Comprehensive evaluation method and system for operation state of electric energy metering system

Abstract

The invention provides an electric energy metering system operation state comprehensive evaluation method and system, and the method comprises the steps: receiving a real-time energy consumption data stream, carrying out the processing through an adaptive threshold anomaly detection algorithm and a Bayesian inference method, and generating a real error range estimation corresponding to each data point; generating a target prediction result by using the time sequence prediction model in combination with historical same-period data, seasonal factors and a reinforcement learning mechanism; comparing the target prediction result with the real-time energy consumption data flow to obtain a comparison result, and generating a comprehensive performance fluctuation evaluation result based on the comparison result and the multi-dimensional state space model; constructing a comprehensive evaluation index system by using an integrated learning algorithm and a fuzzy logic algorithm, and obtaining a comprehensive score corresponding to each ammeter by using the comprehensive evaluation index system; according to the method, the prediction accuracy of the electric energy metering system, the comprehensiveness of health condition evaluation and the scientificity of maintenance decision are improved, and support is provided for efficient management and optimal scheduling of an intelligent power grid.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of smart grid technology, and in particular to a method and system for comprehensively evaluating the operating status of an electric energy metering system. Background Art

[0002] In smart grid and energy management systems, the stability and accuracy of power metering systems are crucial. As power systems become increasingly complex and intelligent, real-time and comprehensive evaluation of the operating status of power metering systems has become an urgent need.

[0003] At present, the evaluation methods for the operating status of electric energy metering systems mainly include traditional threshold detection, single model prediction and simple scoring system. However, these methods have obvious shortcomings. Traditional threshold detection uses a fixed threshold, which is difficult to adapt to changes in different environments, resulting in a high false alarm rate; a single time series prediction model is difficult to handle complex nonlinear relationships and uncertainties, and the prediction accuracy is insufficient; existing health status assessment methods mostly rely on simple rules or single indicators, fail to fully consider multi-dimensional factors, and are difficult to comprehensively evaluate the actual operating status of the meter. Therefore, the existing solutions have great limitations in terms of adaptability, prediction accuracy and comprehensive evaluation. Summary of the invention

[0004] The embodiment of the present invention provides a method and system for comprehensively evaluating the operating status of an electric energy metering system, which are used to solve the problems of poor adaptive capability, insufficient prediction accuracy and incomplete comprehensive evaluation in the prior art.

[0005] In a first aspect, an embodiment of the present invention provides a method for comprehensively evaluating the operating status of an electric energy metering system, comprising:

[0006] Receiving a real-time energy consumption data stream, detecting the real-time energy consumption data stream using an adaptive threshold anomaly detection algorithm to obtain a filtered data stream, and performing uncertainty quantification processing on the filtered data stream using a Bayesian inference method to generate a true error range estimate corresponding to each data point;

[0007] Based on the true error range estimation, the time series forecasting model is used to forecast the short-term load of the system and generate the target forecast result. During the forecasting process, the historical data of the same period, seasonal factors and reinforcement learning mechanism are combined to optimize the forecast;

[0008] Comparing the target prediction result with the real-time energy consumption data stream to obtain a comparison result, and determining an instant deviation metric and a performance fluctuation situation based on the comparison result and a multidimensional state space model to generate a comprehensive performance fluctuation evaluation result;

[0009] Based on the comprehensive performance fluctuation evaluation results, an integrated learning algorithm and a fuzzy logic algorithm are used to construct a comprehensive evaluation index system. Using the comprehensive evaluation index system, the health status of each electric meter in the electric energy metering system is evaluated to obtain a comprehensive score corresponding to each electric meter.

[0010] Optionally, comparing the target prediction result and the real-time energy consumption data stream to obtain a comparison result. Based on the comparison result and a multi-dimensional state space model, an instant deviation metric and a performance fluctuation condition are determined to generate a comprehensive performance fluctuation evaluation result, including:

[0011] Using the dynamic time warping algorithm to align the time series data of the target prediction result and the real-time energy consumption data stream to obtain aligned data. Based on the aligned data, calculate the absolute error between the predicted value and the actual value at each time point to obtain error information. Apply the sliding window technique to analyze the error information to obtain a comparison result;

[0012] Based on the comparison result, construct a multi-dimensional state space model, and use a Kalman filter to estimate the hidden state in the multi-dimensional state space model to obtain a state estimate value. Based on the state estimate value, calculate the instant deviation metrics corresponding to different dimension indicators, and the dimension indicators include power consumption, voltage, and current;

[0013] Analyze the change trend of the instant deviation metrics in different time periods to generate the performance fluctuation conditions of the electric energy metering system corresponding to different time periods;

[0014] Use the grey relational analysis method to analyze the mutual relationship between the dimension indicators to obtain the analysis result of the mutual influence between the dimension indicators;

[0015] Combine the instant deviation metrics, the performance fluctuation conditions, and the analysis result of the mutual influence between the dimension indicators to generate a comprehensive performance fluctuation evaluation result.

[0016] Optionally, based on the comprehensive performance fluctuation evaluation results, an integrated learning algorithm and a fuzzy logic algorithm are used to construct a comprehensive evaluation index system. Using the comprehensive evaluation index system, the health status of each electric meter in the electric energy metering system is evaluated to obtain a comprehensive score corresponding to each electric meter, including:

[0017] Based on multiple evaluation dimensions, use an integrated learning algorithm and a deep neural network to process the comprehensive performance fluctuation evaluation results to obtain a preliminary score reflecting the operating state of the electric energy metering system. The evaluation dimensions include energy consumption efficiency, peak load management level, and stability;

[0018] Introduce the fuzzy logic algorithm and Bayesian network to process the non-linear relationships and uncertain information in the preliminary score, so as to adjust the preliminary score and generate an adjusted score;

[0019] Use the graph neural network to analyze the influence of the power grid topology structure to obtain the structural influence result, and construct a comprehensive evaluation index system based on the structural influence result and the adjusted score;

[0020] Based on the comprehensive evaluation index system and the multi-objective optimization algorithm, comprehensively evaluate the health status of each electric meter in the electric energy metering system to obtain the comprehensive score corresponding to each electric meter.

[0021] Optionally, use the graph neural network to analyze the influence of the power grid topology structure to obtain the structural influence result, and construct a comprehensive evaluation index system based on the structural influence result and the adjusted score, including:

[0022] Based on the adjusted score, use the graph neural network to model the electric meters in the power grid and the connection relationships between the electric meters to generate a target representation vector reflecting the overall structure characteristics of the power grid;

[0023] Based on the target representation vector, analyze the influence of the power grid topology structure on the operating states of each electric meter, identify the key operating risk points, and determine the structural influence result based on the key operating risk points;

[0024] Integrate the structural influence result and the adjusted score to obtain the integrated data, apply the multi-layer perceptron to process the integrated data to obtain the processed data, and construct a comprehensive evaluation index system based on the processed data.

[0025] Optionally, based on the adjusted score, use the graph neural network to model the electric meters in the power grid and the connection relationships between the electric meters to generate a target representation vector reflecting the overall structure characteristics of the power grid, including:

[0026] Based on the adjusted score, use the graph neural network to model the electric meters in the power grid and the connection relationships between the electric meters to generate a preliminary representation vector, and combine the topology structure characteristics of the power grid during the modeling process to obtain the interaction and network effect between the electric meters. The topology structure characteristics include the connection mode and connection distance between network nodes, and the network nodes represent electric meters;

[0027] Introduce an adaptive weight mechanism to dynamically adjust the weights of the connection relationships between the electric meters to update the preliminary representation vector and obtain an updated representation vector;

[0028] Apply the variational autoencoder to perform dimensionality reduction processing on the updated representation vector to generate a target representation vector.

[0029] Optionally, based on the estimated true error range, a time series prediction model is used to predict the short-term load of the system, generating a target prediction result. During the prediction process, historical data for the same period, seasonal factors, and a reinforcement learning mechanism are combined for optimized prediction, including:

[0030] Based on the estimated true error range, a time series prediction model is used to predict the short-term load of the system, generating a preliminary prediction result;

[0031] Based on the preliminary prediction result and historical data for the same period, the preliminary prediction result is adjusted to obtain an adjusted prediction result;

[0032] Based on the adjusted prediction result, seasonal factors are introduced to generate a seasonally adjusted prediction result. Based on the seasonally adjusted prediction result, a reinforcement learning mechanism is applied. By interacting with the environment, the key parameters of the time series prediction model are adjusted to obtain the target prediction result. The reinforcement learning mechanism includes a multi-agent reinforcement learning algorithm and a Q reinforcement learning algorithm.

