Abnormal mode data processing system driven by power marketing big data

Abstract

The invention relates to the technical field of data processing, in particular to an abnormal mode data processing system driven by power marketing big data, which comprises a distributed collaborative acquisition module for constructing a space-time alignment three-dimensional data stream, a multi-modal feature reconstruction module for separating periodic noise and quantizing environmental interference, and a data processing module for processing abnormal mode data. The resistance feature decoupling module generates a purification feature vector set and a noise confidence index through orthogonal projection, and the dynamic algorithm adaptation module dynamically schedules an isolated forest algorithm, a weighted distance measurement algorithm and a sparse self-encoding clustering algorithm according to the noise confidence index. The behavior chain verification module establishes a combined physical rule verification mechanism of an environment temperature threshold value, a load deviation degree and an equipment state, and the closed-loop strategy engine module adaptively adjusts a feature decoupling loss function weight according to a decision boundary offset, so that the accuracy and the environmental adaptability of real electricity consumption abnormity identification in a complex noise environment are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an anomaly pattern data processing system driven by big data in electricity marketing. Background Technology

[0002] Electricity marketing big data refers to the process by which the power industry, within a smart grid environment, uses advanced information technology to collect, process, and analyze large-scale electricity consumption data sets. Its aim is to optimize energy allocation, improve operational efficiency, and enhance customer service experience. Through sophisticated algorithms, it mines electricity consumption patterns, identifies potential demand response opportunities, supports accurate load forecasting and personalized marketing strategy development, thereby driving decisions to optimize resource allocation, reduce network losses, and enhance service reliability. In the development of modern power systems, electricity marketing big data plays a core information-driven role, providing enterprises with market trend insights and a deep understanding of user behavior, helping the industry evolve towards efficiency, intelligence, and sustainability.

[0003] In the context of big data in electricity marketing, the processing of abnormal pattern data suffers from insufficient identification accuracy under high-dimensional noise interference. This is particularly true in massive electricity consumption sequences, where data sparsity and complex temporal characteristics make it difficult for algorithms to accurately distinguish between genuine abnormal patterns, such as electricity fraud, and normal behavioral fluctuations. For example, in the analysis of user electricity consumption time series collected by smart meters, standard clustering models are sensitive to changes in regular usage caused by holidays or sudden weather changes, producing false positives and incorrectly classifying normal users as abnormal, thus weakening the reliability of load forecasting and the effectiveness of demand response strategies. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an abnormal pattern data processing system driven by big data in electricity marketing. This invention solves the problem of insufficient accuracy in identifying abnormal patterns under high-dimensional noise interference in electricity consumption time series, which makes it impossible to effectively distinguish between real abnormal electricity consumption and normal dynamic fluctuations.

[0005] To solve the above-mentioned technical problems, the specific technical solution of the present invention is as follows: The present invention provides a power marketing big data-driven abnormal pattern data processing system, comprising: The distributed collaborative acquisition module collects power time-series data and environmental parameters from smart meters, and outputs a three-dimensional data stream including time, physical quantities, and spatial dimensions. The multimodal feature reconstruction module receives the three-dimensional data stream, applies time window segmentation to extract spectral features and environmental response features, and generates a load feature map; The adversarial feature decoupling module receives the load feature map, performs orthogonal projection of the feature space to separate the fluctuation feature vector and the noise confidence index, and outputs the purified fluctuation feature vector set. The dynamic algorithm adaptation module receives the set of fluctuation feature vectors and the noise confidence index, and dynamically activates and optimizes the algorithm combination according to the feature sparsity and noise confidence to generate a set of suspected abnormal points. The behavior chain verification module receives the set of suspected abnormal points, performs multi-dimensional verification to distinguish between abnormal events and reasonable fluctuations, and outputs a list of verified abnormal events and samples of normal dynamic fluctuations that have passed verification. The closed-loop strategy engine module receives the list of verified abnormal events and the verified normal dynamic fluctuation samples. It triggers marketing audit instructions for events in the list of verified abnormal events, handles real electricity consumption anomalies, imports the verified normal dynamic fluctuation samples into the historical feature library, adjusts the periodic feature extraction parameters of the multimodal feature reconstruction module through an incremental learning mechanism, and adjusts the loss function weight of the adversarial feature decoupling module according to the decision boundary offset.

[0006] Furthermore, the distributed collaborative acquisition module of the power marketing big data-driven abnormal pattern data processing system described in this invention is configured as follows: Terminal devices are deployed at power grid nodes to periodically collect instantaneous values ​​of voltage and current from smart meters, and to obtain meteorological service parameters and the geographical coordinates of the power grid. The three-dimensional data stream includes timestamps, meter identifiers, power values, temperature values, humidity values, longitude values, and latitude values; The periodically collected instantaneous values ​​of voltage and current from the smart meter, along with the acquired meteorological service parameters and power grid geographic coordinates, are input into the multimodal feature reconstruction module for segmentation.

[0007] Furthermore, the multimodal feature reconstruction module of the power marketing big data-driven abnormal pattern data processing system of the present invention is configured as follows: Spectral feature extraction is performed on the input three-dimensional data stream: the daily periodic component and weekly / monthly periodic features are analyzed using fast Fourier transform to generate a standardized periodic feature spectrum; environmental response feature extraction is performed: a nonlinear correlation model between load value and temperature value is constructed to output a set of temperature sensitivity coefficients; the standardized periodic feature spectrum and the set of temperature sensitivity coefficients are fused to generate a basic load feature map; Identify the seasonal dependence patterns in the basic load feature map and add seasonal identification factors to the map; input the load feature map with seasonal identification factors into the adversarial feature decoupling module to perform orthogonal projection processing.

[0008] Furthermore, the adversarial feature decoupling module of the power marketing big data-driven abnormal pattern data processing system of the present invention is configured as follows: The input load feature map is decomposed into a fluctuation feature subspace and a noise feature subspace by an encoder; feature samples are reconstructed using a generator, and the error amplitude is calculated by comparing the reconstructed samples with the original samples by a discriminator; a fluctuation feature vector set and a noise confidence index are generated based on the error amplitude, wherein the fluctuation feature vector set contains 64-dimensional clean features, and the noise confidence index is a continuous value; the fluctuation feature vector set and the noise confidence index are transmitted as output to the dynamic algorithm adaptation module.

[0009] Furthermore, in the power marketing big data-driven abnormal pattern data processing system of the present invention, the dynamic algorithm adaptation module is configured as follows: The system monitors the feature sparsity, kurtosis, and noise confidence index of the input features in real time. If the feature sparsity exceeds a set threshold, the sparse autoencoder clustering algorithm is called to process the features. If the feature kurtosis exceeds a set threshold, the weighted distance metric algorithm is called to process the features. If the noise confidence index reaches a preset threshold, the depth parameters of the isolated forest algorithm are adjusted before detection is performed. All suspected points output by the algorithms are integrated to generate an abnormal suspected point set with marked floating radius decision boundaries. The abnormal suspected point set with decision boundary markings is then output to the behavior chain verification module for multi-dimensional verification.

