features generated by the LSTM are fused in the fully connected layer, and act together on the final load regulation capacity prediction result.

[0011] Furthermore, the load data includes: active power data and reactive power data; the environmental data includes: temperature, humidity, wind direction and wind speed.

[0012] Furthermore, the load data and environmental data are preprocessed to fill in the missing data, and a two-stage data filling method based on a K-nearest neighbor joint expectation maximization algorithm is used.

[0013] Furthermore, the Inception module extracts multi-scale features of the input data by using convolution kernels and pooling operations of different sizes in parallel.

[0014] Furthermore, the MHA module captures the feature correlations of different time steps in parallel through multiple attention heads, and the output is a weighted combination of time series features.

[0015] Furthermore, the process of the SE module processing input features is as follows: performing global average pooling on each channel to capture the global features of the channel; modeling the global features of each channel through two fully connected layers, generating adaptive weight coefficients, applying the generated weights to each channel, and performing weighted adjustment on the original features.

[0016] Furthermore, the loss function of the MS-Inception-LSTM model is:

[0017]

[0018] Among them, y i is the true value, is the predicted value, w i is the weight factor.

[0019] An adjustable load capacity assessment system, comprising:

[0020] Data acquisition and processing module: collects power grid load data and environmental data, and pre-processes the load data and environmental data;

[0021] Model building and training module: Build the MS-Inception-LSTM model and train it using the preprocessed load data and environmental data;

[0022] And, the load capacity assessment module: after preprocessing the collected load data and environmental data, it inputs the trained MS-Inception-LSTM model and outputs the load regulation capacity prediction value;

[0023] The MS-Inception-LSTM model includes: an Inception module, an MHA module, an SE module, an AFF feature fusion module, an LSTM module and a fully connected layer; wherein the Inception module extracts features of the preprocessed load data and then inputs them into the MHA module and the SE module respectively; the AFF feature fusion module fuses the features output by the MHA module and the SE module; the LSTM module models the fused features to capture the long-term dependent relationship in the time series; the preprocessed environmental data and the load data time series features generated by the LSTM are fused in the fully connected layer, and act together on the final load regulation capacity prediction result.

[0024] A computer storage medium stores a readable program, which can execute the adjustable load capacity evaluation method mentioned in the claim when the program is run.

[0025] An electronic device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus;

[0026] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above-mentioned adjustable load capacity evaluation method.

[0027] Beneficial effects of the present invention:

[0028] 1. The model of the present invention combines the advantages of the Inception module, MHA attention mechanism, SE attention mechanism and LSTM, can efficiently extract the features of load data, and enhance the model's capture of time dependency and feature importance through the MHA mechanism and SE mechanism, and further model the timing dependency using LSTM. Finally, an excellent evaluation result of adjustable load capacity is obtained through the fully connected layer; it can capture the local and global features in the load regulation capability, and effectively process the timing relationship, and is suitable for complex load regulation capability evaluation tasks.

[0029] 2. In order to improve the accuracy and efficiency of load capacity prediction, the present invention adopts a two-stage data filling method (KNN-EM): K-Nearest Neighbor (KNN) algorithm combined with Expectation Maximization (EM) algorithm. Aiming at the problem that high-frequency sampled power load data (including active power and reactive power) and multi-dimensional environmental parameters (such as temperature, humidity, wind speed, etc.) are prone to packet loss during long-distance transmission, this method effectively fills in missing data and ensures the integrity and consistency of input data. First, the K-nearest neighbor algorithm uses neighborhood information to preliminarily fill in missing values ​​to ensure that local characteristics are retained; then, the EM algorithm further optimizes the expected iteration and combines the global statistical characteristics to accurately fill in and avoid data jumps and outlier interference. This method significantly reduces the impact of noise on model training and improves data quality while retaining the periodicity and time series characteristics of load data. The filled high-integrity data provides a more comprehensive learning basis for the model, significantly reduces prediction errors, enhances robustness in complex environments, and ultimately provides an efficient and accurate solution for load capacity assessment tasks.

