Intelligent electric meter state evaluation and replacement model

Abstract

The invention relates to the technical field of electric power, and discloses an intelligent electric meter state evaluation and replacement model, an intelligent electric meter state evaluation system is established, the system realizes automation and intelligence of a whole chain from data acquisition to decision output, and powerful support is provided for fine management of the electric power industry; the data acquisition layer is used for continuously acquiring key operation data including current, voltage, power, meter code, clock and the like from the intelligent electric energy meter in real time by utilizing a remote communication technology; the data processing layer is used for cleaning, fusing and preprocessing the original data through a series of algorithms to ensure the accuracy, integrity and consistency of the data, and the model of the scheme is based on the data collected by the intelligent electric meter and alarm events and depends on the big data processing and analysis capability of an electric energy data platform; and real-time monitoring, abnormity diagnosis and decision-making replacement of the state of the electric meter are realized.

Description

Technical Field

[0001] The present invention belongs to the field of electric power technology, and in particular relates to a smart meter status evaluation and replacement model. Background Art

[0002] With the rapid development of smart grids, smart energy meters are indispensable metering equipment in power grids. Their accuracy and stability are directly related to the fairness of energy metering and the safety of power grid operation. However, due to the influence of factors such as natural environment, service life, quality of internal components and human interference, the energy meter may be inaccurate or abnormal during operation. In the existing technology, the status evaluation and replacement of electricity meters often rely on manual inspections and regular replacement strategies. This method is not only inefficient, but also difficult to detect and deal with meter failures or abnormalities in a timely and accurate manner, resulting in waste of power resources and potential safety hazards in electricity use. In addition, the lack of effective meter error calculation and analysis methods makes it impossible to effectively guarantee the metering accuracy. To this end, we propose a smart meter status evaluation and replacement model. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a smart meter status evaluation and replacement model, which effectively solves the above problems.

[0004] To achieve the above objectives, the present invention provides the following technical solutions: a smart meter status evaluation and replacement model, establishing a smart meter status evaluation system, the system realizes the automation and intelligence of the whole chain from data collection to decision output, and provides strong support for the refined management of the power industry;

[0005] Data collection level: Using remote communication technology, key operating data including current, voltage, power, meter code, clock, etc. are obtained from smart electric energy meters in real time and continuously;

[0006] Data processing layer: Through a series of algorithms, the raw data is cleaned, integrated and preprocessed to ensure the accuracy, completeness and consistency of the data;

[0007] Abnormal diagnosis layer: Based on the preset abnormal diagnosis rules, such as meter code flying away, power sudden change, current imbalance, in-depth analysis of the processed data, rapid identification of possible inaccuracy problems of the energy meter, and classification and marking of abnormal types;

[0008] Error calculation layer: further introduce the error formula to calculate the current error value of the electric energy meter and its confidence interval;

[0009] Decision model layer: A comprehensive decision support system for electricity meter replacement was built, which comprehensively considered multi-dimensional information such as meter aging, abnormal conditions, metering accuracy, and prediction robustness, and used the random forest algorithm to achieve efficient classification and prediction. This model not only significantly improved the intelligence level and accuracy of electricity meter replacement decisions, but also ensured the scientificity and effectiveness of the decision;

[0010] First, list all unique category values ​​of "meter abnormality type" and "meter abnormal event". Assume that "meter abnormality type" has n category values, recorded as C1, C2, ..., C n , “Abnormal events of electric meters” has m category values, denoted as E1, E2, ..., E n , create a coding matrix for each feature; for "meter abnormality type", create an N×n matrix (N is the number of samples), where each row corresponds to a sample and each column corresponds to a category value; for "meter abnormal event", create an N×m matrix; traverse each sample in the data set, and according to the category value of its "meter abnormality type" and "meter abnormal event", set the corresponding position in the corresponding coding matrix to 1, and the remaining positions to 0, that is: if the "meter abnormality type" of the i-th sample is C j , then the i-th row and j-th column of the encoding matrix are set to 1. Similarly, similar operations are performed for "electricity meter abnormal events". The two encoding matrices are merged with other numerical features in the data set to form a new data set. Since then, "electricity meter abnormal type" and "electricity meter abnormal events" have been quantified by one-hot encoding. In addition, the lower limit and upper limit of the confidence interval are taken as two independent features as features.

