A metrological equipment whole life cycle decommissioning evaluation and recycling method and storage medium
By combining the random forest algorithm and the long short-term memory network model, the problem of inaccurate performance evaluation in the decommissioning management of metering equipment is solved, enabling scientific prediction and accurate decision-making on the timing of equipment decommissioning, and improving the level of intelligence in equipment management.
Patent Information
- Application Number
- CN202611248590.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-18
- Publication Date
- 2026-11-17
AI Technical Summary
Existing methods for managing the decommissioning of metering equipment cannot identify the differentiated impacts of equipment performance degradation, leading to inaccurate equipment performance assessments, an inability to scientifically predict the timing of decommissioning, and resulting in resource waste or safety hazards.
The random forest algorithm is used for feature fusion, combined with principal component analysis for dimensionality reduction, and a long short-term memory network model is used to capture temporal correlation information. A retirement command is generated through weighted scoring to achieve scientific prediction and accurate decision-making for metering equipment.
It enables scientific prediction and precise decision-making on the timing of metering equipment retirement, improves the level of intelligence in the entire life cycle management of equipment, and ensures the reasonable retirement and reuse of equipment.
Smart Images

Figure CN123118059A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power equipment maintenance technology, specifically to a method and storage medium for the full life cycle decommissioning assessment and reuse of metering equipment. Background Technology
[0002] The safe and stable operation of the power system relies on the accurate measurement of a massive number of metering devices. These devices undertake key tasks such as energy metering, electricity consumption monitoring, and load analysis, directly affecting grid dispatching decisions and the fairness of electricity billing. With the in-depth advancement of smart grid construction, the number of metering devices has exceeded hundreds of millions, and the scale of decommissioned equipment generated each year is enormous. How to scientifically assess the residual value of these devices and achieve efficient reuse has become an important issue that the power industry urgently needs to address.
[0003] Current equipment decommissioning management primarily relies on batch replacement based on equipment lifespan or fixed calibration cycles. This approach ignores the varying impacts of different operating environments on equipment performance degradation. Electricity meters manufactured in the same batch may experience a rate of accuracy degradation several times different when deployed in humid coastal environments versus dry inland areas, but current methods cannot identify this difference. Adding to the complexity, equipment undergoes multiple stages throughout its lifecycle, including procurement, on-site installation, periodic calibration, and fault repair. Performance data, environmental parameters, and maintenance records generated at each stage are scattered across different systems, lacking a comprehensive traceability mechanism. This makes it impossible to accurately reconstruct the equipment's true health status upon decommissioning.
[0004] This situation leads to a mismatch between equipment performance assessments and actual operating conditions. Equipment performance indicators such as metering error, insulation strength, and mechanical life do not evolve independently but are interconnected and dynamically correlated. For example, prolonged overload operation accelerates the aging of electronic components, thus affecting metering accuracy and communication stability. However, traditional assessment methods only check whether individual indicators are qualified or not, failing to capture the correlation and degradation patterns between these performance indicators. When the metering error of an energy meter is close to the limit but still within the acceptable range, if its internal components have already undergone latent degradation due to environmental stress, continued use may lead to sudden inaccuracies in the short term, while premature scrapping results in resource waste. Therefore, how to construct a performance traceability system that integrates multi-source heterogeneous data throughout the entire equipment lifecycle, accurately identify the coupled degradation characteristics between different performance indicators, and then formulate differentiated retirement strategies that balance safety and economy, has become a key issue in improving the efficiency of metering asset management. Summary of the Invention
[0005] This invention aims to solve existing technical problems and provides a method and storage medium for the full life cycle decommissioning assessment and reuse of metering equipment, which can enable scientific prediction and accurate decision-making on the timing of decommissioning metering equipment.
[0006] As one aspect of the present invention, a method for the full life-cycle decommissioning assessment and reuse of metering equipment is provided, which includes the following steps: Step S11: Obtain multi-source heterogeneous operation data of the metering device during its life cycle, and extract a first environmental feature and a first performance feature from the multi-source heterogeneous operation data, wherein the multi-source heterogeneous operation data covers the monitoring records of the metering device at different operating stages. Step S12: Use the random forest algorithm to perform feature fusion processing on the first environmental feature and the first performance feature, and combine the total number of features with the corresponding weight coefficients to obtain a fused feature set; Step S13: Construct a multidimensional feature matrix based on the fused feature set, perform dimensionality reduction on the multidimensional feature matrix using principal component analysis, extract the main components based on the number of retained principal components, and obtain the target feature vector. Step S14: Process the target feature vector through a long short-term memory network model, extract the temporal correlation information in the target feature vector, capture the degradation pattern in the time series data, and determine the corresponding coupling degradation features; Step S15: If the coupling degradation feature is greater than a preset degradation threshold, the metering device is determined to be in a first healthy state; if the coupling degradation feature is less than or equal to the preset degradation threshold, the metering device is determined to be in a second healthy state. Step S16: For the metering device in the first health state, obtain the corresponding first operating condition, and perform a weighted calculation on the coupling degradation features based on the first operating condition to obtain a first degradation score; Step S17: If the first degradation score is greater than a preset score threshold, generate a first retirement instruction; if the first degradation score is less than or equal to the preset score threshold, generate a second retirement instruction. Step S18: For the metering equipment in the second health state, obtain the corresponding second operating condition, and perform a weighted calculation on the coupling degradation features according to the second operating condition to obtain a second degradation score; Step S19: If the second degradation score is greater than a preset score threshold, a third retirement instruction is generated; if the second degradation score is less than or equal to the preset score threshold, a fourth retirement instruction is generated.