[0030] 3. The present invention inputs environmental data into the model after the LSTM module, and does not participate in time series modeling; as a set of static features, environmental data and the time series features of load data generated by LSTM are fused in the subsequent fully connected layer, and act together on the final load regulation capacity prediction results. This design processes load data and environmental data separately, respectively mining the complex dynamic characteristics of load data and the global background information of environmental data, and at the same time fuses the two types of data in the final stage, enhancing the overall prediction ability of the model. This method of separate modeling and late fusion not only reduces computational redundancy, but also effectively avoids the interference of environmental data on the modeling of load data time series features. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0032] Figure 1 It is the framework flow chart of the MS-Inception-LSTM model of the present invention;

[0033] Figure 2 It is the MHA module framework flow chart of the present invention;

[0034] Figure 3 It is the SE module framework flow chart of the present invention. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] Example 1

[0037] A method for evaluating adjustable load capacity comprises the following steps:

[0038] S1, collecting power grid load data and environmental data, and preprocessing the load data and environmental data;

[0039] The grid load data is a multi-dimensional time series, and the data of each time step contains several features; the grid load data includes: active power data, reactive power data; environmental data includes: temperature, humidity, wind direction and wind speed;

[0040] In order to ensure that the input data of the load regulation capacity evaluation model is comprehensive and high-quality, the present invention strictly considers the synchronization and comprehensiveness of power load data and related environmental parameters during the data collection process. The specific process is as follows:

[0041] Load data collection: The load data collected in this study include active power and reactive power data, which are obtained through the intelligent monitoring equipment of the power grid dispatching system. To ensure the timeliness and accuracy of the data, the sampling period is set to 1 minute, so that 60 data points can be obtained within 1 hour. In order to build a prediction model, the load data of the past week (ie 7 days) is used as input each time to predict the load regulation capacity of the next day (the 8th day). After that, a rolling prediction mechanism is formed: the data from the 2nd to 8th days is used to predict the 9th day, and so on. The load data input each time contains 2×7×24×60=20160 data points, covering the full time domain variation characteristics of the power load.

[0042] Environmental data collection: In order to fully reflect the external environmental factors that affect the load regulation capability, the present invention simultaneously collects four key environmental parameters, including temperature, humidity, wind direction and wind speed. Environmental data is collected through accurate and reliable meteorological sensors, and the data sampling period is consistent with the load data, which is 1 minute, thereby ensuring the consistency of the two types of data in the time dimension. The introduction of environmental data helps the model capture the impact of external environmental changes on the load regulation capability and improve the accuracy and practicality of the prediction.

[0043] The preprocessing of load data and environmental data is to fill the missing data: the missing data is filled by using the two-stage data filling method (KNN-EM) of K-Nearest Neighbor (KNN) algorithm combined with expectation maximization (EM) algorithm.

[0044] In the process of collecting and remotely transmitting load data and environmental data, there is a certain risk of data packet loss due to the short sampling period (1 minute) and large amount of data, which leads to partial data loss. In order to ensure data integrity and avoid the impact of missing data on model training and prediction, this paper adopts a two-stage data filling method based on the K-nearest neighbor joint expectation maximization (KNN-EM) algorithm. The specific processing flow is as follows:

[0045] Stage 1: K-nearest neighbor (KNN) filling

[0046] 1) Similarity measurement

[0047] For the time point t where the missing value is located, the observations at the adjacent time points constitute a feature vector X = {x1, x2, ...x n}, select the K neighboring points that are most similar to time point t through similarity measurement (Euclidean distance):

[0048]

[0049] Where: x ij represents the value of the jth dimension at the i-th time point, x tj Indicates the value of the missing time point (initially filled with the field mean or 0).

[0050] 2) Weight calculation

[0051] Calculate the weight w of each neighboring point according to the distance i :

[0052]

[0053] 3) Missing value estimation

[0054] Use the selected K neighboring points to average the missing values:

[0055]

[0056] in, is the initial fill value.

[0057] Phase 2: Expectation-Maximization (EM) Optimization

[0058] 4) Data initialization

[0059] The data matrix after initial filling is recorded as X = {x ij}, where some missing values ​​have been filled by KNN.

[0060] 5) Step E (expected calculation)

[0061] Based on the current data matrix X and the data distribution assumption, calculate the conditional expectation of missing values:

[0062]

[0063] Among them, μ j is the mean of the jth dimension, ∑ oj is the covariance between observed and missing values, ∑ oo is the covariance matrix of the observed values, x o is the observed value vector.

[0064] 6) M step (parameter update)

[0065] Update the model parameters (mean and covariance matrix) using the expected calculation results:

[0066]

[0067] 7) Iterative Update

[0068] Repeat the E and M steps until the following convergence conditions are met:

[0069] ||X (k+1) -X (k) ||<ε

[0070] Among them, ε is the preset convergence threshold, X (k) is the padded data matrix for the kth iteration.