[0011] Preferably, the method comprises the following steps:

[0012] 1. Implementation Method of Data Collection Layer

[0013] In order to obtain the key operating data of smart energy meters in real time and uninterruptedly, remote communication technologies such as LTE-M, NB-IoT or 5G are used. A remote communication module is installed on each smart energy meter. The module can send key data such as current, voltage, power, meter code, clock, etc. to the data center through the wireless network at a fixed time or according to preset conditions (such as data changes exceeding the threshold). The data center deploys a corresponding receiving system to ensure the complete reception and storage of data, providing a solid data foundation for subsequent analysis;

[0014] 2. Data Processing Layer Implementation

[0015] At the data processing layer, a distributed computing architecture is adopted, and multiple high-performance servers are used to process the collected raw data in parallel. First, the data is cleaned. The data cleaning strategy needs to be particularly cautious, because some seemingly "abnormal" data may be a direct reflection of meter failure or error. Basic data cleaning can be performed, such as processing obvious format errors and non-numeric characters, but it does not involve complex denoising or error data correction. For obvious erroneous data (such as readings that obviously exceed the meter range), reasonable default values ​​can be used to replace them, but these replacement operations should be recorded so that they can be identified and processed in subsequent analysis. Then, data fusion technology is used to integrate data from different sources to form a unified data format to improve the efficiency and accuracy of subsequent analysis. The entire processing process uses automated process control to ensure the quality and consistency of data processing;

[0016] 3. Abnormal Diagnosis Layer Implementation Method

[0017] In the meter abnormality diagnosis layer, a series of meter abnormality diagnosis rules are pre-set, such as meter code flying away (meter code changes abnormally quickly), power mutation (power increases or decreases significantly in a short period of time), current imbalance (three-phase current deviation exceeds the set threshold). These rules are formulated based on the professional knowledge and practical experience of the power industry, with high accuracy and reliability. The processed data is input into the abnormality diagnosis system. The system conducts in-depth analysis of the data according to the preset rules, quickly identifies possible misalignment problems of the energy meter, and classifies and marks the abnormal types. At the same time, the system supports abnormal alarm function. Once an abnormality is found, an alarm message will be sent to the management personnel immediately so that timely measures can be taken;

[0018] 4. Error Calculation Layer Implementation

[0019] At the error calculation layer, a professional error calculation formula is introduced:

[0020]

[0021] The current error value of the electric energy meter is calculated based on the total meter power value, sub-meter power value and total line loss power value of the electric energy meter. In order to evaluate the reliability of the error, the confidence interval of the error is also calculated. The confidence interval can help understand the credibility of the error estimate. For example, if the confidence interval of a meter error is very narrow, it means that the error estimate is very accurate, and you can rely on this error value more confidently when making decisions. On the contrary, if the confidence interval is very wide, it means that there is a large uncertainty in the error estimate, and you need to be more cautious when making decisions to consider the impact of this uncertainty on the results;

[0022] By measuring the total meter electricity, the total sub-meter electricity and the total line loss electricity independently for multiple times (for example, measuring in different time periods or under different conditions), multiple sample values ​​of electricity errors can be obtained. Based on these sample values, the sample mean and sample standard deviation can be calculated, and the confidence interval of the electricity error can be further calculated. In order to calculate the confidence interval of the electricity meter error, and considering that the project allows an electricity error within 5%, the following steps can be followed:

[0023] Perform multiple (assuming n) independent measurements, and record the total meter power (T i ), total electricity consumption of each meter (S i ) and total line loss power (L i ), where i represents the number of measurements;

