Metering equipment performance prediction method and device, equipment and medium

Abstract

The invention discloses a metering equipment performance prediction method and device, equipment and a medium. The method comprises the following steps: acquiring a multi-dimensional data set of a target metering equipment set; wherein the multi-dimensional data set comprises equipment attribute data, operation state data, historical fault data and influence factor data; generating a fault association rule set based on the multi-dimensional data set through a preset association rule mining algorithm; performing clustering analysis on the target metering equipment set through a preset clustering algorithm to generate a fault risk grade index; and combining and dividing the target metering equipment set through the fault association rule set and the fault risk level index, and performing performance prediction on each metering equipment subset through a preset performance prediction model to obtain a performance prediction result. According to the method, the problem of analysis data heterogeneity caused by equipment fault inducement diversity and state difference can be solved, and the accuracy of the performance prediction result of the metering equipment is improved.

Description

Technical Field

[0001] This invention relates to the field of equipment performance prediction technology, and in particular to a method, apparatus, equipment and medium for predicting the performance of metering equipment. Background Technology

[0002] With the digital transformation of the energy industry, the number and types of metering assets (such as electricity meters, transformers, and data acquisition terminals) continue to grow, and their stable operation directly affects the fairness of electricity trading and the reliability of energy supply.

[0003] Existing technologies for equipment performance prediction have limitations in their approaches and methods. Common practices include setting static alarm thresholds based on historical fault data or performing simple time-series extrapolation based on the historical operating curves of a single device. More advanced methods may establish dedicated physical or statistical models for specific models or scenarios. However, these methods are essentially targeted predictions for individual devices or highly homogeneous groups of devices. Their model parameters and rules heavily rely on specific data or expert experience, resulting in weak generalization capabilities. Because different devices vary in model, installation environment, operating load, and usage stage, the combinations of causes leading to faults and performance degradation are complex and diverse. This "device-model" or "scenario-threshold" approach makes it difficult to conduct unified and efficient performance prediction and risk management for large-scale, multi-type metering equipment clusters at different life stages. This leads to high deployment and maintenance costs for prediction systems and makes it difficult to adapt to the dynamic changes in assets. Summary of the Invention

[0004] This invention provides a method, apparatus, device, and medium for predicting the performance of metering equipment. It can solve the problem of heterogeneity in analytical data caused by the diversity of causes and state differences of equipment failure by classifying equipment combinations through association rules and clustering results, thereby improving the accuracy of the performance prediction results of metering equipment.

[0005] In a first aspect, embodiments of the present invention provide a method for predicting the performance of metering equipment, comprising: Obtain a multidimensional dataset of the target metering equipment set; wherein the multidimensional dataset includes equipment attribute data, operating status data, historical fault data, and influencing factor data; the influencing factor data includes environmental factor data and service life data; Using a pre-defined association rule mining algorithm, the association between equipment failures and pre-defined key influencing factors is mined based on the multidimensional dataset, generating a set of failure association rules; wherein, the key influencing factors include environmental factors and service life; By using a preset clustering algorithm, cluster analysis is performed on the target metering equipment set based on the fault association rule set and the multidimensional dataset to generate a fault risk level index. The target metering equipment set is divided into several metering equipment subsets by combining the fault association rule set and fault risk level index. Then, the performance of each metering equipment subset is predicted by a preset performance prediction model to obtain the performance prediction results of each metering equipment subset.

[0006] This invention provides a comprehensive and core data foundation for all subsequent analyses, covering key dimensions throughout the entire equipment lifecycle (basic information, operating status, past faults, and external causes), avoiding analytical biases caused by incomplete data and ensuring sufficient data support for subsequent steps. By mining fault association rules, it clarifies the core causes of faults, providing targeted correlation basis for subsequent risk level classification and performance prediction, solving the problem of traditional methods ignoring the potential correlation between faults and multiple factors. Through clustering algorithms, it groups equipment according to the similarity of fault risks, achieving differentiated distinction of equipment risks and avoiding "one-size-fits-all" management, laying the foundation for accurate prediction by subsequent grouping. By subdividing metering equipment according to association rules and risk levels, it avoids mutual interference between equipment data with different risks and different fault causes, improving the targeting and accuracy of predictions (the performance degradation patterns of equipment in the same subset are more consistent). Compared with existing technologies, this application can solve the problem of heterogeneity in analytical data caused by the diversity of equipment fault causes and state differences through equipment combination classification by association rules and clustering results, improving the accuracy of metering equipment performance prediction results.

[0007] Furthermore, before obtaining the multidimensional dataset of the target set of metering devices, the process also includes preprocessing the raw dataset to obtain the multidimensional dataset. Specifically, the pre-acquired raw dataset is preprocessed to obtain a multidimensional dataset, as follows: Based on the preset interval division threshold, the pre-acquired original influencing factor data is discretized to obtain the first influencing factor data; Based on the preset fault type, the pre-acquired original historical fault data is discretized to obtain the first historical fault data; Using the mutual information method, based on the first historical fault data and the first influencing factor data, the mutual information value between the fault type and each influencing factor is calculated, and the first influencing factor data is filtered according to the mutual information value to obtain the second influencing factor data; By integrating the first historical fault data, the second influencing factor data, the pre-acquired equipment attribute data, and the operating status data, a multidimensional dataset is obtained.