[0071] KNN filling provides a reasonable initial value for the EM algorithm, reducing the error caused by random initialization. The EM algorithm can further improve the accuracy of missing value filling through iterative optimization. This method combines local and global information to ensure the efficiency and reliability of data filling.

[0072] S2, build the MS-Inception-LSTM model and use the preprocessed load data and environmental data to train the MS-Inception-LSTM model;

[0073] like Figure 1As shown in the figure, the MS-Inception-LSTM model includes: Inception module, MHA (Multi-Head Attention, multi-head attention mechanism) module, SE (Squeeze-and-Excitation) module, AFF (Attention Feature Fusion) feature fusion module, LSTM module and fully connected layer; Among them, the Inception module extracts features of the preprocessed load data, and then inputs them into the MHA module and SE module respectively. The AFF feature fusion module fuses the features output by the MHA module and the SE module. The LSTM module models the fused features to capture the long-term dependent relationship in the time series. The preprocessed environmental data and the load data time series features generated by LSTM are fused in the fully connected layer, and work together on the final load regulation capacity prediction result.

[0074] 1) Inception module

[0075] The parallel convolution structure of the Inception module is used to extract multi-scale features from the input data (preprocessed data); specifically, the Inception module extracts multi-scale features of the input data by using convolution kernels of different sizes (1x1, 3x3, 5x5) and pooling operations in parallel. This multi-branch structure can effectively capture spatial information at different levels, and take local details and global features into consideration at the same time. Finally, the output features of each branch are concatenated and fused to form a richer feature representation. This multi-scale feature extraction method can improve the model's ability to express the features of the input data and provide high-quality feature input for subsequent processing steps.

[0076] 2) MHA module

[0077] It is used to capture the dependencies between different time steps of load data and enhance the model's ability to model global dependencies in input data. The features extracted by the Inception module are input into the MHA module. The MHA module uses multiple attention heads to capture the feature correlations of different time steps in parallel, helping the model to focus on time series data from different angles. The output is a weighted combination of time series features, capturing global time dependencies.

[0078] like Figure 2 As shown in Figure 2, the process of MHA module processing input features includes:

[0079] Step 1: Map the input features into Query (Q), Key (K), and Value (V) vectors;

[0080] The input features are first mapped into three sets of vectors: Query (Q), Key (K) and Value (V); these vectors are used to calculate the correlation between each element in the input sequence; the input vectors are transformed through three different sets of linear transformations to obtain the corresponding Q, K, and V matrices.

[0081] Step 2: Calculate a single self-attention module;

[0082] For each attention head, the dot product between Query and Key is used to calculate the similarity to measure the relevance of each element in the sequence. The specific formula is:

[0083]

[0084] Among them, d k is the dimension of the Key vector, Used for scaling to avoid gradient instability caused by too large values, and normalize the similarity through the softmax function to ensure that the attention weight is between 0 and 1.

[0085] Step 3: Multiple heads calculate attention in parallel;

[0086] The key to MHA is to perform multiple attention calculations in parallel. The input features are passed through multiple independent attention heads (h), each of which captures features from a different perspective (subspace). These different heads can capture the complex relationships and interdependencies between different parts of the input features.

[0087] Step 4: Splice multiple results;

[0088] The output results of all attention heads will be spliced ​​together after the calculation is completed to form an integrated global feature representation.

[0089] Step 5: Linear transformation output;

[0090] The concatenated multi-head attention results are mapped through a linear transformation (fully connected layer) to obtain the final attention output. This step ensures that the features of different heads are reasonably combined to form a holistic understanding of the input sequence.

[0091] 3) SE module

[0092] According to the importance of different feature channels, the weights of feature channels are dynamically adjusted. Specifically, the SE module compresses the input feature map from a global perspective, generates adaptive channel weights, and weights each channel to enhance channel attention. The SE mechanism can effectively improve the model's ability to identify the importance of feature channels, allowing the model to pay more attention to feature channels that contribute to the task and ignore irrelevant or redundant channel features.

[0093] like Figure 3 As shown, the process of SE module processing input features includes:

[0094] Input features: Assume that the input feature map is X and its dimension is (B, C, H, W), where B is the batch size, C is the number of channels, and H and W are the height and width of the feature map.

[0095] Step 1: Compression;

[0096] Through the global average pooling operation, the global information of each channel is calculated. Specifically, for channel c, its average activation value on the entire feature map is calculated:

[0097]

[0098] Among them, z c It is a compressed representation of the c channel.

