Multi-source heterogeneous flexible load data fusion and processing method and system

Abstract

The invention discloses a multi-source heterogeneous flexible load data fusion and processing method and system, and the method is characterized in that the method comprises the following steps: carrying out the unified data collection and marking of flexible load equipment and sensors from different regions and different types; performing multi-level decomposition on the data through wavelet decomposition, and performing BiTCN feature extraction on the wavelet component of each layer; performing clustering analysis on the features extracted by the BiTCN by using a K-Means clustering algorithm, and dynamically selecting the most suitable wavelet decomposition basis and adaptive filter parameters; and reconstructing the wavelet component subjected to filtering and fusion processing, performing feature extraction on the reconstructed signal, and quantizing the filtering and reconstruction effects.

Description

Technical Field

[0001] The present invention relates to the field of data fusion and processing, and more specifically, to a method and system for fusion and processing of multi-source heterogeneous flexible load data. Background Art

[0002] Traditional signal analysis methods, such as Fourier transforms, often struggle to effectively process transient and non-stationary signals. This is especially true when fusing multi-source, heterogeneous data, which often faces challenges such as signal loss, noise interference, and signal heterogeneity. Therefore, there is an urgent need for advanced methods that can perform efficient time-frequency analysis and possess powerful data fusion capabilities to address the demands of complex signal processing.

[0003] In modern power systems and smart energy management, the widespread integration of flexible loads has led to the generation of massive, multi-source, and heterogeneous power data. This data not only exhibits significant differences in time and frequency domains, as well as in measurement accuracy, but also poses a key challenge in efficiently integrating and processing this heterogeneous data. Traditional data processing methods based on single features or fixed filtering often struggle to achieve high-precision and robust signal separation and feature extraction in complex environments.

[0004] In response to the above problems, a multi-source heterogeneous flexible load data fusion and processing method and system are urgently needed. Summary of the Invention

[0005] In order to solve the deficiencies in the prior art, the present invention provides a data fusion and processing method and system based on BiTCN feature extraction and K-Means clustering combined with wavelet decomposition and adaptive filtering.

[0006] The present invention adopts the following technical solutions.

[0007] The first aspect of the present invention relates to a method for fusion and processing of multi-source heterogeneous flexible load data, which includes the following steps: unified data collection and labeling for flexible load devices and sensors from different regions and different types; multi-level decomposition of the data through wavelet decomposition, and BiTCN feature extraction of the wavelet components of each layer; cluster analysis of the features extracted by BiTCN using the K-Means clustering algorithm, and dynamic selection of the most suitable wavelet decomposition basis and adaptive filter parameters; reconstructing the wavelet components that have been filtered and fused, extracting features from the reconstructed signals, and quantifying the filtering and reconstruction effects.

[0008] Unified data collection and labeling are carried out for flexible load equipment and sensors from different regions and types, including: labeling content including sampling frequency, data source, and noise level; unified formatting and time synchronization of the labeled data, interpolation method is used to fill missing values, and data normalization is performed; low-pass filter or mean filter is used to perform preliminary noise suppression on the pre-processed data.

[0009] The data is decomposed into multiple levels through wavelet decomposition, and BiTCN feature extraction is performed on the wavelet components of each level, including: selecting a wavelet basis according to the local characteristics of the signal, and performing 3 to 5 levels of wavelet decomposition; and BiTCN feature extraction is performed on the wavelet components of each level.

[0010] The K-Means clustering algorithm is used to perform cluster analysis on the features extracted by BiTCN, and the most suitable wavelet decomposition basis and adaptive filter parameters are dynamically selected, including: using the K-Means clustering algorithm to perform cluster analysis on the features extracted by BiTCN, dividing the data in the feature space into several categories; and dynamically selecting the corresponding wavelet decomposition basis and corresponding adaptive filter parameters for each category of data.

[0011] The wavelet components that have been filtered and fused are reconstructed, features are extracted from the reconstructed signals, and the filtering and reconstruction effects are quantified, including: reconstructing the wavelet components that have been filtered and fused through inverse wavelet transform; extracting the mean, variance, peak value, valley value, and slope change point of the signal from the reconstructed signal; and quantifying the filtering and reconstruction effects using signal-to-noise ratio and mean square error.

[0012] The wavelet components that have been filtered and fused are reconstructed, features are extracted from the reconstructed signals, and the filtering and reconstruction effects are quantified, including: optimizing hyperparameters based on the results of signal quality assessment; the hyperparameters include the number of wavelet decomposition layers, thresholds, and filter step sizes.