[0033] Optionally, based on the adjusted prediction result, seasonal factors are introduced to generate a seasonally adjusted prediction result. Based on the seasonally adjusted prediction result, a reinforcement learning mechanism is applied. By interacting with the environment, the key parameters of the time series prediction model are adjusted to obtain the target prediction result. The reinforcement learning mechanism includes a multi-agent reinforcement learning algorithm and a Q reinforcement learning algorithm, including:

[0034] Based on the adjusted prediction result, seasonal factors are introduced and combined with Fourier transform to identify and extract key periodic frequency components. The key periodic frequency components include daily cycle components, weekly cycle components, and special event cycle components;

[0035] Based on the seasonal factors, the key periodic frequency components, and historical load data, a long short-term memory network model is constructed. Using the long short-term memory network model, a seasonally adjusted prediction result is generated;

[0036] The multi-agent reinforcement learning algorithm is applied. By coordinating the work of multiple agents, the key parameters of the time series prediction model and the long short-term memory network model are optimized to generate the target prediction result. During the process of applying the multi-agent reinforcement learning algorithm for optimization, the Q reinforcement learning algorithm is introduced to adjust the behavior strategy of the agents according to the environmental feedback information. The environmental feedback information includes the real-time energy consumption data stream and the operation state change information of the power metering system.

[0037] In a second aspect, an embodiment of the present invention provides a comprehensive evaluation system for the operation state of a power metering system, including:

[0038] A receiving module is used to receive a real-time energy consumption data stream, detect the real-time energy consumption data stream using an adaptive threshold anomaly detection algorithm to obtain a filtered data stream, and perform uncertainty quantification processing on the filtered data stream using a Bayesian inference method to generate a true error range estimate corresponding to each data point;

[0039] A prediction module is used to predict the short-term load of the system based on the true error range estimation using a time series prediction model to generate a target prediction result, and to optimize the prediction by combining historical data of the same period, seasonal factors, and a reinforcement learning mechanism during the prediction process;

[0040] A generation module, used for comparing the target prediction result with the real-time energy consumption data stream to obtain a comparison result, and determining an instant deviation metric and a performance fluctuation situation based on the comparison result and a multidimensional state space model to generate a comprehensive performance fluctuation evaluation result;

[0041] The evaluation module is used to construct a comprehensive evaluation index system based on the comprehensive performance fluctuation evaluation result, using an integrated learning algorithm and a fuzzy logic algorithm, and use the comprehensive evaluation index system to evaluate the health status of each meter in the electric energy metering system to obtain a comprehensive score corresponding to each meter.

[0042] In a third aspect, an embodiment of the present invention provides a computing device, comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a comprehensive evaluation method for the operating status of an electric energy metering system as described in any one of the first aspects.

[0043] In a fourth aspect, an embodiment of the present invention provides a computer storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement a method for comprehensively evaluating the operating status of an electric energy metering system as described in any one of the first aspects.

[0044] In an embodiment of the present invention, a real-time energy consumption data stream is received, and an adaptive threshold anomaly detection algorithm is used to detect the real-time energy consumption data stream to obtain a filtered data stream. A Bayesian inference method is used to perform uncertainty quantification processing on the filtered data stream to generate an estimated true error range corresponding to each data point. Based on the estimated true error range, a time series prediction model is used to predict the short-term load of the system to generate a target prediction result. During the prediction process, historical data of the same period, seasonal factors, and a reinforcement learning mechanism are combined for optimized prediction. The target prediction result is compared with the real-time energy consumption data stream to obtain a comparison result. Based on the comparison result and a multi-dimensional state space model, an instant deviation metric and a performance fluctuation condition are determined to generate a comprehensive performance fluctuation evaluation result. Based on the comprehensive performance fluctuation evaluation result, an integrated learning algorithm and a fuzzy logic algorithm are used to construct a comprehensive evaluation index system. Using the comprehensive evaluation index system, the health status of each electric meter in the electric energy metering system is evaluated to obtain a comprehensive score corresponding to each electric meter. The technical solution provided by the present invention can dynamically adjust the threshold according to the real-time energy consumption data by introducing an adaptive threshold anomaly detection algorithm, avoiding false alarms and missed alarms caused by a fixed threshold. This makes the anomaly detection more accurate, capable of promptly capturing real anomalies, improving the reliability and response speed of the system. The time series prediction model not only considers periodic and non-periodic components but also dynamically adjusts the model parameters through a reinforcement learning mechanism, significantly improving the accuracy and robustness of short-term load prediction. The comprehensive performance fluctuation evaluation result provides a more accurate operation state evaluation to help identify potential problems. The comprehensive evaluation index system not only considers multiple evaluation dimensions but also improves the flexibility and accuracy of the evaluation through intelligent algorithms. Through the comprehensive evaluation of the health status of the electric meters, the system can identify potential problems in advance, formulate reasonable maintenance plans, extend the equipment life, and improve the system reliability. In summary, the above technical solution significantly improves the anomaly detection accuracy, prediction accuracy, comprehensiveness of health status evaluation, and scientific nature of maintenance decision-making of the electric energy metering system through means such as adaptive anomaly detection, uncertainty quantification, optimized prediction mechanism, instant deviation metric, and comprehensive evaluation, thereby providing strong support for the efficient management and optimized scheduling of the smart grid. Among them, through the graph neural network and topological structure features, it is ensured that the evaluation result is closer to the actual operating environment; the adaptive weight mechanism enables the representation vector to more accurately reflect the interaction and network effect between electric meters, especially the changes in different time periods; the variational autoencoder removes redundant information, improving the efficiency and accuracy of subsequent analysis and evaluation.

[0045] These aspects or other aspects of the present invention will be more clearly understood in the following description of the embodiments. Brief Description of the Drawings

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0047] Figure 1 It is a flowchart of a comprehensive evaluation method for the operating state of an electric energy metering system provided by an embodiment of the present invention;

[0048] Figure 2 It is a schematic structural diagram of a comprehensive evaluation system for the operating state of an electric energy metering system provided by an embodiment of the present invention;

[0049] Figure 3 It is a schematic structural diagram of a computing device provided by an embodiment of the present invention. Detailed implementation manners

[0050] In order to enable those skilled in the art to better understand the solutions of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention.

[0051] In some processes described in the specification, claims and above-mentioned accompanying drawings of the present invention, there are multiple operations that appear in a specific order. However, it should be clearly understood that these operations can be executed not in the order in which they appear herein or in parallel. The operation numbers such as 101, 102, etc. are only used to distinguish different operations, and the numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations can be executed in sequence or in parallel. It should be noted that the descriptions such as "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., and do not represent a sequence, nor do they limit that "first" and "second" are of different types.

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0053] Figure 1 An embodiment of the present invention provides a flowchart of a comprehensive evaluation method for the operating state of an electric energy metering system. As Figure 1 shown, the method includes:

[0054] In modern smart grids, accurate evaluation of the operating status of the electric energy metering system is crucial to ensuring the reliability and efficiency of power supply. Traditional evaluation methods are often limited to static data analysis and are difficult to fully reflect the dynamic changes and complexity of the power grid. Based on this, the present invention provides a comprehensive evaluation method for the operating status of the electric energy metering system, such as Figure 1 ,include:

[0055] Step 101: receiving a real-time energy consumption data stream, detecting the real-time energy consumption data stream using an adaptive threshold anomaly detection algorithm to obtain a filtered data stream, and performing uncertainty quantification processing on the filtered data stream using a Bayesian inference method to generate a true error range estimate corresponding to each data point;

[0056] In this step, the real-time energy consumption data stream refers to the energy consumption data collected in real time from the energy metering system, which usually records the energy usage of each meter at fixed time intervals (such as every minute or every hour). The filtered data stream refers to the data stream processed by the anomaly detection algorithm, which removes obviously abnormal data points and retains normal or near-normal energy consumption data.

[0057] Suppose there is a smart grid system that contains multiple electricity meters. These meters send electricity consumption data to the central server every 5 minutes, and the data is obtained to obtain real-time energy consumption data streams. In order to ensure the accuracy of the data, an adaptive threshold anomaly detection algorithm is used to process the data. The algorithm dynamically adjusts the threshold according to the data distribution over a period of time in the past, so as to more accurately identify abnormal data points. For example, at noon on a certain day, a certain meter suddenly reported an extremely high energy consumption value, far exceeding its usual average value. Through the adaptive threshold anomaly detection algorithm, this anomaly is identified and marked as an abnormal data point, and then removed from the data stream, leaving the filtered data stream for subsequent processing. The Bayesian inference method is used to quantify the uncertainty of these data. Specifically, a prior distribution is constructed based on historical data, and combined with the observed values ​​in the filtered data stream, this distribution is updated to obtain an estimate of the true error range of each data point. For example, for an energy consumption data point of a certain meter at a certain moment, the Bayesian inference method can give the true value of the data point that may fall within a certain interval (such as ±5%), which provides a more reliable basis for subsequent prediction and evaluation.