[0008] This invention discretizes influencing factor data to adapt to the input requirements of association rule mining algorithms (which excel at handling discrete data). Continuous data (such as temperature and service life) is transformed into quantifiable intervals, facilitating subsequent mining of the correlation between multiple factors and faults. By discretizing historical fault data according to fault type, a unified standard for fault type representation is established, making fault types clearer and facilitating the calculation of the correlation between faults and influencing factors. This provides a clear analytical object for feature selection and association mining. Furthermore, by calculating the mutual information value between fault types and each influencing factor, highly correlated factors are screened, redundant features are eliminated, and key factors with high mutual information values ​​are retained. This reduces data volume, improves the efficiency of subsequent algorithm operation, and avoids redundant data interfering with the analysis results.

[0009] Furthermore, the step of using a preset association rule mining algorithm to mine the association between equipment failures and preset key influencing factors based on the multidimensional dataset, and generating a set of failure association rules, specifically involves: Based on the multidimensional dataset, several candidate first-order itemsets are generated, and the support of each candidate first-order itemset is calculated. The candidate first-order itemsets are then filtered using the support to obtain frequent first-order itemsets. The frequent first-order itemsets are concatenated to generate several candidate second-order itemsets, and the support of each candidate second-order itemset is calculated. The candidate second-order itemsets are then filtered based on the support to obtain frequent second-order itemsets. The splicing operation and support calculation are continued layer by layer until the preset conditions are met, so as to generate and filter out several frequent itemsets of different orders layer by layer. Fault association rules satisfying a preset rule structure are extracted from frequent itemsets of each order to obtain an initial fault association rule set. The confidence level of each fault association rule in the initial fault association rule set is calculated. The initial fault association rule set is then filtered using the confidence level to obtain a fault association rule set. The rule structure is a unidirectional structure from antecedent to consequent. The antecedent is composed of a combination of discretized intervals of environmental factors and service life. The consequent is composed of a single fault type.

[0010] This invention employs a layer-by-layer concatenation and filtering process to obtain itemsets of various orders, progressively uncovering the correlation between "single factor → failure" and "multiple factor combinations → failure." Higher-order itemsets reflect the complex causes of failures (e.g., failures caused by a combination of age, temperature, and voltage fluctuations), avoiding the limitations of solely focusing on single-factor correlations. Support is used to filter itemsets of each order, ensuring that frequent itemsets have statistical significance (e.g., support ≥ 5% indicates frequent occurrence of the itemset in the data, not accidental), preventing the discovery of association rules without practical reference value. A pre-defined rule structure is used to filter the generated association rules, clarifying their specificity and ensuring that the rules directly explain "which combinations of influencing factors lead to which specific failure," avoiding ambiguous rules (e.g., unclear failure type or influencing factors), facilitating direct application in subsequent risk assessment and prediction. Confidence is calculated to filter association rules, ensuring their effectiveness and credibility (e.g., confidence ≥ 75% indicates high credibility), ensuring reliable results when using these rules for risk level classification and early warning.

[0011] Furthermore, the step of performing cluster analysis on the target metering equipment set using a preset clustering algorithm, based on the fault association rule set and the multidimensional dataset, to generate a fault risk level index, specifically involves: Using a pre-defined K-means algorithm, cluster analysis is performed on the target metering equipment set based on the fault association rule set and the multidimensional dataset. Based on the cluster analysis results, the target metering equipment set is divided into different clusters. The optimal number of clusters in the K-means algorithm is determined by the elbow rule. Based on the characteristics of each metering device within each cluster, a fault risk level index is generated; wherein, the characteristics of each metering device are extracted based on the multidimensional dataset.

[0012] This invention utilizes the K-means algorithm, combined with association rules (fault causes) and multidimensional data (operating status, historical faults), to provide a more comprehensive basis for clustering. The elbow method avoids unreasonable clustering caused by too many or too few clusters, ensuring optimal clustering results. Devices within the same cluster have similar fault causes, operating status, and historical fault characteristics, and their fault risk levels are similar. By classifying high, medium, and low risks based on intra-cluster features, the risk level indicators are ensured to be reasonable and accurate, providing a clear basis for subsequent cluster prediction and differentiated maintenance.

[0013] Furthermore, the performance prediction of each subset of metering equipment is performed using a preset performance prediction model to obtain the performance prediction results for each subset of metering equipment, specifically as follows: For each subset of metering equipment, the corresponding performance index time series data is extracted from the operating status data in the multidimensional dataset, and a linear prediction result is generated based on the performance index time series data using a pre-trained ARIMA model. Using a pre-trained LSTM model, nonlinear residual prediction results are generated based on the linear prediction results and the time series data of the performance indicators. The linear prediction results and the nonlinear residual prediction results are fused to obtain the performance prediction results.

[0014] This invention utilizes the ARIMA model's ability to capture long-term, stable linear changes (such as a slow upward trend in error rate) for linear prediction, providing a basic linear trend support for prediction and solving the problem that a single model cannot handle linear patterns. It also leverages the LSTM model's ability to handle complex, non-linear fluctuating data (such as a sudden increase in error rate due to sudden environmental changes) to compensate for the limitations of the ARIMA model and capture the non-linear fluctuation patterns of performance indicators. Through data fusion, it takes into account both the "long-term linear trend" and "short-term non-linear fluctuations" of equipment performance, making the prediction results more comprehensive and accurate, and avoiding the problem of large prediction biases in single models.

