Data credibility dynamic evaluation method and system fusing multi-dimensional indexes

Abstract

This invention discloses a method and system for dynamic data credibility assessment that integrates multi-dimensional indicators, relating to the field of big data processing. The method includes: formulating multi-dimensional credibility assessment indicators based on business needs; acquiring raw data streams from power acquisition terminals and calculating real-time indicator values; based on an integrated assessment model, merging and calculating the real-time indicator values ​​and weights to output a comprehensive credibility score; triggering an early warning signal when the comprehensive credibility score is lower than a credibility threshold, and locating the specific indicator type and associated equipment causing the low credibility based on indicator contribution analysis; matching and executing a preset repair strategy based on the location results, and iteratively re-executing the assessment process after repair until the comprehensive credibility score meets the credibility threshold. This solves the problems of incomplete data dimension coverage and insufficient dynamic adaptability to massive multi-source data in existing power big data processing scenarios, which lead to one-sided credibility assessment results and data distortion.

Description

Technical Field

[0001] This invention relates to the field of big data processing, specifically to a method and system for dynamic evaluation of data credibility that integrates multi-dimensional indicators. Background Technology

[0002] In the electricity consumption information collection business, the reliability of electricity data such as user electricity consumption information, meter record data, and real-time electricity load data is an important foundation for supporting electricity consumption verification, abnormal electricity consumption analysis, metering dispute resolution, and electricity billing.

[0003] Traditional methods for assessing the credibility of electricity data often focus on a single dimension, failing to fully integrate various influencing indicators such as meter operating status, data collection environment characteristics, and communication link quality within a big data processing framework. Furthermore, they frequently rely on static thresholds, making them prone to biased assessments due to incomplete data coverage. Simultaneously, the lack of dynamic adaptation to real-time data streams hinders the dynamism and accuracy of credibility assessments. This makes it difficult to effectively identify data distortion caused by data collection equipment malfunctions or communication interference in big data scenarios, increasing the risk of incorrect electricity billing, user disputes and complaints, and losses in electricity consumption and revenue.

[0004] In summary, there is a need for an intelligent solution that can integrate multi-dimensional indicators, achieve dynamic and accurate assessment, and automatically locate and repair problems to support the application needs of big data processing and data quality management in the power sector. Summary of the Invention

[0005] This application provides a dynamic evaluation method and system for data credibility that integrates multi-dimensional indicators. It aims to solve the problems of single evaluation dimensions, poor data scenario adaptability, and data distortion in the existing technology of power data in massive, multi-source, and heterogeneous scenarios, and improve the intelligent level of data quality governance and credibility evaluation in the big data environment.

[0006] In view of the above problems, this application provides a dynamic evaluation method for data credibility that integrates multi-dimensional indicators.

[0007] Firstly, this application provides a method for dynamically evaluating data credibility by integrating multi-dimensional indicators, the method comprising: Based on the business needs of the power consumption information collection profession, a multi-dimensional credibility assessment index was developed. The raw data streams from the power acquisition terminal are collected in real time, and the real-time index values ​​of each raw data stream within the evaluation period are calculated based on the multi-dimensional reliability evaluation index. Based on the pre-trained fusion evaluation model, the real-time index values ​​and corresponding weights are fused and calculated to output a comprehensive credibility score for each original data stream. When any comprehensive credibility score falls below the credibility threshold, an early warning signal is triggered, and based on the contribution analysis of the indicators, the specific type of indicator and related equipment that caused the low credibility are located. Based on the location results, a preset repair strategy is automatically matched and executed, and the evaluation process is iterated again after the repair until the comprehensive credibility score meets the credibility threshold.

[0008] Secondly, this invention provides a dynamic data credibility evaluation system that integrates multi-dimensional indicators, including: The data acquisition module is used to develop multi-dimensional credibility assessment indicators based on the business needs of the power consumption information collection profession. The indicator extraction module is used to collect raw data streams from the power acquisition terminal in real time, and calculate the real-time indicator value of each raw data stream within the evaluation period based on the multi-dimensional credibility evaluation indicators. The fusion calculation module is used to perform fusion calculation on the real-time indicator values ​​and corresponding weights based on the pre-trained fusion evaluation model, and output the comprehensive credibility score of each original data stream. The anomaly detection module is used to trigger an early warning signal when any comprehensive credibility score is lower than the credibility threshold, and to locate the specific index type and associated device that caused the low credibility based on the index contribution analysis. The repair evaluation and verification module is used to automatically match and execute a preset repair strategy based on the location results, and to iterate and execute the evaluation process again after repair until the comprehensive credibility score meets the credibility threshold.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application employs multi-dimensional credibility assessment indicators to comprehensively cover key dimensions of the data, laying a solid foundation for subsequent accurate assessments. Secondly, it calculates real-time indicator values ​​using raw data streams and multi-dimensional credibility assessment indicators, standardizing and quantitatively analyzing the data source. Thirdly, a pre-trained fusion assessment model integrates real-time indicator values ​​and corresponding weights, outputting a comprehensive credibility score for each raw data stream. An intelligent model pre-trained based on historical power business data performs weighted fusion and anomaly pattern identification on multi-dimensional indicators, outputting dynamic quantitative scores. Simultaneously, it issues warnings when credibility thresholds are reached and, based on indicator contribution analysis, identifies specific indicator types and associated equipment leading to low credibility, precisely pinpointing problematic indicators and related equipment, thus shortening troubleshooting time. Finally, based on the identification results, it automatically matches and executes preset repair strategies, and iteratively re-executes the assessment process after repair, ensuring data quality and the accuracy of business processing, and achieving continuous and proactive improvement in data credibility. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating the dynamic data credibility assessment method that integrates multi-dimensional indicators as described in this application.

[0012] Figure 2 This is a schematic diagram of the structure of the data credibility dynamic evaluation system that integrates multi-dimensional indicators in this application.

[0013] In the attached diagram, the components represented by each number are as follows: Data acquisition module 11; indicator extraction module 12; fusion calculation module 13; anomaly detection module 14; repair evaluation and verification module 15. Detailed Implementation

[0014] This application provides a dynamic evaluation method for data credibility that integrates multi-dimensional indicators, aiming to solve the problems of single evaluation dimensions, insufficient adaptability to massive multi-source data scenarios, and data distortion faced by existing technologies in power data under big data processing scenarios.

[0015] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0016] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0017] The present invention will now be described in detail with reference to the accompanying drawings.

[0018] Example 1, as Figure 1 As shown, this application provides a method for dynamically evaluating data credibility by integrating multi-dimensional indicators, the method comprising: S10: Based on the business needs of the power consumption information collection profession, formulate multi-dimensional credibility assessment indicators; In this embodiment, a software and hardware system is used to automatically collect, store, and process electricity consumption data from smart meters, various collection terminals, and other devices through an electricity information collection platform; the business requirements of the electricity consumption information collection profession refer to the requirements for data quality; and the multi-dimensional credibility evaluation index refers to the design of quantifiable measurement standards based on different stages and attributes of the data lifecycle.

[0019] Through the power information acquisition platform, data from various business needs are collected simultaneously, and evaluation indicators regarding data integrity and data security are formulated based on these diverse business needs.

[0020] In step S10 of the method provided in this application embodiment, the multi-dimensional credibility evaluation index includes at least data integrity index, communication stability index and numerical reasonableness index, and initial weights and dynamic adjustment rules are configured for each index. The dynamic adjustment rules are to adjust the initial integrity weight, initial stability weight and initial reasonableness weight proportionally according to real-time business scenario information.

[0021] Multi-dimensional credibility assessment indicators should include at least three categories: data integrity, communication stability, and numerical reasonableness. The integrity dimension checks whether all frozen daily electricity consumption data was successfully collected. The stability dimension checks the signal quality and interruption status of the communication link used to collect the meter data on that day. The reasonableness dimension checks whether the total daily electricity consumption is within the user's historical consumption range and whether the time-of-use electricity consumption curve matches their lifestyle. Even if the daily electricity consumption data packet uploaded by the meter is complete, has high data integrity, and there are no interruption records in the communication process, indicating high data stability, but the electricity consumption value far exceeds the historical normal range, the data reasonableness is low, and the data can be considered unreliable. Therefore, it is necessary to establish and simultaneously monitor these three categories of indicators—data integrity, communication stability, and numerical reasonableness—to form an assessment framework for data credibility.

[0022] Based on the business needs of power consumption information collection, credibility assessment indicators were developed regarding data integrity, communication stability, and data security. Through the power information collection platform, various business-requirement data were collected simultaneously, and multi-dimensional credibility assessment indicators were established. Initial weights for each indicator were dynamically configured and adjusted according to business needs and scenarios.