[0099] Step 2: Motivation;

[0100] The compressed feature z is passed through two layers of fully connected networks (usually using ReLU activation function and Sigmoid activation function) to generate channel weights:

[0101] s c =σ(FC2(ReLU(FC1(z))))

[0102] Among them, σ represents the Sigmoid function, and FC1 and FC2 are two fully connected layers.

[0103] Step 3: Weighting;

[0104] Apply the generated weight s to each channel to perform weighted adjustment on the original features:

[0105]

[0106] in, It is the weighted feature map.

[0107] Through the above steps, the SE module can adaptively adjust the importance of each channel, thereby enhancing the network's ability to represent features.

[0108] 4) AFF feature fusion module

[0109] The features output by the MHA and SE modules are fused, combining the feature information extracted by two different attention mechanisms; specifically:

[0110] The AFF feature fusion module achieves more intelligent feature fusion by introducing the attention mechanism, significantly improving the fusion effect; the AFF feature fusion module dynamically adjusts the contribution of each feature by weighting the different input features, enhancing the model's attention to key features and reducing the interference of redundant information. At the same time, the AFF feature fusion module can comprehensively consider the time dependency of the input features and the importance of feature channels to ensure that the most valuable information is retained during the feature fusion process; the fused features provide high-quality input for subsequent time series modeling (LSTM), thereby further improving the overall performance and prediction accuracy of the model.

[0111] 5) LSTM module

[0112] The fused features are modeled through LSTM to capture the long-term dependency relationship in the time series and further improve the prediction accuracy of load regulation capability. Specifically:

[0113] The fused features are input into LSTM. LSTM has memory units that can capture dependencies between long time steps and is particularly suitable for time series data. The output of LSTM is the prediction result or feature representation of the time series.

[0114] 6) Fully connected layer

[0115] The preprocessed environmental data and the time series features of the load data generated by LSTM are fused in the fully connected layer and work together on the final load regulation capacity prediction results.

[0116] Environmental data is placed directly after the LSTM module, which means that it does not need time series modeling and is only used as a set of static features (features that change slowly or are basically constant in time). This is because environmental data can include temperature, humidity, etc., which have a greater impact on load but do not change frequently or show complex time dependencies.

[0117] In the present invention, the environmental data and the load data are fused after LSTM for the following considerations:

[0118] 1) Separation of time series and static features: LSTM has fully modeled the time series of load data. At this time, there is no need to introduce environmental data into time series modeling to avoid interfering with the dynamic learning of load data. Environmental data directly participates in subsequent fusion, which helps to maintain its original information.

[0119] 2. Emphasize environmental data as supplementary information: Environmental data only plays a supporting role, and its main purpose is to provide an external background explanation for load changes. Introducing it after LSTM allows the model to focus more on using environmental features as additional conditions to correct the load prediction results, rather than affecting the main sequence modeling process.

[0120] The fully connected layer receives the temporal features of the load data and the static features of the environmental data to achieve the final fusion and prediction. This design can improve the expressiveness of the model by fusing the two types of data in the final stage, allowing the environmental data to conditionally modify the prediction results. This avoids the interference of environmental data on the time modeling path of the load data. The significance of the fusion of the two types of data is:

[0121] 1) Separate modeling of time series and static data: The temporal characteristics of load data are fully modeled by LSTM, and environmental data serves as additional global background information and directly participates in result prediction.

[0122] 2) Reduce computational complexity: If the environmental data does not have obvious time series characteristics, placing it directly after LSTM can avoid redundant calculations while retaining the original information.

[0123] 3) Clearer information flow: The model is modeled separately according to the dynamic characteristics of load data and the static characteristics of environmental data, and then the two are integrated in the final stage. This design embodies a clear information processing logic.

[0124] This design processes load data and environmental data separately, respectively mining the complex dynamic characteristics of load data and the global background information of environmental data, and at the same time fusing the two types of data in the final stage to enhance the overall prediction ability of the model. This method of separate modeling and post-fusion not only reduces computational redundancy, but also effectively avoids the interference of environmental data on the modeling of load data time series characteristics. The significance of the model is to make full use of the characteristics of different data types to improve the accuracy and stability of load regulation capacity prediction.