[0015] Furthermore, the optimal model parameters of the ARIMA model are determined using the Akaike Information Criterion.

[0016] This invention uses the AIC criterion to balance model fit and complexity, selects optimal parameters, avoids overfitting (too many parameters leading to fitting noise) or underfitting (too few parameters leading to insufficient fitting) of the ARIMA model, ensures the linear prediction accuracy of the ARIMA model, lays a good foundation for subsequent LSTM residual prediction, and improves overall prediction performance.

[0017] Furthermore, after obtaining the performance prediction results of each subset of metering equipment, fault warnings are also included; Specifically, the fault warning process involves: integrating the fault association rule set, fault risk level indicators, and performance prediction results to update multi-level warning thresholds in real time. The operating status data of each metering device in the target metering device set is monitored in real time, and performance prediction is performed based on the operating status data to obtain real-time performance prediction results; Based on the real-time performance prediction results and the current multi-level warning threshold, it is determined whether each metering device has reached the preset trigger condition. If any metering device has reached the trigger condition, a fault warning is issued for the corresponding metering device.

[0018] This invention dynamically optimizes the warning threshold based on the actual condition of the equipment, avoiding "warning delays or false warnings caused by fixed thresholds"; through real-time monitoring, it promptly captures dynamic changes in the equipment's operating status, updates performance prediction results, ensures the real-time nature of warnings, and avoids failures that cannot be intervened in advance due to prediction delays; it transforms performance prediction results into actionable fault warnings, informing maintenance personnel of potential risks in advance, giving them sufficient time to take maintenance measures, and preventing faults from occurring or escalating.

[0019] Secondly, embodiments of the present invention provide a metering equipment performance prediction device, including a data acquisition module, an association rule generation module, a risk level generation module, and a prediction result acquisition module, wherein... The data acquisition module is used to acquire a multidimensional dataset of the target metering equipment set; wherein, the multidimensional dataset includes equipment attribute data, operating status data, historical fault data, and influencing factor data; the influencing factor data includes environmental factor data and service life data; The association rule generation module is used to mine the association relationship between equipment failures and preset key influencing factors based on the multidimensional dataset using a preset association rule mining algorithm, and generate a set of failure association rules; wherein, the key influencing factors include environmental factors and service life; The risk level generation module is used to perform cluster analysis on the target metering equipment set based on the fault association rule set and the multidimensional dataset using a preset clustering algorithm to generate a fault risk level index. The prediction result acquisition module is used to combine and divide the target metering equipment set into several metering equipment subsets by using the fault association rule set and fault risk level index, and to perform performance prediction on each metering equipment subset by using a preset performance prediction model to obtain the performance prediction result of each metering equipment subset.

[0020] This invention provides a comprehensive and core data foundation for all subsequent analyses through a data acquisition module, covering key dimensions of the entire equipment lifecycle (basic information, operating status, past faults, and external causes), avoiding analytical biases caused by incomplete data and ensuring sufficient data support for subsequent steps. Through an association rule generation module, fault association rules are mined to clarify the core causes of faults, providing targeted correlation basis for subsequent risk level classification and performance prediction, solving the problem of traditional methods ignoring the potential correlation between faults and multiple factors. Through a risk level generation module, clustering algorithms are used to group equipment according to the similarity of fault risks, achieving differentiated equipment risk distinction and avoiding "one-size-fits-all" management, laying the foundation for accurate prediction by subsequent grouping. Through a prediction result acquisition module, metering equipment is subsetted according to association rules and risk levels, avoiding mutual interference between equipment data with different risks and fault causes, improving the targeting and accuracy of predictions (the performance degradation patterns of equipment in the same subset are more consistent).

[0021] Thirdly, embodiments of the present invention provide a terminal device, including: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction that causes the processor to perform the operation of the metering equipment performance prediction method as described in any of the above.

[0022] Fourthly, embodiments of the present invention provide a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device or apparatus containing the computer-readable storage medium to perform the metering equipment performance prediction method as described in any of the preceding claims.

[0023] The above description is merely an overview of the technical solutions of the embodiments of the present invention. In order to better understand the technical means of the embodiments of the present invention and to implement them in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the embodiments of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of a method for predicting the performance of metering equipment provided in an embodiment of the present invention; Figure 2 This is a structural diagram of a metering equipment performance prediction device provided in an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Example 1: like Figure 1 As shown, an embodiment of the present invention provides a method for predicting the performance of metering equipment, comprising the following steps: S101, Obtain a multidimensional dataset of the target metering equipment set; wherein, the multidimensional dataset includes equipment attribute data, operating status data, historical fault data, and influencing factor data; the influencing factor data includes environmental factor data and service life data; In this embodiment, before acquiring the multidimensional dataset of the target metering device set, the preprocessing of the pre-acquired original dataset is further included to obtain the multidimensional dataset. Specifically, the preprocessing of the pre-acquired original dataset to obtain the multidimensional dataset involves: discretizing the pre-acquired original influencing factor data according to a preset interval division threshold to obtain first influencing factor data; discretizing the pre-acquired original historical fault data according to a preset fault type to obtain first historical fault data; calculating the mutual information value between the fault type and each influencing factor based on the first historical fault data and the first influencing factor data using the mutual information method, and filtering the first influencing factor data according to the mutual information value to obtain second influencing factor data; and integrating the first historical fault data, the second influencing factor data, the pre-acquired device attribute data, and the operating status data to obtain the multidimensional dataset.