[0023] For example, business requirements and scenarios may include normal power supply, equipment maintenance, and peak-valley load scenarios. Weights are configured based on the real-time business scenario type. In the maintenance scenario, the weight for numerical integrity is 0.4, the weight for communication stability is 0.4, and the weight for data rationality is 0.2. Then, based on the abnormal frequency ratio of each indicator in the current scenario, the initial coefficients are dynamically adjusted by ±10% to 30%. When real-time business scenario information indicates that there is currently a thunderstorm with strong communication interference, the dynamic adjustment rule is triggered. Because the weather warning is red for heavy rain, and the abnormal frequency ratio of the communication stability indicator is 30%, the weight of the communication stability indicator is increased by 30%. The weights of other indicators are dynamically reduced proportionally, ensuring that the sum of the weights of all dimensions' credibility assessment indicators is 1.

[0024] In this embodiment, a multi-dimensional credibility assessment index system is dynamically configured based on business needs and scenarios, clarifying the dimensions and initial rules of the assessment, so that the assessment conclusions are more in line with the actual business and provide unified benchmark data for the future.

[0025] S20: Collect raw data streams from the power acquisition terminal in real time, and calculate the real-time index value of each raw data stream within the evaluation period based on the multi-dimensional reliability evaluation index. In this embodiment of the application, the power acquisition terminal refers to the source device that generates power data; the raw data stream refers to a series of raw messages or records sent from the terminal that include timestamps, device IDs, load data and communication status metadata; the evaluation period refers to the time window for evaluation calculation; and the real-time index value refers to a specific value obtained after calculating the raw data stream within the specified evaluation period.

[0026] By accessing the raw data stream of the terminal and constructing a system based on the credibility assessment indicators, the credibility assessment indicators of each dimension of each data stream are configured, and the scores of each indicator are calculated independently for each data stream, by comparing the input data with the actual amount of data obtained.

[0027] Step S20 in the method provided in this application embodiment includes: During the evaluation period, each raw data stream is cached and preprocessed, wherein the preprocessing includes data cleaning and timestamp alignment; For data integrity metrics, the packet loss rate of the current raw data stream is statistically analyzed in real time, and the corresponding key field fill rate is calculated. The two are then combined into an integrity metric value according to a preset ratio. For communication stability indicators, the average signal strength of the communication link corresponding to the current raw data stream is monitored and calculated in real time, and the frequency of communication interruption is recorded cumulatively. Based on the weighted combination of the average signal strength and the interruption frequency, a stability indicator value is generated. For the numerical rationality index, the load value sequence in the current raw data stream is extracted in real time, the boundary compliance is checked according to the laws of power physics, and the deviation of the current value from the historical normal fluctuation range is calculated. The rationality index value is formed by combining the boundary compliance check result and the deviation. The integrity index value, the stability index value, and the rationality index value are output in a structured format according to a unified format and used as input to the pre-trained fusion evaluation model.

[0028] In this embodiment, each raw data stream is first cached and preprocessed within the evaluation period. The preprocessing includes data cleaning and timestamp alignment. Data cleaning refers to identifying and processing errors, redundancies, invalidities, or outliers in the raw data; timestamp alignment corrects all data points belonging to the same logical event within the same evaluation period to a unified time base.

[0029] Upon receiving the raw data stream for an evaluation period, all data packets within that period are buffered. Then, a preprocessing procedure is initiated. Data packets are cleaned and inspected: abnormal values ​​are marked as invalid or replaced; timestamp alignment calibrates the timestamps of all data points to a unified reference time grid, ensuring temporal consistency and comparability in subsequent calculations.

[0030] For abnormal current data packets, data cleaning removes them and fills the gaps using interpolation from previous and subsequent time points. In the uploaded timestamps, data that deviates from the standard interval is reallocated to the correct standard time slots through timestamp alignment.

[0031] Secondly, for data integrity indicators, the packet loss rate of the current raw data stream is statistically analyzed in real time, and the corresponding key field fill rate is calculated. The two are then combined into an integrity indicator value according to a preset ratio.

[0032] The packet loss rate refers to the ratio of the difference between the total number of packets expected to be received and the number of packets actually successfully received and parsed within the evaluation period, to the expected total number, measuring the overall loss situation at the data transmission level; the critical field fill rate refers to the ratio of the number of core business fields that are not empty or have invalid default values ​​in the successfully received packets to the total number of all critical fields, measuring the integrity at the data content level; the preset ratio refers to a weight pre-set based on business experience, used to balance the contribution of packet loss rate and critical field fill rate to the final integrity indicator value.

[0033] The preset sum of the packet loss rate and the key field fill rate is 1. If the preset ratio of the packet loss rate is higher, it means that more importance is attached to the overall acquisition of packets.

[0034] The system calculates the packet loss rate by comparing the expected number of packets to be returned with the actual number of valid response packets received during past evaluation periods, and also calculates the key field fill rate by counting the number of non-empty key fields and the total number of key fields. Specifically, the packet loss rate = number of lost packets / total number of packets; the key field fill rate = number of non-empty key fields / total number of key fields. Combining the preset ratios of the packet loss rate and the key field fill rate yields the integrity metric value: Integrity metric value = (1 - packet loss rate) × preset ratio of packet loss rate + key field fill rate × preset ratio of key field fill rate.

[0035] For example, if the preset ratio of packet loss rate to key field fill rate is 7:3, and the packet loss rate and key field fill rate are 0% and 90% respectively, then the integrity index value = (1-0%)×(7 / 10)+90%×(3 / 10)=0.97, indicating that the data is basically complete, but there are minor missing contents.

[0036] Furthermore, regarding communication stability indicators, the average signal strength of the communication link corresponding to the current raw data stream is monitored and calculated in real time. Simultaneously, the frequency of communication interruptions is recorded cumulatively. A stability indicator value is generated based on a weighted combination of the average signal strength and the interruption frequency. The average signal strength refers to the arithmetic mean of the signal strengths reported by the communication modules carrying the data stream transmission within the evaluation period, reflecting the communication link quality. The frequency of communication interruptions refers to the cumulative number of times the communication link experiences a complete disconnection event within the evaluation period.

[0037] The system monitors and calculates the signal strength of each record on the communication link within a certain time period in real time, and determines the average signal strength. It also records the frequency of communication interruptions. The average signal strength and interruption frequency are standardized, and corresponding weights are assigned based on the impact of standardization on data stability. The standardized values ​​of the average signal strength and interruption frequency are then weighted and fused together to obtain a stability index value: Stability Index Value = Standardized Average Signal Strength × Standardized Average Signal Strength Weight + Standardized Interruption Frequency × Standardized Interruption Frequency Weight.

[0038] Communication stability is assessed from both sustained quality and burst failure perspectives by using a weighted combination of the mean signal strength and the frequency of interruptions. The standardized formula for the interrupt frequency is: 1 - Interrupt frequency / Maximum allowable interrupt frequency.

[0039] For example, the weight of the signal strength mean after standardization is 0.7, the weight of the interruption frequency after standardization is 0.3, the interruption frequency is 2 times / evaluation cycle, the maximum allowable interruption frequency is preset to 3 times / evaluation cycle, the interruption frequency standardization = 1-2 / 3≈0.33, the signal strength mean is -85dBm, and the standardization is 0.8, then the stability index value = (0.8×0.7)+(0.33×0.3)=0.659.

[0040] Simultaneously, for the numerical rationality index, the load numerical sequence in the current raw data stream is extracted in real time. Boundary compliance verification is performed according to the laws of electrical physics, and the deviation of the current value from the historical normal fluctuation range is calculated. The rationality index value is formed by combining the boundary compliance verification results and the deviation degree. Among them, the load numerical sequence refers to the ordered set of actual business values ​​carried in the raw data stream after removing metadata such as protocol headers and timestamps; the boundary compliance verification is a rule check based on the laws of electrical physics and equipment nameplate parameters; the historical normal fluctuation range is the dynamic range that its value may appear under normal conditions, obtained through statistical learning based on the long-term historical data of the equipment; the deviation degree refers to the quantitative value of the difference between the current value or sequence and the corresponding historical normal fluctuation range.

[0041] Power value sequences are extracted in real time for boundary compliance verification. Historical power data from previous periods is then retrieved to calculate the degree of deviation. The final rationality index value is formed by combining the boundary compliance verification results and the degree of deviation. Appropriate weights are assigned based on the impact of the boundary compliance verification results and the degree of deviation on data rationality. The rationality index value is calculated by weighting the boundary compliance verification results and the degree of deviation. The final rationality index value is calculated as follows: Ratio = Boundary Compliance Verification Result × Boundary Compliance Verification Result Weight + (1 - Degree of Deviation) × Degree of Deviation Weight, where the degree of deviation = |Current Value - Historical Mean| / Historical Standard Deviation. The calculated degree of deviation is mapped to the range [0,1] for standardization. A value of 1 indicates compliance, and 0 indicates non-compliance.