[0007] Optionally, step S12 further includes: Determine the total number of features of the first environmental feature and the first performance feature; For each of the first environmental feature and the first performance feature, a corresponding weighting coefficient is determined; Call the random forest algorithm function, and input the first environmental feature, the first performance feature, and the corresponding weight coefficients; Weighted fusion calculation is performed by combining the corresponding weight coefficients; A comprehensive set of fusion features is generated based on the results of the weighted fusion calculation.
[0008] Optionally, step S13 further includes: The features in the fused feature set are arranged and combined to construct a multidimensional feature matrix; For the aforementioned multidimensional feature matrix, calculate the corresponding covariance matrix; Extract the eigenvalues of the covariance matrix and the corresponding principal component vectors; Sort the components according to their eigenvalues to determine the number of principal components to retain; Based on the principal component vectors corresponding to the number of retained principal components, the multidimensional feature matrix is mapped and transformed to obtain the target feature vector.
[0009] Optionally, step S14 further includes: Input the target feature vector into the long short-term memory network model; Combining the hidden state from the previous time step with the input at the current time step, matrix multiplication is performed between the input layer weight matrix and the hidden layer weight matrix; By introducing a bias term and an activation function to process the results of matrix multiplication, temporal correlation information is extracted from the target feature vector. The corresponding coupling degradation features are determined based on the temporal correlation information.
[0010] Optionally, step S15 further includes: Obtain the preset degradation threshold; The coupling degradation feature is compared with the preset degradation threshold in terms of numerical value; When the coupling degradation characteristic is greater than the preset degradation threshold, the metering device is determined to be in a first health state. If the coupling degradation characteristic is less than or equal to the preset degradation threshold, the metering device is determined to be in a second health state.
[0011] Optionally, step S16 further includes: Identify the metering device in the first health state; Collect the first operating condition corresponding to the metering device in the first health state; Extract the operating parameters from the first operating condition; Set the corresponding weight values according to the operating condition parameters; The first degradation score is obtained by weighting the coupling degradation features using the weight values.
[0012] Optionally, step S17 further includes: Obtain the preset scoring threshold; The first degradation score is numerically compared with the preset score threshold; If the first degradation score is greater than the preset score threshold, a first retirement command is generated; If the first degradation score is less than or equal to the preset score threshold, a second retirement instruction is generated.
[0013] Optionally, step S18 further includes: Identify the metering device in the second health state; Collect the second operating condition corresponding to the metering equipment in the second health state; Extract the operating parameters from the second operating condition; Set the corresponding weight value according to the operating parameters in the second operating condition; The coupling degradation features are weighted and calculated using the weight values corresponding to the second operating condition to obtain a second degradation score.
[0014] Optionally, step S19 further includes: Obtain the initial operating data of the target device; By cleaning the data and removing outliers from the initial running data, standard running data is obtained. Based on the standard operating data, the random forest algorithm is used to extract time series features to obtain the equipment attenuation. Calculate the ratio of the equipment attenuation to the initial reference value to determine the equipment loss. The equipment status parameters are obtained by classifying and categorizing the equipment based on the amount of equipment loss. The support vector machine algorithm is used to perform regression prediction on the device state variables to obtain an initial degradation score; The initial degradation score is processed by a moving average to obtain the second degradation score; The second degradation score is compared with the preset score threshold; If the second degradation score is greater than the preset score threshold, a third retirement instruction is generated; If the second degradation score is less than or equal to the preset score threshold, a fourth retirement instruction is generated.
[0015] Accordingly, as another aspect of the present invention, a computer-readable storage medium is also provided, wherein a computer program is stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the method as described above.
[0016] Implementing the embodiments of the present invention has the following beneficial effects: This invention discloses a method and storage medium for the full lifecycle decommissioning assessment and reuse of metering equipment. Addressing the core problems in the decommissioning decision-making process for metering equipment, such as the difficulty in effectively integrating multi-source heterogeneous data, inaccurate assessment of degradation status, and difficulty in accurately determining the decommissioning timing, this invention proposes a complete solution. First, this invention deeply integrates environmental and performance characteristics during equipment operation using a random forest algorithm. Then, it uses principal component analysis to reduce the dimensionality of the fused multidimensional feature matrix, effectively eliminating redundant information and extracting key features. Subsequently, it uses a long short-term memory network model to capture temporal correlation information, accurately identifying the coupled degradation characteristics of the equipment, and performing differentiated weighted scoring based on the actual operating conditions of the equipment. Finally, it generates corresponding decommissioning instructions based on different health states and degradation scores. Through this organic combination of multi-level feature extraction and intelligent assessment mechanisms, this invention achieves scientific prediction and accurate decision-making regarding the decommissioning timing of metering equipment, effectively improving the intelligence level of equipment lifecycle management and providing reliable technical support for the rational decommissioning and reuse of equipment. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, obtaining other drawings based on these drawings without creative effort still falls within the scope of the present invention. Figure 1 This is a main flowchart of an embodiment of a method for the full life-cycle decommissioning assessment and reuse of metering equipment provided by the present invention; Figure 2 for Figure 1 A more detailed schematic diagram of step S12; Figure 3 for Figure 1 A more detailed diagram of step S14. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings.