[0013] The second aspect of the present invention relates to a multi-source heterogeneous flexible load data fusion and processing system, which includes a data acquisition and annotation module, a wavelet decomposition and feature extraction module, a clustering module, and a reconstruction module; the data acquisition and annotation module performs unified data acquisition and annotation for flexible load equipment and sensors from different regions and different types; the wavelet decomposition and feature extraction module decomposes the data into multiple levels through wavelet decomposition, and performs BiTCN feature extraction on the wavelet components of each layer; the clustering module uses the K-Means clustering algorithm to perform cluster analysis on the features extracted by BiTCN, and dynamically selects the most suitable wavelet decomposition basis and adaptive filter parameters; the reconstruction module reconstructs the wavelet components that have been filtered and fused, extracts features from the reconstructed signals, and quantifies the filtering and reconstruction effects.

[0014] The data collection and annotation module uniformly collects and annotates data from flexible load equipment and sensors of different regions and types, including: the annotation content includes sampling frequency, data source, and noise level; the annotated data is uniformly formatted and time-synchronized, missing values ​​are filled using interpolation methods, and data normalization is performed; and the pre-processed data is subjected to preliminary noise suppression using a low-pass filter or a mean filter.

[0015] The wavelet decomposition and feature extraction module decomposes the data into multiple layers through wavelet decomposition and performs BiTCN feature extraction on the wavelet components of each layer, including: selecting the wavelet basis according to the local characteristics of the signal and performing 3 to 5 layers of wavelet decomposition; and performing BiTCN feature extraction on the wavelet components of each layer.

[0016] The clustering module uses the K-Means clustering algorithm to perform cluster analysis on the features extracted by BiTCN and dynamically selects the most suitable wavelet decomposition basis and adaptive filter parameters, including: using the K-Means clustering algorithm to perform cluster analysis on the features extracted by BiTCN to divide the data in the feature space into several categories; dynamically selecting the corresponding wavelet basis and corresponding adaptive filter parameters for each category of data.

[0017] The reconstruction module reconstructs the wavelet components that have been filtered and fused, extracts features from the reconstructed signal, and quantifies the filtering and reconstruction effects, including: reconstructing the wavelet components that have been filtered and fused through inverse wavelet transform; extracting the mean, variance, peak value, valley value, and slope change point of the signal from the reconstructed signal; and quantifying the filtering and reconstruction effects using signal-to-noise ratio and mean square error.

[0018] The reconstruction module reconstructs the wavelet components that have been filtered and fused, extracts features from the reconstructed signal, and quantifies the filtering and reconstruction effects, including optimizing hyperparameters based on the results of signal quality assessment; the hyperparameters include the number of wavelet decomposition layers, threshold, and filter step size.

[0019] A third aspect of the present invention relates to a terminal, comprising a processor and a storage medium; the storage medium is used to store instructions; and the processor is used to operate according to the instructions to execute the steps of the method described in the first aspect of the present invention.

[0020] A fourth aspect of the present invention relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect of the present invention.

[0021] The beneficial effect of the present invention is that, compared with the existing technology, the multi-source heterogeneous flexible load data fusion and processing method and system in the present invention extracts deep time-frequency features from signals through BiTCN, and uses K-Means clustering to classify different data patterns, and then selects a suitable wavelet decomposition basis and adaptive filter for each type of signal.

[0022] The beneficial effects of the present invention also include:

[0023] 1. Wavelet decomposition can effectively separate load information and interference noise in different frequency bands at multiple scales and resolutions, providing a foundation for high-precision, scalable data analysis. Adaptive filters effectively suppress various random noises and non-stationary interferences by dynamically adjusting filter parameters in real time. The method of the present invention optimizes the signal processing process through feature extraction and classification, and selects the most appropriate wavelet decomposition basis and filtering parameters for different data categories. This enables efficient and accurate fusion of multi-source heterogeneous data, providing reliable data support for optimized scheduling, accurate prediction, and energy consumption behavior analysis of flexible loads. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is the overall flow chart of the system;

[0025] Figure 2 Figure 2 is the BiTCN structure diagram;

[0026] Figure 3 It is the wavelet decomposition and reconstruction diagram;

[0027] Figure 4 It is the structural diagram of the adaptive filter;

[0028] Figure 5 To obtain a new signal map using wavelet decomposition and adaptive filter. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solutions and advantages of the present invention clearer and more accurate, the technical solutions of the present invention are described in detail below through multiple specific embodiments. The embodiments used in the present invention are only used to explain the present invention and are not intended to limit the content of the present invention.