[0058] Step 102: Based on the true error range estimation, the system short-term load is predicted using a time series prediction model to generate a target prediction result, and during the prediction process, historical data of the same period, seasonal factors, and a reinforcement learning mechanism are combined to perform optimized prediction;

[0059] In this step, historical concurrent data refers to historical data corresponding to the current time period, such as data from the same period last year. Seasonal factors refer to considering the impact of different seasons, months, or holidays on power consumption.

[0060] This step is based on an estimation of the true error range and uses a long short-term memory network as a time series prediction model to predict the short-term load of the system. During the prediction process, historical concurrent data (such as load data from the same period last year) and seasonal factors (such as the peak period of air conditioner use in summer) are combined, and the model parameters are dynamically adjusted through a multi-agent reinforcement learning mechanism. For example, if the prediction result shows that a certain day is a weekend and the temperature is high, the model will combine historical data and current weather forecasts to adjust the prediction parameters to more accurately predict the load situation on that day. The final generated target prediction result not only considers periodic and aperiodic components but also improves the robustness and accuracy of the prediction through the reinforcement learning mechanism.

[0061] Step 103: Compare the target prediction result and the real-time energy consumption data stream to obtain a comparison result. Based on the comparison result and the multi-dimensional state space model, determine the immediate deviation metric and the performance fluctuation situation to generate a comprehensive performance fluctuation evaluation result;

[0062] In this step, the immediate deviation metric refers to an indicator that measures the difference between the current predicted value and the actual value. The performance fluctuation situation refers to an indicator that reflects the stability of the system operation state, including the fluctuation amplitude and frequency.

[0063] This step compares the target prediction result with the real-time energy consumption data stream, calculates the difference between the two to obtain a comparison result. Then, based on these comparison results and the multi-dimensional state space model, determine the immediate deviation metric and the performance fluctuation situation. For example, if the actual load of a certain electricity meter during a certain period is much higher than the predicted value, this deviation is recorded, and whether this deviation belongs to normal fluctuation or there are potential problems is analyzed through the multi-dimensional state space model. The final generated comprehensive performance fluctuation evaluation result can comprehensively reflect the operation state of the electricity meter and help identify potential risks.

[0064] Step 104: Based on the comprehensive performance fluctuation evaluation result, use an ensemble learning algorithm and a fuzzy logic algorithm to construct a comprehensive evaluation index system. Use the comprehensive evaluation index system to evaluate the health status of each electricity meter in the power metering system to obtain a comprehensive score corresponding to each electricity meter;

[0065] In this step, the comprehensive score refers to the final score obtained according to the comprehensive evaluation index system, which reflects the health status of the electricity meter.

[0066] This step is based on the comprehensive performance fluctuation evaluation results. Using ensemble learning algorithms (such as random forest) and fuzzy logic algorithms, a comprehensive evaluation index system is constructed. This system includes multiple evaluation dimensions, such as energy consumption efficiency, peak load management level, stability, etc. By processing and weighting the data of each dimension, the comprehensive score of each electricity meter is finally generated. For example, if an electricity meter performs well in terms of energy consumption efficiency and stability but is slightly lacking in peak load management, the comprehensive score will reflect these characteristics. In this way, the health status of each electricity meter can be comprehensively evaluated, potential problems can be discovered in advance, reasonable maintenance plans can be formulated, the equipment life can be extended, and the system reliability can be improved.

[0067] In the embodiment of the present invention, the adaptive threshold anomaly detection algorithm can dynamically adjust the threshold, avoid false alarms and missed alarms, and improve the reliability and response speed of the system; significantly improve the accuracy and robustness of short-term load prediction; through the multi-dimensional state space model and the comprehensive evaluation index system, the health status of the electricity meter can be comprehensively evaluated, providing a scientific basis to support maintenance decisions; through the comprehensive evaluation of the health status of the electricity meter, potential problems can be identified in advance, reasonable maintenance plans can be formulated, the equipment life can be extended, and the system reliability can be improved.

[0068] The present invention provides a specific embodiment, step 103. Compare the target prediction result and the real-time energy consumption data stream to obtain a comparison result. Based on the comparison result and the multi-dimensional state space model, determine the instantaneous deviation metric and the performance fluctuation situation to generate a comprehensive performance fluctuation evaluation result, which specifically includes the following steps:

[0069] Step 301: Use the dynamic time warping algorithm to align the time series data of the target prediction result and the real-time energy consumption data stream to obtain the aligned data. Based on the aligned data, calculate the absolute error between the predicted value and the actual value at each time point to obtain the error information. Apply the sliding window technique to analyze the error information to obtain the comparison result;

[0070] In this step, the aligned data refers to adjusting the time series data of the target prediction result and the real-time energy consumption data stream to the same time axis through the dynamic time warping algorithm to ensure the corresponding relationship between the two in terms of time. The absolute error refers to the absolute value of the difference between the predicted value and the actual value, which is used to measure the prediction accuracy.

[0071] Assume that real-time energy consumption data stream and target prediction results have been obtained. Due to possible time delays or other factors in the actual collection process, the timestamps of the prediction results and the actual data are not exactly the same. Therefore, the dynamic time warping algorithm is used to align these two sets of time series data to ensure that the predicted values and actual values correspond one by one at each time point. For example, the real-time energy consumption data stream of a certain electric meter between 2:00 pm and 4:00 pm on a certain day is [100, 105, 110, 108, 106] kWh, while the target prediction result is [98, 103, 112, 107, 105] kWh. After alignment by the DTW algorithm, the absolute error at each time point is calculated, and the error information obtained is [2, 2, 2, 1, 1] kWh. Applying the sliding window technique with a window size of 3 time points, this error information is analyzed. For the first window (from 2:00 to 2:30), the mean absolute error is (2 + 2 + 2) / 3 = 2 kWh; for the second window (from 2:15 to 2:45), the mean absolute error is (2 + 2 + 1) / 3 ≈ 1.67 kWh; and so on. The final comparison results show the error variation in different time periods.

[0072] Step 302: Based on the comparison results, construct a multi-dimensional state space model, use a Kalman filter to estimate the hidden state in the multi-dimensional state space model to obtain a state estimate value, and based on the state estimate value, calculate the immediate deviation metric corresponding to different dimension indicators, where the dimension indicators include power consumption, voltage, and current;

[0073] In this step, the hidden state refers to the state variable in the system that cannot be directly observed but can be inferred from the observed data.

[0074] In this step, based on the comparison results, a multi-dimensional state space model is constructed, which considers the dimension indicators of power consumption, voltage, and current. To estimate the hidden states of these dimensions, a Kalman filter is used to process the model. Assume that in a certain time period, there are the following observed data: power consumption is 100 kWh; voltage is 220 V; current is 5 A; through the Kalman filter, the hidden states behind these observed data are estimated. For example, the hidden power consumption is 102 kWh; the hidden voltage is 221 V; the hidden current is 4.9 A; based on these state estimate values, the immediate deviation metric corresponding to different dimension indicators is calculated. For example, for power consumption, the immediate deviation metric is |100 - 102| = 2 kWh; for voltage, the immediate deviation metric is |220 - 221| = 1 V; for current, the immediate deviation metric is |5 - 4.9| = 0.1 A. These immediate deviation metrics reflect the operating states of the dimension indicators at the current time point.

[0075] Step 303: Analyze the changing trends of the instant deviation metrics within different time periods to generate the performance fluctuations of the power metering system corresponding to different time periods;

[0076] In this step, the changing trend refers to the law of data change over time, such as rising, falling, or fluctuating.

[0077] This step further analyzes the changing trends of the instant deviation metrics within different time periods. For example, assume that the changing trend from 8:00 am to 8:00 pm in a day is of concern. By analyzing the instant deviation metrics during this period, it can be obtained that during the morning period (8:00 - 12:00), the instant deviation metrics of power consumption are relatively stable, with an average of 1.5 kWh; during the afternoon period (12:00 - 16:00), with the arrival of the electricity consumption peak, the instant deviation metrics of power consumption increase, with an average of 2.5 kWh; during the evening period (16:00 - 20:00), the electricity consumption gradually decreases, and the instant deviation metrics also decrease accordingly, with an average of 1.8 kWh. Through the analysis of these changing trends, the performance fluctuations of the power metering system corresponding to different time periods are generated. For example, it can be concluded that the system operates relatively stably in the morning period, there are certain fluctuations in the afternoon period due to the electricity consumption peak, and the fluctuations gradually decrease in the evening period.