[0019] like Figure 1The diagram shown illustrates the main flow of an embodiment of a method for the full life-cycle decommissioning assessment and reuse of metering equipment provided by the present invention, in conjunction with... Figure 2 and Figure 3 As shown, the method includes the following steps: Step S11: Obtain multi-source heterogeneous operation data of the metering device during its life cycle, and extract a first environmental feature and a first performance feature from the multi-source heterogeneous operation data, wherein the multi-source heterogeneous operation data covers the monitoring records of the metering device at different operating stages. Step S12: Use the random forest algorithm to perform feature fusion processing on the first environmental feature and the first performance feature, and combine the total number of features with the corresponding weight coefficients to obtain a fused feature set; Step S13: Construct a multidimensional feature matrix based on the fused feature set, perform dimensionality reduction on the multidimensional feature matrix using principal component analysis, extract the main components based on the number of retained principal components, and obtain the target feature vector. Step S14: Process the target feature vector through a long short-term memory network model, extract the temporal correlation information in the target feature vector, capture the degradation pattern in the time series data, and determine the corresponding coupling degradation features; Step S15: If the coupling degradation feature is greater than a preset degradation threshold, the metering device is determined to be in a first healthy state; if the coupling degradation feature is less than or equal to the preset degradation threshold, the metering device is determined to be in a second healthy state. Step S16: For the metering device in the first health state, obtain the corresponding first operating condition, and perform a weighted calculation on the coupling degradation features based on the first operating condition to obtain a first degradation score; Step S17: If the first degradation score is greater than a preset score threshold, generate a first retirement instruction; if the first degradation score is less than or equal to the preset score threshold, generate a second retirement instruction. Step S18: For the metering equipment in the second health state, obtain the corresponding second operating condition, and perform a weighted calculation on the coupling degradation features according to the second operating condition to obtain a second degradation score; Step S19: If the second degradation score is greater than a preset score threshold, a third retirement instruction is generated; if the second degradation score is less than or equal to the preset score threshold, a fourth retirement instruction is generated.
[0020] The following will explain each step with specific examples.
[0021] In one example, step S11 involves acquiring multi-source heterogeneous operation data of the metering device during its lifecycle, and extracting a first environmental feature and a first performance feature from the multi-source heterogeneous operation data, wherein the multi-source heterogeneous operation data covers the monitoring records of the metering device at different operating stages. Specifically, the multi-source heterogeneous operational data originates from multiple data channels: during the procurement and warehousing stage, factory parameters such as equipment precision level, initial error coefficient, and tolerance temperature range are recorded; during the on-site installation stage, spatial information such as latitude and longitude, altitude, installation method (indoor / outdoor), and hanging orientation are collected; during the periodic verification stage, verification item data such as basic error, creep, starting current, and insulation resistance are included for each verification; and during the fault repair stage, historical maintenance records such as fault codes, repaired and replaced parts, and post-repair verification pass rate are stored. This data exists in various formats, including structured tables, semi-structured JSON logs, and unstructured images (such as on-site installation photos). It is uniformly extracted and transformed using ETL tools and loaded into a time-series database (such as InfluxDB), and associated with each device by its unique ID.
[0022] From the above data, the first environmental characteristics are extracted, including numerical characteristics such as annual average temperature, daily temperature range, relative humidity, rainfall, salt spray concentration (for coastal areas), dust concentration, and power frequency magnetic field interference intensity; the first performance characteristics are extracted, including metering error (active / reactive), starting current deviation, creep threshold, insulation resistance value, power consumption, communication success rate, and clock accuracy.
[0023] For missing values, linear interpolation or the median of similar equipment is used to fill them; for outliers (such as verification records with errors exceeding ±5%), they are marked and manually reviewed. Those that are verified to be correct are retained and an outlier label is added.
[0024] In step S12, the random forest algorithm is used to perform feature fusion processing on the first environmental feature and the first performance feature. Combining the total number of features with their corresponding weight coefficients, a fused feature set is obtained, specifically as follows: Figure 2 As shown, step S12 includes: Determine the total number of features of the first environmental feature and the first performance feature; For each of the first environmental feature and the first performance feature, a corresponding weighting coefficient is determined; Call the random forest algorithm function, and input the first environmental feature, the first performance feature, and the corresponding weight coefficients; Weighted fusion calculation is performed by combining the corresponding weight coefficients; A comprehensive set of fusion features is generated based on the results of the weighted fusion calculation.
[0025] Specifically, the random forest algorithm is used to fuse the first environmental feature and the first performance feature to obtain a fused feature set, as shown in the following formula:
[0026] in Represents the set of fused features. Indicates the total number of features. The weight coefficients of the k-th feature are represented by RF, which stands for Random Forest algorithm function. This represents the k-th first environmental feature. This represents the k-th primary performance feature. The formula generates a comprehensive fused feature set by weighted fusion of multiple environmental and performance features.
[0027] In practical engineering implementation, the Random Forest algorithm uses 100 decision trees to construct a regression forest. The maximum depth of each tree is limited to 10 layers, the minimum number of samples required for internal node splits is 5, and the minimum number of samples for leaf nodes is 3, to balance fitting ability and generalization performance. Let the first environmental feature dimension be E_dim (e.g., 12 dimensions) and the first performance feature dimension be P_dim (e.g., 9 dimensions), then the total number of features K = E_dim + P_dim = 21. The feature weight coefficients w_k are determined by a combination assignment method: first, the initial importance score is calculated using the built-in feature importance assessment of Random Forest (based on the average reduction of Gini impurity), then adjusted using the entropy weight method (using information entropy to measure the degree of variation of each feature), and finally, the subjective judgment matrix given by three senior econometric experts in the Analytic Hierarchy Process (AHP) is introduced for correction. The fusion ratio of the three is 0.4 (Gini), 0.3 (entropy weight), and 0.3 (AHP). After three to four rounds of iteration and convergence, the final weight coefficients are obtained.