[0030] Wavelet decomposition, as a multi-scale, multi-resolution time-frequency analysis method, can effectively distinguish high-frequency noise from low-frequency trends in signals, supporting the extraction of key features across different time and frequency bands from multi-source, heterogeneous load data. Furthermore, adaptive filters can dynamically update their filtering parameters in real time, continuously optimizing the filtering effect based on changes in the load signal's characteristics, thereby improving the ability to suppress multiple noise sources and non-stationary signals.

[0031] In order to further optimize the processing of multi-source heterogeneous data, the present invention introduces the BiTCN (Binary Convolutional Network) feature extraction method and K-Means clustering algorithm. First, deep time-frequency features are extracted from the load data through BiTCN. BiTCN uses bit-based convolution operations, which can effectively reduce the computational complexity while retaining the key information of the signal. Then, K-Means clustering is used to perform cluster analysis on the extracted features to identify different patterns in the data. Based on the clustering results, the most appropriate wavelet decomposition basis and adaptive filter parameters are dynamically selected for each type of signal to further improve the accuracy and robustness of data processing.

[0032] To address the current challenges of multi-source heterogeneous flexible load data collection and processing, such as susceptibility to noise interference, a wide variety of data types, and significant differences in feature scales, traditional signal processing methods struggle to balance diverse frequency domain characteristics with real-time dynamic requirements. This paper proposes a multi-source heterogeneous flexible load data fusion and processing method based on BiTCN feature extraction and K-Means clustering, combining wavelet decomposition with an adaptive filter. This method optimizes the signal processing process through feature extraction and classification, and selects the most appropriate wavelet decomposition basis and filtering parameters for different data types.

[0033] By combining BiTCN feature extraction with K-Means clustering, the present invention achieves accurate processing of different types of flexible load data, providing strong data support and theoretical basis for subsequent advanced functions such as energy consumption prediction, optimized scheduling, and energy behavior analysis.

[0034] A first aspect of the present invention relates to a method for fusion and processing of multi-source heterogeneous flexible load data, the method comprising the following steps:

[0035] Step 1: Perform unified data collection and labeling for flexible load devices and sensors from different regions and types.

[0036] Figure 1 This is the overall flow chart of the system. Figure 1 , demonstrating the processing process of multi-source heterogeneous flexible load data, including data acquisition, preprocessing, feature extraction, clustering analysis, noise suppression and parameter optimization steps.

[0037] Unified data collection and labeling are carried out for flexible load equipment and sensors from different regions and types, including: labeling content including sampling frequency, data source, and noise level; unified formatting and time synchronization of the labeled data, interpolation method is used to fill missing values, and data normalization is performed; low-pass filter or mean filter is used to perform preliminary noise suppression on the pre-processed data.

[0038] The present invention first collects unified data from flexible load devices and sensors from different regions and types, ensuring coverage of a wide range of energy-consuming devices. These devices may include various household appliances, industrial equipment, or power load monitoring instruments. To ensure coverage of these diverse energy-consuming devices, the collected data includes different types of power usage information.

[0039] Because this data may vary in format, time stamps, and sampling rate, detailed data annotation is required after collection. This annotation includes information such as sampling frequency, data source, and noise level, providing the necessary prior information for subsequent data processing, analysis, and filtering. Sampling frequency: The sampling interval for each data source. Data source: The device or sensor from which the data originates. Noise level: This classifies the level of noise that may be present in the data to facilitate the selection of appropriate processing methods. Next, data processing and fusion are performed through the following steps.

[0040] Data collection and annotation: Data from flexible load devices and sensors from different regions and types is uniformly collected to ensure coverage of various energy-consuming devices. Because this data may vary in format, time stamps, and sampling rates, detailed annotation is required after collection. This annotation includes information such as sampling frequency, data source, and noise level, providing necessary prior information for subsequent data processing, analysis, and filtering.

[0041] Data formatting and time synchronization: Since collected raw data may come from different devices and have different formats, unified formatting is required. Converting all data to a standard format ensures that each data item can be seamlessly integrated with other data. Furthermore, time synchronization is necessary to ensure that data from different sources is consistent across timelines. This standardization facilitates subsequent unified analysis and ensures data consistency.

[0042] Data preprocessing is performed on the labeled time series data, using interpolation methods to fill missing values ​​and perform data normalization to ensure that all data are processed at the same scale.

