Intelligent analysis system for electric energy quality and read data

Abstract

The invention relates to the technical field of data processing, in particular to an intelligent analysis system for electric energy quality and read data, which comprises a data fusion module, a modeling module, an analysis module, a detection module, an optimization module and a control module. The data fusion module aligns data of an electric meter terminal, power grid monitoring equipment and an environment sensor through a sliding time window, and the analysis module calculates a power grid node loss transfer coefficient based on a graph neural network and fuses transformer no-load loss and line contact resistance parameters to generate a dynamic line loss evaluation matrix. And the detection module identifies the power consumption characteristic deviation degree through a random forest classifier, and generates a priority management strategy in combination with a multi-dimensional abnormal scoring model. The control module adaptively selects a communication protocol to execute a regulation and control instruction according to a network state, a heartbeat detection mechanism feeds back operation data of a governance device in real time, model parameters are driven to be iteratively updated, and dynamic cooperation of power supply quality optimization and line loss governance is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to an intelligent analysis system for power quality and meter reading data. Background Art

[0002] The intelligent analysis of power quality and meter reading data is a digital management method for dynamically monitoring and predicting power grid power quality problems by integrating power system operation data and user electricity consumption behavior information, and combining data mining and machine learning technologies. The existing power quality governance and electricity consumption data management mainly rely on manual meter reading combined with a basic automation system. The traditional data collection method has data errors caused by human interference, which affects the accuracy of electricity bill calculation and line loss analysis; the existing line loss management system lacks effective multi-dimensional correlation analysis capabilities for abnormal losses in complex power grid environments and is difficult to accurately identify the key nodes of line losses; at the same time, the existing technology is limited to the basic statistical level in the processing of massive electricity consumption data and fails to construct a dynamic correlation model between data features and power quality indicators, resulting in a lag in early warning of abnormal electricity consumption behavior and optimization of power supply quality. Summary of the Invention

[0003] Aiming at the deficiencies of the prior art, the present invention provides an intelligent analysis system for power quality and meter reading data, which is used to solve the problems of abnormal warning lag and low power supply optimization efficiency caused by errors in power data collection due to manual intervention, inaccurate identification of key line loss nodes, and lack of dynamic correlation between electricity consumption data features and power quality indicators in the prior art.

[0004] To solve the above technical problems, the specific technical solutions of the present invention are as follows: The intelligent analysis system for power quality and meter reading data provided by the present invention includes: A data fusion module configured to collect raw data through meter terminals, power grid monitoring devices, and environmental sensors, transmit the raw data to a central processor through an anti-interference communication protocol, perform timestamp alignment and data cleaning on the raw data, and generate a standardized data set; A modeling module configured to receive the standardized data set, extract the temporal features of user electricity consumption behavior, power grid harmonics, and voltage sags in the standardized data set through a convolutional neural network, generate a feature vector matrix, and construct a dynamic correlation model based on historical power grid operation data, and output the mapping relationship between power quality indicators and electricity consumption patterns in the feature vector matrix; An analysis module, configured to calculate the loss weights between line nodes through a graph neural network according to the mapping relationship output by the dynamic association model, combine the power grid topology structure data and the real-time load distribution, generate a line loss contribution degree evaluation matrix including line impedance and load rate parameters, identify the high-loss area topology paths and associated devices with contribution degrees exceeding the threshold in the line loss contribution degree evaluation matrix, and output the coordinates of the optimization target area and the parameter adjustment range; A detection module, configured to compare the current power consumption characteristics in the eigenvector matrix with the historical normal mode in real time based on the mapping relationship output by the dynamic association model, detect harmonic distortion rate and load mutation deviation data through an ensemble learning model, and generate a set of governance strategies including priority sorting in combination with the coordinates of the optimization target area output by the analysis module, and trigger warning instructions and regulation suggestions; An optimization module, configured to call a long short-term memory network to predict the future load curve according to the deviation data output by the detection module, generate an optimized power consumption time period plan by integrating the time-of-use electricity price policy, and feedback the optimized load distribution data to the analysis module for updating the loss model parameters in the power grid topology structure data; A control module, configured to receive the regulation suggestions of the detection module and the optimization plan of the optimization module, match the current network environment through an adaptive communication module, send device parameter adjustment instructions to the power quality governance device, synchronously collect the operation status data of the power quality governance device and send it back to the data fusion module to form a closed-loop feedback link from data collection to policy execution; Among them, the standardized data set generated by the data fusion module is input into the modeling module, the mapping relationship output by the modeling module drives the analysis module to generate the coordinates of the optimization target area, the detection module triggers a warning instruction according to the coordinates of the optimization target area and the deviation data detected in real time, the optimization plan generated by the optimization module is regulated and controlled through the control module, and the operation status data of the power quality governance device is fed back to the data fusion module to complete the closed-loop data flow.

[0005] Further, for the intelligent analysis system of power quality and meter reading data of the present invention, the data fusion module is further configured to: Perform data segmentation alignment on the electricity consumption data collected by the meter terminal and the voltage and current waveform data collected by the power grid monitoring device using a sliding time window to generate a time-aligned electricity consumption and waveform data set; perform time series fusion on the temperature and humidity data collected by the environmental sensor and the time-aligned electricity consumption and waveform data set through interpolation to generate a spatio-temporal association data set; remove the noise data in the spatio-temporal association data set through a data cleaning algorithm to generate a standardized data set, and output the standardized data set to the modeling module.

[0006] Furthermore, for the intelligent analysis system of power quality and meter reading data according to the present invention, the modeling module is further configured to: Input the standardized data set output by the data fusion module into a multi-layer convolutional neural network to extract the load fluctuation period characteristics and power grid harmonic spectrum characteristics in the user's electricity consumption behavior; Capture the temporal dependence relationship of voltage sag events through a bidirectional long short-term memory network to generate a feature vector matrix including load fluctuations and harmonic characteristics; Perform correlation analysis on the feature vector matrix and the fault events in the historical power grid fault record database to establish a non-linear mapping relationship between the power quality index and the user's electricity consumption pattern.

[0007] Furthermore, for the intelligent analysis system of power quality and meter reading data according to the present invention, the analysis module is further configured to: Based on the non-linear mapping relationship established by the modeling module, analyze the electrical parameters and real-time load distribution of each node in the power grid topology structure data; Calculate the power loss transfer coefficient between adjacent nodes through the adjacency matrix propagation algorithm of the graph neural network to generate an initial loss weight matrix; Generate a dynamic loss weight matrix including line impedance, load rate, and environmental factors according to the initial loss weight matrix and the load rate parameter in the real-time load data, and input the dynamic loss weight matrix into the optimization module.

[0008] Furthermore, for the intelligent analysis system of power quality and meter reading data according to the present invention, the analysis module is further configured to: Superimpose the no-load loss parameter of the transformer and the contact resistance loss parameter of the line in the dynamic loss weight matrix to generate a weight matrix corrected by the correlation factor; Iteratively optimize the weight matrix corrected by the correlation factor through the gradient backpropagation algorithm to generate a dynamically updated line loss contribution degree evaluation matrix; According to the contribution degree threshold of the nodes in the line loss contribution degree evaluation matrix, screen out the device identifiers in the high-loss area with a contribution degree exceeding the preset threshold, and output the device identifiers in the high-loss area to the detection module.

[0009] Furthermore, for the intelligent analysis system of power quality and meter reading data according to the present invention, the detection module is further configured to: Input the current electricity consumption feature vector output by the modeling module into a random forest classifier to calculate the deviation degree between the current electricity consumption feature vector and the historical normal mode; Based on the deviation degree and the line loss contribution degree evaluation matrix generated by the analysis module, construct a multi-dimensional anomaly scoring model including harmonic distortion rate and load mutation weight; Generate a set of priority strategies including governance device identifiers, adjustment parameter ranges, and execution orders based on the output results of the multi-dimensional anomaly scoring model, and output the set of priority strategies to the optimization module.

[0010] Furthermore, for the intelligent analysis system of power quality and collection data according to the present invention, the optimization module is further configured to: Extract the user power consumption pattern deviation characteristics in the set of priority strategies output by the detection module; Input the deviation characteristics into a long short-term memory network to predict the peak-valley change trend of the load curve in the next 24 hours, and generate an initial load prediction result; Combine the time period division rules in the external time-of-use electricity price policy database to optimize the time period of the initial load prediction result, and generate electricity consumption time period optimization suggestions and load transfer plans.

[0011] Furthermore, for the intelligent analysis system of power quality and collection data according to the present invention, the optimization module is further configured to: Perform a joint simulation on the electricity consumption time period optimization suggestions and the updated power grid topology loss model parameters of the analysis module; Perform multi-objective optimization on the load transfer plan through a genetic algorithm to generate a Pareto optimal solution set that meets the requirements of reducing line loss and improving power supply quality; Feed back the Pareto optimal solution set to the analysis module, trigger the real-time update of the line loss contribution degree evaluation matrix, and input the updated evaluation matrix into the detection module.

[0012] Furthermore, for the intelligent analysis system of power quality and collection data according to the present invention, the control module is further configured to: Dynamically select an instruction transmission protocol according to the network delay and bandwidth data detected by the adaptive communication module; Package the regulation suggestions of the detection module and the optimization plan of the optimization module into a data packet that conforms to the communication interface of the target power quality governance device; Real-time monitor the operating status of the power quality governance device through a heartbeat detection mechanism, and transmit the status data including the voltage regulation response time and harmonic suppression efficiency back to the data fusion module.