[0042] For example, if the boundary compliance verification result has a weight of 0.6, the deviation weight is 0.4, the deviation obtained from the standardization process is 0.3, and the boundary compliance verification result is 1, then the reasonableness index value = 1 × 0.6 + (1 - 0.3) × 0.4 = 0.88, indicating that there may be a fundamental logical error or data tampering.

[0043] Finally, the integrity, stability, and rationality index values ​​are output in a structured format according to a unified standard and used as input to the pre-trained fusion evaluation model. The structured output refers to encapsulating and organizing the calculated index values ​​according to a unified format, transforming them from intermediate variables within the program into interface data that can be read by other modules.

[0044] The integrity, stability, and rationality index values ​​are encapsulated in a unified format and output as a structured data credibility feature package. This package serves as input to the subsequent pre-trained fusion evaluation model. Upon receiving this structured input, the model can immediately extract the feature values ​​and begin calculating the comprehensive credibility score.

[0045] In this embodiment, a multi-dimensional sub-indicator fusion calculation logic is designed to ensure data integrity, communication stability, and numerical rationality. This comprehensively covers the quality dimensions of the entire data collection process, avoiding the one-sidedness of traditional single-sub-indicator calculations. Furthermore, by using preset weighting ratios and standardized scoring rules, sub-indicators of different units and ranges are transformed into indicator values ​​of a unified dimension, ensuring the objectivity and comparability of indicator calculations. Finally, through a structured output in a unified format, seamless integration of multi-dimensional indicator data with the subsequent fusion evaluation model is achieved, providing standardized and reliable input data for comprehensive credibility scoring. Compared to the problem of fragmented and unreusable traditional indicator calculation results, this significantly improves data processing efficiency and evaluation accuracy. It ensures the accuracy of electricity billing data from the source of indicator calculation and effectively reduces the risk of data distortion caused by non-standard or incomplete indicator calculations.

[0046] S30: Based on the pre-trained fusion evaluation model, the real-time indicator values ​​and corresponding weights are fused and calculated to output the comprehensive credibility score of each original data stream; In this embodiment, the pre-trained fusion evaluation model is a machine learning model pre-trained using a large amount of historical data; the comprehensive credibility score is a comprehensive score output by the model after fusing and calculating multiple real-time indicator values, representing a quantitative judgment on the overall credibility of the original data stream.

[0047] The real-time values ​​and current weights of various metrics are calculated and input into a pre-trained fusion evaluation model. The model then fuses the real-time metric values ​​and the weights adjusted by dynamic adjustment rules, and outputs a weighted average of the comprehensive credibility score, which incorporates intelligent judgment based on historical experience and complex pattern recognition.

[0048] In step S30 of the method provided in this application embodiment, the pre-training step of the fusion evaluation model includes: Based on the raw data stream collected historically, and in accordance with the multi-dimensional credibility assessment indicators, the historical integrity indicator value, historical stability indicator value, and historical rationality indicator value are calculated and combined with the corresponding historical weight data to form a historical sample set. The historical weight data are the weight values ​​of each indicator obtained after adjustment by dynamic adjustment rules. For each sample in the historical sample set, a corresponding historical comprehensive credibility label is assigned based on historical data quality event records, forming a training label set; Based on machine learning, a network architecture for a fusion evaluation model is constructed, wherein the network architecture is configured to learn the fusion mapping relationship between historical indicator values ​​and historical weight data, and the output layer is the comprehensive credibility score; The fusion evaluation model is trained under supervision using the historical sample set and the training label set until the error converges to a preset range, thus completing the pre-training of the model.

[0049] In this embodiment of the application, firstly, based on the original data stream collected in history, the historical integrity index value, historical stability index value, and historical rationality index value are calculated according to the multi-dimensional credibility evaluation index. Combined with the corresponding historical weight data, they together form a historical sample set. The historical weight data are the weight values ​​of each index obtained after being adjusted by dynamic adjustment rules.

[0050] Historical raw data streams are raw data records continuously collected and stored from various power acquisition terminals over a relatively long period of time. Historical integrity / stability / reasonableness index values ​​are quantitative results of the indicators in each past evaluation period obtained by processing the historical raw data stream using index calculation methods. Historical weight data are the actual effective weight values ​​after adjusting the initial weights of the indicators according to the real-time business scenario information at that time in each historical evaluation period through dynamic adjustment rules. Historical sample sets are the inputs represented by samples in machine learning.

[0051] Based on the raw data streams collected historically, historical integrity, historical stability, and historical rationality indicators are calculated. Simultaneously, historical weight data is retrieved, and the weights are dynamically adjusted according to dynamic adjustment rules. The three indicator values ​​are then combined with the weight data to form a complete historical sample. Similarly, all historical raw data streams are processed to obtain a set of historical samples for training the model.

[0052] Secondly, for each sample in the historical sample set, a corresponding historical comprehensive credibility label is assigned based on historical data quality event records, forming a training label set. Here, historical data quality event records are logs of data quality issues recorded during the same historical period; the historical comprehensive credibility label is a label assigned to each historical sample by the historical data quality event records.

[0053] For each historical sample, query historical data quality event records, perform credibility analysis on all samples, and construct a training label set with the historical sample set based on the credibility results and corresponding credibility labels.

[0054] Furthermore, based on machine learning, a network architecture for a fusion evaluation model is constructed. This architecture is configured to learn the fusion mapping relationship between historical indicator values ​​and historical weight data, and the output is a comprehensive credibility score. The fusion mapping relationship refers to the non-linear interaction between the input features and the output target.

[0055] Finally, the fusion evaluation model is trained under supervision using historical sample sets and training label sets until the error converges to a preset range, thus completing the model's pre-training.

[0056] A backpropagation (BP) neural network is used to construct a fusion evaluation model to predict the adjustment direction of the output parameters. The BP neural network model is a feedforward neural network trained through backpropagation of errors and is commonly used to predict continuous values.

[0057] For example, the steps to construct a fusion evaluation model based on a BP neural network are as follows: Model construction: It mainly consists of an input layer, a hidden layer, and an output layer. The input layer receives a set of historical samples; the hidden layer performs non-linear transformations through activation functions; and the output layer outputs a comprehensive confidence score.

[0058] Finally, model training involves setting the initial learning rate and weights, assigning weights, calculating the error between the predicted and actual results using the mean squared error function, adjusting the weights, and iterating repeatedly until the error is minimized. Parameters are generated through forward propagation and updated through backpropagation. Performance is evaluated using a validation set after each training epoch to avoid overfitting. The model is considered successful when the MSE loss decreases by less than 1e over five consecutive training epochs. -5 When the MSE loss on the validation set stabilizes below 0.01, the model is considered converged, and the fusion evaluation model is obtained.

[0059] For example, the model is trained using a dataset containing 100,000 historical samples. Training stops when the preset convergence criterion is reached, resulting in a fusion evaluation model. The latest real-time metric values ​​[0.97, 0.659, 0.88] and the current weights [0.4, 0.4, 0.2] are input into the model, the mapping relationship is invoked, and the overall confidence score is output as 0.7, accurately reflecting its high-risk status without any human intervention.

[0060] In step S30 of the method provided in this application embodiment, the calculation step of the credibility threshold includes: Obtain historical statistical values ​​of the comprehensive credibility scores obtained by the fusion evaluation model for all raw data streams within a preset historical period, and use them as the initial threshold benchmark; Calculate in real time the status scores of all raw data streams on each credibility assessment indicator within the current assessment period; Based on the current weights of each indicator, the status scores are weighted and fused to generate a dynamic adjustment factor that characterizes the current overall data quality. Based on the dynamic adjustment factor, the initial threshold benchmark is proportionally adjusted to obtain the credibility threshold applied in the current evaluation period.

[0061] In this embodiment, the historical statistical values ​​of the comprehensive credibility score obtained after calculation by the fusion evaluation model of all original data streams within a preset historical period are first acquired, and these historical statistical values ​​of the comprehensive credibility score are used as the initial threshold benchmark. The initial threshold benchmark is the basic value for establishing the historical comprehensive credibility score, reflecting the general level of data credibility that should be achieved or maintained under the condition of no special external interference.

[0062] Select a preset historical period, and for all raw data streams in the preset historical period, recalculate the historical comprehensive credibility score corresponding to each data point using a pre-trained fusion evaluation model, and then perform statistical analysis. Use the 75th percentile of the comprehensive credibility score as the initial threshold benchmark, that is, in all comprehensive credibility data, make 75% of the data less than or equal to the initial threshold benchmark, and only 25% of the data greater than the initial threshold benchmark.

[0063] For example, if the 75th percentile is 0.6, then 0.6 is used as the initial threshold benchmark.