[0043] The purpose of this step is to address some issues that may be encountered during the data collection process, such as missing values ​​and scale differences. Specific methods include: Interpolation methods to fill missing values: When missing values ​​appear in the data, appropriate interpolation methods are used to fill the missing parts to ensure data integrity. Data normalization: In order to process the data on a unified scale, data normalization methods (such as minimum-maximum normalization and Z-score normalization) are used to convert the data to the same dimension. The purpose of preprocessing is to provide high-quality, standardized data for subsequent noise suppression and signal analysis.

[0044] Initial noise suppression is performed on the preprocessed data, using methods such as low-pass filters or mean filters to remove high-frequency noise. After data preprocessing, some high-frequency noise may still remain, which may affect the accuracy of subsequent analysis. Therefore, a low-pass filter or mean filter is used to perform preliminary noise suppression on the data. A low-pass filter can remove high-frequency noise from the data, while a mean filter smoothes the data to reduce the impact of noise on the results. The purpose of this step is to provide a cleaner signal source for subsequent, more in-depth analysis.

[0045] Step 2: Decompose the data into multiple layers through wavelet decomposition, and perform BiTCN feature extraction on the wavelet components of each layer.

[0046] After initial noise suppression, the data is further processed using wavelet decomposition. Wavelet decomposition is an effective time-frequency analysis method that decomposes the signal into multiple frequency bands and extracts information at different time scales. The specific process of this step is as follows: An appropriate wavelet basis is selected based on the local characteristics of the signal. Three to five layers of wavelet decomposition are performed to obtain multiple levels of low-frequency and high-frequency components. Each layer contains characteristics of the signal at different time scales, providing rich information for subsequent feature extraction.

[0047] Traditional signal theory is based on Fourier analysis. However, as a global transformation, the Fourier transform has certain limitations, such as the lack of localized analysis capabilities and the inability to analyze non-stationary signals. Compared with the Fourier transform, the wavelet transform is a localized transform in the time and frequency domains, and can therefore effectively extract information from signals. Through operations such as scaling and translation, it can perform multi-scale and refined analysis of functions or signals, solving many difficult problems that the Fourier transform cannot solve.

[0048] The Haar function is the earliest orthogonal wavelet function with compact support used in wavelet analysis. It is also the simplest wavelet function. It is a single rectangular wave with a support domain in the range of t∈[0,1].

[0049] Daubechies wavelet is abbreviated as dbN, where N is the order of the wavelet. Wavelet function Ψ(t) and scaling function The support area in is 2N-1, and the vanishing moment of Ψ(t) is N. The dbN wavelet has good regularity, meaning that the smoothing error introduced by the wavelet as a sparse basis is not easily detected, resulting in a smoother signal reconstruction process. The characteristic of the dbN wavelet is that the order of the vanishing moment increases with the order (sequence N). A higher vanishing moment improves smoothness, localization in the frequency domain, and better frequency band partitioning. However, this weakens the tight support in the time domain, significantly increases the computational complexity, and degrades real-time performance.

[0050] The Symlet wavelet function, proposed by Ingrid Daubechies, is a nearly symmetric wavelet function that improves upon the db function. Symlet wavelets are typically denoted as symN (N = 2, 3, …, 8). The symN wavelet has a support range of 2N-1, a vanishing moment of N, and good regularity. Compared to the dbN wavelet, this wavelet shares similarities in continuity, support length, and filter length. However, the symN wavelet exhibits greater symmetry, which, to a certain extent, reduces phase distortion during signal analysis and reconstruction.

[0051] Daubechies constructed the Coiflet wavelet, which has a series of coifN (N = 1, 2, 3, 4, 5). The 2N-order moment of the Coiflet wavelet function Ψ(t) is zero, and the scaling function The 2N-1 moment of is zero. Ψ(t) and The support length of the Coiflet is 6N-1. It has better symmetry than dbN.

[0052] To address the incompatibility between symmetry and accurate signal reconstruction, biorthogonal wavelets were introduced. Two wavelets, called duals, are used for signal decomposition and reconstruction, respectively. Biorthogonal wavelets resolve the conflict between linear phase and orthogonality. Because of their linear phase properties, they are primarily used in signal and image reconstruction. A common approach involves using one function for decomposition and another for reconstruction.

[0053] The difference between biorthogonal wavelets and orthogonal wavelets is that orthogonal wavelets satisfy <Ψ j,k ,Ψ l,m >=δ j,k δ l,m , that is, the basis functions of the wavelet function's expansion and translation are completely orthogonal, and the orthogonality satisfied by the biorthogonal wavelet is <Ψ j,k ,Ψ l,m >=δ j,k , that is, there is orthogonality between the wavelet functions under different scale scaling, but there is no orthogonality between the wavelet function systems obtained by translation between the same scales. Therefore, the wavelet used for decomposition and reconstruction is not the same function, and the corresponding filter cannot be generated by the same wavelet.