[0028] The output of the RF function is the regression prediction value of the k-th feature pair for a single decision tree (i.e., the average value of the leaf nodes where the feature pair falls in the tree). The arithmetic mean of the output values of all 100 trees is taken as the standard fusion value of the feature pair. For example, the fusion value of the "temperature-error" feature pair of a three-phase electricity meter after processing by random forest is 0.52, and the output of the "humidity-insulation resistance" feature pair is 0.67. After all 21 feature pairs are fused, a fusion feature vector of length 21 is obtained, and each element comprehensively represents the cross-coupling effect between a single environmental factor and a single performance index.
[0029] In step S13, the step of constructing a multidimensional feature matrix based on the fused feature set, performing dimensionality reduction on the multidimensional feature matrix using principal component analysis, extracting principal components based on the number of retained principal components, and obtaining the target feature vector includes: The features in the fused feature set are arranged and combined to construct a multidimensional feature matrix; For the aforementioned multidimensional feature matrix, calculate the corresponding covariance matrix; Extract the eigenvalues of the covariance matrix and the corresponding principal component vectors; Sort the components according to their eigenvalues to determine the number of principal components to retain; Based on the principal component vectors corresponding to the number of retained principal components, the multidimensional feature matrix is mapped and transformed to obtain the target feature vector.
[0030] In a specific example, a multidimensional feature matrix is constructed based on the fused feature set. Principal component analysis is then used to reduce the dimensionality of the multidimensional feature matrix to obtain the target feature vector, as expressed by the formula:
[0031] in Let represent the target eigenvector, PCA represent Principal Component Analysis, Φ represent the multidimensional feature matrix, and M represent the number of principal components retained. This represents the eigenvalue of the i-th principal component. Let represent the i-th principal component vector. This formula achieves feature dimensionality reduction by extracting the principal components.
[0032] Specifically, assuming the total number of metering devices participating in the evaluation is N (e.g., 2000 smart meters in operation in a regional power grid), and each device corresponds to K-dimensional fusion features, the constructed multidimensional feature matrix Φ is an N×K matrix (2000×21). To eliminate the influence of inconsistent dimensions between different features, each column of Φ is first Z-score standardized before calculating the covariance matrix, so that the mean of each column is 0 and the standard deviation is 1. Then, the covariance matrix C (K×K order) of the standardized matrix is calculated, and the Jacobian eigenvalue decomposition algorithm is used to extract all eigenvalues λ_i (i=1,2,…,K) of C and their corresponding eigenvectors φ_i (length K).
[0033] Arrange all eigenvalues in descending order of size, and calculate the cumulative variance contribution rate (the ratio of the sum of the first m eigenvalues to the sum of all eigenvalues) sequentially. Set a target cumulative variance contribution rate of ≥95%. When m increases from 1, the m that first reaches or exceeds this target value is the number of principal components M to be retained. In typical scenarios, 21-dimensional features typically retain M = 6 to 8 principal components, which can explain more than 95% of the variation information in the original data.
[0034] The final target feature vector Vtarget is obtained by projecting the original matrix Φ onto the principal component direction, with the mapping formula Vtarget = Φ. [ 1, 2,..., [M], where the matrix in parentheses is the K×M projection matrix composed of the first M principal component vectors. After projection, each device is represented by an M-dimensional vector, which significantly reduces the input dimension of the subsequent LSTM network without losing key information, effectively suppressing the curse of dimensionality. The components of the projected target feature vector are normalized and mapped to the [-3,3] interval, which facilitates stable gradient propagation during neural network training.
[0035] Specifically, using real-world data, assuming 500 device samples, with a fusion feature dimension K=20, a 500×20 matrix Φ is constructed. After covariance calculation, the eigenvalues are sorted as [5.2, 3.8, 2.1, 1.5, 0.9, 0.5, ...]. The cumulative variance contribution rate of the first 5 is 5.2+3.8+2.1+1.5+0.9=13.5, with a total variance sum of 20 and a contribution rate of 67.5%, less than 95%. Adding a sixth eigenvalue of 0.5 brings the cumulative contribution to 14.0, a contribution rate of 70%, and so on, until the first 8 principal components reach a cumulative contribution of 19.1, a contribution rate of 95.5%. Therefore, M=8 is retained. After mapping, the target feature vector length for each sample is 8, significantly reducing redundant dimensions.
[0036] In step S14, the target feature vector is processed using a long short-term memory network model to extract temporal correlation information from the target feature vector, capture degradation patterns in the time-series data, and determine the corresponding coupling degradation features. Specific steps are as follows: Figure 3 As shown, it includes: Input the target feature vector into the long short-term memory network model; Combining the hidden state from the previous time step with the input at the current time step, matrix multiplication is performed between the input layer weight matrix and the hidden layer weight matrix; By introducing a bias term and an activation function to process the results of matrix multiplication, temporal correlation information is extracted from the target feature vector. The corresponding coupling degradation features are determined based on the temporal correlation information.
[0037] In a specific example, a Long Short-Term Memory (LSTM) network model is used to process the target feature vector, extracting temporal correlation information from it to determine the corresponding coupling degradation features. The formula is as follows:
[0038] in This indicates coupling degradation characteristics; LSTM stands for Long Short-Term Memory network model. This represents the input target feature vector. This represents the activation function. This represents the hidden layer weight matrix. This indicates the hidden state at the previous moment. Represents the input layer weight matrix. 'b' represents the input at the current time step, and 'b' represents the bias term. This formula uses a neural network to capture the degradation patterns in time-series data.