[0078] Step 304: Use the grey relational analysis method to analyze the mutual relationships between the dimension indicators to obtain the analysis results of the mutual influences between the dimension indicators;

[0079] In this step, the mutual relationship refers to the dependence and interaction between different dimension indicators.

[0080] This step uses the grey relational analysis method to analyze in order to more deeply understand the mutual relationships between power consumption, voltage, and current. Assume that the observed data collected over a period of time is as follows: power consumption is [100, 105, 110, 108, 106] kWh; voltage is [220, 222, 223, 221, 220] V; current is [5, 5.1, 5.2, 5.1, 5] A; through the grey relational analysis method, the correlation degrees between the dimension indicators are calculated. For example, the correlation degree between power consumption and voltage is 0.85, indicating a strong positive correlation between the two; the correlation degree between power consumption and current is 0.92, indicating a stronger correlation between the two. These analysis results better analyze the mutual influences between different dimension indicators and provide a scientific basis for subsequent optimization and improvement.

[0081] Step 305: Combine the instant deviation metrics, the performance fluctuations, and the analysis results of the mutual influences between the dimension indicators to generate a comprehensive performance fluctuation assessment result;

[0082] This step combines the analysis results of the instantaneous deviation metric, performance fluctuations, and the mutual influence among dimensional indicators to generate a comprehensive performance fluctuation assessment result. For example, by integrating all the above information, the following conclusions can be drawn: during the morning period, the overall system operation is relatively stable, the relationship among power consumption, voltage, and current is relatively simple, mainly showing a stable low-load state; during the afternoon period, due to the peak electricity consumption, the power consumption increases significantly, and the voltage and current also fluctuate accordingly, the system enters a high-load state, and the instantaneous deviation metric increases, indicating that the system is under greater pressure; during the evening period, as the electricity consumption decreases, the system gradually returns to a relatively stable state, and both the instantaneous deviation metric and performance fluctuations are improved. In summary, the comprehensive performance fluctuation assessment result fully reflects the operation state of the power metering system at different time periods, and can help identify potential problems and formulate reasonable maintenance plans.

[0083] The embodiment of the present invention ensures the accurate correspondence in time between the predicted value and the actual value through the dynamic time warping algorithm, improving the reliability of error analysis; through the sliding window technique and the Kalman filter, not only the absolute error at each time point is calculated, but also the hidden state in the multi-dimensional state space model is estimated, providing more refined error information; the comprehensive performance fluctuation assessment result fully reflects the operation state of the power metering system, helping to identify potential problems; through the research on the performance fluctuations at different time periods and the mutual relationship among dimensional indicators, it provides a scientific basis for the optimization and improvement of the system, and improves the support level for maintenance decision-making.

[0084] In the power metering system, accurately evaluating the health status of each electric meter is crucial for ensuring the stable operation and efficient management of the power system. Traditional evaluation methods often rely on a single indicator and are difficult to fully reflect the true state of the electric meter and its complex working environment. Based on this, the present invention provides a specific embodiment, step 104, based on the comprehensive performance fluctuation assessment result, using the ensemble learning algorithm and the fuzzy logic algorithm to construct a comprehensive evaluation index system, and using the comprehensive evaluation index system to evaluate the health status of each electric meter in the power metering system to obtain a comprehensive score corresponding to each electric meter, specifically including the following steps:

[0085] Step 401: Based on multiple evaluation dimensions, use the ensemble learning algorithm and the deep neural network to process the comprehensive performance fluctuation assessment result to obtain a preliminary score reflecting the operation state of the power metering system, where the evaluation dimensions include energy consumption efficiency, peak load management level, and stability;

[0086] In this step, the evaluation dimension refers to different aspects used to evaluate the operation state of the power metering system, such as energy consumption efficiency, peak load management level, and stability.

[0087] This step processes the evaluation results of comprehensive performance fluctuations based on three main evaluation dimensions: energy consumption efficiency, peak load management level, and stability. The specific steps are as follows. Energy consumption efficiency processing, which measures the accuracy of the electricity consumption recorded by the electricity meter. For example, the average energy consumption of a certain electricity meter is 100 kWh / day in the past month; Peak load management level processing, which evaluates the ability of the electricity meter to cope with peak electricity demand. For example, the maximum load of a certain electricity meter during peak electricity consumption hours is 300 kW. Stability processing, which checks the stable performance of the electricity meter under different conditions. For example, the voltage fluctuation range of a certain electricity meter is ±2% in the past week. The data of these evaluation dimensions are fused and processed using integrated learning algorithms (such as random forests) and deep neural networks. For example, for a certain electricity meter, the integrated learning algorithm may obtain an energy consumption efficiency score of 85 points, a peak load management level score of 90 points, and a stability score of 88 points based on historical data. Then, the deep neural network further processes these scores and combines non-linear features to generate a preliminary score reflecting the overall operating state of the electricity meter, such as 87 points.

[0088] Step 402: Introduce a fuzzy logic algorithm and a Bayesian network to process the non-linear relationships and uncertain information in the preliminary score, so as to adjust the preliminary score and generate an adjusted score;

[0089] Although the preliminary score in this step has considered multiple evaluation dimensions, there are still some non-linear relationships and uncertain information. Therefore, a fuzzy logic algorithm and a Bayesian network are introduced to further optimize the score. The fuzzy logic algorithm is used to process the fuzzy relationships between evaluation dimensions. For example, when the energy consumption efficiency is high but the stability is low, the fuzzy logic algorithm can balance the influence of the two to ensure a more reasonable score. The Bayesian network is used to process uncertain information. For example, if there are missing or abnormal data in some evaluation dimensions, the Bayesian network can reason based on the data of other relevant dimensions to fill in or correct this information; Through the processing of the fuzzy logic algorithm and the Bayesian network, the preliminary score is adjusted. For example, for the above-mentioned electricity meter, the adjusted score after processing may become 86 points, which reflects a more accurate evaluation of the operating state.

[0090] Step 403: Use a graph neural network to analyze the influence of the power grid topology structure to obtain the structural influence result, and based on the structural influence result and the adjusted score, construct a comprehensive evaluation index system;

[0091] In this step, the structural influence result refers to the result obtained by analyzing the electricity meters and their connection relationships in the power grid through a graph neural network, which reflects the influence of the topology structure on the operating state of the electricity meter. The comprehensive evaluation index system refers to an index system composed of multiple evaluation dimensions, which is used to comprehensively evaluate the health status of the electricity meter.

[0092] To more comprehensively evaluate the health status of electricity meters, it is necessary to consider the topology of the power grid. For this purpose, graph neural networks are used to model the electricity meters and their connection relationships in the power grid. For example, in a certain regional power grid, electricity meter A and electricity meter B are connected by a high-load line, while electricity meter C is connected to electricity meter A by a low-load line. Through graph neural network analysis, key nodes and bottlenecks are identified. For example, high-load lines may cause local power supply instability, thus affecting the operating status of relevant electricity meters; based on the structural impact results and the adjusted scores, a comprehensive evaluation index system is constructed. For example, for electricity meter A, considering its key position on the high-load line, its comprehensive score may be slightly reduced to reflect potential risks; while for electricity meter C, due to its relatively stable connection, its comprehensive score may be relatively high.

[0093] Step 404: Based on the comprehensive evaluation index system and the multi-objective optimization algorithm, comprehensively evaluate the health status of each electricity meter in the power metering system to obtain the comprehensive score corresponding to each electricity meter;

[0094] In this step, the multi-objective optimization algorithm is used to comprehensively evaluate the health status of each electricity meter. Specifically, the comprehensive evaluation index system considers multiple dimensions such as energy consumption efficiency, peak load management level, stability, and the impact of the power grid topology. Through the multi-objective optimization algorithm, the best balance point is found among these dimensions to generate the comprehensive score of each electricity meter. For example, for electricity meter A, its comprehensive score is 84 points, reflecting its potential risk at the high-load line position; for electricity meter B, its comprehensive score is 89 points, indicating its good operating status; for electricity meter C, its comprehensive score is 92 points, showing its relatively stable connection and good operation. These comprehensive scores not only provide a scientific basis to support maintenance decisions but also help the system identify potential problems in advance and formulate reasonable maintenance plans.