[0010] Furthermore, the behavior chain verification module of the power marketing big data-driven abnormal pattern data processing system described in this invention is configured as follows: The system obtains the current ambient temperature value, the instantaneous load value of the power grid, and the operating status code of the power distribution equipment from the real-time monitoring source. It retrieves the historical load baseline data of the target user, analyzes the operating status code of the power distribution equipment together with the historical load baseline data, generates a dynamic constraint rule group, and detects the deviation of the ambient temperature value, the instantaneous load value and the historical load baseline data, as well as whether the operating status code of the power distribution equipment meets the preset legality conditions based on the dynamic constraint rule group. If all conditions are met, the corresponding suspected anomaly is removed from the list; the suspected anomalies that have not been removed from the list are aggregated to generate a list of verification anomaly events, and the list of verification anomaly events is output to the closed-loop strategy engine module to drive the update of the marketing audit strategy library.

[0011] Furthermore, in the power marketing big data-driven abnormal pattern data processing system of the present invention, the closed-loop strategy engine module is configured as follows: Based on the newly imported historical feature library and the verified samples, the phase angle of the periodic feature components is recalculated and the corresponding hash index structure is updated. According to the change magnitude of the recalculated phase angle of the periodic components, the incremental learning mechanism is triggered to generate an instruction to adjust the weight of the loss function of the adversarial feature decoupling module. The adjustment instruction is executed to update the weight parameters of the loss function of the adversarial feature decoupling module. The updated phase angle of the periodic components, hash index structure and weight parameters of the loss function are synchronized to the multimodal feature reconstruction module and the adversarial feature decoupling module to complete the cross-module parameter collaborative refresh.

[0012] Furthermore, the power marketing big data-driven abnormal pattern data processing system of the present invention also includes: The noise confidence index is generated by the adversarial feature decoupling module and input into the dynamic algorithm adaptation module; The dynamic algorithm adaptation module performs dynamic switching of optimization algorithms based on noise confidence, including: preset noise confidence index threshold range; When the noise confidence index falls into the low noise range, the isolated forest algorithm is activated to identify the core suspect point; when the noise confidence index falls into the medium noise range, the weighted distance metric algorithm is activated to perform anti-interference anomaly detection. When the noise confidence index falls into the high noise range, the sparse autoencoder clustering algorithm is activated to perform feature dimensionality reduction and density clustering. The selected optimization algorithm processes the set of fluctuating feature vectors based on the current input features to generate a set of suspected anomalies.

[0013] Furthermore, the power marketing big data-driven abnormal pattern data processing system of the present invention further includes: The behavior chain verification module outputs the verified normal dynamic fluctuation sample to the closed-loop strategy engine module; The closed-loop strategy engine module imports the verified normal dynamic fluctuation samples into the historical feature library and updates the periodic feature extraction parameters of the multimodal feature reconstruction module through an incremental learning mechanism, so that the spectral feature extraction of the multimodal feature reconstruction module adapts to the new sample data, and establishes an operation mode in which the verification results drive the parameter update of the feature reconstruction module.

[0014] Furthermore, in the power marketing big data-driven abnormal pattern data processing system of the present invention, the step of adjusting the loss function weight of the adversarial feature decoupling module according to the decision boundary offset includes: Real-time monitoring of the decision boundary offset of the abnormal suspect point set markers generated by the dynamic algorithm adaptation module; When the decision boundary offset falls into the preset high offset range, the weight coefficient of the fluctuation feature reconstruction loss is reduced. When the decision boundary offset falls into the preset low offset range, the target weight of the noise discrimination loss is increased; The dynamically adjusted loss function weights are synchronized to the output layer executor of the adversarial feature decoupling module.

[0015] Beneficial effects of this invention; This invention eliminates acquisition noise by constructing a spatiotemporally aligned 3D data stream through a distributed collaborative acquisition module, separates periodic background noise and quantifies environmental interference through a multimodal feature reconstruction module, and performs orthogonal projection to generate a purified feature vector set and noise confidence index, thereby reducing high-dimensional noise interference from the source. A dynamic algorithm adaptation module dynamically schedules the Isolation Forest algorithm, Weighted Distance Metric algorithm, and Sparse Autoencoder Clustering algorithm based on the noise confidence index, combined with the multi-parameter physical rule joint detection mechanism of the Behavior Chain Verification module, to achieve accurate differentiation between real abnormal events and environmental response fluctuations. A closed-loop strategy engine module incrementally updates periodic feature parameters based on verification results and adaptively adjusts the feature decoupling loss function weights according to the decision boundary offset, forming a closed-loop learning system of feature extraction, anomaly detection, and feedback optimization. This continuously improves the accuracy and environmental adaptability of anomaly identification in complex noise environments, effectively solving the technical problem of insufficient accuracy in identifying abnormal patterns and normal dynamic fluctuations in power consumption time-series data. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the drawings without creative effort.

[0017] Figure 1 The system architecture diagram of the abnormal pattern data processing system driven by big data for power marketing provided in the embodiments of the present invention is shown. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. The technical solutions provided by various embodiments of this invention will be described in detail below with reference to the accompanying drawings. To better understand the objectives of this invention, it will be described in further detail below.

[0019] Please see Figure 1 The present invention provides an abnormal pattern data processing system driven by big data in power marketing, comprising: The distributed collaborative acquisition module collects power time-series data and environmental parameters from smart meters, and outputs a three-dimensional data stream including time, physical quantities, and spatial dimensions. The multimodal feature reconstruction module receives the three-dimensional data stream, applies time window segmentation to extract spectral features and environmental response features, and generates a load feature map; The adversarial feature decoupling module receives the load feature map, performs orthogonal projection of the feature space to separate the fluctuation feature vector and the noise confidence index, and outputs the purified fluctuation feature vector set. The dynamic algorithm adaptation module receives the set of fluctuation feature vectors and the noise confidence index, and dynamically activates and optimizes the algorithm combination according to the feature sparsity and noise confidence to generate a set of suspected abnormal points. The behavior chain verification module receives the set of suspected abnormal points, performs multi-dimensional verification to distinguish between abnormal events and reasonable fluctuations, and outputs a list of verified abnormal events and samples of normal dynamic fluctuations that have passed verification. The closed-loop strategy engine module receives the list of verified abnormal events and the verified normal dynamic fluctuation samples. It triggers marketing audit instructions for events in the list of verified abnormal events, handles real electricity consumption anomalies, imports the verified normal dynamic fluctuation samples into the historical feature library, adjusts the periodic feature extraction parameters of the multimodal feature reconstruction module through an incremental learning mechanism, and adjusts the loss function weight of the adversarial feature decoupling module according to the decision boundary offset.