[0039] Suppose that each metering device has T historical sampling times during its lifecycle (T equals the number of verifications / monitorings the device has undergone since installation, usually no less than 5 times). Arrange the target feature vectors of each device in chronological order to form a T×M temporal input matrix. The LSTM network adopts a two-layer stacked architecture, with the first layer containing 64 LSTM units and the second layer containing 32 LSTM units, with a time expansion step size equal to T. At each time step t, the LSTM units sequentially perform forget gate, input gate, cell state update, and output gate operations, taking the hidden state h_T at the last moment, passing it through a fully connected linear layer (output dimension 1) and applying Sigmoid activation to obtain the final coupling degradation feature value D_coupling. This value is strictly limited to the (0, 1) interval; the closer the value is to 1, the more severe the degradation, and the closer it is to 0, the healthier the device state.
[0040] The LSTM model was trained using supervised learning. The training set consisted of historical time-series data from 3000 devices retired over the past three years, with the comprehensive remaining life assessment value (annotated by experts) from the last inspection before retirement serving as the regression label. The Huber loss (robust to outliers) was chosen as the loss function, and the Adam optimizer was used (learning rate 0.001, β1=0.9, β2=0.999). The batch size was 64, the maximum training epochs were 300, and an early stopping strategy was employed (training was terminated if the validation set loss did not decrease for 15 consecutive epochs). After training, the model parameters were fixed and used for real-time degradation assessment of the time-series features of new devices during the online inference phase.
[0041] In actual deployment, a device undergoes six periodic checks, each corresponding to a target feature vector (length 8), forming a 6×8 input sequence. The hidden layer weight matrix W_h of the LSTM is 64×64, the input layer weight matrix W_x is 64×8, and the bias term b is 64-dimensional. During forward propagation, h_0 and C_0 are initialized to zero vectors at time t=1, and the calculation continues until t=6. After h_6 passes through the output layer (weights 64×1, bias scalar), the scalar D_coupling is obtained. If this value is 0.62, it indicates that the device degradation is relatively significant.
[0042] In step S15, determining that the metering device is in a first healthy state if the coupling degradation characteristic is greater than a preset degradation threshold, and determining that the metering device is in a second healthy state if the coupling degradation characteristic is less than or equal to the preset degradation threshold, includes: Obtain the preset degradation threshold; The coupling degradation feature is compared with the preset degradation threshold in terms of numerical value; When the coupling degradation characteristic is greater than the preset degradation threshold, the metering device is determined to be in a first health state. If the coupling degradation characteristic is less than or equal to the preset degradation threshold, the metering device is determined to be in a second health state.
[0043] The initial value of the preset degradation threshold was determined by measuring the distribution of coupling degradation characteristic values of a batch of 100 brand-new reference energy meters certified by the National Institute of Metrology of China under standard laboratory conditions (temperature 23±2°C, relative humidity 60±15%RH, no electromagnetic interference) after continuous operation with rated current for 2000 hours. The 80th percentile of this distribution was taken as the initial threshold. Subsequently, this initial threshold was applied to the data of 500 historical devices whose retirement conclusions were clearly established in the field (confirmed by manual disassembly and analysis) for reverse verification and optimization. A confusion matrix-driven F1-score maximization strategy was adopted, and a grid search was performed in the interval [0.30, 0.70] with a step size of 0.01. Finally, the threshold that maximizes the F1-score was selected. According to actual measurements, this threshold is usually between 0.52 and 0.58. In this embodiment, 0.55 is preferred.
[0044] When D_coupling > 0.55, a device is considered to be in the first health state, indicating that the overall performance of the device has significantly deteriorated, and at least two of the core indicators such as insulation strength, metering error, or communication capability are close to or exceed the allowable limits of the regulations. When D_coupling ≤ 0.55, a device is considered to be in the second health state, indicating that although the device has some performance fluctuations, the core indicators are still within the qualified range, and it has the potential to continue operating or be downgraded.
[0045] In step S16, for the metering device in the first health state, the corresponding first operating condition is obtained, and the coupling degradation features are weighted and calculated according to the first operating condition to obtain a first degradation score, including: Identify the metering device in the first health state; Collect the first operating condition corresponding to the metering device in the first health state; Extract the operating parameters from the first operating condition; Set the corresponding weight values according to the operating condition parameters; The first degradation score is obtained by weighting the coupling degradation features using the weight values.
[0046] The first operating condition targets equipment with severe degradation, focusing on collecting external stress parameters it experiences in the current operating environment. These parameters include: daily load fluctuation coefficient (the ratio of daily maximum load to daily average load, reflecting the impact of impulsive loads), total harmonic distortion (THD) of system voltage (reflecting the degree of power quality pollution), three-phase imbalance (reflecting off-center load conditions), average fluctuation of ambient temperature over the past 30 days (the average difference between the daily maximum and minimum temperatures), and vibration intensity at the equipment installation point (mm / s, for equipment near large motors or transformers). These parameters are collected every 5 minutes by intelligent monitoring terminals deployed on the metering equipment side, aggregated into daily averages by edge computing nodes, and then the moving average over the past 90 days is taken as the representative value for the current operating condition to eliminate random fluctuations.
[0047] Weighting coefficients were assigned to the above operating parameters (determined based on the analytic hierarchy process): load fluctuation coefficient weight 0.30, harmonic distortion rate weight 0.25, three-phase imbalance weight 0.15, temperature fluctuation amplitude weight 0.15, and vibration intensity weight 0.15. Each original parameter was first mapped to the [0,1] interval using deviation standardization (Min-Max), then multiplied by its corresponding weight and summed to obtain the operating condition influence factor α (value interval [0,1]).