[0027] In one specific embodiment, the specific process for collecting raw metering equipment data and performing standardization processing is as follows: Four types of core data are collected: basic attribute data, operational status data, environmental impact data, and full life cycle data. The sampling frequency for operational status data is set to once every 15 minutes, and the sampling frequency for environmental impact data is set to once per hour. Specifically, four types of core data are collected through IoT terminals, database interfaces, and sensor networks: Basic attribute data: Device ID, model, manufacturing date, rated parameters (such as rated voltage, accuracy class), installation location (latitude and longitude, substation area); Operating status data: real-time metering error rate, voltage / current RMS values, power consumption, communication success rate (e.g., 4G / power line carrier communication), sampling frequency set to 1 time / 15 minutes; Environmental impact data: Temperature (-40℃~70℃), relative humidity (0%~100%), electromagnetic interference intensity (dBμV / m), and voltage fluctuation frequency (number of times per month ≥±10% of rated voltage) of the installation area are collected by deploying temperature and humidity sensors and electromagnetic monitoring modules, with a sampling frequency of 1 time / hour. Full life cycle data: years of use (current date - manufacturing date), historical fault records (fault type, occurrence time, repaired parts), calibration records (calibration time, error correction value).

[0028] Furthermore, outliers are removed using the "3σ criterion + manual verification," and continuous data is standardized using Min-Max to map it to the [0,1] interval. The standardization formula is as follows: In the formula, x represents the original data. , These are the minimum and maximum values ​​of the indicator, respectively. Furthermore, the service life, environmental data, and fault types are discretized. The service life is divided into A1 (1~3 years), A2 (3~5 years), and A3 (≥5 years); the temperature is divided into B1 (≤0℃), B2 (0℃~30℃), B3 (30℃~45℃), and B4 (≥45℃); the humidity is divided into C1 (≤40%), C2 (40%~70%), and C3 (≥70%); and the fault types are discretized into five categories: D1 (measuring error exceeds the standard), D2 (communication failure), D3 (hardware damage), D4 (power failure), and D5 (software abnormality). Furthermore, the mutual information method is used to calculate the mutual information value between each feature and the fault type. Features with a mutual information value ≥ 0.3 (such as service life, temperature, humidity, and voltage fluctuation frequency) are filtered out, redundant features (such as equipment color and manufacturing batch) are removed, and a standardized dataset is output.

[0029] S102, using a preset association rule mining algorithm, based on the multidimensional dataset, the association relationship between equipment failures and preset key influencing factors is mined to generate a set of failure association rules; wherein, the key influencing factors include environmental factors and service life; In this embodiment, the step of mining the association relationship between equipment failures and preset key influencing factors based on the multidimensional dataset using a preset association rule mining algorithm, and generating a failure association rule set, specifically involves: generating several candidate first-order itemsets based on the multidimensional dataset, and calculating the support of each candidate first-order itemset to filter the candidate first-order itemsets using the support to obtain frequent first-order itemsets; concatenating the frequent first-order itemsets to generate several candidate second-order itemsets, and calculating the support of each candidate second-order itemset to filter the candidate second-order itemsets using the support to obtain frequent second-order itemsets. The process involves layer-by-layer splicing and support calculations until a preset condition is met, thereby generating and filtering several frequent itemsets of different orders. Fault association rules satisfying a preset rule structure are extracted from each frequent itemset to obtain an initial fault association rule set. The confidence level of each fault association rule in the initial fault association rule set is calculated, and the initial fault association rule set is filtered using the confidence level to obtain a fault association rule set. The rule structure is a unidirectional structure from antecedent to consequent. The antecedent is composed of a combination of discretized intervals of environmental factors and service life. The consequent is composed of a single fault type.

[0030] In one specific embodiment, the association rule structure is defined as "antecedent → consequent", where the antecedent (condition itemset) consists of discrete intervals of environmental factors and service life, such as: {A3, B4} (service life ≥ 5 years + temperature ≥ 45℃), {A2, C3} (service life 3-5 years + humidity ≥ 70%). Consequence (Result Item Set): Single fault type, such as: D1 (measuring error exceeds the standard), D2 (communication failure).

[0031] In one specific embodiment, association rule mining is performed using the Apriori algorithm, wherein the parameters of the Apriori algorithm are set as follows: (1) Minimum support Set it to 5% (i.e., the frequency of an itemset in the dataset is greater than or equal to 5% of the total number of samples, to ensure that the itemset is statistically significant). (2) Minimum confidence Set to 75% (i.e., the credibility of the rule "predecessor → consequent" is ≥75% to ensure the validity of the association); (3) Fault association weights W: Weights (D1=1.2, D2=1.1, D3=1.0, D4=0.9, D5=0.8) are set for different fault types to obtain the corrected itemset support: Prioritize retaining association rules for high-impact faults.

[0032] In one specific embodiment, the process of generating frequent itemsets and association rules is as follows: First, generate 1-itemsets (e.g., ), and calculate the corrected support to filter out those that meet the requirements. Frequent 1-itemsets; Furthermore, 2-itemsets and 3-itemsets (such as...) are generated through the "join-pruning" operation. Repeated filtering yields frequent itemsets of different orders; Furthermore, association rules are extracted from the frequent item sets of each order, and confidence scores are calculated: Finally, the selection was made The rules are used to construct a "fault-influencing factor" association library, an example of which is shown in Table 1.

