Fuzzy power consumption prediction method and system based on multi-scale dual supervision driving

Abstract

The application belongs to the technical field of power consumption prediction, and discloses a fuzzy power consumption prediction method and system based on multi-scale double supervision driving, which comprises the following steps: multi-scale feature reconstruction is performed on data samples, reconstruction error is calculated, sample structure guide indexes are obtained by aggregating the reconstruction error, and the samples are screened in combination with structure guide indexes and entropy contribution; the screened samples are subjected to fuzzy division, the membership degrees of all samples are arranged in order to obtain a fuzzy membership degree matrix; structure supervision signals and target supervision signals are fused, and the membership degree matrix is adjusted by using the fused guide signals; the membership degrees in the new membership degree matrix are used as weights to perform weighted summation on the prediction results of each local regression model to obtain the final prediction result. The application designs a fuzzy division method combining a multi-scale structure adaptive mechanism with structure and target double supervision signals, and improves the modeling accuracy of the model for complex energy consumption data, the system adaptive ability and the sample utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of electricity consumption forecasting technology, and in particular to a fuzzy electricity consumption forecasting method and system based on multi-scale dual-supervision driving. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In intelligent electricity consumption behavior analysis and power load forecasting, existing machine learning and fuzzy modeling methods have the following main problems when dealing with data that is structurally complex, behaviorally diverse, and exhibits significant energy consumption fluctuations:

[0004] Lack of structural adaptability: Traditional models struggle to dynamically identify multi-scale structural differences among users in different time periods and regions, resulting in coarse segmentation and unstable models.

[0005] Lack of target feedback mechanism: Existing fuzzy partitioning methods usually only divide based on input features or distribution, ignoring model prediction error and target orientation, resulting in the model convergence direction being inconsistent with the real task.

[0006] Fuzzy weights are fixed and difficult to adjust: Once fuzzy membership is generated, it is no longer updated and cannot be adjusted based on actual residuals or modeling performance, resulting in insufficient system flexibility and interpretability.

[0007] Redundant samples were not removed: There are a large number of samples in the data that have low information content or repetitive structure. These samples were not effectively screened, which reduced the modeling efficiency and made it easy to overfit. Summary of the Invention

[0008] To address the aforementioned issues, this invention proposes a fuzzy electricity consumption prediction method and system based on multi-scale dual supervision. It designs a fuzzy partitioning method that combines a multi-scale structural adaptive mechanism with dual supervision signals of structure and target, thereby improving the model's accuracy in modeling complex energy consumption data, the system's adaptive capability, and the efficiency of sample utilization.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] In a first aspect, the present invention provides a fuzzy electricity consumption prediction method based on multi-scale dual supervision, comprising the following steps:

[0011] Obtain users' historical electricity consumption data;

[0012] Multi-scale feature reconstruction is performed on the acquired data samples, the reconstruction error is calculated, the reconstruction errors at each scale are aggregated to obtain the sample structure guidance index, the entropy contribution of the data samples is calculated, and the data samples are screened by combining the sample structure guidance index and the entropy contribution.

[0013] The filtered samples are fuzzy partitioned to determine the membership degree of each sample to each cluster center. The membership degrees of all samples are arranged in order to obtain the fuzzy membership degree matrix.

[0014] The structural supervision signal is generated based on the sample structure guidance index. The confidence level is constructed using the unsupervised model residuals. The target supervision signal is obtained based on the confidence level. The structural supervision signal and the target supervision signal are fused to obtain the fused guidance signal. The membership matrix is ​​adjusted using the fused guidance signal to obtain a new membership matrix.

[0015] For each fuzzy subset, a local regression model is trained, and the membership degrees in the new membership matrix are used as weights to perform a weighted sum of the prediction results of each local regression model to obtain the final prediction result.

[0016] As an alternative implementation method, the entropy contribution of the data samples is calculated, specifically as follows:

[0017] Calculate the variance of the data sample and select the sample with the largest variance. Each feature dimension constitutes a projection subspace. A joint histogram of the projection subspace is constructed to estimate the distribution and calculate the entropy. For each sample, it is removed from the dataset and the entropy of the dataset is re-estimated. The entropy contribution of the data sample is calculated by combining the original entropy of the dataset with the re-estimated entropy.