[0095] In the embodiment of the present invention, through the integrated learning algorithm and the deep neural network, the comprehensiveness and accuracy of the evaluation are improved; the fuzzy logic algorithm and the Bayesian network are introduced to ensure that the scores are more reasonable and accurate; the impact of the power grid topology is analyzed by combining the graph neural network, making the evaluation results closer to the actual operating environment and enhancing the reliability of the evaluation; the comprehensive scores provide a scientific basis for maintenance decisions, improving the scientificity and predictability of maintenance decisions.

[0096] The present invention provides a specific embodiment. In step 403, the graph neural network is used to analyze the impact of the power grid topology to obtain the structural impact results. Based on the structural impact results and the adjusted scores, a comprehensive evaluation index system is constructed, which specifically includes the following steps:

[0097] Step 411: Based on the adjusted scores, use a graph neural network to model the electricity meters in the power grid and the connection relationships between the electricity meters, so as to generate a target representation vector reflecting the overall structural characteristics of the power grid;

[0098] In this step, the target representation vector refers to a low-dimensional vector generated by a graph neural network that can reflect the overall structural characteristics of the power grid, capturing the information of the electricity meters and their connection relationships.

[0099] To further evaluate the positions and impacts of these electricity meters in the power grid topology, a graph neural network (GNN) is used to model the electricity meters in the power grid and their connection relationships. Specifically, each electricity meter is regarded as a node in the graph, and the physical or logical connections between the electricity meters (such as power transmission paths) are regarded as the edges in the graph. Through the graph neural network, the interactions and network effects between the electricity meters can be captured, generating a target representation vector reflecting the overall structural characteristics of the power grid. For example, in a certain regional power grid, electricity meter A and electricity meter B are connected by a high-load line, while electricity meter C is connected to electricity meter A by a low-load line. Through graph neural network modeling, the target representation vectors of each electricity meter are obtained. For example, for electricity meter A, it is 0.7, 0.2, 0.1; for electricity meter B, it is 0.5, 0.3, 0.2; for electricity meter C, it is 0.6, 0.2, 0.2; these vectors not only contain the characteristics of the electricity meters themselves but also reflect their positions and connection relationships in the network.

[0100] Step 412: Based on the target representation vector, analyze the impact of the power grid topology on the operating states of each electricity meter, identify the key operating risk points, and determine the structural impact results based on the key operating risk points;

[0101] In this step, the power grid topology refers to the graph structure describing the electricity meters in the power grid and their connection relationships, including nodes (electricity meters) and edges (connections). The key operating risk points refer to the key nodes or connection relationships in the power grid that have a significant impact on the operating states of the electricity meters, such as high-load nodes, bottleneck nodes, etc.

[0102] For example, in the above regional power grid, the high-load nodes are that Meter A and Meter B are connected by a high-load line, which may become a bottleneck, resulting in unstable local power supply; the bottleneck node is that Meter A is located at a critical position on the high-load line. If there is a problem with this line, it will seriously affect the normal operation of Meter A and Meter B; the isolated node is that Meter C is connected to Meter A through a low-load line, but if Meter A fails, Meter C may lose its main power supply source, affecting its stability. Through this analysis, the key operation risk points are identified, and the structural impact results are determined based on these risk points. For example, for Meter A, since it is located at a critical position on the high-load line, the structural impact result indicates that it has a relatively high operation risk; for Meter C, although its connection is relatively stable, it depends on the normal operation of Meter A, so there is also a certain risk.

[0103] Step 413: Integrate the structural impact result and the adjusted score to obtain integrated data, apply a multi-layer perceptron to process the integrated data to obtain processed data, and construct a comprehensive evaluation index system based on the processed data;

[0104] This step combines the structural impact result and the adjusted score to form integrated data. For example, for Meter A, its adjusted score is 86 points, and the structural impact result shows that it has a relatively high operation risk. Integrating this information together, the following data points are formed: Meter A is 86, high-load node, critical position 86, high-load node, critical position; Meter B is 90, high-load node, ordinary position 90, high-load node, ordinary position; Meter C is 84, low-load node, dependent on Meter 84, low-load node, dependent on Meter A; Apply a multi-layer perceptron (MLP) to process the integrated data. The multi-layer perceptron can automatically extract high-level features in the data to help better understand the complex relationships between different factors. For example, after being processed by the multi-layer perceptron, the processed data is obtained: Meter A is 0.85, 0.15, 0.05; Meter B is 0.92, 0.08, 0.05; Meter C is 0.80, 0.10, 0.10; Based on these processed data, a comprehensive evaluation index system is constructed, which takes into account multiple dimensions such as energy consumption efficiency, peak load management level, stability, and the impact of the power grid topology structure. For example, for Meter A, the final score given by the comprehensive evaluation index system is 85 points, reflecting its potential risk at the critical position on the high-load line; for Meter B, the final score is 90 points, indicating its good operation status; for Meter C, the final score is 82 points, indicating its dependence on the stability of Meter A.

[0105] In the embodiments of the present invention, a target representation vector reflecting the overall structural characteristics of the power grid is generated through a graph neural network, ensuring that the evaluation results are closer to the actual operating environment; by identifying key operating risk points, the pertinence and accuracy of the evaluation are enhanced; by constructing a comprehensive evaluation index system, a scientific basis is provided to support maintenance decisions, improving the comprehensiveness and reliability of the evaluation; through the comprehensive evaluation of the health status of electric meters, potential problems are identified in advance, a reasonable maintenance plan is formulated, the equipment life is extended, and the system reliability is improved, providing strong support for the efficient management and optimal scheduling of the smart grid.

[0106] In the electric energy metering system, accurately capturing the overall structural characteristics of the power grid is crucial for optimizing management and fault prediction. Traditional methods often struggle to comprehensively reflect the complexity of the electric meters and their connection relationships in the power grid. Based on this, the present invention provides a specific embodiment. Step 411: Based on the adjusted scores, use a graph neural network to model the electric meters and the connection relationships between the electric meters in the power grid to generate a target representation vector reflecting the overall structural characteristics of the power grid, which specifically includes the following steps:

[0107] Step 421: Based on the adjusted scores, use a graph neural network to model the electric meters and the connection relationships between the electric meters in the power grid to generate a preliminary representation vector. During the modeling process, the topological structural characteristics of the power grid are combined to obtain the interactions and network effects between the electric meters. The topological structural characteristics include the connection mode and connection distance between network nodes, and the network nodes represent the electric meters.

[0108] In this step, the topological structural characteristics refer to the characteristics describing the electric meters and their connection relationships in the power grid, including the connection mode and connection distance between network nodes. The network nodes represent the electric meters here.

[0109] This step is to further evaluate the position and impact of these electricity meters in the power grid topology. A graph neural network is used to model the electricity meters and their connection relationships in the power grid. Specifically, each electricity meter is regarded as a node in the graph, and the physical or logical connections between the electricity meters (such as power transmission paths) are regarded as the edges in the graph. During the modeling process, the topological structure characteristics of the power grid are combined, including the connection methods (such as series, parallel) and connection distances (such as physical distance, transmission loss) between network nodes, to obtain the interactions and network effects between the electricity meters. For example, in a certain regional power grid, electricity meter A and electricity meter B are connected by a high-load line with a physical distance of 5 kilometers and a large transmission loss; electricity meter C is connected to electricity meter A by a low-load line with a physical distance of 3 kilometers and a small transmission loss. Through graph neural network modeling, a preliminary representation vector of each electricity meter is generated to capture its position and connection relationship in the network. For example, electricity meter A is 0.7, 0.2, 0.1; electricity meter B is 0.5, 0.3, 0.2; electricity meter C is 0.6, 0.2, 0.2; these vectors not only contain the characteristics of the electricity meters themselves but also reflect their positions and connection relationships in the network.

[0110] More specifically, the present invention provides a calculation formula for the preliminary representation vector to describe how to model the electricity meters and their connection relationships through a graph neural network and the process of generating the preliminary representation vector. The specific calculation formula is as follows:

[0111] H 0 =f GNN (A, X)+μB;

[0112]

[0113] f GNN (A, X)=MPM(σ ( WX + b ) , A ) ;

[0114] Among them, H 0 represents the preliminary representation vector. The preliminary representation vector is a matrix, and each row represents the preliminary feature representation of a node (such as an electricity meter); f GNN(A, X) represents a graph neural network function used to process input data and generate new node representations; A represents the adjacency matrix, which is a square matrix used to represent the connection relationships between nodes in the graph; X represents the node feature matrix, and these features can include but are not limited to power readings, geographical locations, and historical load data, etc.; μ represents a tuning parameter used to control the influence degree of the correction matrix B on the final preliminary feature representation; B represents the correction matrix, which considers external influencing factors (such as weather, seasonal changes, etc.) and aims to enhance the adaptability of the model to external condition changes; K represents the number of external influencing factors; β k represents the weight coefficient of the k-th external influencing factor, which determines the importance of this factor in the correction matrix; F k represents the feature matrix of the k-th external influencing factor, such as time series data of environmental variables like temperature, humidity, etc.; W represents the weight matrix used to map input features to a new space, usually trained through the backpropagation algorithm; b represents the bias vector, which acts on the input features together with the weight matrix to help adjust the position of the output value and is also trained through the backpropagation algorithm; σ represents the activation function; MPM represents the message passing mechanism, which is a core part of the graph neural network and allows nodes to exchange information with their neighbors to update their own representations. It aggregates the information of neighbors according to the adjacency matrix and combines it with the information of the current node.