[0125] 7) Training of MS-Inception-LSTM model

[0126] The preprocessed load data and environmental data obtained in S1 are divided into a training set, a validation set, and a test set; the model is trained end-to-end using the back propagation algorithm, and the model parameters are optimized using the training set data;

[0127] Among them, in the present invention, a weighted MAPE loss function (Weighted MAPE, WMAPE) is designed as the loss function of the MS-Inception-LSTM model. The formula of the weighted MAPE loss function is as follows:

[0128]

[0129] Where: y i is the true value, is the predicted value, w i It is a weight factor that controls the contribution of each data point to the total error and is adjusted according to actual business needs.

[0130] Weight factor w i The design can be flexibly adjusted to meet the needs of different scenarios. The present invention designs a weight based on volatility (gradient): if the load changes drastically during a period of time, these time points are given a higher weight:

[0131] w i =|y i -y i-1 |

[0132] This design emphasizes the performance of the model during periods of fluctuation. By introducing weight factors into the loss function, the model dynamically adjusts the errors at different time points, which is more in line with the needs of actual load forecasting. It also focuses on optimizing key scenarios: such as during peak power consumption or when load fluctuations are large, the model will give priority to reducing the errors in these key scenarios. In addition, it can be extended to other load forecasting tasks to adapt to different business needs.

[0133] After training, the validation set and test set are used to evaluate the performance of the model, and the mean absolute percentage error (MAPE) is used as the evaluation indicator.

[0134] The expression of MAPE is:

[0135]

[0136] Among them, y t is the true value, is the predicted value and n is the total number of samples.

[0137] S3, after preprocessing the collected load data and environmental data, input them into the trained MS-Inception-LSTM model and output the predicted value of load regulation capacity.

[0138] In addition, it should be noted that:

[0139] 1) The present invention combines the Inception module with LSTM to effectively solve the challenges between feature extraction and time series modeling; specifically: when processing time series data, a single feature extraction method may not be able to capture the complexity and diversity of the data. Although LSTM performs well in time series data modeling, it has a strong dependence on features. By using the Inception module to extract multi-scale and multi-dimensional features, it can reduce information redundancy and the loss of important features, and provide a more comprehensive input for LSTM. In time series data, the time dependence of features is often very strong, and traditional methods are difficult to effectively capture long-term and short-term dependencies. The gating mechanism of LSTM can handle these dependencies, and combined with the rich features extracted by the Inception module, it will significantly improve the model's understanding and modeling capabilities of time dependence. In addition, the flexibility and versatility of the model enable it to adapt to different types of data and application scenarios, have strong robustness, and can effectively evaluate adjustable load capacity in complex environments. Finally, through this combination, the decision-making process of the model is more interpretable, and users can more clearly understand how the model makes decisions based on the extracted features, which improves its credibility in practical applications.

[0140] 2) The present invention uses and integrates the two attention mechanisms, MHA and SE, in parallel to ensure that the model can capture key information more comprehensively. Specifically, first, MHA focuses on capturing the global relationship between features, while the SE mechanism focuses on improving the channel representation of important features. Using these two mechanisms in parallel can effectively combine the advantages of both and ensure comprehensive attention to features. Secondly, MHA is good at capturing long-range dependencies in time series data, while the SE mechanism emphasizes the importance of feature channels. Parallel use can achieve a better balance between time series and spatial information. Finally, this parallel structure enables the model to have higher flexibility and adaptability when processing different types of data, and can better cope with various complex inputs.

[0141] Based on similar inventive concepts, an embodiment of the present invention further provides a computer storage medium storing a readable program, which can execute the above-mentioned adjustable load capacity evaluation method when the program is running.

[0142] Based on similar inventive concepts, an embodiment of the present invention provides an electronic device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus;

[0143] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above-mentioned adjustable load capacity evaluation method.

[0144] Based on similar inventive concepts, an embodiment of the present invention further provides a computer program product, including computer instructions, wherein the computer instructions instruct a computing device to execute operations corresponding to the above-mentioned adjustable load capacity evaluation method.

[0145] Example 2

[0146] In this embodiment, an adjustable load capacity evaluation system is proposed, comprising:

[0147] Data acquisition and processing module: collects power grid load data and environmental data, and pre-processes the load data and environmental data;

[0148] Model construction and training module: Construct the MS-Inception-LSTM model and use the preprocessed load data and environmental data to train the MS-Inception-LSTM model; the MS-Inception-LSTM model includes: Inception module, MHA module, SE module, AFF feature fusion module, LSTM module and fully connected layer; among them, the Inception module extracts features from the preprocessed load data and then inputs them into the MHA module and SE module respectively, the AFF feature fusion module fuses the features output by the MHA module and the SE module, the LSTM module models the fused features to capture the long-term dependent relationship in the time series, and the preprocessed environmental data and the load data time series features generated by LSTM are fused in the fully connected layer, which work together on the final load regulation capacity prediction result.