[0033] Table 1. Example of a rule association library S103, using a preset clustering algorithm, cluster analysis is performed on the target metering equipment set based on the fault association rule set and the multidimensional dataset to generate a fault risk level index; In this embodiment, based on the fault association rule set and the multidimensional dataset, cluster analysis is performed on the target metering device set to generate a fault risk level index. Specifically, the target metering device set is clustered using a preset K-means algorithm based on the fault association rule set and the multidimensional dataset, and the target metering device set is divided into different clusters based on the cluster analysis results. The optimal number of clusters in the K-means algorithm is determined by the elbow rule. A fault risk level index is generated based on the characteristics of each metering device within each cluster. The characteristics of each metering device are extracted based on the multidimensional dataset.

[0034] In one specific embodiment, to achieve differentiated maintenance, the K-means++ algorithm (which optimizes the selection of initial cluster centers and avoids local optima) is used to classify the risks of the measured assets. The steps are as follows: Clustering feature selection: Four core features were selected (all from standardized data): Feature 1: Service life (discretized mapped value, such as...) ; Feature 2: Historical failure frequency (number of failures in the last 2 years / 2, mapped to [0,1]); Feature 3: Association rule matching degree (the number of high-confidence rules matched by the device's current environment / years of use, such as 2 matches → 0.8, 0 matches → 0.2); Feature 4: Measurement error volatility (standard deviation of error rate over the past 3 months / average error rate, mapped to [0,1]).

[0035] Furthermore, the optimal number of clusters is determined using the elbow rule. (Calculate the silhouette coefficient for different K values, when...) The contour coefficient is at its maximum at 0.82. Furthermore, K-means++ is used to initialize cluster centers (to avoid clustering bias caused by random initialization), and the Euclidean distance from each device to the cluster center is calculated iteratively to assign the device to the nearest cluster. Finally, clustering results are calculated based on the characteristics of the devices in each cluster, and the clustering results are defined as three risk levels. Differentiated management strategies are then formulated, and examples of the three risk levels are shown in Table 2.

[0036] Table 2 Examples of Three Risk Levels It should be noted that if the equipment's historical records contain multiple faults (such as "measuring error exceeding the standard D1" and "communication failure D2" occurring simultaneously), the specific clustering process is as follows: (1) Fault type discretization: Each fault is classified separately into D1~D5 and is not combined into "compound fault"; (2) Association rule mining: The consequent of the rule is only "single fault type", and multiple faults generate corresponding association rules respectively (e.g., D1 corresponds to {A3,B4}→D1, D2 corresponds to {A2,C3}→D2). (3) Clustering classification: Statistical analysis of “total number of failures in the past 2 years” is mapped to the feature of “historical failure frequency” (e.g. 2 failures → 1.0, 1 failure → 0.5), without distinguishing the specific type of failure.

[0037] S104, the target metering equipment set is combined and divided into several metering equipment subsets by the fault association rule set and fault risk level index, and the performance of each metering equipment subset is predicted by the preset performance prediction model to obtain the performance prediction results of each metering equipment subset.

[0038] In one specific embodiment, the target set of metering devices is combined and divided to obtain several subsets of metering devices. Specifically, they are grouped according to "association rules + risk level": such as "high-risk devices matching the rule {A3,B4}→D1" and "medium-risk devices matching the rule {A2,C3}→D2". Each group has a separate prediction model (to avoid cross-group data interference).

[0039] In one specific embodiment, "metering error rate" and "communication success rate" are selected as core prediction indicators (directly related to fault type), and the time series length is set to 90 days (i.e., predicting the trend of the next 30 days based on the data of the previous 90 days).

[0040] In this embodiment, the performance prediction of each subset of metering equipment is performed using a preset performance prediction model to obtain the performance prediction results for each subset of metering equipment. Specifically, for each subset of metering equipment, the corresponding performance index time series data is extracted from the operating status data in the multidimensional dataset, and a linear prediction result is generated based on the performance index time series data using a pre-trained ARIMA model; a nonlinear residual prediction result is generated based on the linear prediction result and the performance index time series data using a pre-trained LSTM model; and the linear prediction result and the nonlinear residual prediction result are fused to obtain the performance prediction result.

[0041] In this embodiment, the optimal model parameters of the ARIMA model are determined by the Akaike Information Criterion.

[0042] In one specific embodiment, the ARIMA-LSTM fusion model is constructed in advance, and the construction process is as follows: (1) Data stationarity processing: Perform ADF test on the predicted feature time series. If it is not stationary (P value > 0.05), perform d-order differencing (e.g., after 1-order differencing, P value < 0.05, determine d = 1). (2) ARIMA model training: For data that is stationary after differencing, the ARIMA parameters (p=2, d=1, q=1) are determined by the AIC criterion, and the linear trend (such as the long-term slow upward trend of the error rate) is fitted. (3) LSTM model training: The LSTM network structure is set as “input layer (90 time steps) → hidden layer (64 neurons) → dropout layer (0.2) → hidden layer (32 neurons) → output layer (30 time step predictions)”. The optimizer is Adam and the loss function is MSE.

[0043] In one specific embodiment, the data fusion of the linear prediction result and the nonlinear residual prediction result to obtain the performance prediction result specifically involves: The ARIMA linear prediction results are superimposed with the LSTM nonlinear residual prediction results to obtain the final performance trend curve: .