[0064] Secondly, the status scores of all raw data streams on each credibility assessment indicator are calculated in real time within the current assessment period. Here, the current assessment period refers to the time window in which the credibility assessment is being conducted; the status score is a macro-level value representing the overall health of the entire monitored system on a specific credibility dimension.

[0065] The overall status score is obtained by comparing the real-time values ​​of data integrity, communication stability and numerical rationality indicators with their respective historical normal ranges. The higher the score, the better the overall performance in that dimension.

[0066] Next, based on the current weights of each indicator, the state scores are weighted and fused to generate a dynamic adjustment factor that represents the overall data quality. Weighted fusion refers to the mathematical operation of combining multiple state scores into a single value according to their weight ratios; the dynamic adjustment factor is a scalar coefficient generated through weighted fusion.

[0067] Based on the weights adjusted according to the dynamic adjustment rules, the status scores of each credibility assessment indicator are weighted and fused to obtain a dynamic feature factor. The dynamic adjustment factor is calculated as follows: Dynamic Adjustment Factor = Completeness Indicator Status Score × Completeness Indicator Status Score Weight + Stability Indicator Status Score × Stability Indicator Status Score Weight + Reasonableness Indicator Status Score × Reasonableness Indicator Status Score Weight. A dynamic adjustment factor greater than 1 indicates that the current overall data quality environment is better than historical norms; a dynamic adjustment factor less than 1 indicates that it is worse than historical norms. The dynamic adjustment factor allows for the quantification of the current macroeconomic environment's quality level.

[0068] For example, according to the dynamic adjustment rules under severe weather conditions, the adjusted weights for stability and integrity are 0.5, integrity and rationality are 0.3, and rationality and rationality are 0.2. The status scores for integrity, stability, and rationality are 0.7, resulting in a dynamic adjustment factor of 0.7 × 0.3 + 0.6 × 0.5 + 0.7 × 0.2 = 0.65.

[0069] Finally, based on the dynamic adjustment factor, the initial threshold benchmark is proportionally adjusted to obtain the credibility threshold applied within the current evaluation period. The proportional adjustment involves multiplying the initial threshold benchmark by the dynamic adjustment factor to obtain a new threshold; the credibility threshold, after dynamic adjustment, is the final threshold used to determine the overall credibility score of all individual data streams within the current period.

[0070] Based on the dynamic adjustment factor, the baseline is proportionally adjusted. The current confidence threshold is calculated as: initial threshold baseline × dynamic adjustment factor. This current confidence threshold is used for adaptive adjustments. In harsh environments, a lower and more applicable threshold is used to determine the status of meters with non-communication-related anomalies, making alarms more targeted.

[0071] For example, if the initial threshold baseline is 0.6, then the current confidence threshold = 0.6 × 0.65 = 0.39. A more stringent standard is adopted in high-quality environments to encourage continuous optimization. In low-quality environments, such as when the meter's overall score is only 0.5 due to its own malfunction, it will not trigger an alarm at the old fixed threshold of 0.7, but will trigger an alarm at the new dynamic threshold of 0.39, thus ensuring safety.

[0072] In step S30 of the method provided in this application embodiment, the status score of all raw data streams on each credibility evaluation index is calculated in real time within the current evaluation period, including: For data integrity indicators, based on the integrity indicator values ​​of all original data streams within the current assessment period, the degree of conformity with the historical normal range of integrity is calculated to obtain an integrity status score; For communication stability indicators, based on the stability indicator values ​​of all raw data streams in the current evaluation period, the degree of conformity with the historical normal stability range is calculated to obtain a stability status score. For the numerical reasonableness index, based on the reasonableness index values ​​of all raw data streams in the current evaluation period, the degree of conformity with the historical reasonableness normal range is calculated to obtain the reasonableness status score.

[0073] In this embodiment, firstly, for the data integrity index, based on the integrity index values ​​of all original data streams within the current evaluation period, the degree of conformity with the historical normal range of integrity is calculated to obtain an integrity status score. The historical normal range of integrity is a numerical interval determined by statistical methods through analysis of the integrity index values ​​of all data streams over past periods, representing the boundary of normal fluctuations in integrity index values ​​under normal circumstances. The degree of conformity measures the degree of matching between the overall distribution of the current integrity index value and the historical normal range of integrity; it represents the proportion of the current value falling within the normal range.

[0074] The system integrates integrity metric values ​​from all raw data streams within the current evaluation period. It then calls upon pre-calculated historical integrity ranges to calculate the degree of conformity between the current value and the historical range. Finally, the percentage is directly mapped or converted into an integrity status score to represent the current score on the integrity dimension.

[0075] For example, the range derived from historical data is 0.75-0.95. However, due to intermittent fiber optic interruptions between concentrators, the data packet loss rate of the electricity meter surged, and the integrity index value dropped to 0.7.

[0076] Secondly, regarding communication stability indicators, based on the stability indicator values ​​of all raw data streams within the current evaluation period, the degree of conformity with the historical normal stability range is calculated to obtain a stability status score. Similarly, the stability indicator values ​​of all raw data streams within the current evaluation period are integrated, and then the pre-calculated historical normal integrity range is invoked to calculate the degree of conformity between the current value and the historical range. Finally, the percentage is directly mapped or converted into a stability status score to represent the current score on the integrity dimension.

[0077] In a regional interference scenario, severe weather caused the stability value of 3000 wireless links to drop to 0.55. At this point, a large number of low values ​​appear in the current value distribution, and the conformity with the historical normal range may drop to 50%. The obtained stability status score is only 0.6.

[0078] Finally, for the numerical reasonableness indicators, based on the reasonableness indicator values ​​of all raw data streams within the current assessment period, the degree of conformity between the current daily electricity consumption data and the historical reasonable and normal range is calculated to obtain a reasonableness status score. All current reasonableness indicator values ​​are then integrated. Next, the current reasonableness indicator values ​​are compared with the historical reasonable and normal range to calculate the conformity. The conformity measures how many users' data behavior is currently normal. Finally, a reasonableness status score is obtained based on the conformity.

[0079] For example, an anomaly appeared in the distribution of previous values ​​around 0.5, which is significantly inconsistent with the historical reasonableness range, resulting in a very low calculated compliance score. Therefore, the obtained reasonableness status score may drop from the normal 0.88 to 0.7.

[0080] In this embodiment, a fusion evaluation model is constructed and trained to understand nonlinear correlations and scenario differences, significantly improving the accuracy and intelligence level of credibility scoring and providing a reliable basis for all subsequent decisions. A dynamic adjustment mechanism is built to achieve real-time co-evolution of evaluation standards and the system environment, capturing systemic risks and enabling early warning standards to be dynamically adjusted according to the overall data quality environment, improving alarm accuracy while ensuring the practicality of alarms and the effective investment of operational resources.

[0081] S40: When any comprehensive credibility score is lower than the credibility threshold, an early warning signal is triggered, and based on the indicator contribution analysis, the specific indicator type and related equipment that caused the low credibility are located. In this embodiment, the early warning signal is an automatically generated alarm notification that may include information such as level, time, and content; the indicator contribution analysis is a diagnostic analysis technique used to quantify the contribution of each lower-level indicator in the comprehensive credibility score; the associated device is the source power acquisition terminal that is located and generates low-credibility data.

[0082] When the overall credibility score falls below the current credibility threshold, an alert is automatically triggered, and an indicator contribution analysis is performed. The algorithm calculates the contribution of rationality, stability, and integrity indicators in the score loss, thereby identifying the main problem indicator types and binding them to related devices to achieve precise problem location.

[0083] Step S40 in the method provided in this application embodiment includes: For the original data stream where the overall credibility score is lower than the credibility threshold, obtain the real-time values ​​of each indicator and their corresponding current weights; The gradient values ​​of the fusion evaluation model for each real-time indicator value are obtained through the backpropagation algorithm and used as the model gradient contribution. By comparing the real-time values ​​of various indicators with their corresponding historical normal ranges, the degree of abnormal deviation of each indicator is obtained as the indicator anomaly degree. The current weights of each indicator are used as the business importance coefficients; The contribution of each indicator is obtained by weighting and fusing the model gradient contribution, the indicator anomaly, and the business importance coefficient.

[0084] In this embodiment of the application, for the original data stream whose overall credibility score is lower than the credibility threshold, the real-time index values ​​and corresponding current weights of each indicator are first obtained.

[0085] Once an object with a comprehensive credibility score below the credibility threshold is identified, it is marked as an object requiring in-depth diagnosis: obtain all the input information used by the data stream when calculating the comprehensive score, namely the three real-time indicator values ​​and the current weights used in the calculation. By focusing on the specific problem, complete data is obtained during instance evaluation, avoiding errors that may be caused by rough inference, and providing accurate input for subsequent quantitative attribution from multiple dimensions.