[0149] And, the load capacity evaluation module: after preprocessing the collected load data and environmental data, it is input into the trained MS-Inception-LSTM model to output the load regulation capacity prediction value.

[0150] The method of the present invention may be implemented in hardware, firmware, or as software or computer code that may be stored in a recording medium (such as a CDROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code that is originally stored in a remote recording medium or a non-temporary machine-readable medium downloaded over a network and will be stored in a local recording medium, so that the method described herein may be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that a computer, processor, microprocessor controller, or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code, and when the software or computer code is accessed and executed by a computer, processor, or hardware, the method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the method shown herein, the execution of the code converts the general-purpose computer into a dedicated computer for executing the method shown herein.

[0151] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A method for evaluating adjustable load capacity, characterized in that: The following steps are involved: Collect power grid load data and environmental data, and pre-process the load data and environmental data; Build the MS-Inception-LSTM model and train it using the preprocessed load data and environmental data; After preprocessing, the collected load data and environmental data are input into the trained MS-Inception-LSTM model to output the predicted value of load regulation capacity; The MS-Inception-LSTM model includes: an Inception module, an MHA module, an SE module, an AFF feature fusion module, an LSTM module and a fully connected layer; wherein the Inception module extracts features of the preprocessed load data and then inputs them into the MHA module and the SE module respectively; the AFF feature fusion module fuses the features output by the MHA module and the SE module; the LSTM module models the fused features to capture the long-term dependent relationship in the time series; the preprocessed environmental data and the load data time series features generated by the LSTM are fused in the fully connected layer, and act together on the final load regulation capacity prediction result.

2. The adjustable load capacity evaluation method according to claim 1, characterized in that: The load data includes: active power data and reactive power data; the environmental data includes: temperature, humidity, wind direction and wind speed.

3. The adjustable load capacity evaluation method according to claim 2, characterized in that: The load data and environmental data are preprocessed to fill in the missing data, and a two-stage data filling method based on a K-nearest neighbor joint expectation maximization algorithm is used.

4. The adjustable load capacity evaluation method according to claim 1, characterized in that: The Inception module extracts multi-scale features of input data by using convolution kernels of different sizes and pooling operations in parallel.

5. The adjustable load capacity evaluation method according to claim 1, characterized in that: The MHA module captures the feature correlations of different time steps in parallel through multiple attention heads, and the output is a weighted combination of time series features.

6. The adjustable load capacity evaluation method according to claim 1, characterized in that: The process of the SE module processing input features is as follows: global average pooling is performed on each channel to capture the global features of the channel; The global features of each channel are modeled through two fully connected layers to generate adaptive weight coefficients, which are then applied to each channel to perform weighted adjustment on the original features.

7. The adjustable load capacity evaluation method according to claim 1, characterized in that: The loss function of the MS-Inception-LSTM model is: Among them, y i is the true value, is the predicted value, w i is the weight factor.

8. An adjustable load capacity assessment system, characterized in that: include: Data acquisition and processing module: collects power grid load data and environmental data, and pre-processes the load data and environmental data; Model building and training module: Build the MS-Inception-LSTM model and train it using the preprocessed load data and environmental data; And, the load capacity assessment module: after preprocessing the collected load data and environmental data, it inputs the trained MS-Inception-LSTM model and outputs the load regulation capacity prediction value; The MS-Inception-LSTM model includes: an Inception module, an MHA module, an SE module, an AFF feature fusion module, an LSTM module and a fully connected layer; wherein the Inception module extracts features of the preprocessed load data and then inputs them into the MHA module and the SE module respectively; the AFF feature fusion module fuses the features output by the MHA module and the SE module; the LSTM module models the fused features to capture the long-term dependent relationship in the time series; the preprocessed environmental data and the load data time series features generated by the LSTM are fused in the fully connected layer, and act together on the final load regulation capacity prediction result.

9. A computer storage medium storing a readable program, characterized in that: When the program is running, it can execute the adjustable load capacity evaluation method described in any one of claims 1 to 7.

10. An electronic device, characterized in that: include: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the adjustable load capacity evaluation method as described in any one of claims 1-7.

Images