[0020] The distributed collaborative data acquisition module periodically collects instantaneous values ​​of voltage and current from electricity meters through intelligent terminal devices deployed at power grid nodes. It simultaneously accesses a meteorological service interface to obtain temperature and humidity parameters and extracts spatial coordinate data using the power grid geographic information system. This module employs timestamp alignment technology to integrate multi-source heterogeneous data, generating a standardized three-dimensional data stream that includes time, physical quantity, and spatial dimensions.

[0021] After receiving the 3D data stream, the multimodal feature reconstruction module first applies a sliding time window segmentation technique to divide the data into units. For each data unit, a Fast Fourier Transform is performed to extract the daily and weekly / monthly periodic feature spectra. Simultaneously, a load-temperature nonlinear regression model is constructed, and the temperature sensitivity coefficient matrix is ​​calculated. The periodic feature spectra and the temperature sensitivity coefficient matrix are then fused using tensors to generate a basic load feature map. A seasonal pattern recognition algorithm is used to annotate the basic map with seasonal identifiers, forming a load feature map with seasonal characteristics.

[0022] The adversarial feature decoupling module employs an encoder-generator-discriminator architecture to process the load feature map. The encoder performs orthogonal basis vector decomposition, separating the fluctuation feature subspace and the noise feature subspace. After the generator reconstructs the feature samples, the discriminator calculates the Euclidean distance between the original sample and the reconstructed sample as the error magnitude. Based on the error magnitude, a fluctuation feature vector set containing 64-dimensional cleaned features and a continuous noise confidence index are generated.

[0023] The dynamic algorithm adaptation module monitors the sparsity distribution, kurtosis coefficient, and noise confidence index of the input features in real time. When the feature sparsity exceeds a set threshold, the sparse autoencoder clustering algorithm is activated; when the feature kurtosis exceeds a threshold, the weighted Mahalanobis distance algorithm is invoked; and the tree depth parameters of the isolated forest algorithm are dynamically adjusted based on the threshold range of the noise confidence index. The suspected points output by each algorithm are processed by the decision boundary fusion module to generate a set of abnormal suspected points with floating radius labels.

[0024] The behavior chain verification module retrieves the real-time status codes of power distribution equipment and constructs dynamic constraint rules by associating them with the user's historical load baseline. The physical logic chain verification engine detects the compliance of ambient temperature values, load deviation rates, and equipment status codes, removing anomaly markers from suspicious points that meet the constraint conditions. Suspicious points that are not removed from the markers are aggregated to generate a list of verification anomaly events.

[0025] The closed-loop strategy engine module imports validated normal samples into the historical feature library, triggering an incremental learning mechanism to recalculate the periodic feature phase angle and update the hash index structure. Based on the phase angle change magnitude, it generates a loss function weight adjustment instruction to adjust the weight ratio of the fluctuation feature reconstruction loss and noise discrimination loss in the adversarial feature decoupling module. The updated periodic component parameters, hash index, and loss weights are synchronously fed back to the feature reconstruction module and the feature decoupling module, completing the collaborative optimization of system parameters.

[0026] The modules form a closed-loop technical path: 3D data flow drives feature reconstruction; feature maps are decoupled to generate purified feature vectors; dynamic algorithms are adapted to generate a set of suspected points; behavioral chain verification outputs verification results; and the strategy engine module optimizes feature extraction and decoupling parameters based on the verification results. The system achieves continuous evolution of processing capabilities through an incremental learning mechanism, solving the challenge of identifying abnormal patterns and normal fluctuations in electricity marketing data.

[0027] Specifically, the distributed collaborative acquisition module of the power marketing big data-driven abnormal pattern data processing system described in this invention is configured as follows: Terminal devices are deployed at power grid nodes to periodically collect instantaneous values ​​of voltage and current from smart meters, and to obtain meteorological service parameters and the geographical coordinates of the power grid. The three-dimensional data stream includes timestamps, meter identifiers, power values, temperature values, humidity values, longitude values, and latitude values; The periodically collected instantaneous values ​​of voltage and current from the smart meter, along with the acquired meteorological service parameters and power grid geographic coordinates, are input into the multimodal feature reconstruction module for segmentation.

[0028] The distributed collaborative data acquisition module periodically collects instantaneous voltage and current data from smart meters through intelligent monitoring terminal equipment deployed in power grid substations and distribution transformer areas, with the sampling frequency synchronized with the power grid's power frequency cycle. This module obtains environmental parameters such as temperature and humidity in real time through a meteorological service interface and extracts the longitude and latitude spatial coordinates of the equipment installation location by calling the power grid geographic information system interface. A high-precision time synchronization protocol is used to align the timestamps of multi-source data, forming a structured three-dimensional data stream containing fields such as timestamp, meter identifier, power value, temperature value, humidity value, longitude value, and latitude value.

[0029] During data acquisition, instantaneous voltage and current values ​​from smart meters are transmitted to edge computing nodes via power line carrier communication, and power values ​​are calculated by integrating these instantaneous values. Meteorological service parameters are obtained through an application programming interface (API), and the power grid's geographic coordinates are retrieved in real-time from a spatial database. The three-dimensional data stream uses timestamps as the primary index key to establish a data cube structure with time, physical quantity, and spatial dimensions. This data cube is transmitted to the multimodal feature reconstruction module via a message queue, providing standardized input for subsequent time window segmentation.

[0030] After receiving the 3D data stream, the multimodal feature reconstruction module first divides the time window into sliding time windows based on the timestamp sequence. Within each window, power values ​​are extracted using a Fast Fourier Transform to obtain spectral features, while temperature and humidity values ​​are used to construct an environmental response model. Longitude and latitude values ​​participate in spatial grid encoding to achieve spatial alignment of multidimensional features. The data cube structure ensures the continuity of the time series, the integrity of physical quantities, and the consistency of spatial location, establishing a data foundation for feature reconstruction.

[0031] This invention establishes a complete data acquisition chain: the terminal device layer acquires physical signals, the data transmission layer fuses multi-source heterogeneous data, and the data modeling layer generates a standardized 3D data stream. Each link adopts an industrial Internet of Things (IoT) architecture, conforms to power system data acquisition specifications, and provides highly reliable input for subsequent feature processing.