[0115] The above preliminary representation vector formula combines the feature data of meter nodes in the power grid with their physical connections, captures the complex relationships between nodes through the graph neural network, and generates a more informative representation vector. At the same time, considering that the operation of the power grid is affected by external factors such as weather and seasonal changes, a correction matrix is added to enhance the adaptability of the model to external conditions. In addition, the message passing mechanism and multi-layer perceptron architecture are adopted, enabling the model to process data at multiple levels, aggregate neighbor information, and dynamically update node representations, thereby comprehensively reflecting the actual connection situation and load changes of the power grid. The application of the formula significantly improves the richness and accuracy of node representations. It can not only capture the non-linear relationships in the input data but also enhance the model's learning ability for complex patterns in the power grid environment. Through layer-by-layer feature extraction and dynamic information update, the model exhibits higher robustness and generalization ability, providing a solid foundation for subsequent analysis and prediction. This method effectively improves the prediction accuracy and stability, ensuring good adaptability in the face of changes in the power grid topology and operating state.

[0116] Step 422: Introduce an adaptive weight mechanism to dynamically adjust the weights of the connection relationships between meters to update the preliminary representation vector and obtain an updated representation vector;

[0117] In order to more accurately capture the interactions and network effects between meters, an adaptive weight mechanism is introduced to dynamically adjust the weights of the connection relationships between meters. For example, during a certain period of time, due to peak electricity consumption, the weights of some connections may increase, reflecting the increased importance of the area; while in other periods, the weights may decrease, reflecting the reduced relative importance of the area. Assume that during the peak electricity consumption in the afternoon of a certain day, the connection between meter A and meter B becomes more critical, so the connection weights between them are dynamically adjusted. Specifically, the connection weight between meter A and meter B is adjusted from 0.7 to 0.8; the connection weight between meter C and meter A remains unchanged at 0.6; based on these adjusted weights, the preliminary representation vector is updated to obtain the updated representation vector. For example, meter A is 0.75, 0.2, 0.05; meter B is 0.6, 0.3, 0.1; meter C is 0.6, 0.2, 0.2; these updated representation vectors more accurately reflect the interactions and network effects between meters, especially the changes in different time periods.

[0118] Step 423: Apply a variational autoencoder to perform dimensionality reduction processing on the updated representation vector to generate a target representation vector;

[0119] In order to further optimize the representation vector, a variational autoencoder (VAE) is applied to reduce the dimension of the updated representation vector. The variational autoencoder can automatically extract the key features in the representation vector and remove redundant information, thereby generating a more compact target representation vector. After the variational autoencoder reduces the dimension, the target representation vector is obtained. These target representation vectors not only retain the key features of the meter in the network, but also remove unnecessary redundant information, making subsequent analysis and evaluation more efficient and accurate.

[0120] The embodiments of the present invention ensure that the evaluation results are closer to the actual operating environment through graph neural network and topological structure features; the adaptive weight mechanism enables the representation vector to more accurately reflect the interaction and network effects between electricity meters, especially the changes in different time periods; the variational autoencoder removes redundant information and improves the efficiency and accuracy of subsequent analysis and evaluation.

[0121] In order to improve the prediction accuracy and reliability of the electric energy metering system, the dynamic changes of the load and its characteristics of being affected by multiple factors must be fully considered. Traditional prediction methods often find it difficult to fully capture these complex factors. Based on this, the present invention provides a specific embodiment, step 102, based on the true error range estimation, using the time series prediction model to predict the short-term load of the system, generate the target prediction result, and combine historical data of the same period, seasonal factors and reinforcement learning mechanism to optimize the prediction during the prediction process, specifically including the following steps:

[0122] Step 201: Based on the real error range estimation, use a time series prediction model to predict the short-term load of the system and generate a preliminary prediction result;

[0123] In this step, a long short-term memory network is used as the time series prediction model to predict the short-term load of the system. Specifically, based on the historical data within the past week, the electricity consumption in the next 24 hours is predicted. The preliminary prediction result of this step reflects the expected electricity consumption in each time period within the next 24 hours, but the influence of other factors has not been considered yet.

[0124] Step 202: Based on the preliminary prediction result and the historical data of the same period, adjust the preliminary prediction result to obtain an adjusted prediction result;

[0125] To improve the prediction accuracy, the preliminary prediction result is compared with the historical data of the same period and corresponding adjustments are made. By comparing the preliminary prediction result and the historical data of the same period, obvious differences can be found in some time periods. For example, between 0:00 and 1:00, the preliminary prediction results are 100 kWh and 98 kWh respectively, while the historical data of the same period are 105 kWh and 102 kWh respectively. Therefore, the prediction values of these time periods are appropriately adjusted to make them closer to the historical data. The adjusted prediction result more accurately reflects the actual load situation and reduces the prediction error.

[0126] Step 203: Based on the adjusted prediction result, introduce seasonal factors to generate a seasonally adjusted prediction result. Based on the seasonally adjusted prediction result, apply a reinforcement learning mechanism. By interacting with the environment, adjust the key parameters of the time series prediction model to obtain a target prediction result. The reinforcement learning mechanism includes a multi-agent reinforcement learning algorithm and a Q reinforcement learning algorithm;

[0127] In this step, the key parameters refer to the model architecture parameters, hyperparameters, time window parameters, etc. of the time series prediction model.

[0128] To further optimize the prediction results, seasonal factors are introduced to consider the impact of different seasons, months, or holidays on power consumption. For example, the peak period of air conditioner usage in summer may lead to a significant increase in power consumption, while the heating demand in winter may have a different impact. Suppose today is a working day in summer. Considering the high temperature, it is expected that the power consumption will peak in the afternoon. Therefore, based on the adjusted prediction results, seasonal factors are further introduced for adjustment to generate the seasonally adjusted prediction results. Next, a reinforcement learning mechanism is applied to dynamically adjust the key parameters of the time series prediction model by interacting with the environment to obtain a more accurate target prediction result. Specifically, multi-agent reinforcement learning algorithms and Q-reinforcement learning algorithms are used to optimize the model parameters. Assume that multiple electricity meters in the power grid act as a group of agents. Each electricity meter selects the optimal action (such as adjusting the load) based on its own and its neighbors' states and optimizes the overall prediction accuracy through group collaboration. For a single electricity meter, the Q-value table is continuously updated through the Q-reinforcement learning algorithm to select the best action (such as adjusting the prediction parameters) to maximize the long-term reward (such as minimizing the prediction error). For example, assume that the actual load of a certain electricity meter at a certain time period is much higher than the predicted value. Through the multi-agent reinforcement learning algorithm, this electricity meter and its neighboring electricity meters will adjust their own prediction parameters to make the next prediction more accurate. At the same time, through the Q-reinforcement learning algorithm, this electricity meter will select the optimal parameters according to the current state and action to further optimize the prediction result. The final obtained target prediction result not only considers historical data and seasonal factors but also dynamically adjusts the model parameters through the reinforcement learning mechanism, improving the accuracy and robustness of the prediction.

[0129] In the embodiments of the present invention, by introducing the estimation of the true error range and using the time series prediction model to generate the preliminary prediction results, the basic data for prediction is ensured to be more reliable; the prediction accuracy is further improved by combining with historical data of the same period; seasonal factors are introduced to generate the seasonally adjusted prediction results, making the prediction results better reflect the actual load conditions, especially the changes in different seasons and special periods; the reinforcement learning mechanism is applied to dynamically adjust the key parameters of the time series prediction model by interacting with the environment, improving the robustness and adaptability of the prediction. Especially the application of multi-agent reinforcement learning algorithms and Q-reinforcement learning algorithms enables the model to better cope with the complex and changing environment; through the accurate prediction of the short-term load of the system, it provides a scientific basis for power dispatching and load management, improves the support level for maintenance decision-making, helps reduce the failure rate, lower the maintenance cost, and ensure the efficient operation of the power system.