[0044] It should be noted that ARIMA predictions can fit the long-term linear trend of time series (such as the slow upward trend of measurement error rate with the number of years of use). In practice, this can guide "long-term maintenance plans"—for example, predicting that the error rate of a certain device will approach the ±1.5% threshold after one year and making a replacement plan for the parts in advance. LSTM prediction results can fit the short-term nonlinear residuals of ARIMA models (such as abnormal fluctuations in error rate caused by sudden high temperature or voltage fluctuations). In practice, this means capturing "short-term sudden risks"—for example, if the error rate of a certain device suddenly increases by 0.3% within 3 days due to high temperature, it can trigger an emergency warning and prevent the fault from escalating. The combination of the two technologies covers both "long-term performance degradation" (ARIMA) and "short-term abnormal fluctuations" (LSTM), enabling fault warnings to balance "foresight" (an average of 18 days in advance) and "accuracy" (90.6% accuracy), thus solving the problem of single models "missing long-term risks" or "misjudging short-term fluctuations".

[0045] It should be noted that the collected data from each set of metering equipment operation data is divided into training set, validation set, and test set in a ratio of 7:1.5:1.5. The validation set is used to optimize model parameters, and the test set is used to verify performance. The test set is required to have a mean absolute error (MAE) ≤ 0.15% (metering error rate prediction) and an MAE ≤ 2% (communication success rate prediction).

[0046] In this embodiment, after obtaining the performance prediction results of each subset of metering devices, a fault warning is also included. Specifically, the fault warning process involves: integrating the fault association rule set, fault risk level indicators, and performance prediction results to update multi-level warning thresholds in real time; monitoring the operating status data of each metering device in the target metering device set in real time, and performing performance prediction based on the operating status data to obtain real-time performance prediction results; and determining whether each metering device has reached a preset trigger condition based on the real-time performance prediction results and the current multi-level warning thresholds. If any metering device reaches the trigger condition, a fault warning is issued for the corresponding metering device.

[0047] In one specific embodiment, a three-layer early warning mechanism is constructed by integrating the association rule base, clustering risk level, and time series prediction results, specifically as follows: (1) Warning threshold setting: Level 1 Threshold (Emergency Warning): Time series predicted values ​​exceed the equipment's rated thresholds (e.g., metering error rate ≥ ±1.5%, communication success rate ≤ 90%). Level 2 Threshold (Important Warning): Although the time series forecast value does not exceed the rated threshold, the rate of increase / decrease within 30 days is greater than or equal to the threshold (e.g., the error rate increases by 0.05% per day, and the communication success rate decreases by 0.5% per day). Level 3 threshold (general warning): The device matches ≥1 high-confidence association rule (e.g., {A3,B4}→D1), and the clustering level is medium to high risk.

[0048] (2) Early warning triggering and output: Real-time monitoring: Collects current device status data every hour, matches and associates rules, updates risk levels, and calls predictive models; Warning Judgment: If any warning threshold is met, the corresponding level of warning is triggered, and a warning report is generated, which includes the device ID, installation location, association rule matching results, and risk level; Prediction results: Performance trend curve for the next 30 days, and time to exceed warning threshold (e.g., "The error rate is expected to reach 1.5% in 15 days"). Finally, decision recommendations are generated, which include maintenance priorities (urgent / important / general), recommended maintenance measures (such as "replace the error compensation module" or "adjust the temperature and humidity of the installation environment"), and regional load linkage prompts (such as "the power load of the transformer area to which this equipment belongs is expected to increase by 10% in the next month, and it is recommended to perform maintenance in advance to avoid the failure from affecting the power supply").

[0049] This invention provides a comprehensive and core data foundation for all subsequent analyses, covering key dimensions throughout the equipment's entire lifecycle (basic information, operating status, past faults, and external causes), avoiding analytical biases caused by incomplete data and ensuring sufficient data support for subsequent steps. By mining fault association rules, it clarifies the core causes of faults, providing targeted correlation basis for subsequent risk level classification and performance prediction, solving the problem of traditional methods ignoring the potential correlation between faults and multiple factors. Through clustering algorithms, it groups equipment according to the similarity of fault risk, achieving differentiated distinction of equipment risks and avoiding "one-size-fits-all" management, laying the foundation for accurate prediction by subsequent grouping. By subdividing metering equipment according to association rules and risk levels, it avoids mutual interference between equipment data with different risks and fault causes, improving the targeting and accuracy of predictions (the performance degradation patterns of equipment in the same subset are more consistent). Compared with existing technologies, this invention can solve the problem of heterogeneity in analytical data caused by the diversity of equipment fault causes and state differences through equipment combination classification by association rules and clustering results, improving the accuracy of metering equipment performance prediction results.