[0086] Secondly, the gradient values ​​of the fusion evaluation model for each real-time indicator value are obtained through the backpropagation algorithm, serving as the model gradient contribution. The backpropagation algorithm is a core algorithm for training machine learning models such as neural networks; the gradient value refers to the partial derivative of the calculated comprehensive credibility score with respect to each input indicator value after the real-time indicator value is input into the fusion evaluation model; the model gradient contribution is the relative value obtained after standardizing the calculated gradient values, used to characterize the direct impact of each indicator on the current decrease in the comprehensive credibility score. A larger gradient value means a greater impact of the indicator change on the score, and therefore its contribution may be higher.

[0087] The system acquires the real-time values ​​of various metrics from the corresponding raw data streams and inputs them into a pre-trained fusion evaluation model to obtain the overall credibility score. Then, it initiates a backpropagation algorithm to perform reverse computation within the model's complex computational graph, obtaining the gradient value of the output score with respect to each input value. This gradient value is used as the original measure of the model's gradient contribution; the absolute value of the gradient represents the sensitivity or influence of that metric on the final score near the current value. If the gradient value of a certain real-time metric is significantly higher than the other two, it indicates that in the model's current decision function, a small change in that real-time metric will have a relatively large impact on the final score; that is, the impact of that real-time metric on the overall credibility score loss is greater than that of other real-time metric values.

[0088] For example, the real-time index value of meter A is [0.9, 0.7, 0.3], and the calculated gradient value is [0.1, 0.18, 0.4]. Among them, the gradient value of the rationality index is extremely large, while the gradient values ​​of other indices are small. This indicates that the model judges that the main reason for the low overall credibility score is the rationality of the data.

[0089] Next, by comparing the real-time values ​​of each indicator with their corresponding historical normal ranges, the degree of abnormal deviation of each indicator is obtained as the indicator anomaly degree.

[0090] Each real-time indicator value is compared with its respective historical normal range. The degree of abnormal deviation for each current value is calculated and standardized to obtain the anomaly degree for each indicator. The anomaly degree measures how much the current real-time indicator value deviates from its historical normal range. The anomaly degree for each indicator is calculated and normalized using min-max normalization; a larger value indicates a greater anomaly. The deviation degree is calculated by subtracting the mean of the historical normal range from the real-time indicator value and then dividing by the standard deviation of the historical indicator values, i.e., Deviation degree = (Real-time indicator value - Mean of historical normal range) / Standard deviation of historical indicator values.

[0091] For example, the real-time index value of meter A is [0.9, 0.7, 0.3], and the historical normal range is 0.75-0.95. Among them, 0.9 is between 0.75 and 0.95 with a deviation of 0. The deviations of 0.7 and 0.3 are (0.75-0.7) / ((0.95+0.75) / 2)≈0.06 and (0.75-0.3) / ((0.95+0.75) / 2)≈0.53, respectively.

[0092] Meanwhile, the current weights of each indicator are used as business importance coefficients. These business importance coefficients represent the significance of each indicator in the overall evaluation under the current business scenario and strategy. A higher weight indicates greater business emphasis on that dimension.

[0093] The weights of each indicator are directly used as the business importance coefficients for completeness, stability, and rationality indicators. These business importance coefficients determine the degree of importance the business places on rationality indicators; higher business importance corresponds to a greater contribution.

[0094] For example, the business importance coefficients for the stability, integrity, and rationality indicators of meter A are 0.5, 0.3, and 0.2, respectively. During the final weighted fusion calculation, the contribution of the integrity issue will be significantly amplified, thus being identified as the primary problem.

[0095] Finally, the model gradient contribution, indicator anomaly, and business importance coefficient are weighted and fused to obtain the indicator contribution of each indicator.

[0096] The model gradient contribution, indicator anomaly, and business importance coefficient are normalized separately and then weighted and fused according to their respective weights. The final indicator contribution = model gradient contribution × model gradient contribution weight + indicator anomaly × model gradient contribution weight + business importance coefficient × model gradient contribution weight.

[0097] For example, the model gradient contribution, indicator anomaly, and business importance coefficient are normalized to [0,1] and then weighted and fused using a 3:4:3 weighting. For the stability indicator: gradient contribution 0.1, indicator anomaly 0, business importance coefficient 0.5. Final contribution = 0.3×0.1 + 0.4×0 + 0.3×0.5 = 0.18. For the integrity indicator: gradient contribution 0.18, indicator anomaly 0.06, business importance coefficient 0.3. Final contribution = 0.3×0.18 + 0.4×0.06 + 0.3×0.3 = 0.168. For the rationality indicator: high gradient contribution 0.4, extremely high indicator anomaly 0.53, business importance coefficient 0.2. Final contribution = 0.3×0.4 + 0.4×0.53 + 0.3×0.2 = 0.392. After normalization, the contribution of each indicator is approximately: 5% for completeness, 40% for stability, and 55% for rationality.

[0098] Step S40 in the method provided in this application embodiment further includes: Based on the calculation results of the contribution of the indicators, the contribution of each indicator is ranked. The indicator with the highest contribution in the ranking is identified as the main problem indicator type that causes the overall credibility score to be lower than the credibility threshold; By parsing the device identification information contained in the original data stream, the main problem indicator types are associated with the power acquisition terminal that generated the current original data stream, and the associated devices are located. The output includes a location report containing the main problem indicator types and associated device identifiers.

[0099] In this embodiment, the contribution of each indicator is first ranked according to the calculation results. The indicators are arranged in descending order of value, visually demonstrating the relative importance of each indicator in causing the problem, providing a direct basis for further judgment.

[0100] For example, the contribution of indicators is sorted from largest to smallest, resulting in the following order: rationality indicator contribution (55%) > stability indicator contribution (40%) > integrity indicator contribution (5%).

[0101] Secondly, the indicator with the highest contribution in the ranking is identified as the main problematic indicator type causing the overall credibility score to fall below the credibility threshold. The indicator with the highest contribution in the ranking, i.e., the first-ranked indicator, is considered by the calculation model, data statistics, and business rules to be the dimension with the greatest responsibility among all possible reasons for the current low credibility score. The first-ranked indicator, the one with the highest contribution, is thus identified as the main problematic indicator type causing the current low score.

[0102] For example, for meter A, after contribution calculation and sorting, the contribution of the integrity index is 5%, the contribution of the stability index is 40%, and the contribution of the rationality index is 55%. The main problem index type is determined to be communication rationality.

[0103] Next, by parsing the device identification information contained in the raw data stream, the main problem indicator types are associated with the power acquisition terminal that generated the current raw data stream, and the associated devices are located. The device identification information is a field in the raw data stream that identifies the data source. It typically includes: meter asset number or concentrator number and port number, etc.

[0104] By parsing the raw data stream, key device identification information is obtained. Then, an association mapping is performed: the main problem indicator types are associated with the parsed device identifiers, binding the main problem indicator types to the physical entities of the data. After locating the associated devices, the specific physical device causing the problem is finally determined.

[0105] For example, the primary issue type of the data stream is determined to be communication legitimacy. This is achieved by parsing the raw data stream, extracting the meter asset number, and the accessed base station information. During the association mapping, communication legitimacy is mapped to the meter, and may also be mapped to a specific communication module and its current network environment. This facilitates the determination of remote module resets and greatly improves the targeted nature of the handling process.

[0106] Finally, a location report is output, containing the main problem indicator types and associated device identifiers. The main problem indicator types and associated device identifiers are summarized to output the location report. In addition, the report usually includes supplementary information such as timestamps, original overall credibility scores, and details of each contribution, forming a complete diagnostic profile.

[0107] For example, for meter 1, the output location report is as follows: main problem indicator type: communication rationality; associated device identifier: asset number, concentrator to which it belongs.

[0108] In this embodiment, by integrating three key dimensions—model gradient contribution, indicator anomaly, and business importance coefficient—multi-perspective attribution of low-reliability data is achieved, overcoming the one-sidedness of single-perspective attribution and providing credible quantitative evidence for subsequent accurate positioning. Furthermore, through a standardized sorting, judgment, association, and reporting output process, the contribution values ​​are transformed into specific positioning instructions, making the warning signals less directional. This significantly improves the efficiency and accuracy of operational response and enhances the processing capabilities of the entire data quality control system.

[0109] S50: Based on the positioning results, automatically match and execute the preset repair strategy, and iterate the evaluation process again after the repair until the comprehensive credibility score meets the credibility threshold.

[0110] In this embodiment, the repair strategy is an automated or semi-automated processing scheme pre-designed for different types of problems; automatic matching is to automatically select the most suitable response scheme from the strategy library based on the problem indicator type and associated device in the location report.