[0013] Furthermore, for the intelligent analysis system of power quality and collection data according to the present invention, the control module is further configured to: Extract the voltage regulation response time and harmonic suppression efficiency parameters from the operating status data of the power quality governance device; Compare and analyze the harmonic suppression efficiency parameter with the expected targets in the priority policy set to generate a regulation effect evaluation report; Adjust the weight parameters of the dynamic association model in the modeling module according to the evaluation report, and feedback the adjusted model parameters to the analysis module to complete the model self-optimization of the closed-loop feedback link.

[0014] Advantages of the present invention; The present invention achieves significant technical effects through a multi-module collaborative data processing architecture and a closed-loop feedback mechanism. The data fusion module uses a sliding time window alignment and interpolation method fusion technology to eliminate data timing misalignment and noise interference caused by manual intervention, and generates a highly complete standardized data set to provide a reliable input for subsequent analysis; the modeling module extracts the timing characteristics of load fluctuations, harmonic spectra, and voltage sag events through a convolutional neural network and a bidirectional long short-term memory network, and constructs a dynamic association model between power quality indicators and power consumption patterns to solve the problem of the lack of association between data features and quality indicators in traditional technologies; the analysis module calculates the grid node loss transfer coefficient based on a graph neural network, superimposes the no-load loss of the transformer and the line contact resistance parameters, and generates a dynamically updated line loss contribution degree evaluation matrix to accurately locate the key equipment nodes in the high-loss area, with a significant improvement in line loss identification accuracy; the detection module uses a random forest classifier and a multi-dimensional anomaly scoring model to identify harmonic distortion and load mutation anomalies in real time, and generates a set of priority governance strategies to shorten the abnormal warning response time; the optimization module combines a long short-term memory network to predict the load curve and a genetic algorithm for multi-objective optimization to generate a Pareto optimal solution set that takes into account both line loss reduction and power supply quality improvement; the control module executes the regulation instructions through an adaptive communication protocol and a heartbeat detection mechanism and real-time feedbacks the device status data, driving the iterative update of the weight parameters of the dynamic association model, forming a full-link closed-loop feedback from data acquisition, feature modeling, anomaly detection to strategy execution, and finally achieving the synergistic effect of improving the power supply quality optimization efficiency and reducing the line loss governance cost. Brief Description of the Drawings

[0015] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for the embodiments will be briefly introduced below. Obviously, for those of ordinary skill in the art, other drawings can be obtained according to the drawings without creative labor.

[0016] Figure 1 It is a system architecture diagram of the intelligent analysis system for power quality and meter reading data provided by the embodiments of the present invention. Detailed Embodiments

[0017] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts belong to the scope of protection of the present invention. The following describes in detail the technical solutions provided by each embodiment of the present invention with reference to the drawings. To better understand the objectives of the present invention, the present invention is further described in detail below.

[0018] Please refer to Figure 1 , the intelligent analysis system for power quality and meter reading data provided by the present invention includes: A data fusion module, configured to collect raw data through meter terminals, power grid monitoring devices, and environmental sensors, transmit the raw data to a central processor through an anti-interference communication protocol, perform timestamp alignment and data cleaning on the raw data, and generate a standardized data set; The data fusion module collects user power consumption data through meter terminals, records the instantaneous values of current and voltage and power parameters, the power grid monitoring device synchronously obtains the voltage waveform, current harmonic spectrum, and frequency fluctuation data of the power grid node, and the environmental sensor monitors the temperature and humidity parameters in real time to quantify the impact of environmental factors on line resistance and transformer losses. The data collected by different devices is transmitted to the central processor through an anti-interference communication protocol, and the communication protocol dynamically adjusts the coding method according to the data transmission distance and electromagnetic environment to reduce external interference during signal transmission.

[0019] There are sampling frequency differences in the asynchronous data collected by the meter terminal and the power grid monitoring device. The sliding time window is used to segment the raw data, and the power consumption, voltage, and current waveform data collected by different devices are divided into data segments with a unified time reference at fixed time intervals to eliminate the timing misalignment problem caused by device clock deviation or communication delay. The division of the time window is dynamically adjusted according to the highest sampling frequency of the device. For example, the high-frequency waveform data and low-frequency power consumption data are aligned with the second-level timestamp to generate a power consumption and waveform data set with consistent time dimensions.

[0020] The temperature and humidity data collected by the environmental sensor has a lower sampling frequency than the power data. The interpolation method is used to complement the discrete environmental parameters to the same time resolution as the power data. The linear interpolation algorithm generates a continuous time series based on the environmental parameter values at adjacent time points and fuses it with the power consumption and waveform data sets aligned with time to form a spatio-temporal correlation data set including power consumption behavior, power quality, and environmental parameters.

[0021] The spatio-temporal correlated dataset may contain outliers caused by sensor signal interference or communication packet loss. The noise data deviating from the normal fluctuation range is identified and removed through data cleaning algorithms. Wavelet transform performs multi-scale decomposition on high-frequency noise components to separate the effective signal from electromagnetic interference pulses; the median filtering algorithm eliminates instantaneous outliers in the data, smooths the data fluctuations, and retains the effective features related to the grid operation state. The cleaned dataset removes redundant noise and generates a standardized dataset, which includes time-synchronized power consumption, waveforms, and environmental parameters, providing highly consistent input data for the subsequent modeling and analysis modules.

[0022] After the standardized dataset is output to the modeling module, it triggers the subsequent feature extraction and dynamic correlation model construction processes, forming a complete preprocessing link from data acquisition, alignment, cleaning to standardization. The timestamp alignment and interpolation fusion technology improves the consistency of data in the time and space dimensions, and the data cleaning algorithm enhances the effectiveness and integrity of the dataset, providing a reliable basis for the modeling and analysis of subsequent modules. Each step forms a progressive data integration logic through sequential processing and spatial parameter fusion, supporting the system's collaborative processing ability for multi-source heterogeneous data.

[0023] The modeling module is configured to receive the standardized dataset, extract the temporal features of the user's electricity consumption behavior, grid harmonics, and voltage sags in the standardized dataset through a convolutional neural network, generate a feature vector matrix, and construct a dynamic correlation model based on historical grid operation data, and output the mapping relationship between the power quality indicators and the electricity consumption pattern in the feature vector matrix; After receiving the standardized dataset, the modeling module inputs it into a multi-layer convolutional neural network for feature extraction. The first-layer convolutional kernel captures local features for the load fluctuation cycle characteristics of the user's electricity consumption behavior, identifies the peak-valley change pattern of the daily load curve through convolutional operations, and extracts the periodic law in the low-frequency load fluctuations. The second-layer convolutional kernel focuses on the grid harmonic spectrum features, uses convolutional kernels of different scales to separate high-frequency harmonic components, and identifies harmonic distortion signals in specific frequency bands, such as abnormal fluctuations in the amplitudes of the 3rd and 5th harmonics. The feature codes output by the convolutional layer are processed by the pooling layer for dimensionality reduction, retaining the key frequency band information, and forming a preliminary feature set of load fluctuations and harmonic spectra.

[0024] The bidirectional long short-term memory network receives the feature codes output by the convolutional neural network. The forward layer processes the gradual change trend of the current before the voltage sag event in chronological order, capturing the slow current drop feature at the start stage of the voltage dip; the backward layer processes the residual harmonic fluctuations after the voltage recovery in reverse order, identifying the recovery delay feature of the harmonic amplitude after the sag ends. The hidden states of the two-way time steps are fused through a concatenation operation to generate a temporal feature vector containing the complete temporal dependence relationship of the voltage sag event, covering the whole process dynamic features before, during, and after the event.

[0025] The eigenvector matrix is dynamically time-warped and matched with the fault events in the historical power grid fault record database. The dynamic time warping algorithm is used to align the time series patterns of the real-time eigenvectors and historical fault events, eliminating the problem of sequence length mismatch caused by differences in event durations. Through kernel function mapping, high-dimensional features such as the harmonic distortion rate and voltage sag duration in the eigenvector matrix are nonlinearly correlated with the load mutation amplitude and peak-valley difference parameters in the user's power consumption pattern, establishing a dynamic mapping relationship between power quality indicators and power consumption patterns. The set of mapping relationship parameters includes the correlation weight between the harmonic distortion threshold and the load mutation amplitude, and the quantization coefficient of the impact of voltage sag duration on power supply quality.

[0026] The dynamic association model updates the mapping relationship parameters through iterative training. The fault event labels in the historical power grid operation data are used as supervision signals, and the backpropagation algorithm adjusts the weight parameters of the convolutional neural network and the bidirectional long short-term memory network to optimize the sensitivity of the feature extraction process to power quality abnormal events. The updated model parameters drive the adjustment of the fusion weights of the load fluctuation, harmonic spectrum, and voltage sag features in the eigenvector matrix, forming dynamic association rules adapted to changes in the power grid operation state. After the mapping relationship is output to the analysis module, it provides a quantitative association basis for the feature and quality indicators for the line loss contribution assessment, supporting subsequent anomaly detection and strategy optimization processes.