[0018] As an alternative implementation method, the screened samples are subjected to fuzzy segmentation, specifically as follows:

[0019] We randomly initialize C cluster centers. For each sample, we calculate its distance to each cluster center. In each iteration, we update the membership degree of the sample to each category based on the distance of the sample to the cluster center, and at the same time update the cluster centers.

[0020] As an alternative implementation method, the target supervision signal is:

[0021]

[0022] in, For confidence level, and It refers to adjusting parameters.

[0023] As an alternative implementation method, the structural monitoring signal is:

[0024]

[0025] in, Indicators guiding the sample structure. and It is an adjustable parameter.

[0026] As an alternative implementation method, the membership matrix is ​​adjusted using the fusion guidance signal, specifically as follows:

[0027] Multiply the fusion guidance signal value corresponding to the sample by the membership degree of the sample to which it belongs in the membership matrix to obtain a weighted matrix. Then, perform a normalization operation on the weighted matrix to obtain a new membership matrix.

[0028] Secondly, the present invention provides a fuzzy electricity consumption prediction system based on multi-scale dual-supervision driving, comprising:

[0029] The data acquisition module is configured to acquire users' historical electricity consumption data.

[0030] The sample screening module is configured to: perform multi-scale feature reconstruction on the acquired data samples, calculate the reconstruction error, aggregate the reconstruction errors at each scale, obtain the sample structure guidance index, calculate the entropy contribution of the data samples, and screen the data samples by combining the sample structure guidance index and the entropy contribution.

[0031] The membership matrix calculation module is configured to: perform fuzzy partitioning on the filtered samples, determine the membership degree of each sample to each cluster center, arrange the membership degrees of all samples in order, and obtain the fuzzy membership matrix;

[0032] The membership matrix adjustment module is configured to: generate a structural supervision signal based on the sample structure guidance index, construct a confidence score using the unsupervised model residuals, obtain a target supervision signal based on the confidence score, fuse the structural supervision signal and the target supervision signal to obtain a fused guidance signal, and adjust the membership matrix using the fused guidance signal to obtain a new membership matrix;

[0033] The model prediction module is configured to: train a local regression model for each fuzzy subset, use the membership degrees in the new membership matrix as weights, and perform a weighted summation of the prediction results of each local regression model to obtain the final prediction result.

[0034] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0035] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.

[0036] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0038] This invention proposes a fuzzy electricity consumption prediction method based on multi-scale dual supervision. Addressing the shortcomings of existing fuzzy modeling and machine learning methods in intelligent electricity consumption prediction, such as insufficient modeling accuracy, poor structural adaptability, and lack of feedback adjustment capability in fuzzy partitioning, this invention proposes an adaptive fuzzy ensemble system that integrates multi-scale feature extraction, information entropy-driven sample selection, and a dual supervision guidance mechanism. This system can dynamically capture the structural differences in electricity consumption behavior among different users or at different time periods while maintaining model interpretability. It also optimizes fuzzy partitioning and model weights in real time by combining prediction residual feedback information, thereby improving the model's ability to model complex electricity consumption data and its generalization performance. This method is suitable for energy consumption behavior analysis and prediction tasks in households, power grids, or regions. Compared with traditional fuzzy systems or machine learning models, this invention has the following significant advantages:

[0039] Strong structural adaptability: Multi-scale fusion of reconstruction errors significantly enhances the ability to model the periodicity and non-stationarity of electrical energy behavior.

[0040] Improved prediction accuracy: Feedback optimization is performed using target-oriented supervision signals to guide the model to converge toward the direction of minimum residual.

[0041] The model has good interpretability: the fuzzy subset partitioning has structural and semantic basis and supports the output of "IF-THEN" rules.

[0042] High modeling efficiency: Reduces training burden and enhances modeling robustness through information entropy evaluation and sample selection.

[0043] Embedded deployment: Suitable for smart buildings and home energy systems, and can be used in conjunction with embedded energy management systems.