[0054] The wavelet function and scaling function of Meyer wavelet are both defined in the frequency domain. It is not compactly supported, but it converges quickly.

[0055] Dmeyer is the discrete Meyer wavelet, which is an FIR-based approximation of the Meyer wavelet and is used for the calculation of fast discrete wavelet transform.

[0056] Gaussian wavelet is the differential form of Gaussian density function. It is a non-orthogonal and non-biorthogonal wavelet without scaling function.

[0057] The Mexican Hat function is the second-order derivative of the Gaussian function. Because its shape resembles the cross-section of a Mexican hat, we call it the Mexican Hat function. It has good localization in both time and frequency, but lacks a scaling function, so this wavelet function is not orthogonal.

[0058] Morlet wavelet is a single-frequency sine function under a Gaussian envelope, has no scaling function, and is a non-orthogonal decomposition.

[0059] The ComplexGaussian wavelet belongs to a class of complex wavelets and has no scaling function.

[0060] A spline function is a type of function that is piecewise smooth and has a certain degree of smoothness at the intersection of each segment. It is referred to as a spline.

[0061] Morlet wavelet is a single-frequency complex sinusoidal modulated Gaussian wave and is also the most commonly used complex-valued wavelet. This wavelet has good resolution in both time and frequency domains. The modification of this wavelet is particularly suitable for the analysis of seismic data.

[0062] Figure 3 The figure shows the wavelet decomposition and reconstruction diagram, which decomposes a relatively complex signal x(t) into several simple signals for processing. After processing, the several simple signals can be synthesized into a signal that is fuller than the original signal.

[0063] A type of Doyle wavelet decomposition is as follows

[0064]

[0065] Where: t represents the time of the sequence signal; Ψ o,r (t) is the wavelet basis vector with scale factor o and shift factor r; FDWT(o, r) represents the wavelet coefficient, which is the sum of x(t) and Ψ o,r (t) is the result of the dot product, which means x(t) in the wavelet basis vector Ψ o,r (t) is the projection length onto the

[0066] The wavelet basis is selected according to the local characteristics of the signal, and 3 to 5 layers of wavelet decomposition are performed to obtain several layers of low-frequency and high-frequency components.

[0067] Figure 4Figure 1 is the structural diagram of an adaptive filter. The input signal x(n) passes through a parameter-adjustable digital filter to generate an output signal y(n). This signal is compared with the desired signal d(n) to form an error signal e(n). The filter parameters are adjusted through an adaptive algorithm to minimize the mean square value of e(n). The filter parameters obtained at the previous moment are used to automatically adjust the filter parameters at the current moment to adapt to the unknown or time-varying statistical characteristics of the signal and noise, thereby achieving optimal filtering.

[0068] The LMS algorithm formula is as follows:

[0069] w(n+1)=w(n)+μe(n)x(n)

[0070] Where: w(n) is the weight vector of the filter. μ is the step size, which controls the update speed of the filter.

[0071] e(n)=d(n)-y ^ (n) is the error signal, representing the difference between the desired output and the actual filtered output. x(n) is the input signal vector. y^(n) = w(n) x(n) is the output of the filter.

[0072] Figure 2 This is the BiTCN structure diagram. The forward and reverse path structures are used to extract and fuse forward and reverse information. BiTCN feature extraction is performed on the wavelet components of each layer. BiTCN uses convolution operations to extract deep time-frequency features from the signal, reducing computational complexity while retaining key information, providing effective features for subsequent classification and clustering.

[0073] From each layer of components after wavelet decomposition, BiTCN (Binary Convolutional Network) is used to extract deep time-frequency features. BiTCN is a feature extraction method based on convolutional neural networks (CNNs). Through convolution operations, it can extract valuable time-frequency features from signals. The advantages of this process are: Reduced computational complexity: BiTCN reduces the amount of computation by bit-based convolution operations. Preserving key information: While maintaining efficient computation, it can retain key information in the signal. The extracted features will provide effective input for subsequent K-Means clustering.

[0074] Step 3: Use the K-Means clustering algorithm to perform cluster analysis on the features extracted by BiTCN and dynamically select the most suitable wavelet decomposition basis and adaptive filter parameters.