[0027] The analysis module is configured to, according to the mapping relationship output by the dynamic association model, combine the power grid topology structure data and the real-time load distribution, calculate the loss weights between line nodes through a graph neural network, generate a line loss contribution assessment matrix including line impedance and load rate parameters, identify the high-loss area topological paths and associated devices with contribution degrees exceeding the threshold in the line loss contribution assessment matrix, and output the optimized target area coordinates and parameter adjustment ranges; After receiving the mapping relationship output by the dynamic association model, the analysis module analyzes the electrical connection relationships and electrical parameters of each node in the power grid topology structure data. Based on the adjacency matrix of the power grid topology, the connection weights between nodes are constructed, and the current transmission paths and impedance matching parameters of adjacent nodes are aggregated through a message passing mechanism to calculate the power loss transfer coefficient between lines. The adjacency matrix propagation algorithm weights and fuses the voltage, current parameters of nodes and line impedance to quantify the energy loss distribution ratio between adjacent nodes, generating an initial loss weight matrix reflecting the basic loss distribution.

[0028] The load rate parameter in the real-time load distribution data is dynamically integrated with the initial loss weight matrix. The load rate parameter characterizes the actual power transmission demand of the node. Combining the line impedance parameter and the temperature data collected by the environmental sensor, the line resistance is dynamically corrected to generate a multi-dimensional dynamic loss weight matrix including the line operation state, load rate, and environmental factors. Each element in the matrix corresponds to the loss weight value of the power grid node, reflecting the contribution degree of each node to the overall line loss under the current operating conditions.

[0029] The dynamic loss weight matrix is superimposed with the no-load loss parameter of the transformer and the contact resistance loss parameter of the line to correct the additional loss caused by equipment aging or loose connection. The no-load loss parameter of the transformer is calculated based on the core material characteristics and the operating duration, and the contact resistance loss parameter of the line is dynamically adjusted according to the oxidation degree and fastening state of the connection point. The weight matrix corrected by the correlation factor is iteratively optimized through the gradient backpropagation algorithm. Taking the actual line loss measurement value as the objective function, the weight distribution ratio of each node in the matrix is adjusted to reduce the interference of environmental temperature fluctuations on the resistance correction model, and a dynamically updated line loss contribution degree evaluation matrix is generated.

[0030] The contribution degree threshold of the node in the line loss contribution degree evaluation matrix is set based on historical line loss data and power grid operation standards. The nodes with contribution degrees exceeding the preset value are identified through the threshold screening algorithm, and the electrical connection path and associated equipment in the high-loss area are determined by combining the power grid topology path analysis. The topology path analysis traces the upstream power supply path and downstream load distribution of the loss node. The associated equipment includes the transformer number, the position of the line sectional switch, and the coordinates of the nodes with abnormal load rates, and outputs the equipment identifier and topology connection relationship in the high-loss area.

[0031] The optimized target area coordinates and parameter adjustment ranges are generated according to the screening results. The coordinates locate the physical positions of the high-loss nodes, and the parameter adjustment ranges include the transformer tap adjustment gear, the line impedance compensation amount, and the load rate control threshold. The output data provides targeted optimization basis for the subsequent module, such as adjusting the transformer voltage regulation strategy or switching the reactive power compensation device, forming a closed-loop logic link from loss assessment to governance strategy. Each step realizes the accurate positioning of the key line loss nodes through topology analysis, dynamic parameter fusion, iterative optimization, and path tracing, supporting the system's adaptive governance ability for complex power grid environments.

[0032] The detection module is configured to, based on the mapping relationship output by the dynamic association model, compare the current power consumption characteristics in the eigenvector matrix with the historical normal mode in real time, detect the harmonic distortion rate and load mutation deviation data through the integrated learning model, and generate a set of governance strategies including priority ranking in combination with the optimized target area coordinates output by the analysis module, triggering a warning instruction and a regulation suggestion; After the detection module receives the mapping relationship output by the dynamic association model, it extracts the current power consumption features in the feature vector matrix, including the load fluctuation period, harmonic spectrum distribution, and voltage sag time series features. The historical normal mode library stores the power consumption feature vectors under typical working conditions. By using a sliding time window to intercept the real-time data segment and dynamically compare it with the historical mode, the dynamic time warping algorithm is used to eliminate the difference in the length of the time series, and the deviation degree of the current feature vector from the historical benchmark in dimensions such as harmonic distortion rate and load mutation rate is calculated.

[0033] The integrated learning model constructs multiple decision trees based on the random forest classifier and processes each dimension data of the current power consumption features in parallel. Each decision tree generates classification rules according to the training data of the historical normal mode, and the voting mechanism synthesizes the judgment results of each tree to output the deviation score of the current power consumption features from the normal mode. The scoring result is marked as a continuous value within the range of 0 to 1, representing the probability intensity of the occurrence of an anomaly. The higher the scoring value, the greater the degree of deviation from the historical normal mode.

[0034] The multi-dimensional anomaly scoring model integrates the deviation score and the node weight parameters of the line loss contribution degree evaluation matrix. The harmonic distortion rate weight coefficient reflects the influence degree of harmonics on the power grid quality, and the load mutation weight coefficient represents the abnormal level of the load change rate. The two are weighted and fused with the line loss contribution degree parameter to generate a comprehensive anomaly score. According to the power grid topology path and equipment association relationship, the scoring model maps the anomaly signal to the optimized target coordinates in the high-loss area, and locates the identifiers of the transformer, line section switch, or capacitor bank corresponding to the anomaly event.

[0035] The priority governance strategy set is generated by sorting the comprehensive anomaly scores in descending order. Anomaly events with scores exceeding the preset threshold trigger warning instructions, which include anomaly type codes, positioning coordinates, and associated device identifiers. The regulation suggestions are matched with the preset governance rule library based on the anomaly type. For example, capacitor switching instructions correspond to harmonic distortion events, and the transformer tap adjustment range is generated for load mutation events. After the strategy set is output to the optimization module, it drives the load prediction and multi-objective optimization process, forming a closed-loop linkage of anomaly detection and strategy execution.

[0036] The warning instructions and regulation suggestions are distributed to the target devices through the adaptive communication protocol. The warning instructions contain fields such as timestamp, anomaly level, and processing time limit. The regulation suggestions are encapsulated into instruction data packets that can be parsed by the device, such as the register address and write value in the Modbus protocol format. The strategy execution status data is transmitted back through the heartbeat detection mechanism, triggering the iterative update of the weight parameters of the dynamic association model, improving the accuracy and response timeliness of subsequent anomaly detection. Each step realizes the accurate identification and hierarchical governance of anomaly events through feature comparison, integrated learning judgment, multi-dimensional scoring fusion, and strategy matching, supporting the system's dynamic response ability to power quality problems.

[0037] An optimization module, configured to call a long short-term memory network to predict a future load curve according to the deviation data output by the detection module, generate an optimized power consumption time period plan by integrating the time-of-use electricity price policy, and feed back the optimized load distribution data to the analysis module for updating the loss model parameters in the grid topology structure data; After receiving the deviation data output by the detection module, the optimization module extracts feature parameters such as the load mutation amplitude, harmonic distortion duration, and voltage sag occurrence frequency that deviate from the historical normal fluctuation range in the user's power consumption pattern. After the deviation data is normalized, it is input into the long short-term memory network. The network captures the temporal correlation of the historical load curve through the time step recurrent unit, learns the periodic law of the daily load peak and valley changes, predicts the load demand distribution trend at each time point within the next 24 hours, and generates an initial load prediction curve.

[0038] The initial load prediction curve is dynamically matched with the time period division rules in the external time-of-use electricity price policy database. The database stores the peak-valley electricity price time period division standard and the time-of-use electricity price gradient parameters. Through the time period alignment algorithm, the peak time period in the predicted load curve is correlated with the electricity price peak time period for analysis, and the high-cost time periods with concentrated load distribution are identified. The dynamic programming algorithm calculates the feasible paths for load transfer and generates optimized suggestions for electricity consumption time periods, including the specific target values and adjustment amplitudes for transferring some peak loads to the low electricity price time periods, such as adjusting the production plan of industrial users or enabling energy storage devices to suppress load fluctuations.

[0039] The load transfer plan is jointly simulated and verified with the grid topology loss model parameters. The simulation engine calls the real-time loss weight matrix and topology structure data provided by the analysis module to simulate the line loss change trend and voltage stability index under different load distribution scenarios, and evaluates the impact of the optimization plan on the grid operation efficiency. The genetic algorithm takes minimizing line loss and suppressing voltage fluctuation as the multi-objective optimization direction, generates multiple groups of load distribution candidate plans through population initialization, and iteratively screens out the solution set that meets the Pareto optimal conditions through crossover and mutation operations to balance the dual requirements of economy and power supply quality.

[0040] The optimized load distribution data is fed back to the analysis module after being converted into a standardized format. The data packet includes the adjusted load time period distribution, target node coordinates, and parameter adjustment range, triggering the real-time update of the line loss contribution evaluation matrix. The updated matrix integrates the latest load distribution and line impedance correction parameters, recalculates the loss transfer weight between nodes, and provides dynamic input for subsequent anomaly detection and strategy optimization. The feedback data synchronously drives the adjustment of the weight parameters of the dynamic association model in the modeling module, such as enhancing the association weight of the load mutation feature to the harmonic distortion rate, and improving the sensitivity of the model to abnormal events.