[0048] The first degradation score is calculated as Score1 = Dcoupling × (1 + α). The physical meaning of this formula is that for severely degraded equipment, harsh operating conditions will further accelerate its performance failure process, thus imposing an amplified penalty on top of the degradation characteristics. For example, for a piece of equipment with D_coupling = 0.62, and α = 0.45 after standardization and weighted calculation of its various operating parameters, then Score_1 = 0.62 × 1.45 = 0.899. This score not only reflects the degree of physical degradation of the equipment itself but also incorporates the accelerating effect of external environmental factors, providing a more realistic quantitative basis for decommissioning decisions.
[0049] In step S17, generating a first retirement instruction if the first degradation score is greater than a preset scoring threshold, and generating a second retirement instruction if the first degradation score is less than or equal to the preset scoring threshold, includes: Obtain the preset scoring threshold; The first degradation score is numerically compared with the preset score threshold; If the first degradation score is greater than the preset score threshold, a first retirement command is generated; If the first degradation score is less than or equal to the preset score threshold, a second retirement instruction is generated.
[0050] Specifically, the preset scoring threshold is determined based on the economic analysis model of the power grid company's metering asset management center. This model comprehensively calculates the following cost items: (a) the product of the expected failure probability of the equipment continuing operation for one year (obtained by fitting historical failure data using a Weibull distribution) and the average repair cost per failure; (b) the expected cost of correcting electricity billing errors caused by excessive metering errors; and (c) the procurement and installation costs of replacing the equipment with new components. When the expected total cost of continued operation exceeds a certain proportion (usually 1.2 times) of the replacement cost, the corresponding degradation score is the threshold. After multiple rounds of financial simulation calculations, this threshold is set to 0.75 in this embodiment.
[0051] If Score_1 > 0.75, a first decommissioning instruction (i.e., an "emergency forced decommissioning" instruction) is generated. This instruction, in addition to carrying the equipment asset ID, also includes a detailed decommissioning execution plan, including suggested dismantling and recycling channels, safety precautions during dismantling (such as capacitor discharge requirements), and the latest completion time for decommissioning (within 48 hours of the instruction being issued). This instruction is simultaneously pushed to the maintenance team's handheld terminal and the scrap warehouse in the warehouse management system via message middleware.
[0052] If Score_1 ≤ 0.75, a second retirement instruction (i.e., "observation buffer retirement" instruction) is generated. This instruction requires the equipment to continue operating but shortens the verification cycle, reducing the original one-year verification cycle to three months. The instruction also clearly marks key monitoring items (such as "automatically compare errors every 10 days") and generates a performance trend early warning report, which is sent to the regional metrology specialist for manual review.
[0053] In step S18, for the metering equipment in the second health state, the corresponding second operating condition is obtained, and the coupling degradation characteristics are weighted and calculated according to the second operating condition to obtain a second degradation score, including: Identify the metering device in the second health state; Collect the second operating condition corresponding to the metering equipment in the second health state; Extract the operating parameters from the second operating condition; Set the corresponding weight value according to the operating parameters in the second operating condition; The coupling degradation features are weighted and calculated using the weight values corresponding to the second operating condition to obtain a second degradation score.
[0054] Specifically, the second operating condition is designed for equipment with less degradation, focusing on characterizing the rate at which current load and actual operating load consume the remaining lifespan. Collected parameters include: daily average load rate (the ratio of actual current to rated current), daily average operating time (actual daily power-on time), the combined environmental temperature and humidity stress over the past 30 days (using a weighted product model of temperature and humidity), and the voltage qualification rate of the transformer substation where the equipment is located (reflecting power supply quality). These parameters are uploaded every 15 minutes through the electricity consumption information collection system, and daily statistics are obtained through streaming calculations on a big data platform. The weighted moving average over the past 60 days (with higher weights for more recent dates and a decay coefficient of 0.95) is then used as the current representative value for the second operating condition.
[0055] The weights of the assigned operating parameters are as follows: load rate weight 0.35 (excessive load accelerates thermal aging of electronic components), operating time weight 0.25 (longer cumulative operating time leads to more significant mechanical wear), combined temperature and humidity stress weight 0.25 (insulation degradation is aggravated under the combined effects of high temperature and high humidity), and voltage qualification rate weight 0.15 (long-term undervoltage or overvoltage affects the lifespan of the switching power supply). After Min-Max normalization, a weighted sum is obtained to obtain the operating condition influence factor β (value range [0,1]).
[0056] The formula for calculating the second degradation score is: Score2 = Dcoupling × (1 (β×0.6). Unlike the first health state, this uses a "negative weighting" logic: for equipment with minor degradation, if the current operating conditions are very mild (small β), the equipment can continue to serve for a long time, and the score is appropriately reduced to indicate low retirement urgency; if the operating conditions are severe (large β), the score is relatively higher, indicating that attention is needed. The coefficient 0.6 is an empirical attenuation factor used to control the maximum correction of the score by the operating conditions to no more than 60%. For example, for a certain equipment, D_coupling=0.43, β=0.30, then Score_2=0.43×(1-0.18)=0.3526. This score achieves fine-grained ranking among equipment within the same health state category, effectively distinguishing between the two sub-scenarios of "healthy but in poor operating conditions" and "healthy and in mild operating conditions".