[0075] After feature extraction, the K-Means clustering algorithm is used to perform cluster analysis on the extracted features. K-Means clustering divides the data in the feature space into several clusters, identifying different patterns in the signal. Based on the clustering results, the data is divided into different categories. Based on the clustering results, the most appropriate wavelet decomposition basis and adaptive filter are selected for each data category. The clustering results can help select the appropriate processing method for different signal patterns, thereby optimizing signal processing results.

[0076] The K-means clustering algorithm is a classic unsupervised learning method widely used in data analysis and machine learning. Its core idea is to partition a dataset into K clusters such that the sum of the distances between each data point and the centroid of its cluster is minimized.

[0077] The goal of K-means is to partition a dataset into K clusters, such that each data point belongs to the nearest cluster center. By repeatedly adjusting the positions of cluster centers, K-means continuously optimizes the density within the cluster, resulting in clusters that are as compact and separated as possible. K-means clusters data points by minimizing the sum of squared distances within the cluster. A cluster is a collection of data points that are, in some sense, "similar to one another." The cluster center is the average of all the points in the cluster and represents the center of the cluster. K-means adjusts the distribution of clusters through repeated iterations, minimizing the distance between the data points within the cluster and the centroid, and gradually converges.

[0078] The K-means clustering process is divided into two main steps: assignment and update. The following are the detailed steps:

[0079] Select K value: Set the number of clusters.

[0080] Initialize cluster centers: Randomly select K data points as initial cluster centers (centroids).

[0081] Assignment Step: For each point in the dataset, assign it to the cluster corresponding to the nearest cluster center. The "distance" here is usually the Euclidean distance.

[0082] Update Step: Based on the current cluster assignment, recalculate the center of each cluster, that is, calculate the mean of all points in the cluster as the new cluster center.

[0083] Repeat the assignment and update steps: Repeat the assignment and update steps until the cluster centers no longer change (converge) or the specified maximum number of iterations is reached.

[0084] Step 4: reconstruct the wavelet components that have been filtered and fused, extract features from the reconstructed signal, and quantify the filtering and reconstruction effects.

[0085] After selecting an appropriate wavelet decomposition basis and filter, the filtered and fused wavelet components are reconstructed. Using the inverse wavelet transform (IDWT), the low-frequency and high-frequency components are merged to restore the complete load signal. The reconstructed signal significantly removes high-frequency noise while retaining key signal information, providing high-quality data for subsequent feature extraction and analysis.

[0086] The formula for reconstructing the signal using inverse wavelet transform is as follows:

[0087]

[0088] c n is the reconstruction coefficient, ψ n (t) is the wavelet component after filtering and fusion processing.

[0089] Figure 5 This is an example of using wavelet decomposition and adaptive filters to obtain a new signal graph. The high-frequency signal and low-frequency signal obtained by wavelet decomposition and adaptive filters are used as new eigenvalue input signals.

[0090] The reconstructed signal is subjected to feature extraction to extract important statistical features, such as the signal's mean, variance, peak value, valley value, slope change point, etc. These features will be stored in a feature library and provide data support for subsequent applications such as load forecasting and optimized scheduling.

[0091] Evaluate the quality of the reconstructed signal using metrics such as signal-to-noise ratio (SNR) and mean square error (MSE) to quantify the filtering and reconstruction effects. Based on these evaluation results, further adjustments to processing parameters are made.

[0092] Based on the evaluation results, the number of wavelet decomposition layers, threshold selection, filter step size and other hyperparameters are optimized to ensure the accuracy and efficiency of data processing.

[0093] Compared with traditional data processing methods, the present invention shows significant advantages in multi-scale time-frequency analysis and real-time dynamic noise suppression. By combining BiTCN feature extraction and K-Means clustering method, the present invention can effectively improve the processing accuracy and robustness of multi-source heterogeneous flexible load data. In particular, when processing complex signals with instantaneous, non-stationary and noise interference, the present invention can achieve high-precision data fusion and feature extraction. In addition, the combination of wavelet decomposition and adaptive filters can dynamically adapt to changes in load signals, further enhancing the suppression effect of noise and interference. Ultimately, the present invention provides reliable data support for the optimal scheduling, accurate prediction and energy consumption behavior analysis of flexible loads, and provides a strong technical guarantee for intelligent energy management in related fields.