[0041] The optimization plan is sent to the power quality governance device for execution through the control module. The load transfer target value is encapsulated into a control instruction recognizable by the device, such as adjusting the tap position of the transformer or switching the reactive power compensation capacitor. The instruction data packet is transmitted to the target node through the adaptive communication protocol. The execution result data is transmitted back to the data fusion module through the heartbeat detection mechanism, forming a closed-loop control link from load prediction, strategy optimization to execution feedback, supporting the continuous optimization and adaptive adjustment of the system to the grid operation state. Each step realizes the intelligent scheduling of load distribution through time series prediction, multi-objective optimization and dynamic feedback, improving the stability of power supply quality while reducing line losses.

[0042] The control module is configured to receive the regulation suggestions of the detection module and the optimization plan of the optimization module, match the current network environment through the adaptive communication module, send the device parameter adjustment instruction to the power quality governance device, synchronously collect the operation status data of the power quality governance device and transmit it back to the data fusion module, forming a closed-loop feedback link from data collection to strategy execution; After the control module receives the regulation suggestions of the detection module and the optimization plan of the optimization module, the adaptive communication module real-time detects the delay fluctuation and bandwidth occupancy rate of the current network environment. Dynamically select the instruction transmission protocol according to the network state parameters, for example, switch to the transmission control protocol with a retransmission mechanism in case of network congestion, and adopt the user datagram protocol to improve the transmission efficiency when the network is stable, balancing the real-time and reliability of instruction transmission.

[0043] The regulation suggestions and optimization plan are converted into a data format recognizable by the target power quality governance device through the protocol adaptation module. The regulation instructions include the transformer tap adjustment position, the capacitor switching time sequence and the load transfer target value, encapsulate the data packet according to the Modbus or IEC 61850 communication standard, add the device address identifier and the cyclic redundancy check code field, and generate a control instruction stream that conforms to the interface specification of the target device. The data packet is transmitted to the target node through the selected protocol, adapting to the communication protocol differences of devices from different manufacturers and improving the instruction compatibility.

[0044] After receiving the instruction, the governance device performs the parameter adjustment operation. The control module polls the device operation status at a fixed time interval through the heartbeat detection mechanism. The status query instruction collects the actual response time of voltage regulation, the harmonic suppression efficiency and the load rate change parameters, and generates a real-time operation status data set including the time stamp. The status data is transmitted back to the data fusion module after encryption processing, triggering the dynamic update of the original data set, and synchronously correcting the feature extraction weight of the modeling module and the line loss evaluation parameter of the analysis module.

[0045] The operating status data is compared and analyzed with the expected regulation targets to generate an evaluation report on the regulation effect. The report includes the analysis of the reasons for voltage regulation delay, the rating of harmonic suppression effect, and the equipment health status indicators, driving the iterative adjustment of the feature weights in the dynamic association model. For example, when the voltage regulation response lags, the association weight of the load mutation feature is enhanced; when the harmonic suppression is insufficient, the detection sensitivity of the harmonic spectrum feature is improved, forming a closed-loop feedback link from command execution to model optimization.

[0046] The data flow in the feedback link realizes the dynamic coordination of the whole system parameters. The optimized load distribution data updates the power grid topology loss model, the adjusted feature weights enhance the anomaly detection accuracy, and the iterated communication protocol parameters improve the subsequent command transmission efficiency. Each step supports the adaptive optimization of the power quality governance strategy of the system through the progressive logic of protocol adaptation, status monitoring, effect evaluation, and parameter iteration, maintaining the dynamic balance of the power grid operation efficiency and power supply quality.

[0047] Among them, the standardized data set generated by the data fusion module is input into the modeling module, the mapping relationship output by the modeling module drives the analysis module to generate the coordinates of the optimized target area, the detection module triggers an early warning command according to the deviation data between the coordinates of the optimized target area and the real-time detection, and the optimization scheme generated by the optimization module is used for regulation through the control module, and the operating status data of the power quality governance device is fed back to the data fusion module to complete the closed-loop data flow.

[0048] The intelligent analysis system for power quality and meter reading data provided by the present invention realizes the full-link closed-loop of data acquisition, processing, analysis, and control through the cooperation of multiple modules. The data fusion module collects the original data through the meter terminals, power grid monitoring devices, and environmental sensors, and transmits it to the central processor using an anti-interference communication protocol. The original data generates a standardized data set after timestamp alignment and data cleaning, eliminating the timing misalignment problem caused by the difference in the sensor sampling frequency, and fusing environmental parameters such as temperature and humidity through the interpolation method to form a spatio-temporal correlation data set. The data cleaning algorithm further removes the noise data to ensure the high integrity and consistency of the data input to the subsequent modules.

[0049] After receiving the standardized data set, the modeling module extracts the load fluctuation cycle features of the user's electricity consumption behavior through a multi-layer convolutional neural network and identifies the power grid harmonic spectrum features. The bidirectional long short-term memory network captures the temporal dependence relationship of voltage sag events and fuses the multi-dimensional features into a feature vector matrix. Combining the fault events in the historical power grid fault record database, the dynamic association model establishes a non-linear mapping relationship between the power quality indicators and the user's electricity consumption pattern, providing basic parameters for subsequent analysis.

[0050] Based on the mapping relationship output by the dynamic association model, the analysis module combines the power grid topological structure data and the real-time load distribution, and uses the adjacency matrix propagation algorithm of the graph neural network to calculate the power loss transfer coefficient between adjacent nodes. The initial loss weight matrix is fused with line impedance, load rate and environmental factors to generate a dynamic loss weight matrix, and further superimposed with the no-load loss of the transformer and the line contact resistance loss parameters. Through iterative optimization of the gradient backpropagation algorithm, a dynamically updated line loss contribution evaluation matrix is generated, and the device identifiers in the high-loss area with a contribution exceeding the preset threshold are screened out, and the coordinates of the optimized target area and the parameter adjustment range are output.

[0051] The detection module compares the current power consumption characteristics in the feature vector matrix with the historical normal mode in real time, and detects the deviation data of the harmonic distortion rate and load mutation through the integrated learning model. Combining the coordinates of the optimized target area output by the analysis module, a multi-dimensional anomaly scoring model including harmonic distortion weight and load mutation weight is constructed, and a set of governance strategies sorted by priority is generated. The strategy set includes governance device identification, parameter adjustment range and execution order, triggers a warning instruction and generates a regulation suggestion.

[0052] The optimization module extracts the deviation characteristics of the user power consumption pattern in the set of governance strategies, and calls the long short-term memory network to predict the peak-valley change trend of the load curve in the next 24 hours. Combining the time period division rules of the time-of-use electricity price policy database, the initial load prediction result is optimized in time period, and the electricity consumption time period optimization suggestion and load transfer plan are generated. Through the genetic algorithm, the multi-objective optimization of the load transfer plan is carried out to generate a Pareto optimal solution set that meets the requirements of reducing line loss and improving power supply quality, and the solution set is fed back to the analysis module to update the line loss contribution evaluation matrix.

[0053] The control module dynamically selects the instruction transmission protocol according to the network environment parameters detected by the adaptive communication module, and encapsulates the regulation suggestion and the optimization scheme into a data packet that conforms to the interface of the target governance device. Through the heartbeat detection mechanism, the running state of the governance device is monitored in real time, and the voltage regulation response time and harmonic suppression efficiency parameters are collected to form a data stream including the device running state. After the state data is transmitted back to the data fusion module, it triggers the adjustment of the weight parameters of the dynamic association model, realizing the closed-loop feedback from data acquisition to strategy execution and then to effect verification. The optimized load distribution data synchronously updates the parameters of the power grid topology loss model to ensure that the system adaptively optimizes the power quality governance strategy.

[0054] Specifically, for the intelligent analysis system of power quality and collection data described in the present invention, the data fusion module is further configured as: For the electricity consumption data collected by the electricity meter terminal and the voltage and current waveform data collected by the power grid monitoring device, a sliding time window is used to segment and align the data to generate a time-aligned electricity consumption and waveform data set; the temperature and humidity data collected by the environmental sensor and the time-aligned electricity consumption and waveform data set are fused in time series through interpolation to generate a spatio-temporal correlation data set; the noise data in the spatio-temporal correlation data set is removed through a data cleaning algorithm to generate a standardized data set, and the standardized data set is output to the modeling module.

[0055] The data fusion module realizes the integration of multi-source heterogeneous data through multi-step processing. The electricity consumption data collected by the electricity meter terminal and the voltage and current waveform data obtained by the power grid monitoring device have mismatched time series due to sampling frequency differences. A sliding time window is used to segment and align the data, and the electricity consumption and waveform data collected by different devices are divided into data segments at fixed time intervals to eliminate the timestamp offset problem and generate a time-aligned electricity consumption and waveform data set. The temperature and humidity data collected by the environmental sensor have inconsistent collection frequencies with the electricity data. Through interpolation, the discrete environmental parameter data is complemented to the same time resolution as the electricity consumption and waveform data set to generate a spatio-temporal correlation data set with a unified time reference. The spatio-temporal correlation data set may include outliers caused by sensor signal interference or communication packet loss. A data cleaning algorithm is used to identify and remove the noise data that deviates from the normal fluctuation range, such as separating high-frequency noise components based on wavelet transform or eliminating pulse interference through median filtering, and finally outputting a standardized data set without noise interference. The standardized data set includes time-synchronized electricity consumption, waveforms, and environmental parameters, providing highly consistent input data for the modeling module to support subsequent feature extraction and model construction.