[0044] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0045] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0046] Figure 1 The flowchart shows the fuzzy electricity consumption prediction method based on multi-scale dual supervision provided in Embodiment 1 of the present invention. Detailed Implementation

[0047] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0048] It should be noted that the following detailed description is exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0049] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. Furthermore, it should be understood that 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 includes other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0050] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0051] Example 1

[0052] like Figure 1 As shown, this embodiment provides a fuzzy electricity consumption prediction method based on multi-scale dual supervision, including the following steps:

[0053] Obtain users' historical electricity consumption data;

[0054] Multi-scale feature reconstruction is performed on the acquired data samples, the reconstruction error is calculated, the reconstruction errors at each scale are aggregated to obtain the sample structure guidance index, the entropy contribution of the data samples is calculated, and the data samples are screened by combining the sample structure guidance index and the entropy contribution.

[0055] The filtered samples are fuzzy partitioned to determine the membership degree of each sample to each cluster center. The membership degrees of all samples are arranged in order to obtain the fuzzy membership degree matrix.

[0056] The structural supervision signal is generated based on the sample structure guidance index. The confidence level is constructed using the unsupervised model residuals. The target supervision signal is obtained based on the confidence level. The structural supervision signal and the target supervision signal are fused to obtain the fused guidance signal. The membership matrix is ​​adjusted using the fused guidance signal to obtain a new membership matrix.

[0057] For each fuzzy subset, a local regression model is trained, and the membership degrees in the new membership matrix are used as weights to perform a weighted sum of the prediction results of each local regression model to obtain the final prediction result.

[0058] The present invention will now be explained in detail.

[0059] Step (1) First, obtain the user's historical electricity consumption data.

[0060] Step (2) Perform multi-scale feature reconstruction on the acquired data samples, calculate the reconstruction error, aggregate the reconstruction errors at each scale, obtain the sample structure guidance index, calculate the entropy contribution of the data samples, and screen the data samples by combining the sample structure guidance index and entropy contribution.

[0061] Multi-scale feature reconstruction:

[0062] In the fields of data processing and feature analysis, multi-scale feature reconstruction is an important method for mining deep structural information from data. This section uses multiple multilayer perceptrons (MLPs) with different hidden layer dimensions to construct a compressed reconstruction model. ,in

[0063] Compression and reconstruction process:

[0064] Traditional data compression and reconstruction methods are often limited to a single scale or simple linear transformations, making it difficult to fully extract the features of data at different levels. The multilayer perceptron used here, however, is a neural network structure with powerful nonlinear mapping capabilities. It consists of an input layer, multiple hidden layers, and an output layer. Neurons in the hidden layers perform nonlinear transformations on the input through activation functions. For the original input data... Different hidden layer dimensions The MLP processes the data in a unique way. The input data first enters the input layer, then undergoes a series of weighted summations and activation operations in the hidden layers, gradually mapping the data to different feature spaces. Finally, the reconstructed data is obtained in the output layer. This neural network-based compression and reconstruction method, compared to traditional methods, can automatically learn the complex nonlinear structure of data and capture richer feature information.

[0065] Methods for extracting structural information at different scales:

[0066] MLPs with different hidden layer dimensions are equivalent to observing data at different scales. MLPs with larger hidden layer dimensions can capture more macroscopic and abstract features of the data, similar to observing an object from a distance, seeing its overall outline and general characteristics; while MLPs with smaller hidden layer dimensions focus more on mining local, detailed features of the data, like observing an object up close, seeing more intricate details. By combining multiple sets of MLPs with different hidden layer dimensions, structural information can be extracted from multiple scales. Compared with existing single-scale or fixed-structure feature extraction techniques, this method can more comprehensively and flexibly adapt to the characteristics of different data, extracting more diverse and complementary features.

[0067] For each sample At each scale Below, the reconstruction error is calculated. To measure the effectiveness of the reconstruction, here It is the Euclidean norm. Finally, the reconstruction errors at various scales are aggregated to obtain the sample structure guiding index. This comprehensively reflects the reconstruction characteristics of the sample at multiple scales.

[0068] Information entropy-driven sample selection:

[0069] Information entropy, as an important indicator for measuring data uncertainty and information content, plays a crucial role in sample selection. This section aims to select samples that are more valuable for subsequent modeling through the calculation and analysis of information entropy.

[0070] First, the selected samples are constructed into a feature matrix. In this matrix, each row corresponds to a sample, and each column corresponds to an original feature dimension. Then, the feature dimension is selected based on its variance. Variance reflects the dispersion of the data; a larger variance means that the data distribution along that dimension is more dispersed, and potentially contains richer information. Therefore, the feature dimension with the largest variance is selected. Each feature dimension constitutes a projection subspace. Compared to methods that randomly select feature dimensions or select based on a single statistic, variance-based selection can more effectively retain features with high information content.