[0032] Specifically, the multimodal feature reconstruction module of the power marketing big data-driven abnormal pattern data processing system of the present invention is configured as follows: Spectral feature extraction is performed on the input three-dimensional data stream: the daily periodic component and weekly / monthly periodic features are analyzed using fast Fourier transform to generate a standardized periodic feature spectrum; environmental response feature extraction is performed: a nonlinear correlation model between load value and temperature value is constructed to output a set of temperature sensitivity coefficients; the standardized periodic feature spectrum and the set of temperature sensitivity coefficients are fused to generate a basic load feature map; Identify the seasonal dependence patterns in the basic load feature map and add seasonal identification factors to the map; input the load feature map with seasonal identification factors into the adversarial feature decoupling module to perform orthogonal projection processing.

[0033] After receiving the 3D data stream output by the distributed collaborative acquisition module, the multimodal feature reconstruction module first performs a spectral feature extraction operation. This operation applies the Fast Fourier Transform algorithm to process the power value time series, decomposing it into daily and weekly / monthly periodic components. A standardized periodic feature spectrum is generated through spectral energy normalization, eliminating amplitude deviations caused by differences in electricity consumption levels among different users. This periodic feature spectrum characterizes the inherent regularity of power load changes over time.

[0034] Simultaneously, environmental response feature extraction is performed to construct a nonlinear regression model of load and temperature values. Kernel function mapping technology is used to establish the correlation between temperature changes and load fluctuations, outputting a set of temperature sensitivity coefficients. This coefficient set contains sensitivity parameters for load changes within different temperature ranges, quantifying the impact of environmental factors on power consumption. The temperature sensitivity coefficient set and the standardized periodic feature spectrum have the same time dimension alignment benchmark.

[0035] The feature fusion stage involves tensor superposition of the standardized periodic feature spectrum and the temperature sensitivity coefficient set. A feature dimensionality expansion technique maps these two heterogeneous features to a unified high-dimensional space, generating a baseline load feature map. This map integrates time-periodic features and environmental response characteristics to form a feature benchmark model of user electricity consumption behavior. Each feature node in the map contains a timestamp, spatial coordinates, and a multi-dimensional feature vector.

[0036] Seasonal pattern recognition is performed based on the baseline load feature map, and time series pattern mining algorithms are used to analyze the seasonal dependence patterns in historical data. The recognition results are labeled as seasonality identifiers as discrete categorical variables and appended to the corresponding time nodes of the baseline load feature map. The seasonality identifiers characterize the feature shift of electricity consumption patterns within different seasonal cycles, enhancing the temporal context information of the feature map.

[0037] Finally, a load feature map with seasonality indicators is generated and transmitted to the adversarial feature decoupling module via a data interface. This map serves as input data for orthogonal projection processing, and its multidimensional feature structure is adapted to the input requirements of the feature space decomposition algorithm. The periodicity, environmental sensitivity, and seasonality features fused in the feature map provide a complete information foundation for subsequent feature decoupling.

[0038] This invention establishes a progressive relationship in feature processing: temporal periodic features are extracted from raw data through spectral transformation; the influence of external factors is quantified through environmental modeling; a comprehensive representation model is constructed through feature fusion; and seasonality markers enhance the temporal dimension information. The outputs of each stage maintain temporal and spatial alignment, forming a structured feature evolution process.

[0039] Specifically, the adversarial feature decoupling module of the power marketing big data-driven abnormal pattern data processing system of the present invention is configured as follows: The input load feature map is decomposed into a fluctuation feature subspace and a noise feature subspace by an encoder; feature samples are reconstructed using a generator, and the error amplitude is calculated by comparing the reconstructed samples with the original samples by a discriminator; a fluctuation feature vector set and a noise confidence index are generated based on the error amplitude, wherein the fluctuation feature vector set contains 64-dimensional clean features, and the noise confidence index is a continuous value; the fluctuation feature vector set and the noise confidence index are transmitted as output to the dynamic algorithm adaptation module.

[0040] After receiving the load feature map with seasonality indicators, the adversarial feature decoupling module first performs an orthogonal decomposition operation of the feature space through the encoder network. This encoder employs orthogonal basis vector projection technology to map the input feature map to two mutually exclusive subspaces: a fluctuation feature subspace representing the essential laws of electricity consumption behavior, and a noise feature subspace reflecting environmental noise and random interference. The subspace decomposition process satisfies orthogonality constraints, ensuring the completeness of feature decoupling.

[0041] The generator network reconstructs feature samples based on the fluctuation feature subspace and rebuilds the load feature representation through deconvolution. The discriminator network simultaneously receives the original load feature map and the reconstructed feature samples, and uses a feature distance calculation algorithm to quantify the degree of difference between the two. This difference value serves as the reconstruction error magnitude, reflecting the degree of information integrity loss during feature decoupling.

[0042] The error amplitude data stream drives a feature cleansing mechanism, where the fluctuating feature subspace undergoes dimensionality compression to generate a fluctuating feature vector set. This vector set contains cleaned feature vectors of fixed dimensions, eliminating high-dimensional noise interference from the original data. Simultaneously, the reconstruction error amplitude is converted into a normalized noise confidence index, which is a continuous value representing the probability level of the current sample being contaminated by noise.

[0043] The final output consists of a data combination containing a fluctuation feature vector set and a noise confidence index, which is transmitted to the dynamic algorithm adaptation module via a high-speed data bus. The fluctuation feature vector set serves as the core feature input, while the noise confidence index acts as an auxiliary parameter for algorithm selection; together, they form the data foundation for the downstream anomaly detection process. The feature dimensions and data types are strictly adapted to the input specifications of the dynamic algorithm adaptation module.

[0044] This invention constructs a closed-loop verification mechanism for feature processing: spatial decomposition establishes the foundation for feature separation, generation and discrimination structures verify the decoupling effect, and error analysis drives the generation of clean features. Subspace orthogonality constraints ensure the theoretical completeness of feature decoupling, and feature distance calculation provides a quantifiable basis for evaluating the decoupling effect. The output data combination simultaneously includes clean features and noise evaluation indicators, forming a multi-dimensional feature quality description system.

[0045] Specifically, in the power marketing big data-driven abnormal pattern data processing system of the present invention, the dynamic algorithm adaptation module is configured as follows: The system monitors the feature sparsity, kurtosis, and noise confidence index of the input features in real time. If the feature sparsity exceeds a set threshold, the sparse autoencoder clustering algorithm is called to process the features. If the feature kurtosis exceeds a set threshold, the weighted distance metric algorithm is called to process the features. If the noise confidence index reaches a preset threshold, the depth parameters of the isolated forest algorithm are adjusted before detection is performed. All suspected points output by the algorithms are integrated to generate an abnormal suspected point set with marked floating radius decision boundaries. The abnormal suspected point set with decision boundary markings is then output to the behavior chain verification module for multi-dimensional verification.

[0046] The dynamic algorithm adaptation module receives the fluctuating feature vector set and noise confidence index input in real time. Through the feature analysis engine, it continuously monitors the sparsity distribution, statistical kurtosis, and noise confidence index of the feature vectors. Feature sparsity reflects the distribution density of effective information points in the feature space, statistical kurtosis characterizes the steepness of the feature value distribution, and noise confidence index measures the degree of interference with the quantified data. This monitoring process employs a sliding time window update mechanism to ensure the timeliness of parameter evaluation.