[0056] Specifically, for the intelligent analysis system of power quality and meter reading data of the present invention, the modeling module is further configured to: Input the standardized data set output by the data fusion module into a multi-layer convolutional neural network to extract the load fluctuation cycle features and power grid harmonic spectrum features in the user's electricity consumption behavior; Capture the temporal dependence relationship of voltage sag events through a bidirectional long short-term memory network to generate a feature vector matrix including load fluctuation and harmonic features; Perform correlation analysis on the feature vector matrix and the fault events in the historical power grid fault record database to establish a non-linear mapping relationship between the power quality index and the user's electricity consumption pattern.

[0057] The modeling module performs multi-level feature extraction and pattern analysis on the standardized data set. The standardized data set output by the data fusion module is input into a multi-layer convolutional neural network, and the low-frequency load fluctuation cycle features and high-frequency power grid harmonic spectrum features in the user's power consumption behavior are extracted through different convolutional kernels respectively. The convolutional layer gradually reduces the data dimension and retains the key frequency band information, and the pooling layer screens out the significant features to form the feature codes of the load fluctuation and harmonic spectrum. The bidirectional long short-term memory network receives the feature codes output by the convolutional neural network. The forward layer captures the current gradual change features before the occurrence of the voltage sag event, and the backward layer identifies the harmonic residual features after the voltage recovery. The time step splicing generates a feature vector matrix including the time series dependence relationship of the complete voltage sag event. The feature vector matrix is subjected to similarity matching with the fault events in the historical power grid fault record database, and the dynamic time warping algorithm is used to align the time series patterns of the fault events and the real-time feature vectors, and a non-linear mapping relationship between the harmonic distortion rate in the power quality index, the voltage sag duration and the load mutation amplitude in the user's power consumption pattern is established. This mapping relationship is mapped to a high-dimensional feature space through a kernel function to form a set of quantifiable correlation parameters, providing dynamic model support for the line loss contribution calculation of the analysis module.

[0058] Specifically, for the intelligent analysis system of power quality and meter reading data described in the present invention, the analysis module is further configured to: Based on the non-linear mapping relationship established by the modeling module, analyze the electrical parameters and real-time load distribution of each node in the power grid topology structure data; Through the adjacency matrix propagation algorithm of the graph neural network, calculate the power loss transfer coefficient between adjacent nodes to generate an initial loss weight matrix; According to the initial loss weight matrix and the load rate parameter in the real-time load data, generate a dynamic loss weight matrix including line impedance, load rate and environmental factors, and input the dynamic loss weight matrix into the optimization module.

[0059] Based on the dynamic association relationship established by the modeling module, the analysis module analyzes the voltage, current parameters and electrical connection relationships of each node in the power grid topology, and determines the power transmission path of the nodes in combination with the real-time load distribution data. The graph neural network constructs the connection weights between nodes according to the adjacency matrix of the power grid topology, aggregates the current and impedance parameters of adjacent nodes through the message passing mechanism, and calculates the power loss transfer coefficient between lines. During the generation process of the initial loss weight matrix, the adjacency matrix propagation algorithm integrates the line impedance parameters and the node load distribution, quantifies the energy loss transfer ratio between adjacent nodes, and forms a weight matrix reflecting the basic loss distribution. The dynamic loss weight matrix integrates the initial loss weight and the load rate parameter in the real-time load data, introduces the temperature and humidity data collected by the environmental sensor to dynamically correct the line resistance parameter, and generates a dynamic matrix including multi-dimensional parameters of the line operation state. After receiving the dynamic matrix, the optimization module iteratively adjusts the matrix weight parameters through the backpropagation algorithm, optimizes the line loss calculation model, provides real-time updated input data for the subsequent line loss contribution degree evaluation, and forms a progressive processing link from topology analysis to dynamic parameter optimization.

[0060] Specifically, for the intelligent analysis system of power quality and meter reading data according to the present invention, the analysis module is further configured to: Superimpose the no-load loss parameter of the transformer and the contact resistance loss parameter of the line in the dynamic loss weight matrix to generate a weight matrix corrected by the correlation factor; Iteratively optimize the weight matrix corrected by the correlation factor through the gradient backpropagation algorithm to generate a dynamically updated line loss contribution degree evaluation matrix; According to the contribution degree threshold of the nodes in the line loss contribution degree evaluation matrix, screen out the device identifiers of the high-loss area where the contribution degree exceeds the preset threshold, and output the device identifiers of the high-loss area to the detection module.

[0061] During the line loss assessment process, the analysis module corrects and optimizes the multi-dimensional parameters of the dynamic loss weight matrix. The dynamic loss weight matrix already includes parameters such as line impedance, load rate, and environmental factors. The no-load loss parameters of the transformer are superimposed to reflect the iron loss and eddy current loss of the transformer in the no-load state. At the same time, the line contact resistance loss parameter is introduced to quantify the additional resistance loss caused by the oxidation or loosening of the connection points. The weight matrix after the correlation factor correction forms an enhanced matrix covering the loss characteristics of the entire life cycle of grid equipment by weighted fusion of the transformer and line contact losses. The gradient backpropagation algorithm takes the actual measured line loss value as the objective function, iteratively adjusts the weight distribution ratio of each node in the matrix, adaptively reduces the interference weight of the environmental temperature on the line resistance during the optimization process, and generates a dynamically updated line loss contribution degree evaluation matrix. The node contribution degree threshold in the evaluation matrix is set based on the analysis of historical line loss data. When screening high-loss nodes with a contribution degree exceeding the threshold, the equipment identifier is determined by combining the grid topology path analysis. The identifier includes the transformer number, the position of the line section switch, and the coordinates of the nodes with abnormal load rates, and is output to the detection module to trigger a targeted anomaly detection process. The dynamically updated evaluation matrix is synchronously fed back to the modeling module to adjust the mapping relationship parameters between the feature vector matrix and the power quality indicators, forming a two-way optimization mechanism for loss assessment and feature modeling.

[0062] Specifically, for the intelligent analysis system of power quality and meter reading data described in the present invention, the detection module is further configured to: Input the current power consumption feature vector output by the modeling module into a random forest classifier, and calculate the deviation degree between the current power consumption feature vector and the historical normal mode; Based on the deviation degree and the line loss contribution degree evaluation matrix generated by the analysis module, construct a multi-dimensional anomaly scoring model including the harmonic distortion rate and the load mutation weight; According to the output result of the multi-dimensional anomaly scoring model, generate a priority policy set including the governance equipment identifier, the adjustment parameter range, and the execution order, and output the priority policy set to the optimization module.

[0063] The detection module constructs an abnormal detection and governance strategy generation link based on the output data of the modeling module and the analysis module. The current power consumption feature vector output by the modeling module includes the load fluctuation period, harmonic spectrum, and time series features. After being input into the random forest classifier, the deviation degrees of each dimension of the feature vector from the typical samples in the historical normal mode library are calculated in parallel by multiple decision trees, and a scoring value quantifying the deviation degree is generated. The line loss contribution degree evaluation matrix provides the node loss weight parameters. Combining with the deviation degree scoring value, a multi-dimensional abnormal scoring model with the harmonic distortion rate as the horizontal axis and the load mutation weight as the vertical axis is constructed. The model correlates the harmonic distortion amplitude, load mutation rate, and line loss contribution degree through a weighted fusion algorithm and outputs a comprehensive abnormal score. The area where the comprehensive abnormal score exceeds the set threshold triggers the strategy generation process. According to the power grid topology path and device association relationship, the governance device identifier and its adjustable parameter range are matched, and a priority strategy set is generated in descending order of the scoring value. The strategy set includes the transformer voltage regulation gear position, capacitor switching instruction, and load transfer target value. After being output to the optimization module, it drives the generation of the load prediction and regulation plan, forming a closed-loop processing link from abnormal detection to strategy execution.

[0064] Specifically, for the intelligent analysis system of power quality and collection data according to the present invention, the optimization module is further configured to: Extract the user power consumption pattern deviation features in the priority strategy set output by the detection module; Input the deviation features into the long short-term memory network to predict the peak-valley change trend of the load curve in the next 24 hours and generate an initial load prediction result; Combine the time period division rules in the external time-of-use electricity price policy database to optimize the time period of the initial load prediction result and generate a time-of-use electricity optimization suggestion and a load transfer plan.

[0065] Based on the priority policy set output by the detection module, the optimization module extracts the deviation features in the user's power consumption pattern that deviate from the historical normal fluctuation range, including the load mutation amplitude, the harmonic distortion duration, and the voltage sag occurrence frequency. After the deviation features are input into the long short-term memory network, the network captures the temporal correlation of the load curve through the time-step recurrent unit, learns the variation rules of the peak and valley periods, predicts the load demand at each time point within the next 24 hours, and generates an initial prediction result including the peak load period and the valley load period. The initial prediction result is matched with the time period division rules in the external time-of-use electricity price policy database, the load prediction data during the peak electricity price period is correlated with the adjustable capacity during the valley period, the feasible path of load transfer is calculated through the dynamic programming algorithm, and an optimized suggestion for the electricity consumption period is generated. The load transfer plan combines the real-time parameters of the power grid topology loss model, determines the target equipment for load adjustment and the adjustment amplitude, and outputs an optimized plan set including the capacitor switching time, the transformer voltage regulation instruction, and the distributed power source scheduling strategy, providing an executable regulation parameter set for the control module and forming a technical closed-loop from load prediction to strategy generation.