[0057] In step S19, generating a third retirement instruction if the second degradation score is greater than a preset scoring threshold, and generating a fourth retirement instruction if the second degradation score is less than or equal to a preset scoring threshold, includes: Obtain the initial operating data of the target device; By cleaning the data and removing outliers from the initial running data, standard running data is obtained. Based on the standard operating data, the random forest algorithm is used to extract time series features to obtain the equipment attenuation. Calculate the ratio of the equipment attenuation to the initial reference value to determine the equipment loss. The equipment status parameters are obtained by classifying and categorizing the equipment based on the amount of equipment loss. The support vector machine algorithm is used to perform regression prediction on the device state variables to obtain an initial degradation score; The initial degradation score is processed by a moving average to obtain the second degradation score; The second degradation score is compared with the preset score threshold; If the second degradation score is greater than the preset score threshold, a third retirement instruction is generated; If the second degradation score is less than or equal to the preset score threshold, a fourth retirement instruction is generated.
[0058] Specifically, the second operating condition is designed for equipment with less degradation, focusing on characterizing the rate at which current load and actual operating load consume the remaining lifespan. Collected parameters include: daily average load rate (the ratio of actual current to rated current), daily average operating time (actual daily power-on time), the combined environmental temperature and humidity stress over the past 30 days (using a weighted product model of temperature and humidity), and the voltage qualification rate of the transformer substation where the equipment is located (reflecting power supply quality). These parameters are uploaded every 15 minutes through the electricity consumption information collection system, and daily statistics are obtained through streaming calculations on a big data platform. The weighted moving average over the past 60 days (with higher weights for more recent dates and a decay coefficient of 0.95) is then used as the current representative value for the second operating condition.
[0059] The weights of the assigned operating parameters are as follows: load rate weight 0.35 (excessive load accelerates thermal aging of electronic components), operating time weight 0.25 (longer cumulative operating time leads to more significant mechanical wear), combined temperature and humidity stress weight 0.25 (insulation degradation is aggravated under the combined effects of high temperature and high humidity), and voltage qualification rate weight 0.15 (long-term undervoltage or overvoltage affects the lifespan of the switching power supply). After Min-Max normalization, a weighted sum is obtained to obtain the operating condition influence factor β (value range [0,1]).
[0060] The formula for calculating the second degradation score is: Score2 = Dcoupling × (1 (β×0.6). Unlike the first health state, this uses a "negative weighting" logic: for equipment with minor degradation, if the current operating conditions are very mild (small β), the equipment can continue to serve for a long time, and the score is appropriately reduced to indicate low retirement urgency; if the operating conditions are severe (large β), the score is relatively higher, indicating that attention is needed. The coefficient 0.6 is an empirical attenuation factor used to control the maximum correction of the score by the operating conditions to no more than 60%. For example, for a certain equipment, D_coupling=0.43, β=0.30, then Score_2=0.43×(1-0.18)=0.3526. This score achieves fine-grained ranking among equipment within the same health state category, effectively distinguishing between the two sub-scenarios of "healthy but in poor operating conditions" and "healthy and in mild operating conditions".
[0061] For example, in the automated decommissioning assessment process of a battery energy storage system, the central control server first calculates the second degradation score by calling a weighted summation algorithm to determine the rate of increase in internal resistance and the rate of capacity decay of the battery pack. For instance, the second degradation score of the current battery pack No. 1 is 85.6 points. The system then compares this score with a preset score threshold of 80.0 points stored in the configuration library. Since 85.6 points is greater than the preset score threshold of 80.0 points, the system determines that the internal structure of the battery pack has suffered irreversible and severe damage, and immediately triggers the third decommissioning instruction generation module. This module runs a material recycling assessment algorithm, calculates the recycling value index of the metal elements in the battery pack to be 92, and then automatically generates a third decommissioning instruction containing a deep dismantling path, hazardous waste treatment code, and material recycling list. This instruction is then directly sent to the programmable logic controller of the automated dismantling line via a data interface. Conversely, if the second degradation score of battery pack No. 2 obtained by the same algorithm is 72.4 points, and the system confirms after comparison and analysis that it is less than or equal to the preset score threshold of 80.0 points, it is determined that it still has tiered utilization value, and the fourth retirement instruction generation module is triggered. This module immediately runs the degradation matching algorithm, combines the current load demand data of the backup power supply of the communication base station, and automatically generates a fourth retirement instruction containing step-down operation parameters, reconfigured topology structure, and target application scenario code, and synchronously transmits it to the manufacturing execution system of the flexible reconfiguration workshop to start the subsequent automated reconfiguration process.
[0062] In another aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement as follows: Figures 1 to 3 The steps of the described method. For more details, please refer to and combine with the foregoing descriptions. Figures 1 to 3 The description of that will not be repeated here.
[0063] Implementing the embodiments of the present invention has the following beneficial effects: This invention discloses a method and storage medium for the full lifecycle decommissioning assessment and reuse of metering equipment. Addressing the core problems in the decommissioning decision-making process for metering equipment, such as the difficulty in effectively integrating multi-source heterogeneous data, inaccurate assessment of degradation status, and difficulty in accurately determining the decommissioning timing, this invention proposes a complete solution. First, this invention deeply integrates environmental and performance characteristics during equipment operation using a random forest algorithm. Then, it uses principal component analysis to reduce the dimensionality of the fused multidimensional feature matrix, effectively eliminating redundant information and extracting key features. Subsequently, it uses a long short-term memory network model to capture temporal correlation information, accurately identifying the coupled degradation characteristics of the equipment, and performing differentiated weighted scoring based on the actual operating conditions of the equipment. Finally, it generates corresponding decommissioning instructions based on different health states and degradation scores. Through this organic combination of multi-level feature extraction and intelligent assessment mechanisms, this invention achieves scientific prediction and accurate decision-making regarding the decommissioning timing of metering equipment, effectively improving the intelligence level of equipment lifecycle management and providing reliable technical support for the rational decommissioning and reuse of equipment.
[0064] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 Units that specify functions within one or more boxes.