[0047] When the feature sparsity exceeds a preset threshold, the sparse autoencoder clustering algorithm is activated. This algorithm extracts key features through dimensionality reduction in the encoding layer and performs density clustering analysis in the hidden layer space, effectively identifying anomalous clustering patterns in high-dimensional sparse features. The algorithm outputs a preliminary set of suspected points labeled with cluster outlier coefficients.

[0048] When the kurtosis of the feature distribution exceeds a preset threshold, the weighted distance metric algorithm is invoked. This algorithm is an improvement on Mahalanobis distance calculation, assigning differentiated weight coefficients to different dimensions based on the feature variance distribution, thereby enhancing the robustness of anomaly detection for non-Gaussian distributed features. The algorithm outputs a set of candidate anomalies with distance deviation.

[0049] When the noise confidence index reaches a preset threshold, the Isolation Forest algorithm's depth adjustment mechanism is triggered. By dynamically increasing the decision tree depth parameter, the algorithm's tolerance for high-noise data is improved, reducing the false alarm rate while maintaining anomaly detection accuracy. After the adjusted Isolation Forest algorithm performs detection, it outputs an anomaly score set based on path length.

[0050] The outputs of multiple algorithms are processed by the decision boundary fusion module, which integrates the suspect tags generated by different algorithms using fuzzy logic rules. The fusion process generates a dynamically floating radius decision boundary based on clustering outlier coefficients, distance deviation, and anomaly scores. This boundary adaptively adjusts the anomaly determination threshold range. Finally, a set of anomaly suspect points with floating radius tags is generated, where each suspect point includes spatial coordinates, a timestamp, and anomaly probability value.

[0051] The set of suspected anomalies is transmitted to the behavior chain verification module through a standardized interface, with the data format adapted to the input specifications of the multi-dimensional verification engine. The transmission process retains temporal and spatial dimension information, ensuring the traceability of the suspected anomaly set to the original power data.

[0052] This invention establishes a logical hierarchy for algorithm scheduling: a monitoring layer continuously quantifies feature characteristics, a decision layer triggers algorithm instances based on the quantification results, an execution layer outputs preliminary detection results, and a fusion layer generates a unified judgment standard. The multi-algorithm parallel processing architecture achieves optimized resource allocation through feature attribute discrimination, and a floating boundary mechanism improves the environmental adaptability of anomaly detection.

[0053] Specifically, the behavior chain verification module of the power marketing big data-driven abnormal pattern data processing system described in this invention is configured as follows: The system obtains the current ambient temperature value, the instantaneous load value of the power grid, and the operating status code of the power distribution equipment from the real-time monitoring source. It retrieves the historical load baseline data of the target user, analyzes the operating status code of the power distribution equipment together with the historical load baseline data, generates a dynamic constraint rule group, and detects the deviation of the ambient temperature value, the instantaneous load value and the historical load baseline data, as well as whether the operating status code of the power distribution equipment meets the preset legality conditions based on the dynamic constraint rule group. If all conditions are met, the corresponding suspected anomaly is removed from the list; the suspected anomalies that have not been removed from the list are aggregated to generate a list of verification anomaly events, and the list of verification anomaly events is output to the closed-loop strategy engine module to drive the update of the marketing audit strategy library.

[0054] The behavior chain verification module first obtains the current ambient temperature, instantaneous load value of the power grid, and operating status codes of power distribution equipment in real time from the power grid monitoring system. Simultaneously, it retrieves the target user load baseline data stored in the historical feature database, which contains typical electricity consumption patterns under the same seasonal identifier. The equipment operating status codes include discrete variables such as equipment operating status flags and maintenance status flags.

[0055] The rule engine maps the operating status codes of power distribution equipment to historical load baseline data to generate dynamic constraint rule sets. These rule sets establish a correspondence between ambient temperature threshold ranges, permissible load deviation intervals, and permissible equipment status conditions, forming a quantifiable chain of physical constraints. Rule parameters are dynamically adjusted based on equipment type to adapt to the verification needs of different power distribution environments.

[0056] Multi-parameter joint detection is performed based on dynamic constraint rule groups: calculating the relative deviation between the instantaneous load value and historical load baseline data; comparing whether the ambient temperature value is within the preset permissible range; and verifying whether the equipment operating status code conforms to the normal operation indicator. All three detections must simultaneously meet preset legality conditions, defined as: temperature value within the threshold range, load deviation not exceeding the permissible range, and no abnormal flags in the equipment status code.

[0057] When the associated data of a suspected anomaly fully meets the validity conditions, the anomaly status mark of that point in the suspected point set is removed. The mark removal operation updates the suspected point's status attributes and records the environmental temperature value, load deviation degree, and other parameters on which the removal decision was based. Suspected points that do not meet the conditions retain their original anomaly marks and confidence scores.

[0058] Finally, all suspected anomalies that have not been de-marked are aggregated and grouped by time and space to generate a verification anomaly event list. This list includes event timestamps, geographic coordinates, anomaly classification codes, and verification decision logs. The verification anomaly event list is transmitted to the closed-loop strategy engine module via a data interface, triggering an incremental update operation of the marketing audit strategy library.

[0059] This invention constructs a traceable verification logic chain: real-time data and historical baselines provide verification basis, a dynamic rule engine generates computable constraints, multi-parameter joint detection achieves cross-validation, and a tag update mechanism ensures that the state is traceable. The verification result output structure adapts to the downstream strategy update requirements, forming a complete closed loop for anomaly event decision-making.

[0060] Specifically, in the power marketing big data-driven abnormal pattern data processing system of the present invention, the closed-loop strategy engine module is configured as follows: Based on the newly imported historical feature library and the verified samples, the phase angle of the periodic feature components is recalculated and the corresponding hash index structure is updated. According to the change magnitude of the recalculated phase angle of the periodic components, the incremental learning mechanism is triggered to generate an instruction to adjust the weight of the loss function of the adversarial feature decoupling module. The adjustment instruction is executed to update the weight parameters of the loss function of the adversarial feature decoupling module. The updated phase angle of the periodic components, hash index structure and weight parameters of the loss function are synchronized to the multimodal feature reconstruction module and the adversarial feature decoupling module to complete the cross-module parameter collaborative refresh.

[0061] The closed-loop strategy engine module first receives the verified samples output by the behavior chain verification module and incrementally imports them into the historical feature database. Based on the newly imported sample set, the phase angle parameters of the load cycle feature components are recalculated. This calculation process uses a spectral analysis algorithm to identify feature phase shifts. The hash index structure is simultaneously updated, and the feature index is reconstructed using an improved locality-sensitive hashing method to improve the retrieval efficiency of similar samples.