[0066] Specifically, for the intelligent analysis system of power quality and collection data according to the present invention, the optimization module is further configured to: Perform a joint simulation of the optimized suggestion for the electricity consumption period and the parameters of the power grid topology loss model updated by the analysis module; Perform multi-objective optimization on the load transfer plan through a genetic algorithm to generate a Pareto optimal solution set that meets the requirements of reducing line loss and improving power supply quality; Feed back the Pareto optimal solution set to the analysis module, trigger the real-time update of the line loss contribution degree evaluation matrix, and input the updated evaluation matrix into the detection module.

[0067] The optimization module realizes the dynamic adjustment of the load strategy through a multi-stage optimization and feedback mechanism. The optimization suggestions for the electricity consumption period include the load transfer target value and the period division scheme. When jointly simulating with the parameters of the power grid topology loss model updated by the analysis module, the simulation engine calls the power grid topology structure data and the real-time load distribution parameters to simulate the line loss change trend and the probability of voltage sag occurrence under different load transfer paths, and evaluates the impact of the optimization suggestions on the power supply quality. The genetic algorithm receives the data sets of line loss and power supply quality indicators output by the simulation, takes the minimization of line loss and the maximization of voltage stability as the optimization objectives, generates multiple groups of load distribution schemes through population initialization, and iteratively screens out the Pareto optimal solution set that satisfies the double-objective constraints through crossover and mutation operations. The Pareto optimal solution set includes the load transfer scheme and the corresponding line loss threshold parameters, which are fed back to the analysis module to drive the real-time update of the line loss contribution degree evaluation matrix. The updated matrix integrates the latest load distribution and line loss weight parameters. The updated evaluation matrix is input to the detection module, triggering the recalibration of the harmonic distortion rate and the load mutation weight in the abnormal scoring model, forming a closed-loop feedback link from strategy optimization to detection parameter adjustment, and enhancing the system's dynamic response ability to changes in the power grid operation state.

[0068] Specifically, for the intelligent analysis system of power quality and meter reading data described in the present invention, the control module is further configured to: Dynamically select an instruction transmission protocol according to the network delay and bandwidth data detected by the adaptive communication module; Package the regulation suggestions of the detection module and the optimization scheme of the optimization module into data packets that conform to the communication interface of the target power quality governance device; Real-time monitor the operating state of the power quality governance device through a heartbeat detection mechanism, and transmit the state data including the voltage regulation response time and the harmonic suppression efficiency back to the data fusion module.

[0069] The control module achieves the precise execution of regulation instructions through multi-stage protocol adaptation and status monitoring. The adaptive communication module continuously detects the latency fluctuations and bandwidth occupancy rates of the network transmission path, and dynamically selects a low-latency transmission protocol or a high-reliability protocol. For example, when the network is congested, it switches to the Transmission Control Protocol with a retransmission mechanism, and when the network status is stable, it uses the User Datagram Protocol to improve transmission efficiency. The regulation suggestions of the detection module include device adjustment instructions and parameter thresholds, and the load transfer scheme of the optimization module includes the target value of time period adjustment. The control module encodes and converts these two types of data according to the communication interface specifications of the target power quality control device, encapsulates them into a data packet format compliant with the Modbus or IEC 61850 standard, adds device address identification and checksum fields, and generates an instruction data stream that can be parsed by the control device. The heartbeat detection mechanism sends status query instructions to the control device at fixed time intervals, collects the actual time taken for voltage regulation and the difference in waveform distortion rates before and after harmonic suppression in the device response data, and calculates real-time regulation efficiency parameters. The status data is timestamped and then transmitted back to the data fusion module, triggering the dynamic update of the original data set, synchronously correcting the feature extraction weights of the modeling module, forming a closed-loop control link from instruction issuance to effect feedback, and maintaining the system's adaptive regulation ability for the grid operation status.

[0070] Specifically, for the intelligent analysis system of power quality and metering data according to the present invention, the control module is further configured to: Extract the voltage regulation response time and harmonic suppression efficiency parameters from the operation status data of the power quality control device; Compare and analyze the harmonic suppression efficiency parameter with the expected target in the priority strategy set to generate a regulation effect evaluation report; Adjust the weight parameters of the dynamic association model in the modeling module according to the evaluation report, and feedback the adjusted model parameters to the analysis module to complete the model self-optimization of the closed-loop feedback link.

[0071] The control module realizes continuous optimization of the control strategy through multi-dimensional data analysis and model iteration. The operating status data of the control device includes the actual time difference between the time when the voltage regulation instruction is issued and the time when the target voltage is reached. The time difference is extracted as the voltage regulation response time parameter; the harmonic suppression efficiency parameter is calculated by comparing the amplitude attenuation ratio of the specific harmonic component in the current waveform before and after the control, reflecting the device's ability to suppress harmonic distortion. The expected goals preset in the priority strategy set include the response time threshold and the harmonic suppression rate target value. The deviation between the actual parameters and the expected goals is calculated during the comparative analysis process, and a control effect evaluation report including response delay cause analysis, harmonic suppression effect evaluation and equipment health rating is generated. After the evaluation report is input into the modeling module, the weight parameters of the dynamic correlation model are adjusted in a targeted manner according to the reasons for the voltage regulation lag and the frequency band characteristics of insufficient harmonic suppression identified in the report, such as increasing the feature weight in the load mutation scenario or reducing the weight ratio of the ambient temperature interference item. The adjusted model parameters are synchronously fed back to the analysis module to drive the recalculation of node loss weights in the line loss contribution evaluation matrix. The updated matrix data is input into the detection module to trigger the calibration of the abnormal scoring model, forming a closed-loop self-optimization link from effect evaluation to model parameter update, thereby improving the system's adaptability to dynamic changes in the power grid.

[0072] The technical features of the technical solution of the present invention are explained as follows: Data fusion module: Electric meter terminal: used to collect user electricity consumption data, record instantaneous values ​​of current and voltage and power parameters, and provide basic data for load analysis.

[0073] Power grid monitoring equipment: acquire voltage waveforms, current harmonic spectra and frequency fluctuation data of power grid nodes in real time, and monitor abnormal power quality indicators.

[0074] Environmental sensors: collect environmental parameters such as temperature and humidity, and quantify the impact of environmental factors on line resistance and transformer losses.

[0075] Sliding time window alignment: Align the asynchronous data collected by different devices in segments at fixed time intervals (such as 1 second) to eliminate clock deviations between devices.

[0076] Interpolation fusion: Linearly interpolate low-frequency environmental data (such as temperature collected every minute) to complete it to the same time resolution as high-frequency electrical energy data (such as voltage per second).

[0077] Wavelet transform denoising: Identify and filter out high-frequency noise (such as electromagnetic interference pulses) in sensor signals through multi-scale decomposition.

[0078] Modeling module: Convolutional Neural Network (CNN): The first-layer convolutional kernels extract the characteristics of the load fluctuation period (such as daily / weekly load curves), and the second-layer convolutional kernels capture the harmonic spectrum characteristics (such as the amplitudes of the 3rd / 5th / 7th harmonics).

[0079] Bidirectional Long Short-Term Memory Network (Bi-LSTM): The forward layer analyzes the gradual change trend of the current before the voltage sag, and the backward layer identifies the residual harmonic characteristics after the sag, generating the complete time-series characteristics of the voltage sag event.

[0080] Dynamic Association Model: Through the mapping of kernel functions, the harmonic distortion rate, voltage sag duration, and the amplitude of user load mutation are associated to form a parameter set of non-linear mapping relationships.

[0081] Analysis Module: Graph Neural Network (GNN): Based on the adjacency matrix of the power grid topology, calculate the current transmission path and impedance matching relationship between adjacent nodes, and quantify the power loss transfer weight.

[0082] Dynamic Loss Weight Matrix: Integrate line impedance, load factor, and temperature correction resistance parameters to reflect the line loss distribution under real-time operating conditions.

[0083] Gradient Backpropagation Optimization: Taking the actual line loss measurement value as the objective function, iteratively adjust the matrix weights to reduce the interference weight of the ambient temperature on the resistance value.

[0084] Detection Module: Random Forest Classifier: Through the voting mechanism of multiple decision trees, calculate the deviation score (in the range of 0 - 1) between the current load fluctuation, harmonic distortion, etc. and the historical normal mode.

[0085] Multi-Dimensional Anomaly Scoring Model: Construct a two-dimensional scoring space with the line loss contribution degree as the horizontal axis and the harmonic distortion rate as the vertical axis, and output a comprehensive anomaly index after weighted fusion.

[0086] Priority Policy Set: Generate a governance instruction sequence according to the sorted score values, for example, preferentially adjust the transformer tap in the high-loss area or switch the compensation capacitor.

[0087] Optimization Module: Long Short-Term Memory Network (LSTM): Predict the peak and valley periods of the load in the next 24 hours and output the time-segmented load curve (such as the peak period from 14:00 to 16:00).

[0088] Genetic Algorithm Multi-Objective Optimization: Taking the minimization of line loss and the maximization of voltage stability as the objectives, generate the Pareto optimal solution set (such as load transfer schemes A / B / C) through crossover and mutation.

[0089] Dynamic Feedback Update: Input the optimized load distribution data into the analysis module to drive the line loss evaluation matrix to recalculate the weights according to the latest topology parameters.

[0090] Control Module: Adaptive Communication Protocol: Dynamically adjusts the instruction transmission strategy according to network latency (selects UDP for < 100ms and switches to TCP for ≥ 100ms).