[0050] Example 2: like Figure 2 As shown, this embodiment provides a metering equipment performance prediction device, including a data acquisition module 201, an association rule generation module 202, a risk level generation module 203, and a prediction result acquisition module 204, wherein... The data acquisition module 201 is used to acquire a multidimensional dataset of the target metering equipment set; wherein, the multidimensional dataset includes equipment attribute data, operating status data, historical fault data, and influencing factor data; the influencing factor data includes environmental factor data and service life data; The association rule generation module 202 is used to mine the association relationship between equipment failure and preset key influencing factors based on the multidimensional dataset using a preset association rule mining algorithm, and generate a set of failure association rules; wherein, the key influencing factors include environmental factors and service life; In this embodiment, the association rule generation module 202, based on the multidimensional dataset, mines the association relationship between equipment faults and preset key influencing factors using a preset association rule mining algorithm, and generates a fault association rule set. Specifically, the association rule generation module 202 generates several candidate first-order itemsets based on the multidimensional dataset, and calculates the support of each candidate first-order itemset. The candidate first-order itemsets are then filtered using the support to obtain frequent first-order itemsets. The frequent first-order itemsets are then concatenated to generate several candidate second-order itemsets, and the support of each candidate second-order itemset is calculated. The candidate second-order itemsets are then filtered using the support. The process involves filtering to obtain frequent second-order itemsets; continuing the splicing operation and support calculation layer by layer until a preset condition is met, thereby generating and filtering several frequent itemsets of different orders layer by layer; extracting fault association rules that satisfy the preset rule structure from each order of frequent itemsets to obtain an initial fault association rule set, and calculating the confidence of each fault association rule in the initial fault association rule set, so as to filter the initial fault association rule set through the confidence score to obtain a fault association rule set; wherein, the rule structure is a unidirectional structure from antecedent to consequent; the antecedent is composed of a combination of discretized intervals of environmental factors and service life; the consequent is composed of a single fault type.

[0051] The risk level generation module 203 is used to perform cluster analysis on the target metering equipment set based on the fault association rule set and the multidimensional dataset using a preset clustering algorithm to generate a fault risk level index. In this embodiment, the risk level generation module 203 performs cluster analysis on the target metering equipment set based on the fault association rule set and the multidimensional dataset using a preset clustering algorithm to generate a fault risk level index. Specifically, the risk level generation module 203 performs cluster analysis on the target metering equipment set based on the fault association rule set and the multidimensional dataset using a preset K-means algorithm, and divides the target metering equipment set into different clusters based on the cluster analysis results. The optimal number of clusters in the K-means algorithm is determined by the elbow rule. A fault risk level index is generated based on the characteristics of each metering equipment in each cluster. The characteristics of each metering equipment are extracted based on the multidimensional dataset.

[0052] The prediction result acquisition module 204 is used to combine and divide the target metering equipment set into several metering equipment subsets by using the fault association rule set and fault risk level index, and to perform performance prediction on each metering equipment subset by using a preset performance prediction model to obtain the performance prediction result of each metering equipment subset.

[0053] In this embodiment, the prediction result acquisition module 204 performs performance prediction on each subset of metering equipment using a preset performance prediction model to obtain the performance prediction results for each subset of metering equipment. Specifically, for each subset of metering equipment, the prediction result acquisition module 204 extracts the corresponding performance index time series data from the operating status data in the multidimensional dataset, and generates a linear prediction result based on the performance index time series data using a pre-trained ARIMA model; generates a nonlinear residual prediction result based on the linear prediction result and the performance index time series data using a pre-trained LSTM model; and fuses the linear prediction result and the nonlinear residual prediction result to obtain the performance prediction result.

[0054] For a more detailed explanation of the working principle and procedures of this embodiment, please refer to the relevant description in Embodiment 1.

[0055] This invention, through its data acquisition module 201, provides a comprehensive and core data foundation for all subsequent analyses, covering key dimensions of the entire equipment lifecycle (basic information, operating status, past faults, and external causes), avoiding analytical biases caused by incomplete data and ensuring sufficient data support for subsequent steps. Through its association rule generation module 202, it mines fault association rules, clarifies the core causes of faults, and provides targeted association basis for subsequent risk level classification and performance prediction, solving the problem of traditional methods ignoring the potential correlation between faults and multiple factors. Through its risk level generation module 203, it uses clustering algorithms to divide equipment groups according to the similarity of fault risks, achieving differentiated equipment risk distinction and avoiding "one-size-fits-all" management, laying the foundation for accurate prediction by subsequent grouping. Through its prediction result acquisition module 204, it divides metering equipment into subsets based on association rules and risk levels, avoiding mutual interference between equipment data with different risks and different fault causes, improving the targeting and accuracy of predictions (the performance degradation patterns of equipment in the same subset are more consistent).

[0056] Example 3: This embodiment provides a terminal device, including: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction that causes the processor to perform the operation of the metering equipment performance prediction method as described in any of the above.

[0057] Example 4: This invention provides a computer-readable storage medium including a stored computer program, wherein the computer program, when running, controls the device or apparatus containing the computer-readable storage medium to execute the metering equipment performance prediction method as described in any of the above embodiments.

[0058] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0059] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for predicting the performance of metering equipment, characterized in that, include: Obtain a multidimensional dataset of the target metering equipment set; wherein the multidimensional dataset includes equipment attribute data, operating status data, historical fault data, and influencing factor data; the influencing factor data includes environmental factor data and service life data; Using a pre-defined association rule mining algorithm, the association between equipment failures and pre-defined key influencing factors is mined based on the multidimensional dataset, generating a set of failure association rules; wherein, the key influencing factors include environmental factors and service life; By using a preset clustering algorithm, cluster analysis is performed on the target metering equipment set based on the fault association rule set and the multidimensional dataset to generate a fault risk level index. The target metering equipment set is divided into several metering equipment subsets by combining the fault association rule set and fault risk level index. Then, the performance of each metering equipment subset is predicted by a preset performance prediction model to obtain the performance prediction results of each metering equipment subset.