[0071] Next, construct the projection subspace. Joint histogram estimation distribution And calculate the entropy. ,in It is a very small constant used to avoid the bias in logarithmic operations. This leads to undefined situations. The higher the entropy value, the greater the uncertainty of the data under that distribution, meaning the greater the amount of information it contains.

[0072] For each sample After removing it from the dataset, the entropy is re-estimated. By calculating the contribution of information entropy This is used to measure the degree of influence of a sample on the overall data information entropy. The greater the contribution of a sample to information entropy, the more important it is in data distribution and information representation.

[0073] The structural information of the comprehensive sample (from step (2)) (Reflection) and entropy contribution Through formula Calculate the importance score, where It is a weighting coefficient used to balance the proportion of structural information and entropy contribution in the importance assessment. Ultimately, the former is retained. High-scoring samples are used for subsequent modeling. Compared to random selection or selection based on a single factor, this screening method can more comprehensively and scientifically select the most valuable samples for model building.

[0074] Step (3) Perform fuzzy partitioning on the filtered samples, determine the membership degree of each sample to each cluster center, and arrange the membership degrees of all samples in order to obtain the fuzzy membership degree matrix.

[0075] Fuzzy clustering and membership generation:

[0076] Fuzzy clustering is a technique that divides data points into different categories in a fuzzy manner. Compared to traditional hard clustering (such as K-Means hard partitioning, where a data point can only belong to one category), it allows data points to belong to multiple categories to varying degrees, which is more in line with the complex characteristics of data in the real world.

[0077] Specific process of fuzzy partitioning:

[0078] In this section, the KMeans or Fuzzy C-Means (FCM) algorithms are used to perform fuzzy partitioning on the samples filtered in step (2). Taking the FCM algorithm as an example, its goal is to minimize an objective function based on the distance between a sample and its cluster center. First, random initialization... Cluster centers Then, for each sample Calculate its distance to each cluster center. distance , here It also uses the Euclidean norm. In each iteration, the membership degree of a sample to each category is updated based on the distance of the sample to the cluster center. Simultaneously update the cluster centers. The formula for updating membership is: ,in It is a fuzzy parameter that controls the degree of fuzziness in clustering. The larger the value, the stronger the clustering fuzziness, and the more evenly the data points belong to multiple categories. Through continuous iteration until the objective function converges, a stable set of cluster centers is obtained. And the membership degree of each sample relative to each cluster center. Compared with traditional K-Means hard clustering, FCM can better handle data with fuzzy boundaries and overlapping data, and more meticulously characterize the distribution features of the data.

[0079] The process of obtaining the membership matrix:

[0080] After completing the fuzzy partitioning, the membership degree of each sample to each cluster center is determined. Then, by arranging the membership degrees of all samples in order, the fuzzy membership degree matrix can be obtained. ,in It is the sample size. This represents the number of cluster centers. Each row in the matrix corresponds to the degree to which a sample belongs to each category, and each column corresponds to a cluster center, showing the membership degree of all samples to that cluster center. This method of generating a membership matrix based on fuzzy clustering provides a foundation for subsequent tasks such as data analysis and pattern recognition using fuzzy information. Compared to the category label matrix obtained by traditional hard clustering, it contains richer information about the relationship between samples and categories.

[0081] Step (4) Generate structural supervision signals based on sample structural guidance indicators, construct confidence levels using unsupervised model residuals, obtain target supervision signals based on confidence levels, fuse structural supervision signals and target supervision signals to obtain fused guidance signals, adjust membership matrices using fused guidance signals to obtain new membership matrices.

[0082] Two-way supervision guides the adjustment of membership degree:

[0083] In data processing and analysis, in order to make the membership degree of a sample more accurately reflect its characteristics, this section introduces a two-way supervision mechanism to adjust the membership degree.

[0084] Structural monitoring signals:

[0085] Structural monitoring signals aim to provide monitoring information from the structural level of data.