[0111] Based on the location results, a preset repair strategy is matched from the strategy library, and an instruction is sent to the main data collection station to execute supplementary data collection or instantaneous data reading. After acquiring new data, the process iterates again: calculating the indicator value of the new data stream → inputting it into the model for evaluation → obtaining a new comprehensive credibility score. If the comprehensive credibility score rises above the credibility threshold, the process ends, and the data is released for settlement; if it is still below the credibility threshold, a more advanced strategy may be matched, and iteration continues until the problem is resolved or the maximum number of iterations is reached.

[0112] Step S50 in the method provided in this application embodiment includes: For issues related to data integrity metrics, implement strategies such as data retransmission requests or missing data supplementation. To address issues with communication stability metrics, strategies such as communication link switching or communication parameter optimization are implemented. For issues related to numerical reasonableness indicators, strategies include implementing data correction or initiating manual review processes; The matching process for the repair strategy is as follows: based on the main problem indicator types and associated device identifiers contained in the location report, the corresponding repair strategy is retrieved from the preset repair strategy library and executed. After executing the repair strategy, update the repair status record of the corresponding power acquisition terminal and start the next round of evaluation cycle for the current power acquisition terminal.

[0113] In this embodiment, the first step is to address the issue of data integrity by implementing a strategy of data retransmission request or missing data supplementation. A data retransmission request refers to proactively re-issuing a data reading instruction to the power data acquisition terminal or its intermediate aggregation device that generated the data, requesting it to re-report the data at the specified time point. Missing data supplementation refers to the system adjusting the acquisition task plan to specifically collect previously missing data points in subsequent acquisition cycles.

[0114] When a location report indicates that a device's main problem is data integrity, it means that data was lost during transmission. A data retransmission request can be attempted: send a repeated command through the main data acquisition station, requesting immediate reporting of the power consumption data for the assessment period. If the retransmission is successful, the data will be completed. If the retransmission fails, the system may initiate a missing data replenishment strategy. This involves increasing the acquisition frequency over a subsequent period, or proactively retrieving historical cached data during the first window after communication is restored to fill in the missing data segments.

[0115] Secondly, regarding communication stability issues, strategies such as communication link switching or communication parameter optimization should be implemented. When the location report indicates that the main problem with a device is communication stability, it suggests that the issue lies in the data transmission path. A communication link switching strategy should be implemented, switching the communication mode remotely via command. If switching is not supported, communication parameter optimization strategies can be employed: remotely restarting the communication module, refreshing network registration, or fine-tuning radio frequency parameters to establish a stable connection.

[0116] Secondly, regarding issues with numerical reasonableness indicators, strategies include implementing data correction or initiating manual review processes. When a location report indicates that the main problem with a device is numerical reasonableness, it signifies a data issue. The solution involves checking if conditions exist for automatic data correction to generate reasonable data. If automatic correction is not available, a manual review process is initiated to acquire the data.

[0117] Meanwhile, the matching process for the repair strategy is as follows: based on the main problem indicator types and associated device identifiers contained in the location report, the corresponding repair strategy is retrieved from the preset repair strategy library and executed. Upon receiving the location report, the system searches the preset repair strategy library based on the main problem indicator types in the report and the associated device identifiers. Finally, the repair strategy is executed, and a power anomaly verification work order is created.

[0118] Finally, after executing the repair strategy, the repair status record of the corresponding power acquisition terminal is updated, and the next round of evaluation cycle for the current power acquisition terminal is initiated. After creating a manual review work order, the repair status record is updated immediately, and the next round of evaluation cycle is initiated. The new round of evaluation will calculate a new comprehensive credibility score. If the score rises above the credibility threshold, the loop is closed, and the warning can be lifted; if the score remains low, and the same problem is identified, it is recorded, and a higher-level strategy is triggered until the comprehensive credibility score meets the credibility threshold.

[0119] In this embodiment, a precise and hierarchical remediation strategy library is pre-set for different root causes, ensuring the targetedness and effectiveness of the remediation measures. By using the problem type and device identifier in the location report as input, the remediation strategy is automatically matched and executed, shortening the remediation response time. After remediation, the device status is immediately updated and the next round of evaluation is initiated, forming continuous iteration, significantly improving the reliability of the data supply chain, reducing electricity bill disputes and operational risks, and optimizing the allocation efficiency of operation and maintenance resources.

[0120] The embodiments of this application, through the above specific implementation methods, achieve the following technical effects: In this embodiment, a multi-dimensional credibility assessment index system is first dynamically configured based on business needs and scenarios, clarifying the dimensions and initial rules of the assessment, so that the assessment conclusions are more in line with the actual business and provide unified benchmark data for the future.

[0121] Secondly, by designing a multi-dimensional sub-indicator fusion calculation logic that integrates data integrity, communication stability, and numerical rationality, the system comprehensively covers the quality dimensions of the entire data collection process, avoiding the one-sidedness of traditional single sub-indicator calculations. Furthermore, through preset weighting ratios and standardized scoring rules, sub-indicators from different units and ranges are transformed into unified-dimensional indicator values, achieving standardized processing and feature fusion of multi-source heterogeneous data and ensuring the objectivity and comparability of indicator calculations. Finally, through a unified format of structured output, seamless integration of multi-dimensional indicator data with subsequent fusion evaluation models is achieved, providing standardized and reliable input data for comprehensive credibility scoring. Compared to the fragmented and unreusable results of traditional indicator calculations, this significantly improves data processing efficiency, data reuse capabilities, and evaluation accuracy in big data scenarios. It ensures the accuracy of electricity billing data from the source of indicator calculation, effectively reducing the risk of data distortion caused by non-standard or incomplete indicator calculations.

[0122] Furthermore, by constructing a fusion evaluation model and training a model capable of understanding nonlinear correlations and scenario differences, the accuracy and intelligence level of credibility scoring are significantly improved, providing a reliable basis for all subsequent decisions. A dynamic adjustment mechanism is constructed to achieve real-time co-evolution of evaluation standards and the system environment, capturing systemic risks and enabling early warning standards to be dynamically adjusted according to the overall data quality environment, improving alarm accuracy while ensuring the usability of alarms and the effective investment of operational resources.

[0123] Meanwhile, by integrating three key dimensions—model gradient contribution, indicator anomaly, and business importance coefficient—multi-perspective attribution of low-reliability data is achieved, overcoming the one-sidedness of single-perspective attribution and providing credible quantitative evidence for subsequent accurate positioning. Furthermore, through standardized sorting, judgment, correlation, and reporting output processes, contribution values ​​are transformed into specific positioning instructions, making early warning signals less directional. This significantly improves the efficiency and accuracy of operational response and enhances the processing capabilities of the entire big data quality control system.

[0124] Ultimately, a precise and hierarchical remediation strategy library was pre-set for different root causes, ensuring the targeted effectiveness of the remediation measures. By using the problem type and device identifier in the location report as input, the remediation strategy was automatically matched and executed, shortening the response time for remediation. After remediation, the device status was immediately updated and the next round of evaluation was initiated, forming a continuous iteration that significantly improved the reliability of the data supply chain, reduced electricity bill disputes and operational risks, and optimized the allocation efficiency of operation and maintenance resources.

[0125] In summary, compared to existing technologies, this application constructs a dynamic evaluation system that integrates multi-dimensional indicators such as data integrity, communication stability, and numerical rationality. It also integrates an intelligent fusion evaluation model pre-trained based on power business scenarios, achieving comprehensive and accurate quantitative scoring for the collection, processing, storage, analysis, and quality governance of massive heterogeneous power data. This overcomes the limitations of traditional single-dimensional static threshold judgments, not only capturing data anomaly patterns caused by equipment failures, communication interference, and other factors in real time, but also dynamically adjusting evaluation thresholds and indicator weights according to real-time business scenarios, significantly improving the adaptability and accuracy of the evaluation.

[0126] Meanwhile, this application establishes a complete closed-loop control mechanism from anomaly identification and precise location to automatic repair: through refined indicator contribution analysis, the source can be quickly traced to the specific problem indicator type and related equipment; subsequently, iterative review is conducted to ensure that the problem is effectively corrected. This ensures the quality of power data from the source, significantly reduces data distortion, and achieves continuous and intelligent improvement in the reliability of power information acquisition system data.