2. The method for predicting the performance of metering equipment as described in claim 1, characterized in that, Before obtaining the multidimensional dataset of the target set of metering devices, the process also includes preprocessing the raw dataset to obtain the multidimensional dataset. Specifically, the pre-acquired raw dataset is preprocessed to obtain a multidimensional dataset, as follows: Based on the preset interval division threshold, the pre-acquired original influencing factor data is discretized to obtain the first influencing factor data; Based on the preset fault type, the pre-acquired original historical fault data is discretized to obtain the first historical fault data; Using the mutual information method, based on the first historical fault data and the first influencing factor data, the mutual information value between the fault type and each influencing factor is calculated, and the first influencing factor data is filtered according to the mutual information value to obtain the second influencing factor data; By integrating the first historical fault data, the second influencing factor data, the pre-acquired equipment attribute data, and the operating status data, a multidimensional dataset is obtained.

3. The method for predicting the performance of metering equipment as described in claim 1, characterized in that, The step involves using a preset association rule mining algorithm to mine the association between equipment failures and preset key influencing factors based on the multidimensional dataset, generating a set of failure association rules. Specifically: Based on the multidimensional dataset, several candidate first-order itemsets are generated, and the support of each candidate first-order itemset is calculated. The candidate first-order itemsets are then filtered using the support to obtain frequent first-order itemsets. The frequent first-order itemsets are concatenated to generate several candidate second-order itemsets, and the support of each candidate second-order itemset is calculated. The candidate second-order itemsets are then filtered based on the support to obtain frequent second-order itemsets. The splicing operation and support calculation are continued layer by layer until the preset conditions are met, so as to generate and filter out several frequent itemsets of different orders layer by layer. Fault association rules satisfying a preset rule structure are extracted from frequent itemsets of each order to obtain an initial fault association rule set. The confidence level of each fault association rule in the initial fault association rule set is calculated. The initial fault association rule set is then filtered using the confidence level to obtain a fault association rule set. The rule structure is a unidirectional structure from antecedent to consequent. The antecedent is composed of a combination of discretized intervals of environmental factors and service life. The consequent is composed of a single fault type.

4. The method for predicting the performance of metering equipment as described in claim 1, characterized in that, The step involves using a preset clustering algorithm to perform cluster analysis on the target metering equipment set based on the fault association rule set and the multidimensional dataset to generate a fault risk level index. Specifically: Using a pre-defined K-means algorithm, cluster analysis is performed on the target metering equipment set based on the fault association rule set and the multidimensional dataset. Based on the cluster analysis results, the target metering equipment set is divided into different clusters. The optimal number of clusters in the K-means algorithm is determined by the elbow rule. Based on the characteristics of each metering device within each cluster, a fault risk level index is generated; wherein, the characteristics of each metering device are extracted based on the multidimensional dataset.

5. The method for predicting the performance of metering equipment as described in claim 1, characterized in that, The process involves using a pre-defined performance prediction model to predict the performance of each subset of metering equipment, resulting in performance prediction results for each subset. For each subset of metering equipment, the corresponding performance index time series data is extracted from the operating status data in the multidimensional dataset, and a linear prediction result is generated based on the performance index time series data using a pre-trained ARIMA model. Using a pre-trained LSTM model, nonlinear residual prediction results are generated based on the linear prediction results and the time series data of the performance indicators. The linear prediction results and the nonlinear residual prediction results are fused to obtain the performance prediction results.

6. The method for predicting the performance of metering equipment as described in claim 5, characterized in that, The optimal model parameters of the ARIMA model are determined using the Akaike Information Criterion.

7. The method for predicting the performance of metering equipment as described in claim 1, characterized in that, After obtaining the performance prediction results of each subset of metering equipment, fault warning is also included; Specifically, the fault warning process involves: integrating the fault association rule set, fault risk level indicators, and performance prediction results to update multi-level warning thresholds in real time. The operating status data of each metering device in the target metering device set is monitored in real time, and performance prediction is performed based on the operating status data to obtain real-time performance prediction results; Based on the real-time performance prediction results and the current multi-level warning threshold, it is determined whether each metering device has reached the preset trigger condition. If any metering device has reached the trigger condition, a fault warning is issued for the corresponding metering device.

8. A device for predicting the performance of metering equipment, characterized in that, It includes a data acquisition module, a association rule generation module, a risk level generation module, and a prediction result acquisition module. The data acquisition module is used to acquire a multidimensional dataset of the target metering equipment set; wherein, the multidimensional dataset includes equipment attribute data, operating status data, historical fault data, and influencing factor data; the influencing factor data includes environmental factor data and service life data; The association rule generation module is used to mine the association relationship between equipment failures and preset key influencing factors based on the multidimensional dataset using a preset association rule mining algorithm, and generate a set of failure association rules; wherein, the key influencing factors include environmental factors and service life; The risk level generation module is used to perform cluster analysis on the target metering equipment set based on the fault association rule set and the multidimensional dataset using a preset clustering algorithm to generate a fault risk level index. The prediction result acquisition module is used to combine and divide the target metering equipment set into several metering equipment subsets by using the fault association rule set and fault risk level index, and to perform performance prediction on each metering equipment subset by using a preset performance prediction model to obtain the performance prediction result of each metering equipment subset.

9. A terminal device, characterized in that, include: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction that causes the processor to perform the operation of the metering equipment performance prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device or apparatus containing the computer-readable storage medium to perform the metering equipment performance prediction method as described in any one of claims 1 to 7.

Images