[0086] Here, the sample structure guidance index obtained in step (2) is used. As a structural indicator, this indicator is a synthesis of reconstruction errors across multiple scales, reflecting the structural expressive power of a sample during feature reconstruction. It is compressed using the Sigmoid function to obtain... The Sigmoid function compresses the input value to... The characteristics of the interval allow the reconstruction error to be converted into a structure monitoring signal value within a reasonable range. Among these, and It is an adjustable parameter. The slope of the control function, The central position of the control functions can be adjusted to flexibly generate appropriate monitoring signals based on the reconstruction error.

[0087] Target monitoring signals:

[0088] The target supervision signal provides supervision from the perspective of the model's target prediction. It utilizes the residuals of the unsupervised model. To construct the confidence level. Here... It is the sample structure guiding index obtained in step (2). It is an estimate of a sample by an unsupervised model, the residual. This measures the difference between the actual structural indicators of the sample and the estimates from the unsupervised model. The formula is used to... Construct confidence ,in This is a very small constant to prevent the denominator from being zero. This confidence level reflects the reliability of the sample under the unsupervised model. Further, through the formula... Obtain target supervision signal ,in and It's about adjusting parameters. control Weights in the calculation, The shape of the control functions can be adjusted to make the target monitoring signal more consistent with the data characteristics.

[0089] Fusion guidance signal:

[0090] To combine the advantages of both structural and target monitoring signals, an exponential decay fusion method is employed. This is achieved through the formula... To integrate, among which Here It is the attenuation coefficient. This could be related variables such as the number of training iterations. With Changes, It will dynamically adjust to achieve reasonable fusion of the two monitoring signals at different stages, thus enabling the fusion of guidance signals. It can reflect both data structure information and information related to the model objective.

[0091] Adjust the membership matrix:

[0092] Using fusion guidance signals For the original membership matrix elements in We perform weighted average to obtain Then, for the newly obtained matrix Perform normalization operation, i.e. Through this series of operations, the membership matrix can be better adapted to the inherent structure of the data and the target requirements of the model, providing a higher quality data representation for subsequent analysis and modeling.

[0093] Meaning of formula parameters:

[0094] Structural monitoring signal formula ;

[0095] : Controls the slope of the Sigmoid function. The larger the absolute value of the function, the steeper the change near the center; the smaller the absolute value, the smoother the change. It determines the reconstruction error. The rate of change when converted into a structural monitoring signal value.

[0096] : Controls the center position of the Sigmoid function, i.e., when hour, It determines the value of the reconstruction error at which the structural monitoring signal is in an intermediate state.

[0097] Target supervision signal formula and ;

[0098] : is a very small constant added to the denominator to prevent excessive sizing. To ensure confidence levels in cases where the denominator is zero, the following conditions must be considered. The validity of the calculation.

[0099] In calculation At that time, control the confidence level The weights in the formula adjust the target monitoring signal pair. The degree of dependence.

[0100] Control function The shape of the target monitoring signal affects the value of the target monitoring signal as confidence level increases. The trend of change, such as the concavity or convexity of a function.

[0101] Fusion guidance signal formula middle

[0102] It is the exponential decay coefficient, which determines... Follow The rate of decay of the change. The larger, Follow The more it increases, the faster it decays; The smaller the value, the slower the decay.

[0103] These can typically be variables related to the training process, such as the number of training iterations or the time step, used to measure different stages of the training process and dynamically adjust accordingly. This allows for the adjustment of the fusion ratio between the two monitoring signals.

[0104] The method of adjusting the membership matrix using a fusion guiding signal:

[0105] 1. First, calculate the fusion guidance signal according to the formulas above. Each sample Each corresponds to one value.

[0106] 2. Then, for each element in the original membership matrix U... Perform a weighted operation, that is This is where the sample is placed. Corresponding fusion guidance signal value The sample belongs to the category in the membership matrix. membership degree Multiplication causes the membership degree to change due to the influence of the fusion guiding signal.

[0107] 3. Finally, the weighted matrix... Perform a normalization operation, denoted as The purpose of normalization is to ensure that the elements in the adjusted membership matrix satisfy certain value ranges and statistical properties (e.g., the sum of the elements in each row is 1, depending on the normalization method used), thus guaranteeing the effectiveness and rationality of the adjusted membership matrix in subsequent applications.

[0108] Step (5) For each fuzzy subset, train a local regression model, use the membership degree in the new membership matrix as weights, and sum the prediction results of each local regression model to obtain the final prediction result.