[0062] The recalculated phase angle of the periodic component is compared with the historical baseline value to quantify the magnitude of the phase angle change. When the magnitude of the change exceeds a preset sensitivity threshold, an incremental learning mechanism is triggered to generate a weight adjustment instruction. This instruction contains the weight adjustment coefficients for the fluctuation feature reconstruction loss function and the noise discrimination loss function in the adversarial feature decoupling module. The adjustment coefficients are dynamically generated as positive or negative adjustments based on the phase shift direction.

[0063] When executing the adjustment command, the weight parameters of the loss function in the adversarial feature decoupling module are updated. The weight ratio of the fluctuation feature reconstruction loss term and the noise discrimination loss term in the neural network loss layer is adjusted through the parameter configuration interface. The weight update process uses a gradient smoothing algorithm to avoid abrupt parameter changes and maintain the stability of model training.

[0064] The updated periodic component phase angle parameters, optimized hash index structure, and adjusted loss function weight parameters are encapsulated into a parameter update package. This package is synchronously transmitted to the multimodal feature reconstruction module and the adversarial feature decoupling module via a cross-module communication protocol. The multimodal feature reconstruction module receives the updated periodic component phase angles to correct the feature extraction benchmark, and the hash index structure to optimize the feature retrieval process. The adversarial feature decoupling module loads the new loss function weight parameters to update the network structure, achieving coordinated parameter updates between the two core processing modules.

[0065] This invention constructs a closed-loop optimization mechanism: sample import updates the basic data of the feature library, phase angle recalculation quantifies changes in feature distribution, the magnitude of change triggers incremental learning decisions, weight adjustment optimizes feature decoupling effects, and cross-module synchronous updates of system-level parameters are achieved. Each step is connected through standardized data interfaces, forming an adaptive system optimization closed loop to ensure continuous evolution of processing capabilities.

[0066] Specifically, the power marketing big data-driven abnormal pattern data processing system of the present invention further includes: The noise confidence index is generated by the adversarial feature decoupling module and input into the dynamic algorithm adaptation module; The dynamic algorithm adaptation module performs dynamic switching of optimization algorithms based on noise confidence, including: preset noise confidence index threshold range; When the noise confidence index falls into the low noise range, the isolated forest algorithm is activated to identify the core suspect point; when the noise confidence index falls into the medium noise range, the weighted distance metric algorithm is activated to perform anti-interference anomaly detection. When the noise confidence index falls into the high noise range, the sparse autoencoder clustering algorithm is activated to perform feature dimensionality reduction and density clustering. The selected optimization algorithm processes the set of fluctuating feature vectors based on the current input features to generate a set of suspected anomalies.

[0067] The noise confidence index generated by the adversarial feature decoupling module is used as a key control parameter input to the dynamic algorithm adaptation module. This index is a continuous numerical variable that quantifies the severity of noise interference on the data. The dynamic algorithm adaptation module presets a discretization threshold range for the noise confidence index, mapping continuous values ​​to three decision domains: low noise, medium noise, and high noise.

[0068] When the noise confidence index is in the low-noise range, the Isolation Forest algorithm is activated. This algorithm constructs isolation trees based on a random forest architecture and uses the path length principle to identify core anomaly suspects, making it suitable for scenarios with high data quality. During algorithm execution, the default tree depth parameter is maintained, focusing on the identification of essential anomaly patterns.

[0069] When the noise confidence index enters the medium noise range, the weighted distance metric algorithm is invoked. This algorithm introduces a dimensionality weighting mechanism, assigning higher weights to key dimensions through feature importance assessment, thereby enhancing the algorithm's robustness to local disturbances. The distance calculation process employs a variant of Mahalanobis distance, effectively suppressing the impact of medium-level noise on anomaly detection.

[0070] When the noise confidence index reaches the high-noise range, the sparse autoencoder clustering algorithm is triggered. This algorithm reduces the dimensionality of the feature space through encoder compression, eliminates high-dimensional noise interference, and performs density clustering analysis in the hidden layer feature space. The algorithm prioritizes ensuring the robustness of the feature representation before performing outlier cluster detection.

[0071] Each activated optimization algorithm independently processes the input set of fluctuating feature vectors, generating an intermediate set of suspected anomalies. All intermediate results are input into the decision fusion engine, which integrates the output labels of different algorithms using temporal-space alignment techniques. The fused result generates a unified set of suspected anomalies, including floating radius decision boundary markers and anomaly probability scores.

[0072] This invention establishes a mapping logic between noise level and algorithm characteristics: a high-efficiency core anomaly detection algorithm is used in low-noise environments, an anti-interference distance metric algorithm is enabled in medium-noise environments, and feature dimensionality reduction and re-detection are prioritized in high-noise scenarios. The noise confidence index serves as an algorithm selector, enabling dynamic adaptation between processing strategies and data quality. Finally, an integrated set of suspected anomalies is output and transmitted to the behavior chain verification module via a standardized interface.

[0073] Specifically, the power marketing big data-driven abnormal pattern data processing system of the present invention further includes: The behavior chain verification module outputs the verified normal dynamic fluctuation sample to the closed-loop strategy engine module; The closed-loop strategy engine module imports the verified normal dynamic fluctuation samples into the historical feature library and updates the periodic feature extraction parameters of the multimodal feature reconstruction module through an incremental learning mechanism, so that the spectral feature extraction of the multimodal feature reconstruction module adapts to the new sample data, and establishes an operation mode in which the verification results drive the parameter update of the feature reconstruction module.

[0074] After completing multi-dimensional verification, the behavior chain verification module marks the verified normal dynamic fluctuation samples with timestamps, spatial coordinates, and load feature vectors, and transmits them to the closed-loop strategy engine module through a data pipeline. This sample contains reasonable fluctuation data verified by the physical constraint logic chain, serving as effective input for the system's incremental learning.

[0075] After receiving samples, the closed-loop strategy engine module performs a historical feature library update operation, importing new samples into the feature library storage partition using an append-only storage mode. During the import process, an association index is established between the samples and seasonality markers, preserving complete spatiotemporal information. A feature library version management mechanism records data change statuses, providing a data foundation for incremental learning.

[0076] The incremental learning process is triggered by the newly imported sample set, and the periodic feature extraction parameters of the multimodal feature reconstruction module are recalculated. The phase angle parameters of the daily periodic components and weekly / monthly periodic features are updated using spectral recalibration technology, and the parameter transition is smoothed using a sliding window mean method. The changes in feature extraction parameters are recorded using hash digests to generate lightweight parameter update instructions.