[0127] Example 2, as Figure 2As shown, based on the same inventive concept as the data credibility dynamic evaluation method integrating multi-dimensional indicators provided in Embodiment 1, this embodiment of the invention also provides a data credibility dynamic evaluation system integrating multi-dimensional indicators, including: Data acquisition module 11 is used to formulate multi-dimensional credibility evaluation indicators based on the business needs of the power consumption information acquisition profession. The indicator extraction module 12 is used to collect raw data streams from the power acquisition terminal in real time, and calculate the real-time indicator value of each raw data stream within the evaluation period based on the multi-dimensional credibility evaluation indicators. The fusion calculation module 13 is used to perform fusion calculation on the real-time index values ​​and corresponding weights based on the pre-trained fusion evaluation model, and output the comprehensive credibility score of each original data stream. The anomaly detection module 14 is used to trigger an early warning signal when any comprehensive credibility score is lower than the credibility threshold, and to locate the specific index type and associated device that caused the low credibility based on the index contribution analysis. The repair evaluation and verification module 15 is used to automatically match and execute a preset repair strategy based on the location result, and to re-iterate the evaluation process after repair until the comprehensive credibility score meets the credibility threshold.

[0128] In one embodiment, the data acquisition module 11 is used to: include at least data integrity indicators, communication stability indicators and numerical reasonableness indicators in the multi-dimensional credibility evaluation indicators, and configure initial weights and dynamic adjustment rules for each indicator, wherein the dynamic adjustment rules are to adjust the initial integrity weight, initial stability weight and initial reasonableness weight proportionally according to real-time business scenario information.

[0129] In one embodiment, the indicator extraction module 12 is used for: During the evaluation period, each raw data stream is cached and preprocessed, wherein the preprocessing includes data cleaning and timestamp alignment; For data integrity metrics, the packet loss rate of the current raw data stream is statistically analyzed in real time, and the corresponding key field fill rate is calculated. The two are then combined into an integrity metric value according to a preset ratio. For communication stability indicators, the average signal strength of the communication link corresponding to the current raw data stream is monitored and calculated in real time, and the frequency of communication interruption is recorded cumulatively. Based on the weighted combination of the average signal strength and the interruption frequency, a stability indicator value is generated. For the numerical rationality index, the load value sequence in the current raw data stream is extracted in real time, the boundary compliance is checked according to the laws of power physics, and the deviation of the current value from the historical normal fluctuation range is calculated. The rationality index value is formed by combining the boundary compliance check result and the deviation. The integrity index value, the stability index value, and the rationality index value are output in a structured format according to a unified format and used as input to the pre-trained fusion evaluation model.

[0130] In one embodiment, the fusion computing module 13 is used for: Based on the raw data stream collected historically, and in accordance with the multi-dimensional credibility assessment indicators, the historical integrity indicator value, historical stability indicator value, and historical rationality indicator value are calculated and combined with the corresponding historical weight data to form a historical sample set. The historical weight data are the weight values ​​of each indicator obtained after adjustment by dynamic adjustment rules. For each sample in the historical sample set, a corresponding historical comprehensive credibility label is assigned based on historical data quality event records, forming a training label set; Based on machine learning, a network architecture for a fusion evaluation model is constructed, wherein the network architecture is configured to learn the fusion mapping relationship between historical indicator values ​​and historical weight data, and the output layer is the comprehensive credibility score; The fusion evaluation model is trained under supervision using the historical sample set and the training label set until the error converges to a preset range, thus completing the pre-training of the model.

[0131] The calculation steps for the credibility threshold include: Obtain historical statistical values ​​of the comprehensive credibility scores obtained by the fusion evaluation model for all raw data streams within a preset historical period, and use them as the initial threshold benchmark; Calculate in real time the status scores of all raw data streams on each credibility assessment indicator within the current assessment period; Based on the current weights of each indicator, the status scores are weighted and fused to generate a dynamic adjustment factor that characterizes the current overall data quality. Based on the dynamic adjustment factor, the initial threshold benchmark is proportionally adjusted to obtain the credibility threshold applied in the current evaluation period.

[0132] The real-time calculation of the status scores of all raw data streams on each credibility assessment indicator within the current assessment period includes: For data integrity indicators, based on the integrity indicator values ​​of all original data streams within the current assessment period, the degree of conformity with the historical normal range of integrity is calculated to obtain an integrity status score; For communication stability indicators, based on the stability indicator values ​​of all raw data streams in the current evaluation period, the degree of conformity with the historical normal stability range is calculated to obtain a stability status score. For the numerical reasonableness index, based on the reasonableness index values ​​of all raw data streams in the current evaluation period, the degree of conformity with the historical reasonableness normal range is calculated to obtain the reasonableness status score.

[0133] In one embodiment, the anomaly detection module 14 is used for: For the original data stream where the overall credibility score is lower than the credibility threshold, obtain the real-time values ​​of each indicator and their corresponding current weights; The gradient values ​​of the fusion evaluation model for each real-time indicator value are obtained through the backpropagation algorithm and used as the model gradient contribution. By comparing the real-time values ​​of various indicators with their corresponding historical normal ranges, the degree of abnormal deviation of each indicator is obtained as the indicator anomaly degree. The current weights of each indicator are used as the business importance coefficients; The contribution of each indicator is obtained by weighting and fusing the model gradient contribution, the indicator anomaly, and the business importance coefficient.

[0134] The step of identifying the specific indicator types and associated devices that lead to low credibility based on indicator contribution analysis includes: Based on the calculation results of the contribution of the indicators, the contribution of each indicator is ranked. The indicator with the highest contribution in the ranking is identified as the main problem indicator type that causes the overall credibility score to be lower than the credibility threshold; By parsing the device identification information contained in the original data stream, the main problem indicator types are associated with the power acquisition terminal that generated the current original data stream, and the associated devices are located. The output includes a location report containing the main problem indicator types and associated device identifiers.

[0135] In one embodiment, the repair evaluation and verification module 15 is used for: For issues related to data integrity metrics, implement strategies such as data retransmission requests or missing data supplementation. To address issues with communication stability metrics, strategies such as communication link switching or communication parameter optimization are implemented. For issues related to numerical reasonableness indicators, strategies include implementing data correction or initiating manual review processes; The matching process for the repair strategy is as follows: based on the main problem indicator types and associated device identifiers contained in the location report, the corresponding repair strategy is retrieved from the preset repair strategy library and executed. After executing the repair strategy, update the repair status record of the corresponding power acquisition terminal and start the next round of evaluation cycle for the current power acquisition terminal.

[0136] The embodiments of this application, through the above specific implementation methods, achieve the following technical effects: In this embodiment, firstly, the data acquisition module 11 dynamically configures a multi-dimensional credibility evaluation index system according to business needs and scenarios, clarifies the evaluation dimensions and initial rules, so that the evaluation conclusions are more in line with the actual business and provide unified benchmark data for the future.

[0137] Secondly, through the indicator extraction module 12, a multi-dimensional sub-indicator fusion calculation logic was designed to ensure data integrity, communication stability, and numerical rationality. This comprehensively covers the quality dimensions of the entire data collection process, avoiding the one-sidedness of traditional single sub-indicator calculations. Furthermore, by using preset weighting ratios and standardized scoring rules, sub-indicators of different units and ranges are transformed into indicator values ​​of a unified dimension, ensuring the objectivity and comparability of indicator calculations. Finally, through a structured output in a unified format, seamless integration of multi-dimensional indicator data with the subsequent fusion evaluation model is achieved, providing standardized and reliable input data for comprehensive credibility scoring. Compared to the problem of fragmented and unreusable traditional indicator calculation results, this significantly improves data processing efficiency and evaluation accuracy. It ensures the accuracy of electricity billing data from the source of indicator calculation and effectively reduces the risk of data distortion caused by non-standard or incomplete indicator calculations.

[0138] Furthermore, through the fusion computing module 13, a fusion evaluation model is constructed and trained to understand nonlinear correlations and scenario differences, significantly improving the accuracy and intelligence level of credibility scoring and providing a reliable basis for all subsequent decisions. A dynamic adjustment mechanism is constructed to achieve real-time co-evolution of evaluation standards and the system environment, capturing systemic risks and enabling early warning standards to be dynamically adjusted according to the overall data quality environment, improving alarm accuracy while ensuring the practicality of alarms and the effective investment of operational resources.

[0139] Meanwhile, through the anomaly detection module 14, three key dimensions—model gradient contribution, indicator anomaly, and business importance coefficient—are integrated to achieve multi-perspective attribution of low-reliability data. This overcomes the one-sidedness of single-perspective attribution and provides reliable quantitative evidence for subsequent accurate positioning. Furthermore, through standardized sorting, judgment, correlation, and reporting output processes, the contribution values ​​are transformed into specific positioning instructions, making the early warning signals less directional. This significantly improves the efficiency and accuracy of operational response and enhances the processing capabilities of the entire data quality control system.

[0140] Finally, through the repair assessment and verification module 15, a precise and hierarchical repair strategy library was preset for different root causes, ensuring the pertinence and effectiveness of the handling measures. By using the problem type and device identifier in the location report as input, the repair strategy was automatically matched and executed, shortening the response time for repair implementation. After repair, the device status was updated immediately and the next round of assessment was initiated, forming a continuous iteration, significantly improving the reliability of the data supply chain, reducing electricity bill disputes and operational risks, and optimizing the allocation efficiency of operation and maintenance resources.