[0091] Modbus / IEC 61850 Encapsulation: Converts the regulation instructions into a data packet format recognizable by the target governance device (such as register address mapping, CRC check).

[0092] Heartbeat Detection Mechanism: Polls the status of the governance device at 1-second intervals, collects the voltage regulation response time (such as the time taken from instruction issuance to reaching the standard ≤ 200ms) and the harmonic suppression rate (such as from 15% to 5%).

[0093] Closed-loop Feedback Link: Regulation Effect Evaluation Report: Compares the actual response time with the expected target (such as the 200ms threshold), and analyzes the reasons for voltage regulation lag (such as communication delay or equipment aging).

[0094] Model Parameter Iteration: Adjusts the weight coefficient of the harmonic distortion rate in the dynamic association model according to the evaluation report to improve the sensitivity of anomaly detection.

[0095] Full-link Collaboration: After the data fusion module receives the status data of the governance device, it triggers the update of the feature extraction weights of the modeling module, forming a self-loop system from data collection to strategy optimization.

[0096] Technical Model of the Data Fusion Module: Sliding Time Window Alignment: For asynchronous data collected by the electricity meter terminal and the power grid monitoring device (such as second-level electricity consumption and millisecond-level waveform data), divides the data segments according to a fixed time window (such as 1 second), aligns the timestamps, eliminates the clock deviation between devices, and generates a dataset of electricity consumption and waveforms with consistent time dimensions.

[0097] Interpolation Method Fusion: Adopts a linear interpolation algorithm for low-frequency environmental data (such as temperature collected per minute), complements it to the same time resolution as the high-frequency electricity data, forms a spatio-temporal correlation dataset, and retains the dynamic influence characteristics of environmental parameters on line resistance.

[0098] Wavelet Transform Denoising: Identifies high-frequency noise (such as electromagnetic interference pulses) in the sensor signal through multi-scale decomposition, filters out the invalid components, retains the effective electricity data, and improves the signal purity of the subsequent modeling input.

[0099] Neural Network Model of the Modeling Module: Multi-layer Convolutional Neural Network (CNN): The first-layer convolutional kernel: Extracts the characteristics of the user load fluctuation period (such as the peak-valley pattern of the daily load curve), and captures the local load change pattern through convolution operations.

[0100] The second-layer convolutional kernel: Focuses on the power grid harmonic spectrum characteristics (such as the amplitudes of the 3rd and 5th harmonics), and identifies the distorted signals in specific frequency bands.

[0101] Bidirectional Long Short-Term Memory Network (Bi-LSTM): The forward layer: Analyzes the gradual change trend of the current before the occurrence of a voltage sag event (such as the timing characteristics of the current slowly dropping to the threshold).

[0102] The reverse layer: Captures the harmonic residual characteristics after the voltage recovery (such as the residual fluctuations of the harmonic amplitudes after the sag), and generates a complete time series feature vector of the voltage sag event.

[0103] The dynamic correlation model: Matches the feature vector matrix with the historical fault event library, and establishes a non-linear relationship between the harmonic distortion rate and the load mutation through kernel function mapping (such as when the distortion rate exceeds 5%, the load mutation amplitude threshold is correlated).

[0104] The graph neural network model of the analysis module: The adjacency matrix propagation algorithm: Based on the power grid topology (such as node connection relationships and electrical parameters), constructs an adjacency matrix to represent the electrical connection strength between nodes, and aggregates the current and impedance parameters of adjacent nodes through a message passing mechanism to calculate the power loss transfer coefficient.

[0105] The dynamic loss weight matrix: Integrates the load factor in the real-time load data (such as node load factor ≥ 80%), line impedance (such as 0.5 Ω / km), and temperature correction resistance parameters (such as the resistance increases by 0.4% for every 1°C increase in temperature) to generate a line loss weight distribution reflecting the current operating state.

[0106] Gradient backpropagation optimization: Taking the actual line loss measurement value as the target (such as the measured loss of a certain node is 5 kW), iteratively adjusts the matrix weight parameters to optimize the model's adaptability to complex power grid environments.

[0107] The anomaly scoring model of the detection module: Random forest classifier: Constructs multiple decision trees, and votes to calculate the deviation score (in the range of 0 - 1, ≥ 0.7 is determined as abnormal) of the current power consumption characteristics (such as load fluctuation amplitude, harmonic distortion rate) from the historical normal pattern.

[0108] Multidimensional anomaly scoring model: Taking the line loss contribution degree (such as node contribution degree ≥ 15%) as the horizontal axis and the harmonic distortion rate (such as distortion rate ≥ 8%) as the vertical axis, constructs a two-dimensional scoring space, and outputs a comprehensive anomaly index after weighted fusion (such as an index ≥ 0.8 triggers an alarm).

[0109] Priority strategy generation: Generate a governance instruction sequence sorted by the scoring value (e.g., preferentially adjust the transformer tap or switch the compensation capacitor in the high-loss area) to ensure that the resource allocation matches the risk level.

[0110] Prediction and optimization model of the optimization module: Long Short-Term Memory Network (LSTM): Input the historical load curve and deviation characteristics, predict the peak and valley periods of the load in the next 24 hours (e.g., predict 14:00 - 16:00 as the peak period), and output the time-segmented load distribution curve.

[0111] Multi-objective optimization of genetic algorithm: Aiming at minimizing the line loss (e.g., reducing it to below 4%) and maximizing the voltage stability (e.g., voltage fluctuation ≤ ±5%), generate a Pareto optimal solution set (e.g., load transfer schemes A / B / C) through crossover and mutation operations to balance economy and power supply quality.

[0112] Dynamic feedback update: Input the optimized load distribution data into the analysis module, drive the line loss evaluation matrix to recalculate the weights according to the latest topological parameters, and achieve the dynamic synchronization of the model and the grid state.

[0113] Communication and feedback model of the control module: Adaptive communication protocol selection: Dynamically adjust the transmission strategy according to the network delay (e.g., select the UDP protocol when the delay < 100ms, switch to the TCP protocol when ≥ 100ms) to ensure the real-time nature of the instructions.

[0114] Modbus / IEC 61850 encapsulation: Convert the regulation instructions into a data packet format recognizable by the target governance device (e.g., register address mapping, CRC check code) to ensure the compatibility and reliability of the instructions.

[0115] Heartbeat detection mechanism: Poll the status of the governance device at a fixed interval (e.g., 1 second), collect the voltage regulation response time (e.g., the time from instruction issuance to reaching the standard ≤ 200ms) and the harmonic suppression rate (e.g., from 15% to 5%), generate a device operation status data set, and feedback it to the data fusion module to trigger model iteration.

[0116] Technical logic of the closed-loop feedback link: Regulation effect evaluation: Compare the actual response parameters (e.g., voltage regulation time 220ms) with the expected target (threshold 200ms), analyze the reasons for the delay (e.g., communication link congestion or device response slowness), and generate an evaluation report.

[0117] Model parameter iteration: Adjust the weight coefficient of the harmonic distortion rate in the dynamic correlation model according to the evaluation report (e.g., increase the distortion rate weight from 0.6 to 0.8) to enhance the sensitivity of anomaly detection.

[0118] Full-link collaboration: After the data fusion module receives the status data of the governance device, it triggers the update of the feature extraction weights of the modeling module, forming an adaptive closed-loop system of "data collection → feature modeling → anomaly detection → policy optimization → instruction execution → effect feedback", improving the system's real-time response ability to the dynamic changes of the power grid.

[0119] In the scenarios of power quality monitoring and line loss governance of the power grid, the present invention realizes the technical solution through the collaboration of multi-source heterogeneous data fusion and intelligent algorithms. The data fusion module collects the user power consumption data through the meter terminal, the power grid monitoring equipment synchronously obtains the voltage and current waveform data, and the environmental sensor monitors the temperature and humidity parameters in real time. The sliding time window is used to align the power consumption and waveform data according to the second-level time stamp, eliminating the data time series misalignment caused by the sampling frequency difference of different devices. After interpolating and fusing the environmental parameter data, a spatio-temporal correlation data set is generated. The high-frequency noise components are separated by wavelet transform, and the impulse interference is eliminated by median filtering. The standardized data set is output to the modeling module to solve the problem of insufficient data integrity caused by manual meter reading.

[0120] The modeling module inputs the standardized data set into a multi-layer convolutional neural network. The first-layer convolutional kernel extracts the load fluctuation period features, the second-layer convolutional kernel captures the harmonic spectrum features, and the bidirectional long short-term memory network analyzes the change law of the current waveform before and after the voltage sag event, generating a feature vector matrix including load, harmonic, and voltage sag features. The feature vector matrix is dynamically time warped and matched with the fault events in the historical power grid fault database to establish a non-linear mapping relationship between the harmonic distortion rate and the load mutation amplitude, solving the problem of the lack of dynamic association between power consumption data and power quality indicators. The analysis module calculates the power loss transfer coefficient between nodes through the adjacency matrix propagation algorithm of the graph neural network based on the power grid topology structure data, superimposes the no-load loss parameters of the transformer and the line contact resistance loss parameters, generates a dynamic loss weight matrix, and combines the load rate parameter in the real-time load data to identify the coordinates of the high-loss area where the line loss contribution exceeds 15%, and the positioning error is controlled within 5% of the power grid topology node spacing.