[0077] The updated periodic feature extraction parameters are synchronized to the multimodal feature reconstruction module via the configuration interface, adjusting the baseline parameters of its spectrum analysis unit. This adjustment enables the feature reconstruction module to automatically adapt to the new sample data features in subsequent processing, improving the accuracy of daily periodic component identification and the ability to capture weekly and monthly periodic features. The parameter update adopts a hot deployment mechanism to ensure that the system continues to operate without interruption.

[0078] This invention establishes a closed-loop learning mechanism: verification results provide high-quality learning samples, the feature library is incrementally stored to update the data foundation, parameter recalculation optimizes the feature extraction model, and hot deployment enables dynamic upgrades to system capabilities. The verification results and feature extraction capabilities form a positive feedback loop, enabling the system to continuously adapt to the evolution of electricity consumption patterns and improve the environmental adaptability of anomaly detection.

[0079] Specifically, the abnormal pattern data processing system driven by big data in power marketing according to the present invention, wherein adjusting the loss function weight of the adversarial feature decoupling module according to the decision boundary offset includes: Real-time monitoring of the decision boundary offset of the abnormal suspect point set markers generated by the dynamic algorithm adaptation module; When the decision boundary offset falls into the preset high offset range, the weight coefficient of the fluctuation feature reconstruction loss is reduced. When the decision boundary offset falls into the preset low offset range, the target weight of the noise discrimination loss is increased; The dynamically adjusted loss function weights are synchronized to the output layer executor of the adversarial feature decoupling module.

[0080] The closed-loop strategy engine module receives the set of suspected anomaly points output by the dynamic algorithm adaptation module in real time and continuously monitors the offset parameter of the decision boundary marked in the point set. This offset represents the change in distance between the anomaly judgment boundary and the normal feature distribution, and the trend value of the offset is obtained by calculating the average value of the offset through a sliding window. The monitoring process uses time series analysis technology to capture the dynamic evolution characteristics of the offset.

[0081] The system predefines the discretized intervals of the decision boundary offset, setting two key decision domains: a high offset interval and a low offset interval. When the offset trend value is detected to be continuously within the high offset interval, a weight reduction instruction is generated for the fluctuation feature reconstruction loss. The weight reduction operation reduces the coefficient weight of the feature reconstruction term in the loss function, thereby reducing the model's constraint on feature fidelity and enhancing the model's adaptability to distribution shifts.

[0082] When the offset trend value remains in the low offset range, a target weight adjustment instruction is generated for the noise discrimination loss. The weight adjustment operation strengthens the model's ability to identify noisy features and improves the robustness of feature decoupling by increasing the coefficient weight of the noise discrimination term in the loss function. The weight adjustment amount is dynamically calculated based on the magnitude of the offset deviation from the baseline value.

[0083] The dynamically adjusted loss function weights are synchronized to the output layer actuator of the adversarial feature decoupling module via a configuration interface. This actuator injects the new weights into the neural network loss calculation layer, updating the weight ratio of fluctuating feature reconstruction loss and noise discrimination loss in real time. Gradient smoothing techniques are used for parameter updates to avoid training oscillations and maintain model convergence stability.

[0084] This invention constructs a dynamic feedback control mechanism: boundary offset monitoring provides system state awareness, interval division establishes decision-making benchmarks, weight adjustment realizes model behavior control, and parameter synchronization completes closed-loop adjustment. Through adaptive adjustment of the loss function weights, the feature decoupling module continuously optimizes its adaptability to changes in power data distribution, ensuring the continuity of anomaly detection accuracy.

[0085] This invention constructs a spatiotemporally aligned 3D data stream through a distributed collaborative acquisition module, eliminating temporal misalignment noise from multi-source heterogeneous data. A multimodal feature reconstruction module applies Fast Fourier Transform to separate periodic background noise, and environmental response modeling quantifies external interference factors such as temperature. An adversarial feature decoupling module uses orthogonal projection technology to decompose the load feature map into mutually exclusive fluctuation feature subspaces and noise feature subspaces, generating a 64-dimensional clean feature vector set and a noise confidence index. This technical approach achieves physical isolation and quantitative evaluation of noise components, reducing the interference of high-dimensional noise on anomaly detection from the feature representation level.

[0086] The dynamic algorithm adaptation module dynamically schedules and optimizes algorithms based on noise confidence indices: in low-noise environments, it activates the isolated forest algorithm to capture core anomaly patterns; in medium-noise scenarios, it employs a weighted distance metric algorithm to enhance resistance to local interference; and in high-noise conditions, it performs feature dimensionality reduction and clustering to suppress noise impact. The behavior chain verification module establishes joint constraint rules based on environmental temperature thresholds, permissible load deviation ranges, and equipment status codes. When all three simultaneously meet preset physical legitimacy conditions, the anomaly marker is removed. This scheme achieves accurate differentiation between real anomaly events and reasonable environmental responses through dynamic algorithm adaptation and multi-dimensional physical rule verification.

[0087] The closed-loop strategy engine module imports verified normal sample increments into the historical feature library, triggering spectral recalibration to update the periodic feature phase angle parameters. Based on the decision boundary offset output by the dynamic algorithm adaptation module, it adjusts the loss function weight coefficients of the adversarial feature decoupling module: in the high offset interval, it reduces the feature reconstruction loss weight to enhance model adaptability, and in the low offset interval, it increases the noise discrimination loss weight to strengthen feature decoupling capability. The updated periodic feature parameters, hash index, and loss weights are synchronously fed back to the feature reconstruction module and the decoupling module, forming a closed-loop learning system of feature extraction, anomaly detection, result feedback, and parameter optimization, continuously improving the recognition robustness in complex noisy environments.

Claims

1. A data processing system for abnormal patterns driven by big data in electricity marketing, characterized in that: include: The distributed collaborative acquisition module collects power time-series data and environmental parameters from smart meters, and outputs a three-dimensional data stream including time, physical quantities, and spatial dimensions. The multimodal feature reconstruction module receives the three-dimensional data stream, applies time window segmentation to extract spectral features and environmental response features, and generates a load feature map; The adversarial feature decoupling module receives the load feature map, performs orthogonal projection of the feature space to separate the fluctuation feature vector and the noise confidence index, and outputs the purified fluctuation feature vector set. The dynamic algorithm adaptation module receives the set of fluctuation feature vectors and the noise confidence index, and dynamically activates and optimizes the algorithm combination according to the feature sparsity and noise confidence to generate a set of suspected abnormal points. The behavior chain verification module receives the set of suspected abnormal points, performs multi-dimensional verification to distinguish between abnormal events and reasonable fluctuations, and outputs a list of verified abnormal events and samples of normal dynamic fluctuations that have passed verification. The closed-loop strategy engine module receives the list of verified abnormal events and the verified normal dynamic fluctuation samples. It triggers marketing audit instructions for events in the list of verified abnormal events, handles real electricity consumption anomalies, imports the verified normal dynamic fluctuation samples into the historical feature library, adjusts the periodic feature extraction parameters of the multimodal feature reconstruction module through an incremental learning mechanism, and adjusts the loss function weight of the adversarial feature decoupling module according to the decision boundary offset.