[0094] The second aspect of the present invention relates to a multi-source heterogeneous flexible load data fusion and processing system, which is implemented using the method described in the first aspect of the present invention; the system includes a data acquisition and annotation module, a wavelet decomposition and feature extraction module, a clustering module, and a reconstruction module; the data acquisition and annotation module performs unified data acquisition and annotation for flexible load equipment and sensors from different regions and different types; the wavelet decomposition and feature extraction module decomposes the data into multiple levels through wavelet decomposition, and performs BiTCN feature extraction on the wavelet components of each layer; the clustering module uses the K-Means clustering algorithm to perform cluster analysis on the features extracted by BiTCN, and dynamically selects the most suitable wavelet decomposition basis and adaptive filter parameters; the reconstruction module reconstructs the wavelet components that have been filtered and fused, extracts features from the reconstructed signals, and quantifies the filtering and reconstruction effects.

[0095] The data collection and annotation module uniformly collects and annotates data from flexible load equipment and sensors of different regions and types, including: the annotation content includes sampling frequency, data source, and noise level; the annotated data is uniformly formatted and time-synchronized, missing values ​​are filled using interpolation methods, and data normalization is performed; and the pre-processed data is subjected to preliminary noise suppression using a low-pass filter or a mean filter.

[0096] The wavelet decomposition and feature extraction module decomposes the data into multiple layers through wavelet decomposition and performs BiTCN feature extraction on the wavelet components of each layer, including: selecting the wavelet basis according to the local characteristics of the signal and performing 3 to 5 layers of wavelet decomposition; and performing BiTCN feature extraction on the wavelet components of each layer.

[0097] The clustering module uses the K-Means clustering algorithm to perform cluster analysis on the features extracted by BiTCN and dynamically selects the most suitable wavelet decomposition basis and adaptive filter parameters, including: using the K-Means clustering algorithm to perform cluster analysis on the features extracted by BiTCN to divide the data in the feature space into several categories; dynamically selecting the corresponding wavelet decomposition basis and corresponding adaptive filter parameters for each category of data.

[0098] The reconstruction module reconstructs the wavelet components that have been filtered and fused, extracts features from the reconstructed signal, and quantifies the filtering and reconstruction effects, including: reconstructing the wavelet components that have been filtered and fused through inverse wavelet transform; extracting the mean, variance, peak value, valley value, and slope change point of the signal from the reconstructed signal; and quantifying the filtering and reconstruction effects using signal-to-noise ratio and mean square error.

[0099] The reconstruction module reconstructs the wavelet components that have been filtered and fused, extracts features from the reconstructed signal, and quantifies the filtering and reconstruction effects, including optimizing hyperparameters based on the results of signal quality assessment; the hyperparameters include the number of wavelet decomposition layers, threshold, and filter step size.

[0100] A third aspect of the present invention relates to a terminal, comprising a processor and a storage medium; the storage medium is used to store instructions; and the processor is used to operate according to the instructions to execute the steps of the method described in the first aspect of the present invention.

[0101] A fourth aspect of the present invention relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect of the present invention.

[0102] Finally, it should be noted that the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art will appreciate that the technical solutions of the present invention still include modifications or equivalent substitutions that may be made to the specific embodiments of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention are intended to be covered by the claims of the present invention.

Claims

1. A multi-source heterogeneous flexible load data fusion and processing method, characterized in that: The method comprises the following steps: Unified data collection and labeling for flexible load equipment and sensors from different regions and types; The data is decomposed into multiple layers through wavelet decomposition, and BiTCN features are extracted for the wavelet components of each layer; Use K-Means clustering algorithm to perform cluster analysis on the features extracted by BiTCN and dynamically select the most suitable wavelet decomposition basis and adaptive filter parameters; The wavelet components that have been filtered and fused are reconstructed, features are extracted from the reconstructed signals, and the filtering and reconstruction effects are quantified.

2. The multi-source heterogeneous flexible load data fusion and processing method according to claim 1 is characterized by: The unified data collection and labeling for flexible load devices and sensors from different regions and types includes: The annotation content includes sampling frequency, data source, and noise level; The labeled data is uniformly formatted and time-synchronized, missing values ​​are filled using interpolation methods, and data normalization is performed; Use a low-pass filter or a mean filter to perform preliminary noise suppression on the preprocessed data.

3. The multi-source heterogeneous flexible load data fusion and processing method according to claim 2 is characterized by: The method of performing multi-level decomposition of the data by wavelet decomposition and performing BiTCN feature extraction on the wavelet components of each layer includes: Select the wavelet basis according to the local characteristics of the signal and perform 3 to 5 layers of wavelet decomposition; BiTCN feature extraction is performed on the wavelet components of each layer.