[0121] The detection module calculates the deviation score of the current power consumption feature vector from the historical normal mode through a random forest classifier, and constructs a multi-dimensional anomaly scoring model in combination with the line loss contribution degree evaluation matrix. When the comprehensive score exceeds the threshold, a set of priority strategies including transformer voltage regulation positions and capacitor switching times is generated. The optimization module uses a long short-term memory network to predict the peak and valley load changes in the next 24 hours, generates a load transfer plan in combination with the time-of-use electricity price policy, and outputs a Pareto optimal solution set after genetic algorithm optimization, triggering the update of the line loss evaluation matrix. The control module dynamically selects the TCP or UDP protocol to send instructions according to the network delay. The response data of the governance device is transmitted back through the heartbeat detection mechanism. The voltage regulation response time error is less than 200 ms, and the harmonic suppression efficiency data drives the iterative update of the dynamic association model weights, forming a full-link collaboration from data collection, anomaly warning to closed-loop regulation, and realizing the improvement of the timeliness of power supply quality optimization.

[0122] The present invention solves the defects of the prior art through a multi-module collaborative data processing architecture. The data fusion module integrates the electric meter terminal, power grid monitoring equipment and environmental sensors, and uses the sliding time window alignment and interpolation method fusion technology to eliminate the timestamp misalignment and data missing problems caused by manual meter reading. The anti-interference communication protocol transmits the original data to the central processor, cleans the noise through the wavelet transform and median filtering algorithm, and generates a highly complete standardized data set, reducing the human intervention error from the data collection source.

[0123] The modeling module extracts the load fluctuation period and harmonic spectrum features based on a convolutional neural network, captures the temporal dependence relationship of voltage sag events in combination with a bidirectional long short-term memory network, and constructs a dynamic association model between power quality indicators and power consumption patterns. The analysis module analyzes the power grid topology through a graph neural network, calculates the power loss transfer coefficient between nodes, superimposes the no-load loss of the transformer and the line contact resistance parameters, and generates a dynamically updated line loss contribution degree evaluation matrix to accurately locate the key equipment nodes in the high-loss area.

[0124] The detection module uses a random forest classifier and a multi-dimensional anomaly scoring model to compare the current power consumption characteristics with the historical mode deviation in real time, and generates a priority governance strategy in combination with the line loss evaluation result. The optimization module predicts the load curve through a long short-term memory network, generates a time period optimization plan by integrating the electricity price policy, and the control module adaptively selects the transmission protocol based on the network state to execute the regulation instruction. The operation state data of the governance device is transmitted back to the data fusion module, driving the iterative update of the dynamic association model weights, forming a full-link closed-loop feedback from data collection, anomaly warning to strategy execution, and realizing the real-time response of power supply quality optimization and line loss governance.

Claims

1. An intelligent analysis system for power quality and meter reading data, characterized in that, Including: A data fusion module, configured to collect raw data through electricity meter terminals, power grid monitoring devices, and environmental sensors, perform timestamp alignment and data cleaning on the raw data, and generate a standardized dataset; A modeling module, configured to extract the temporal sequence features of user electricity consumption behavior, power grid harmonics, and voltage sags in the standardized dataset through a convolutional neural network, generate a feature vector matrix, and construct a dynamic association model based on historical power grid operation data, and output the mapping relationship between power quality indicators and electricity consumption patterns in the feature vector matrix; An analysis module, configured to combine power grid topology structure data and real-time load distribution, calculate the loss weights between line nodes through a graph neural network, generate a line loss contribution degree evaluation matrix, identify the high-loss area topology paths and associated devices with contribution degrees exceeding the threshold in the line loss contribution degree evaluation matrix, and output the coordinates of the optimization target area and the parameter adjustment range; A detection module, configured to compare the current electricity consumption characteristics in the feature vector matrix with the historical normal mode in real time, detect harmonic distortion rate and load mutation deviation data through an integrated learning model, and combine the coordinates of the optimization target area output by the analysis module to generate a set of governance strategies including priority sorting, trigger warning instructions and regulation suggestions; An optimization module, configured to call a long short-term memory network to predict the future load curve according to the deviation data output by the detection module, fuse the time-of-use electricity price policy to generate an optimized electricity consumption period plan, and feedback the optimized load distribution data to the analysis module for updating the loss model parameters in the power grid topology structure data; A control module, configured to match the current network environment through an adaptive communication module, send device parameter adjustment instructions to the power quality governance device, and synchronously collect the operation status data of the power quality governance device and transmit it back to the data fusion module.

2. The intelligent analysis system for power quality and meter reading data according to claim 1, characterized in that The data fusion module is further configured to: Perform data segmentation alignment on the electricity consumption data collected by the electricity meter terminal and the voltage and current waveform data collected by the power grid monitoring device using a sliding time window to generate a time-aligned electricity consumption and waveform dataset; Fuse the temperature and humidity data collected by the environmental sensor with the time-aligned electricity consumption and waveform dataset through interpolation to generate a spatio-temporal correlation dataset; remove the noise data in the spatio-temporal correlation dataset through a data cleaning algorithm to generate a standardized dataset, and output the standardized dataset to the modeling module.

3. The intelligent analysis system for power quality and meter reading data according to claim 1, wherein The modeling module is further configured to: Input the standardized dataset output by the data fusion module into a multi-layer convolutional neural network to extract the load fluctuation cycle features and power grid harmonic spectrum features in user electricity consumption behavior; Capture the temporal dependence relationship of voltage sag events through a bidirectional long short-term memory network to generate a feature vector matrix including load fluctuations and harmonic features; Perform correlation analysis on the feature vector matrix and the fault events in the historical power grid fault record database to establish a non-linear mapping relationship between the power quality indicators and the user electricity consumption pattern.

4. The intelligent analysis system for power quality and meter reading data according to claim 3, wherein The analysis module is further configured to: Based on the non - linear mapping relationship established by the modeling module, analyze the electrical parameters and real - time load distribution of each node in the power grid topology data; Through the adjacency matrix propagation algorithm of the graph neural network, calculate the power loss transfer coefficient between adjacent nodes and generate an initial loss weight matrix; According to the initial loss weight matrix and the load rate parameter in the real - time load data, generate a dynamic loss weight matrix including line impedance, load rate and environmental factors, and input the dynamic loss weight matrix into the optimization module.

5. The intelligent analysis system for power quality and meter reading data according to claim 4, wherein The analysis module is further configured as: Superimpose the no - load loss parameter of the transformer and the contact resistance loss parameter of the line in the dynamic loss weight matrix to generate a weight matrix corrected by the correlation factor; Iteratively optimize the weight matrix corrected by the correlation factor through the gradient backpropagation algorithm to generate a dynamically updated line loss contribution degree evaluation matrix; According to the contribution degree threshold of the nodes in the line loss contribution degree evaluation matrix, screen out the device identifiers in the high - loss area where the contribution degree exceeds the preset threshold, and output the device identifiers in the high - loss area to the detection module.

6. The intelligent analysis system for power quality and meter reading data according to claim 1, wherein The detection module is further configured as: Input the current power consumption feature vector output by the modeling module into a random forest classifier to calculate the deviation degree between the current power consumption feature vector and the historical normal mode; Based on the deviation degree and the line loss contribution degree evaluation matrix generated by the analysis module, construct a multi - dimensional anomaly scoring model including harmonic distortion rate and load mutation weight; According to the output result of the multi - dimensional anomaly scoring model, generate a priority policy set including governance device identification, adjustment parameter range and execution order, and output the priority policy set to the optimization module.

7. The intelligent analysis system for power quality and meter reading data according to claim 6, characterized in that, The optimization module is further configured as: Extract the user power consumption mode deviation features in the priority policy set output by the detection module; Input the deviation features into a long - short - term memory network to predict the peak - valley change trend of the load curve in the next 24 hours and generate an initial load prediction result; Combined with the time - period division rules in the external time - of - use electricity price policy database, optimize the time - period of the initial load prediction result to generate electricity - using time - period optimization suggestions and load transfer schemes.

8. The intelligent analysis system for power quality and meter reading data according to claim 7, characterized in that The optimization module is further configured as: Conduct a joint simulation of the electricity - using time - period optimization suggestions and the updated power grid topology loss model parameters of the analysis module; Perform multi - objective optimization on the load transfer scheme through a genetic algorithm to generate a Pareto optimal solution set that meets the requirements of line loss reduction and power supply quality improvement; Feed back the Pareto optimal solution set to the analysis module, trigger the real - time update of the line loss contribution degree evaluation matrix, and input the updated evaluation matrix into the detection module.

9. The intelligent analysis system for power quality and meter reading data according to claim 8, characterized in that The control module is further configured as: Dynamically select an instruction transmission protocol according to the network delay and bandwidth data detected by the adaptive communication module; Package the regulation suggestions of the detection module and the optimization scheme of the optimization module into a data packet that conforms to the communication interface of the target power quality governance device; The operating status of the power quality management device is monitored in real time through a heartbeat detection mechanism, and status data including voltage regulation response time and harmonic suppression efficiency are transmitted back to the data fusion module.

10. The intelligent analysis system for power quality and meter reading data according to claim 9, characterized in that, The control module is further configured to: Extracting voltage regulation response time and harmonic suppression efficiency parameters from the operating status data of the power quality control device; Compare and analyze the harmonic suppression efficiency parameter with the expected target in the priority strategy set to generate a regulation effect evaluation report; The weight parameters of the dynamic association model in the modeling module are adjusted according to the evaluation report, and the adjusted model parameters are fed back to the analysis module.

Images