[0130] In order to improve the accuracy and adaptability of short-term load forecasting in the power energy metering system, it is necessary to fully consider the influence of seasonal variations and other dynamic factors. Traditional forecasting methods often struggle to cope with these complex changes. Based on this, the present invention provides a specific embodiment. In step 203, based on the adjusted forecasting result, seasonal factors are introduced to generate a seasonally adjusted forecasting result. Based on the seasonally adjusted forecasting result, a reinforcement learning mechanism is applied. By interacting with the environment, the key parameters of the time series forecasting model are adjusted to obtain the target forecasting result. The reinforcement learning mechanism includes a multi-agent reinforcement learning algorithm and a Q reinforcement learning algorithm, and specifically includes the following steps:

[0131] Step 211: Based on the adjusted forecasting result, introduce seasonal factors, and in combination with Fourier transform, identify and extract key periodic frequency components. The key periodic frequency components include daily cycle components, weekly cycle components, and special event cycle components;

[0132] In this step, the key periodic frequency components refer to the periodic features extracted from the time series data, including daily cycle components, weekly cycle components, and special event cycle components.

[0133] To further optimize the forecasting, it is necessary to introduce seasonal factors and in combination with Fourier transform to identify and extract key periodic frequency components. Considering the influence of different seasons, months, or holidays on power consumption, seasonal factors are added to the forecasting model. Through Fourier transform, the adjusted forecasting result is transformed from the time domain to the frequency domain to identify and extract key periodic frequency components. Specifically, the following periodic components are identified: daily cycle components, the periodic changes within 24 hours of each day, such as the peak power consumption during the day and the low valley at night; weekly cycle components, the periodic changes within 7 days of each week, such as the differences between weekdays and weekends; special event cycle components, the periodic changes brought about by specific dates or events (such as holidays, large-scale activities). These periodic components can help the system better analyze the patterns of power consumption, thereby improving the accuracy of forecasting.

[0134] Step 212: Based on the seasonal factors, the key periodic frequency components, and the historical load data, construct a long short-term memory network model, and use the long short-term memory network model to generate a seasonally adjusted forecasting result;

[0135] This step constructs a long short-term memory network model based on seasonal factors, key periodic frequency components, and historical load data. This model can handle and predict long-term dependencies and is very suitable for time series prediction. Using seasonal factors, key periodic frequency components, and historical load data as input features, the long short-term memory network model is trained. Using the long short-term memory network model, seasonally adjusted prediction results are generated. These results not only consider historical data and seasonal factors but also capture complex periodic and aperiodic components through the long short-term memory network model, improving the accuracy and reliability of the prediction.

[0136] Step 213: Apply a multi-agent reinforcement learning algorithm. Through the coordinated work of multiple agents, optimize the key parameters of the time series prediction model and the long short-term memory network model to generate a target prediction result. During the process of applying the multi-agent reinforcement learning algorithm for optimization, introduce the Q reinforcement learning algorithm to adjust the behavior strategy of the agents according to the environmental feedback information, where the environmental feedback information includes the real-time energy consumption data stream and the operation state change information of the power metering system.

[0137] In this step, the environmental feedback information refers to the actual feedback information from the environment, such as the real-time energy consumption data stream and the operation state change information of the power metering system.

[0138] Suppose multiple electric meters in the power grid form a group of agents. Each electric meter selects the optimal action (such as adjusting the load) based on its own and its neighbors' states and optimizes the overall prediction accuracy through group collaboration. For example, when the actual load of a certain electric meter in a certain time period is much higher than the predicted value, this electric meter and its neighbors will adjust their prediction parameters to make the next prediction more accurate. For a single electric meter, continuously update the Q-value table through the Q reinforcement learning algorithm and select the best action (such as adjusting the prediction parameters) to maximize the long-term reward (such as minimizing the prediction error). For example, suppose the actual load of a certain electric meter in a certain time period is much higher than the predicted value. Through the Q reinforcement learning algorithm, this electric meter will select the optimal parameters according to the current state and action to further optimize the prediction result. During the optimization process, introduce the environmental feedback information, including the real-time energy consumption data stream and the operation state change information of the power metering system. This information can help the agents understand the current environmental state and adjust their behavior strategies accordingly. For example, if the actual load of a certain electric meter suddenly increases in a certain time period, the agent will adjust the prediction parameters according to this change to adapt to the new load pattern. Through the synergistic effect of the multi-agent reinforcement learning algorithm and the Q reinforcement learning algorithm, dynamically adjust the key parameters of the time series prediction model and the long short-term memory network model to generate a more accurate target prediction result.

[0139] The embodiment of the present invention identifies and extracts key periodic frequency components through Fourier transform, ensuring that the prediction model can capture complex time series characteristics; the prediction results after seasonal adjustment significantly improve the accuracy and reliability of the prediction; the application of multi-agent reinforcement learning algorithm and Q reinforcement learning algorithm makes the prediction results more accurate and more robust.

[0140] Figure 2 A schematic diagram of a comprehensive evaluation system for the operation status of an electric energy metering system is provided in accordance with an embodiment of the present invention. Figure 2 As shown, the system includes:

[0141] A receiving module 21 is used to receive a real-time energy consumption data stream, detect the real-time energy consumption data stream using an adaptive threshold anomaly detection algorithm to obtain a filtered data stream, and perform uncertainty quantification processing on the filtered data stream using a Bayesian inference method to generate a true error range estimate corresponding to each data point;

[0142] A prediction module 22 is used to predict the short-term load of the system based on the true error range estimation using a time series prediction model to generate a target prediction result, and to optimize the prediction by combining historical data of the same period, seasonal factors and a reinforcement learning mechanism during the prediction process;

[0143] A generating module 23, used for comparing the target prediction result with the real-time energy consumption data stream to obtain a comparison result, and determining an instant deviation metric and a performance fluctuation situation based on the comparison result and a multi-dimensional state space model to generate a comprehensive performance fluctuation evaluation result;

[0144] The evaluation module 24 is used to construct a comprehensive evaluation index system based on the comprehensive performance fluctuation evaluation result, using an integrated learning algorithm and a fuzzy logic algorithm, and use the comprehensive evaluation index system to evaluate the health status of each meter in the electric energy metering system to obtain a comprehensive score corresponding to each meter.

[0145] Figure 2 The comprehensive evaluation system for the operation status of an electric energy metering system can be executed Figure 1 The implementation principle and technical effect of the comprehensive evaluation method for the operation status of an electric energy metering system described in the embodiment are not described in detail. The specific manner in which each module and unit performs operations in the comprehensive evaluation system for the operation status of an electric energy metering system in the above embodiment has been described in detail in the embodiment of the method, and will not be described in detail here.

[0146] In one possible design, Figure 2 The comprehensive evaluation system for the operation status of an electric energy metering system of the embodiment shown can be implemented as a computing device, such as Figure 3As shown, the computing device may include a storage component 31 and a processing component 32;

[0147] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0148] The processing component 32 is used to: receive a real-time energy consumption data stream, detect the real-time energy consumption data stream using an adaptive threshold anomaly detection algorithm to obtain a filtered data stream, and use a Bayesian inference method to quantify the uncertainty of the filtered data stream to generate a true error range estimate corresponding to each data point; based on the true error range estimate, use a time series prediction model to predict the system's short-term load to generate a target prediction result, and optimize the prediction by combining historical data of the same period, seasonal factors, and a reinforcement learning mechanism during the prediction process; compare the target prediction result with the real-time energy consumption data stream to obtain a comparison result, and determine the instant deviation metric and performance fluctuation based on the comparison result and a multidimensional state space model to generate a comprehensive performance fluctuation evaluation result; based on the comprehensive performance fluctuation evaluation result, use an integrated learning algorithm and a fuzzy logic algorithm to construct a comprehensive evaluation index system, and use the comprehensive evaluation index system to evaluate the health status of each meter in the electric energy metering system to obtain a comprehensive score corresponding to each meter.

[0149] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0150] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0151] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0152] The input / output interface provides an interface between the processing component and the peripheral interface module, and the above-mentioned peripheral interface module can be an output device, an input device, etc.

[0153] The communication component is configured to facilitate communication between the computing device and other devices in a wired or wireless manner, etc.

[0154] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. At this time, the computing device can refer to a cloud server, and the above-mentioned processing component, storage component, etc. can be basic server resources rented or purchased from a cloud computing platform.

[0155] The embodiment of the present invention also provides a computer storage medium storing a computer program, and when the computer program is executed by a computer, it can implement the above-mentioned Figure 1 comprehensive evaluation method for the operation state of an electric energy metering system shown in the embodiment.