4. The multi-source heterogeneous flexible load data fusion and processing method according to claim 3 is characterized by: The K-Means clustering algorithm is used to perform cluster analysis on the features extracted by BiTCN, and dynamically select the most suitable wavelet decomposition basis and adaptive filter parameters, including: The K-Means clustering algorithm is used to perform cluster analysis on the features extracted by BiTCN, and the data in the feature space are divided into several categories; The corresponding wavelet decomposition basis and the corresponding adaptive filter parameters are dynamically selected for each type of data.

5. The multi-source heterogeneous flexible load data fusion and processing method according to claim 4 is characterized by: The process of reconstructing the wavelet components after filtering and fusion processing, extracting features from the reconstructed signals, and quantifying the filtering and reconstruction effects includes: The wavelet components that have been filtered and fused are reconstructed through inverse wavelet transform; In the reconstructed signal, extract the signal mean, variance, peak value, valley value, and slope change point; The signal-to-noise ratio and mean square error are used to quantify the filtering and reconstruction effects.

6. The multi-source heterogeneous flexible load data fusion and processing method according to claim 5 is characterized by: The process of reconstructing the wavelet components after filtering and fusion processing, extracting features from the reconstructed signals, and quantifying the filtering and reconstruction effects includes: Optimize hyperparameters based on the results of signal quality assessment; The hyperparameters include the number of wavelet decomposition layers, the threshold, and the step size of the filter.

7. A multi-source heterogeneous flexible load data fusion and processing system, characterized by: The system includes a data acquisition and annotation module, a wavelet decomposition and feature extraction module, a clustering module, and a reconstruction module; The data collection and annotation module performs unified data collection and annotation for flexible load devices and sensors from different regions and types; The wavelet decomposition and feature extraction module decomposes the data into multiple layers through wavelet decomposition and performs BiTCN feature extraction on the wavelet components of each layer; The clustering module uses the K-Means clustering algorithm to perform cluster analysis on the features extracted by BiTCN and dynamically selects the most suitable wavelet decomposition basis and adaptive filter parameters; The reconstruction module reconstructs the wavelet components that have been filtered and fused, extracts features from the reconstructed signals, and quantifies the filtering and reconstruction effects.

8. The multi-source heterogeneous flexible load data fusion and processing system according to claim 7 is characterized by: The data collection and annotation module collects and annotates data uniformly for flexible load devices and sensors from different regions and types, including: The annotation content includes sampling frequency, data source, and noise level; The labeled data is uniformly formatted and time-synchronized, missing values ​​are filled using interpolation methods, and data normalization is performed; Use a low-pass filter or a mean filter to perform preliminary noise suppression on the preprocessed data.

9. The multi-source heterogeneous flexible load data fusion and processing method according to claim 8 is characterized by: The wavelet decomposition and feature extraction module decomposes the data into multiple layers through wavelet decomposition and performs BiTCN feature extraction on the wavelet components of each layer, including: Select the wavelet basis according to the local characteristics of the signal and perform 3 to 5 layers of wavelet decomposition; BiTCN feature extraction is performed on the wavelet components of each layer.

10. The multi-source heterogeneous flexible load data fusion and processing method according to claim 9 is characterized by: The clustering module uses the K-Means clustering algorithm to perform cluster analysis on the features extracted by BiTCN and dynamically selects the most suitable wavelet decomposition basis and adaptive filter parameters, including: The K-Means clustering algorithm is used to perform cluster analysis on the features extracted by BiTCN, and the data in the feature space are divided into several categories; The corresponding wavelet basis and the corresponding adaptive filter parameters are dynamically selected for each type of data.

11. The multi-source heterogeneous flexible load data fusion and processing method according to claim 10, characterized in that: The reconstruction module reconstructs the wavelet components that have been filtered and fused, extracts features from the reconstructed signal, and quantifies the filtering and reconstruction effects, including: The wavelet components that have been filtered and fused are reconstructed through inverse wavelet transform; In the reconstructed signal, extract the signal mean, variance, peak value, valley value, and slope change point; The signal-to-noise ratio and mean square error are used to quantify the filtering and reconstruction effects.

12. The multi-source heterogeneous flexible load data fusion and processing method according to claim 11, characterized in that: The reconstruction module reconstructs the wavelet components that have been filtered and fused, extracts features from the reconstructed signal, and quantifies the filtering and reconstruction effects, including: Optimize hyperparameters based on the results of signal quality assessment; The hyperparameters include the number of wavelet decomposition layers, the threshold, and the step size of the filter.

13. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 6.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images