[0109] Fusion of fuzzy rule training and prediction output:

[0110] This section focuses on using fuzzy subsets for local model training and achieving more accurate predictions through fusion.

[0111] Training a fuzzy subset local regression model:

[0112] For each fuzzy subset Train a local regression model Methods such as Ridge Regression and Support Vector Regression (SVR) can be chosen. Ridge Regression adds a layer of error to the loss function of ordinary least squares. Regularization terms effectively prevent overfitting and perform well when dealing with data exhibiting multicollinearity. Support vector regression, based on the principles of support vector machines, finds the optimal hyperplane to fit the data and has good handling capabilities for nonlinear data. The appropriate regression model is selected based on the specific characteristics and needs of the data, and trained on the corresponding fuzzy subset to uncover local features and patterns within that subset.

[0113] Integrated prediction:

[0114] For any input sample Through its adjusted fuzzy membership matrix Membership degree Perform integrated prediction. The prediction formula is as follows: This involves weighting and summing the predictions of each local regression model based on the degree to which a sample belongs to each fuzzy subset. This ensemble prediction method fully utilizes the local information of different fuzzy subsets and, compared to single-model prediction, can more comprehensively consider the diversity and complexity of the data, thereby improving the accuracy and stability of the prediction.

[0115] For the integrated prediction formula in step (5)

[0116] : is the membership matrix after adjustment in step (4) The elements in the sample represent the sample. Belongs to a fuzzy subset The degree of association reflects the strength of the correlation between the sample and each fuzzy subset.

[0117] It is for fuzzy subsets. The trained local regression model (such as ridge regression, SVR, etc.) is effective for the samples. The predicted output value. Different local regression models learn specific features and patterns on their respective fuzzy subsets, and their output values ​​represent the prediction results of the samples based on the knowledge of that subset.

[0118] Final output It is by taking samples Membership degree of each fuzzy subset As weights, for the corresponding local regression model The output values ​​are obtained by weighted summation. That is, the final predicted value is obtained by comprehensively considering the membership relationship of the sample in each fuzzy subset and the local model prediction results corresponding to each subset, using a weighted fusion method. In this way, the local model output corresponding to the fuzzy subset with higher membership degree has a larger proportion in the final result, reflecting the comprehensive prediction of the sample under different local feature spaces.

[0119] Example 2

[0120] This embodiment provides a fuzzy electricity consumption forecasting system based on multi-scale dual supervision, including:

[0121] The data acquisition module is configured to acquire users' historical electricity consumption data.

[0122] The sample screening module is configured to: perform multi-scale feature reconstruction on the acquired data samples, calculate the reconstruction error, aggregate the reconstruction errors at each scale, obtain the sample structure guidance index, calculate the entropy contribution of the data samples, and screen the data samples by combining the sample structure guidance index and the entropy contribution.

[0123] The membership matrix calculation module is configured to: perform fuzzy partitioning on the filtered samples, determine the membership degree of each sample to each cluster center, arrange the membership degrees of all samples in order, and obtain the fuzzy membership matrix;

[0124] The membership matrix adjustment module is configured to: generate a structural supervision signal based on the sample structure guidance index, construct a confidence score using the unsupervised model residuals, obtain a target supervision signal based on the confidence score, fuse the structural supervision signal and the target supervision signal to obtain a fused guidance signal, and adjust the membership matrix using the fused guidance signal to obtain a new membership matrix;

[0125] The model prediction module is configured to: train a local regression model for each fuzzy subset, use the membership degrees in the new membership matrix as weights, and perform a weighted summation of the prediction results of each local regression model to obtain the final prediction result.

[0126] It should be noted that the above modules correspond to the steps in Embodiment 1, and the examples and application scenarios implemented by the above modules and their corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should also be noted that the above modules can be executed in a computer system as part of the system.

[0127] In further embodiments, the following is also provided:

[0128] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0129] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0130] A computer-readable storage medium for storing computer instructions that, when executed by a processor, perform the method of Embodiment 1.

[0131] The method in Example 1 can be directly executed by a hardware processor, or it can be executed by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0132] A computer program product includes a computer program that, when executed by a processor, implements the method in Embodiment 1.