[0156] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0157] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0158] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A comprehensive evaluation method for the operating status of an electric energy metering system, characterized in that: include: Receiving a real-time energy consumption data stream, detecting the real-time energy consumption data stream using an adaptive threshold anomaly detection algorithm to obtain a filtered data stream, and performing uncertainty quantification processing on the filtered data stream using a Bayesian inference method to generate a true error range estimate corresponding to each data point; Based on the true error range estimation, the time series forecasting model is used to forecast the short-term load of the system and generate the target forecast result. During the forecasting process, the historical data of the same period, seasonal factors and reinforcement learning mechanism are combined to optimize the forecast; Comparing the target prediction result with the real-time energy consumption data stream to obtain a comparison result, and determining an instant deviation metric and a performance fluctuation situation based on the comparison result and a multidimensional state space model to generate a comprehensive performance fluctuation evaluation result; Based on the comprehensive performance fluctuation evaluation results, an integrated learning algorithm and a fuzzy logic algorithm are used to construct a comprehensive evaluation index system. The comprehensive evaluation index system is used to evaluate the health status of each meter in the electric energy metering system to obtain a comprehensive score corresponding to each meter.

2. The method according to claim 1, characterized in that Comparing the target prediction result with the real-time energy consumption data stream to obtain a comparison result, and determining an instant deviation metric and a performance fluctuation situation based on the comparison result and the multidimensional state space model to generate a comprehensive performance fluctuation evaluation result, including: Using a dynamic time warping algorithm, the target prediction result and the time series data of the real-time energy consumption data stream are aligned to obtain aligned data. Based on the aligned data, the absolute error between the predicted value and the actual value at each time point is calculated to obtain error information. The sliding window technology is used to analyze the error information to obtain a comparison result. Based on the comparison result, a multidimensional state space model is constructed, and an implicit state in the multidimensional state space model is estimated using a Kalman filter to obtain a state estimation value, and based on the state estimation value, an instant deviation metric corresponding to different dimensional indicators is calculated, where the dimensional indicators include power consumption, voltage, and current; Analyze the changing trend of the instant deviation measurement in different time periods to generate performance fluctuations of the electric energy metering system corresponding to different time periods; The mutual relationship between the dimensional indicators is analyzed by using the grey correlation analysis method to obtain the analysis result of the mutual influence between the dimensional indicators; A comprehensive performance fluctuation evaluation result is generated by combining the instant deviation metric, the performance fluctuation situation, and the analysis results of the mutual influence between the dimensional indicators.

3. The method according to claim 1, characterized in that Based on the comprehensive performance fluctuation evaluation results, an integrated learning algorithm and a fuzzy logic algorithm are used to construct a comprehensive evaluation index system. The comprehensive evaluation index system is used to evaluate the health status of each meter in the electric energy metering system to obtain a comprehensive score corresponding to each meter, including: Based on multiple evaluation dimensions, the comprehensive performance fluctuation evaluation results are processed using an integrated learning algorithm and a deep neural network to obtain a preliminary score reflecting the operating status of the electric energy metering system, wherein the evaluation dimensions include energy efficiency, peak load management level, and stability; Introducing fuzzy logic algorithms and Bayesian networks to process nonlinear relationships and uncertain information in the preliminary scores, so as to adjust the preliminary scores and generate adjusted scores; Analyze the influence of the topological structure of the power grid by using a graph neural network to obtain structural influence results, and construct a comprehensive evaluation index system based on the structural influence results and the adjusted scores; Based on the comprehensive evaluation index system and the multi-objective optimization algorithm, a comprehensive evaluation is performed on the health status of each electric meter in the electric energy metering system to obtain a comprehensive score corresponding to each electric meter.

4. The method according to claim 3, characterized in that The influence of the grid topology is analyzed by using graph neural network to obtain the structural influence results. Based on the structural influence results and the adjusted scores, a comprehensive evaluation index system is constructed, including: Based on the adjusted scores, using a graph neural network, modeling the electricity meters in the power grid and the connection relationships between the electricity meters to generate a target representation vector that reflects the overall structural characteristics of the power grid; Based on the target representation vector, analyzing the influence of the power grid topology on the operating state of each electric meter, identifying key operating risk points, and determining the structural influence results based on the key operating risk points; The structural impact results and the adjusted scores are integrated to obtain integrated data, a multi-layer perceptron is used to process the integrated data to obtain processed data, and a comprehensive evaluation index system is constructed based on the processed data.

5. The method according to claim 4, characterized in that Based on the adjusted scores, the graph neural network is used to model the electricity meters in the power grid and the connection relationships between the electricity meters to generate a target representation vector reflecting the overall structural characteristics of the power grid, including: Based on the adjusted scores, the graph neural network is used to model the electricity meters in the power grid and the connection relationship between the electricity meters to generate a preliminary representation vector, and the topological structure characteristics of the power grid are combined in the modeling process to obtain the interaction and network effect between the electricity meters, wherein the topological structure characteristics include the connection mode and connection distance between network nodes, and the network nodes represent the electricity meters; Introducing an adaptive weight mechanism to dynamically adjust the weight of the connection relationship between the electric meters to update the preliminary representation vector to obtain an updated representation vector; A variational autoencoder is applied to perform dimensionality reduction processing on the updated representation vector to generate a target representation vector.

6. The method according to claim 1, characterized in that Based on the true error range estimation, the time series forecasting model is used to forecast the short-term load of the system and generate the target forecast result. In the forecasting process, historical data of the same period, seasonal factors and reinforcement learning mechanism are combined to optimize the forecast, including: Based on the true error range estimation, a time series forecasting model is used to forecast the short-term load of the system to generate a preliminary forecast result; Based on the preliminary forecast results and historical data for the same period, the preliminary forecast results are adjusted to obtain adjusted forecast results; Based on the adjusted prediction results, seasonal factors are introduced to generate seasonally adjusted prediction results. Based on the seasonally adjusted prediction results, a reinforcement learning mechanism is applied to adjust the key parameters of the time series prediction model through interaction with the environment to obtain the target prediction results. The reinforcement learning mechanism includes a multi-agent reinforcement learning algorithm and a Q reinforcement learning algorithm.

7. The method according to claim 6, characterized in that Based on the adjusted forecast results, seasonal factors are introduced to generate seasonally adjusted forecast results. Based on the seasonally adjusted forecast results, a reinforcement learning mechanism is applied to adjust key parameters of the time series forecasting model by interacting with the environment to obtain a target forecast result. The reinforcement learning mechanism includes a multi-agent reinforcement learning algorithm and a Q reinforcement learning algorithm, including: Based on the adjusted forecast results, seasonal factors are introduced, and Fourier transform is combined to identify and extract key periodic frequency components, wherein the key periodic frequency components include daily periodic components, weekly periodic components, and special event periodic components; Based on the seasonal factors, the key periodic frequency components and the historical load data, a long short-term memory network model is constructed, and the long short-term memory network model is used to generate seasonally adjusted forecast results; A multi-agent reinforcement learning algorithm is applied to optimize the key parameters of the time series prediction model and the long short-term memory network model through coordination between multiple agents to generate target prediction results. A Q reinforcement learning algorithm is introduced in the process of optimization using the multi-agent reinforcement learning algorithm to adjust the behavior strategy of the agent according to environmental feedback information, wherein the environmental feedback information includes the real-time energy consumption data stream and the operating status change information of the electric energy metering system.

8. A comprehensive evaluation system for the operation status of an electric energy metering system, characterized in that: include: A receiving module is used to receive a real-time energy consumption data stream, detect the real-time energy consumption data stream using an adaptive threshold anomaly detection algorithm to obtain a filtered data stream, and perform uncertainty quantification processing on the filtered data stream using a Bayesian inference method to generate a true error range estimate corresponding to each data point; A prediction module is used to predict the short-term load of the system based on the true error range estimation using a time series prediction model to generate a target prediction result, and to optimize the prediction by combining historical data of the same period, seasonal factors, and a reinforcement learning mechanism during the prediction process; A generation module, used for comparing the target prediction result with the real-time energy consumption data stream to obtain a comparison result, and determining an instant deviation metric and a performance fluctuation situation based on the comparison result and a multidimensional state space model to generate a comprehensive performance fluctuation evaluation result; The evaluation module is used to construct a comprehensive evaluation index system based on the comprehensive performance fluctuation evaluation result, using an integrated learning algorithm and a fuzzy logic algorithm, and use the comprehensive evaluation index system to evaluate the health status of each meter in the electric energy metering system to obtain a comprehensive score corresponding to each meter.

9. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a comprehensive evaluation method for the operating status of an electric energy metering system as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, a comprehensive evaluation method for the operating status of an electric energy metering system as claimed in any one of claims 1 to 7 is implemented.

Images