[0141] In summary, compared to existing technologies, this application constructs a dynamic evaluation system that integrates multi-dimensional indicators such as data integrity, communication stability, and numerical rationality. It also integrates an intelligent fusion evaluation model pre-trained based on power business scenarios, achieving comprehensive and accurate quantitative scoring of the credibility of massive heterogeneous power data. This overcomes the limitations of traditional single-dimensional static threshold judgments, not only capturing data anomaly patterns caused by equipment failures, communication interference, and other factors in real time, but also dynamically adjusting evaluation thresholds and indicator weights according to real-time business scenarios, significantly improving the adaptability and accuracy of the evaluation.

[0142] Meanwhile, this application establishes a complete closed-loop control mechanism from anomaly identification and precise location to automatic repair: through refined indicator contribution analysis, the source can be quickly traced to the specific problem indicator type and related equipment; subsequently, iterative review is conducted to ensure that the problem is effectively corrected. This ensures the quality of power data from the source, significantly reduces data distortion, and achieves continuous and intelligent improvement in the reliability of power information acquisition system data.

[0143] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0144] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0145] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for dynamically evaluating data credibility by fusing multi-dimensional indicators, characterized in that, The method comprises: According to the business needs of the power electricity information collection profession, multi-dimensional credibility evaluation indexes are formulated; Real-time acquisition of raw data streams from power collection terminals, and calculation of real-time index values of each raw data stream in the evaluation period according to the multi-dimensional credibility evaluation indexes; Based on the pre-trained fusion evaluation model, the real-time index values and the corresponding weights are fused and calculated to output the comprehensive credibility score of each raw data stream; When any comprehensive credibility score is lower than the credibility threshold, a warning signal is triggered, and based on the index contribution degree analysis, the specific index type and associated equipment causing low credibility are located; According to the positioning result, the pre-set repair strategy is automatically matched and executed, and after repair, the evaluation process is re-iterated until the comprehensive credibility score meets the credibility threshold. 2.The method of claim 1, wherein, The multi-dimensional credibility evaluation indexes at least include data integrity indexes, communication stability indexes and numerical rationality indexes, and initial weights and dynamic adjustment rules are configured for each index, wherein the dynamic adjustment rule is to proportionally adjust the initial complete weight, the initial stable weight and the initial reasonable weight according to real-time business scene information. 3.The method of claim 1, wherein, Real-time acquisition of raw data streams from power collection terminals, and calculation of real-time index values of each raw data stream in the evaluation period according to the multi-dimensional credibility evaluation indexes, including: In the evaluation period, each raw data stream is cached and preprocessed, wherein the preprocessing includes data cleaning and timestamp alignment; For the data integrity index, the data packet loss rate of the current raw data stream is calculated in real time, and the corresponding key field filling rate is calculated, and the two are combined into an integrity index value according to a preset proportion; For the communication stability index, the average signal strength of the current raw data stream corresponding to the communication link is monitored and calculated in real time, and the frequency of communication interruption is recorded, and based on the weighted combination of the average signal strength and the interruption frequency, a stability index value is generated; For the numerical rationality index, the load value sequence in the current raw data stream is extracted in real time, the boundary compliance is checked according to the law of electricity physics, and the deviation degree of the current value relative to the historical normal fluctuation interval is calculated, and the boundary compliance check result and the deviation degree are combined to form a rationality index value; The integrity index value, the stability index value and the rationality index value are structured and output in a unified format as the input of the pre-trained fusion evaluation model.

4. The method of claim 1, wherein the method further comprises: The pre-training step of the fusion evaluation model comprises: According to the multi-dimensional credibility evaluation indexes, historical integrity index values, historical stability index values and historical rationality index values are calculated based on the historical raw data streams, and combined with the corresponding historical weight data to form a historical sample set, wherein the historical weight data is the index weight value obtained after dynamic adjustment rule adjustment; For each sample in the historical sample set, a corresponding historical comprehensive credibility label is given based on the historical data quality event record to form a training label set; Based on machine learning, a network architecture for a fusion evaluation model is constructed, wherein the network architecture is configured to learn the fusion mapping relationship between historical indicator values ​​and historical weight data, and the output layer is the comprehensive credibility score; The fusion evaluation model is trained under supervision using the historical sample set and the training label set until the error converges to a preset range, thus completing the pre-training of the model.

5. The method of claim 1, wherein the method further comprises: The steps for calculating the credibility threshold include: Obtain historical statistical values ​​of the comprehensive credibility scores obtained by the fusion evaluation model for all raw data streams within a preset historical period, and use them as the initial threshold benchmark; Calculate in real time the status scores of all raw data streams on each credibility assessment indicator within the current assessment period; Based on the current weights of each indicator, the status scores are weighted and fused to generate a dynamic adjustment factor that characterizes the current overall data quality. Based on the dynamic adjustment factor, the initial threshold benchmark is proportionally adjusted to obtain the credibility threshold applied in the current evaluation period.

6. The method of claim 5, wherein the method further comprises: Calculate in real time the status scores of all raw data streams on each credibility assessment metric within the current assessment period, including: For data integrity indicators, based on the integrity indicator values ​​of all original data streams within the current assessment period, the degree of conformity with the historical normal range of integrity is calculated to obtain an integrity status score; For communication stability indicators, based on the stability indicator values ​​of all raw data streams in the current evaluation period, the degree of conformity with the historical normal stability range is calculated to obtain a stability status score. For the numerical reasonableness index, based on the reasonableness index values ​​of all raw data streams in the current evaluation period, the degree of conformity with the historical reasonableness normal range is calculated to obtain the reasonableness status score.

7. The method of claim 1, wherein the method further comprises: The calculation steps for the contribution of the indicator include: For the original data stream where the overall credibility score is lower than the credibility threshold, obtain the real-time values ​​of each indicator and their corresponding current weights; The gradient values ​​of the fusion evaluation model for each real-time indicator value are obtained through the backpropagation algorithm and used as the model gradient contribution. By comparing the real-time values ​​of various indicators with their corresponding historical normal ranges, the degree of abnormal deviation of each indicator is obtained as the indicator anomaly degree. The current weights of each indicator are used as the business importance coefficients; The contribution of each indicator is obtained by weighting and fusing the model gradient contribution, the indicator anomaly, and the business importance coefficient.

8. The method of claim 1, wherein the method further comprises: Based on the contribution analysis of indicators, the specific types of indicators and related devices that lead to low credibility were identified, including: Based on the calculation results of the contribution of the indicators, the contribution of each indicator is ranked. The indicator with the highest contribution in the ranking is identified as the main problem indicator type that causes the overall credibility score to be lower than the credibility threshold; By parsing the device identification information contained in the original data stream, the main problem indicator types are associated with the power acquisition terminal that generated the current original data stream, and the associated devices are located. The output includes a location report containing the main problem indicator types and associated device identifiers.

9. The method of claim 1, wherein the method further comprises: The repair strategy includes at least the following: For issues related to data integrity metrics, implement strategies such as data retransmission requests or missing data supplementation. To address issues with communication stability metrics, strategies such as communication link switching or communication parameter optimization are implemented. For issues related to numerical reasonableness indicators, strategies include implementing data correction or initiating manual review processes; The matching process for the repair strategy is as follows: based on the main problem indicator types and associated device identifiers contained in the location report, the corresponding repair strategy is retrieved from the preset repair strategy library and executed. After executing the repair strategy, update the repair status record of the corresponding power acquisition terminal and start the next round of evaluation cycle for the current power acquisition terminal.

10. A data credibility dynamic evaluation system fusing multi-dimensional indexes, characterized in that, The system for implementing the dynamic data credibility assessment method integrating multi-dimensional indicators as described in any one of claims 1-9, the system comprising: The data acquisition module is used to develop multi-dimensional credibility assessment indicators based on the business needs of electricity information collection professionals. The indicator extraction module is used to collect raw data streams from the power acquisition terminal in real time, and calculate the real-time indicator value of each raw data stream within the evaluation period based on the multi-dimensional credibility evaluation indicators. The fusion calculation module is used to perform fusion calculation on the real-time indicator values ​​and corresponding weights based on the pre-trained fusion evaluation model, and output the comprehensive credibility score of each original data stream. The anomaly detection module is used to trigger an early warning signal when any comprehensive credibility score is lower than the credibility threshold, and to locate the specific index type and related device that caused the low credibility based on the index contribution analysis. The repair evaluation and verification module is used to automatically match and execute a preset repair strategy based on the location results, and to iteratively execute the evaluation process again after repair until the comprehensive credibility score meets the credibility threshold.

Images