2. The abnormal pattern data processing system driven by big data in power marketing as described in claim 1, characterized in that, The distributed collaborative acquisition module is configured as follows: Terminal devices are deployed at power grid nodes to periodically collect instantaneous values ​​of voltage and current from smart meters, and to obtain meteorological service parameters and the geographical coordinates of the power grid. The three-dimensional data stream includes timestamps, meter identifiers, power values, temperature values, humidity values, longitude values, and latitude values; The periodically collected instantaneous values ​​of voltage and current from the smart meter, along with the acquired meteorological service parameters and power grid geographic coordinates, are input into the multimodal feature reconstruction module for segmentation.

3. The abnormal pattern data processing system driven by big data in power marketing as described in claim 2, characterized in that, The multimodal feature reconstruction module is configured as follows: Spectral feature extraction is performed on the input 3D data stream: Fast Fourier Transform is applied to analyze the daily periodic components and weekly / monthly periodic features to generate a standardized periodic feature spectrum; Environmental response feature extraction: Construct a nonlinear correlation model between load value and temperature value, and output a set of temperature sensitivity coefficients; By fusing the standardized periodic characteristic spectrum with the temperature sensitivity coefficient set, a basic load characteristic map is generated; Identify seasonal dependence patterns in the basic load feature map and add seasonality identification factors to the map; The load feature map with seasonality identifier is input into the adversarial feature decoupling module to perform orthogonal projection processing.

4. The abnormal pattern data processing system driven by big data in electricity marketing as described in claim 3, characterized in that, The adversarial feature decoupling module is configured as follows: The encoder decomposes the input load feature map into a fluctuation feature subspace and a noise feature subspace. The generator reconstructs the feature samples, and the discriminator compares the reconstructed samples with the original samples to calculate the error magnitude. Based on the error amplitude, a set of fluctuation feature vectors and a noise confidence index are generated, wherein the set of fluctuation feature vectors contains 64-dimensional purification features and the noise confidence index is a continuous value. The set of fluctuation feature vectors and the noise confidence index are transmitted as output to the dynamic algorithm adaptation module.

5. The abnormal pattern data processing system driven by big data in electricity marketing as described in claim 4, characterized in that, The dynamic algorithm adaptation module is configured as follows: Real-time monitoring of input feature sparsity, kurtosis, and noise confidence index; If the feature sparsity exceeds the set threshold, the sparse autoencoder clustering algorithm is called to process the features. If the kurtosis of the feature distribution exceeds a set threshold, the weighted distance metric algorithm is invoked to process the feature. If the noise confidence index reaches the preset threshold, the detection is performed after adjusting the depth parameters of the isolated forest algorithm. Integrate all the suspicious points output by the algorithms to generate a set of anomalous suspicious points that mark the decision boundary of the floating radius. Output the set of anomalous suspicious points marked with the decision boundary to the behavior chain verification module to perform multi-dimensional verification.

6. The abnormal pattern data processing system driven by big data in power marketing as described in claim 5, characterized in that, The behavior chain verification module is configured as follows: The system obtains the current ambient temperature value, the instantaneous load value of the power grid, and the operating status code of the power distribution equipment from the real-time monitoring source. It retrieves the historical load baseline data of the target user, analyzes the operating status code of the power distribution equipment together with the historical load baseline data, generates a dynamic constraint rule group, and detects the deviation of the ambient temperature value, the instantaneous load value and the historical load baseline data, as well as whether the operating status code of the power distribution equipment meets the preset legality conditions based on the dynamic constraint rule group. If all conditions are met, the corresponding suspicious point will be removed from the list. The system compiles a list of suspected anomalies that have not been de-marked, generates a list of verification anomalies, and outputs the list of verification anomalies to the closed-loop strategy engine module to drive the update of the marketing audit strategy library.

7. The abnormal pattern data processing system driven by big data in electricity marketing as described in claim 6, characterized in that, The closed-loop strategy engine module is configured as follows: Based on the newly imported historical feature library, the verification samples are used to recalculate the phase angle of the periodic feature components and update the corresponding hash index structure. Based on the recalculated phase angle change amplitude of the periodic component, the incremental learning mechanism is triggered to generate a loss function weight adjustment instruction for the adversarial feature decoupling module; The adjustment instruction is executed to update the weight parameters of the loss function of the adversarial feature decoupling module; The updated periodic component phase angle, hash index structure, and loss function weight parameters are synchronized to the multimodal feature reconstruction module and the adversarial feature decoupling module to complete cross-module parameter collaborative refresh.

8. The abnormal pattern data processing system driven by big data in electricity marketing as described in claim 7, characterized in that, Also includes: The noise confidence index is generated by the adversarial feature decoupling module and input into the dynamic algorithm adaptation module; The dynamic algorithm adaptation module performs dynamic switching of optimization algorithms based on noise confidence, including: preset noise confidence index threshold range; When the noise confidence index falls into the low noise range, the isolated forest algorithm is activated to perform core suspect identification. When the noise confidence index falls into the medium noise range, the weighted distance metric algorithm is activated to perform anti-interference anomaly detection. When the noise confidence index falls into the high noise range, the sparse autoencoder clustering algorithm is activated to perform feature dimensionality reduction and density clustering. The selected optimization algorithm processes the set of fluctuating feature vectors based on the current input features to generate a set of suspected anomalies.

9. The abnormal pattern data processing system driven by big data in electricity marketing as described in claim 8, characterized in that, Also includes: The behavior chain verification module outputs the verified normal dynamic fluctuation sample to the closed-loop strategy engine module; The closed-loop strategy engine module imports the verified normal dynamic fluctuation samples into the historical feature library and updates the periodic feature extraction parameters of the multimodal feature reconstruction module through an incremental learning mechanism, so that the spectral feature extraction of the multimodal feature reconstruction module adapts to the new sample data, and establishes an operation mode in which the verification results drive the parameter update of the feature reconstruction module.

10. The abnormal pattern data processing system driven by big data in power marketing as described in claim 9, wherein adjusting the loss function weights of the adversarial feature decoupling module according to the decision boundary offset further includes: Real-time monitoring of the decision boundary offset of the abnormal suspect point set markers generated by the dynamic algorithm adaptation module; When the decision boundary offset falls into the preset high offset range, the weight coefficient of the fluctuation feature reconstruction loss is reduced. When the decision boundary offset falls into the preset low offset range, the target weight of the noise discrimination loss is increased; The dynamically adjusted loss function weights are synchronized to the output layer executor of the adversarial feature decoupling module.

Images