[0133] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0134] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0135] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0136] Those skilled in the art will recognize that the units and algorithm steps described in conjunction with the embodiments herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0137] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A fuzzy electricity consumption forecasting method based on multi-scale dual supervision, characterized in that, It includes the following steps: Obtain the user's historical electricity consumption data; Perform multi-scale feature reconstruction on the obtained electricity consumption data, calculate the reconstruction error, aggregate the reconstruction errors at each scale to obtain a sample structure guiding index, calculate the entropy contribution degree of the electricity consumption data, and screen the electricity consumption data by comprehensively considering the sample structure guiding index and the entropy contribution degree; Perform fuzzy partitioning on the screened electricity consumption data, determine the membership degrees of each electricity consumption data to each cluster center, and arrange the membership degrees of all electricity consumption data in sequence to obtain a fuzzy membership matrix; Generate a structure supervision signal based on the sample structure guiding index, construct a confidence level using the unsupervised model residual, obtain a target supervision signal based on the confidence level, fuse the structure supervision signal and the target supervision signal to obtain a fused guiding signal, and use the fused guiding signal to adjust the membership matrix to obtain a new membership matrix; For each fuzzy subset, train a local regression model, use the membership degrees in the new membership matrix as weights, and perform weighted summation on the prediction results of each local regression model to obtain the final electricity consumption prediction result; The target supervision signal is: in, For confidence level, and It refers to adjusting parameters; The structure supervision signal is: in, Indicators guiding the sample structure. and It is an adjustable parameter; Using the fused guiding signal to adjust the membership matrix specifically means: Multiply the value of the fused guiding signal corresponding to the electricity consumption data by the membership degree of the electricity consumption data belonging to the category in the membership matrix to obtain a weighted matrix, and perform a normalization operation on the weighted matrix to obtain a new membership matrix.

2. The fuzzy electricity consumption forecasting method based on multi-scale dual supervision as described in claim 1, characterized in that, Calculate the entropy contribution degree of the electricity consumption data specifically as follows: Calculate the variance of electricity consumption data and filter out those with the largest variance. Each feature dimension constitutes a projection subspace. A joint histogram of the projection subspace is constructed to estimate the distribution and calculate the entropy. For each electricity consumption data point, it is removed from the dataset and the entropy of the dataset is re-estimated. The entropy contribution of the electricity consumption data is calculated by combining the original entropy of the dataset with the re-estimated entropy.

3. The fuzzy electricity consumption forecasting method based on multi-scale dual supervision as described in claim 1, characterized in that, Perform fuzzy partitioning on the screened electricity consumption data specifically as follows: Randomly initialize C cluster centers. For each electricity consumption data, calculate its distances to each cluster center. In each iteration, update the membership degrees of the electricity consumption data belonging to each category according to the distances of the electricity consumption data to the cluster centers, and at the same time update the cluster centers.

4. A fuzzy electricity consumption forecasting system based on multi-scale dual supervision, characterized in that, Adopt the fuzzy electricity consumption prediction method based on multi-scale dual supervision drive as described in any one of claims 1-3, including: A data acquisition module configured to: obtain the user's historical electricity consumption data; A sample screening module configured to: perform multi-scale feature reconstruction on the obtained electricity consumption data, calculate the reconstruction error, aggregate the reconstruction errors at each scale to obtain a sample structure guiding index, calculate the entropy contribution degree of the electricity consumption data, and screen the electricity consumption data by comprehensively considering the sample structure guiding index and the entropy contribution degree; A membership matrix calculation module configured to: perform fuzzy partitioning on the screened electricity consumption data, determine the membership degrees of each electricity consumption data to each cluster center, and arrange the membership degrees of all electricity consumption data in sequence to obtain a fuzzy membership matrix; A membership matrix adjustment module configured to: generate a structure supervision signal based on the sample structure guiding index, construct a confidence level using the unsupervised model residual, obtain a target supervision signal based on the confidence level, fuse the structure supervision signal and the target supervision signal to obtain a fused guiding signal, and use the fused guiding signal to adjust the membership matrix to obtain a new membership matrix; The model prediction module is configured to: train a local regression model for each fuzzy subset, use the membership degree in the new membership matrix as weights, and perform a weighted summation of the prediction results of each local regression model to obtain the final electricity consumption prediction result.

5. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-3.

6. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the method described in any one of claims 1-3.

7. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the method described in any one of claims 1-3.

Images