[0065] The above description is merely a preferred embodiment of the present invention and should not be construed as limiting the scope of the invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.
Claims
1. A metrology equipment full life cycle decommissioning assessment and re-use method, characterized in that, Includes the following steps: Step S11: Obtain multi-source heterogeneous operation data of the metering device during its life cycle, and extract a first environmental feature and a first performance feature from the multi-source heterogeneous operation data, wherein the multi-source heterogeneous operation data covers the monitoring records of the metering device at different operating stages. Step S12: Use the random forest algorithm to perform feature fusion processing on the first environmental feature and the first performance feature, and combine the total number of features with the corresponding weight coefficients to obtain a fused feature set; Step S13: Construct a multidimensional feature matrix based on the fused feature set, perform dimensionality reduction on the multidimensional feature matrix using principal component analysis, extract the main components based on the number of retained principal components, and obtain the target feature vector. Step S14: Process the target feature vector through a long short-term memory network model, extract the temporal correlation information in the target feature vector, capture the degradation pattern in the time series data, and determine the corresponding coupling degradation features; Step S15: If the coupling degradation feature is greater than a preset degradation threshold, the metering device is determined to be in a first healthy state; if the coupling degradation feature is less than or equal to the preset degradation threshold, the metering device is determined to be in a second healthy state. Step S16: For the metering device in the first health state, obtain the corresponding first operating condition, and perform a weighted calculation on the coupling degradation features based on the first operating condition to obtain a first degradation score; Step S17: If the first degradation score is greater than a preset score threshold, generate a first retirement instruction; if the first degradation score is less than or equal to the preset score threshold, generate a second retirement instruction. Step S18: For the metering equipment in the second health state, obtain the corresponding second operating condition, and perform a weighted calculation on the coupling degradation features according to the second operating condition to obtain a second degradation score; Step S19: If the second degradation score is greater than a preset score threshold, a third retirement instruction is generated; if the second degradation score is less than or equal to the preset score threshold, a fourth retirement instruction is generated.
2. The method of claim 1, wherein, Step S12 further includes: Determine the total number of features of the first environmental feature and the first performance feature; For each of the first environmental feature and the first performance feature, a corresponding weighting coefficient is determined; Call the random forest algorithm function, and input the first environmental feature, the first performance feature, and the corresponding weight coefficients; Weighted fusion calculation is performed by combining the corresponding weight coefficients; A comprehensive set of fusion features is generated based on the results of the weighted fusion calculation.
3. The method of claim 2, wherein, Step S13 further includes: The features in the fused feature set are arranged and combined to construct a multidimensional feature matrix; For the aforementioned multidimensional feature matrix, calculate the corresponding covariance matrix; Extract the eigenvalues of the covariance matrix and the corresponding principal component vectors; Sort the components according to their eigenvalues to determine the number of principal components to retain; Based on the principal component vectors corresponding to the number of retained principal components, the multidimensional feature matrix is mapped and transformed to obtain the target feature vector.
4. The method of claim 3, wherein, Step S14 further includes: Input the target feature vector into the long short-term memory network model; Combining the hidden state from the previous time step with the input at the current time step, matrix multiplication is performed between the input layer weight matrix and the hidden layer weight matrix; By introducing a bias term and an activation function to process the results of matrix multiplication, temporal correlation information is extracted from the target feature vector. The corresponding coupling degradation features are determined based on the temporal correlation information.
5. The method of claim 4, wherein, Step S15 further includes: Obtain the preset degradation threshold; The coupling degradation feature is compared with the preset degradation threshold in terms of numerical value; When the coupling degradation characteristic is greater than the preset degradation threshold, the metering device is determined to be in a first health state. If the coupling degradation characteristic is less than or equal to the preset degradation threshold, the metering device is determined to be in a second health state.
6. The method of claim 5, wherein, Step S16 further includes: Identify the metering device in the first health state; Collect the first operating condition corresponding to the metering device in the first health state; Extract the operating parameters from the first operating condition; Set the corresponding weight values according to the operating condition parameters; The first degradation score is obtained by weighting the coupling degradation features using the weight values.
7. The method of claim 6, wherein, Step S17 further includes: Obtain the preset scoring threshold; The first degradation score is numerically compared with the preset score threshold; If the first degradation score is greater than the preset score threshold, a first retirement command is generated; If the first degradation score is less than or equal to the preset score threshold, a second retirement instruction is generated.
8. The method of claim 7, wherein, Step S18 further includes: Identify the metering device in the second health state; Collect the second operating condition corresponding to the metering equipment in the second health state; Extract the operating parameters from the second operating condition; Set the corresponding weight value according to the operating parameters in the second operating condition; The coupling degradation features are weighted and calculated using the weight values corresponding to the second operating condition to obtain a second degradation score.
9. The method of claim 8, wherein, Step S19 further includes: Obtain the initial operating data of the target device; By cleaning the data and removing outliers from the initial running data, standard running data is obtained. Based on the standard operating data, the random forest algorithm is used to extract time series features to obtain the equipment attenuation. Calculate the ratio of the equipment attenuation to the initial reference value to determine the equipment loss. The equipment status parameters are obtained by classifying and categorizing the equipment based on the amount of equipment loss. The support vector machine algorithm is used to perform regression prediction on the device state variables to obtain an initial degradation score; The initial degradation score is processed by a moving average to obtain the second degradation score; The second degradation score is compared with the preset score threshold; If the second degradation score is greater than the preset score threshold, a third retirement instruction is generated; If the second degradation score is less than or equal to the preset score threshold, a fourth retirement instruction is generated.
10. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 9.