[0024] For each measurement, the meter error (E i ):

[0025]

[0026] This will result in a series of error sample values ​​E1, E2, ..., E n ;

[0027] Sample mean

[0028]

[0029] The standard equation of the sample (S E ):

[0030]

[0031] Select the confidence level, usually 95% (or other level selected according to needs); at the 95% confidence level, the corresponding z value is z a / 2 =1.96 (from the standard normal distribution table);

[0032] Confidence interval width (ME):

[0033]

[0034] Confidence interval upper and lower limits (LB and UB):

[0035]

[0036]

[0037] 5. Decision Model Layer Implementation Method

[0038] At the decision model level, the present invention deeply integrates big data analysis and intelligent optimization algorithms to construct a comprehensive decision model to scientifically guide the replacement decision of electric energy meters; the model comprehensively considers the age of the meter (the number of years of use is used as an indicator of equipment aging), abnormal types and events (identifying common abnormal types and abnormal events through abnormal diagnosis rules), meter errors (evaluating measurement accuracy) and confidence interval data (ensuring the robustness of prediction results). To achieve this goal, the random forest algorithm is used as the basic classifier because it can effectively process multi-dimensional features and has good generalization ability. After the model training is completed, the F1 score evaluation index is used to comprehensively evaluate the performance and decision accuracy of the model. This process not only improves the intelligence and accuracy of the replacement decision, but also provides solid data support and decision-making basis for the long-term stable operation of the power system;

[0039] For data sets containing categorical features, such as "electricity meter abnormality type" and "electricity meter abnormality event", one-hot encoding is usually used for quantization. One-hot encoding is a technique for converting categorical variables into numerical variables. Simply put, it is achieved by creating a new binary column for each category value. These columns are 1 at the category value of the corresponding sample and 0 at the rest. It can avoid simply comparing category values ​​as numerical values, thereby preventing potential misunderstandings and incorrect analysis.

[0040] Preferably, the data set is further divided into training set, validation set and test set, which is usually done by random sampling to ensure that the distribution of the training set, validation set and test set on the target variable is similar. The algorithm of the model, random forest, is an ensemble learning method that improves prediction performance by building multiple decision trees and outputting their mode (for classification problems) or average (for regression problems);

[0041] Decision trees are the basic building blocks of random forests. When building decision trees, some criteria are usually used to select the best split point to minimize the loss function. For classification problems, commonly used criteria include Gini impurity and information gain;

[0042] Gini impurity: For a set T, its Gini impurity G(T) is defined as:

[0043]

[0044] Where C is the number of categories, p i is the proportion of samples belonging to the i-th class in the set T;

[0045] When splitting nodes, select features and split points that minimize Gini impurity.

[0046] Information gain: For a set T, its entropy H(T) is defined as:

[0047]

[0048] Information gain is the difference between the entropy of the parent node and the weighted average entropy of all child nodes.

[0049] Random forest improves performance by building multiple decision trees and aggregating their prediction results. When building each tree, random forest randomly extracts samples from the original data set with replacement to build a training set for each tree. When selecting the split feature for each node, instead of considering all features, a subset is randomly selected from all features and the best split feature is selected from this subset.

[0050] The output of a random forest is usually the support of each category. These probabilities are calculated by aggregating the predictions of all decision trees. Specifically, for each category, the proportion of predictions of that category in all decision trees is calculated, and the category with the highest proportion is selected as the final prediction result.

[0051] The F1 value is used to evaluate the model because the main focus is on the performance of classification tasks, especially on unbalanced datasets. The F1 value is the harmonic mean of precision and recall, which provides a trade-off between the two to more comprehensively evaluate the classification ability of the model.

[0052] Precision: The proportion of samples that are actually positive among all samples predicted by the model to be positive.

[0053]

[0054] Recall rate: The proportion of samples correctly predicted as positive by the model among all samples that are actually positive.

[0055]

[0056] Among them, P(TP,FP) represents precision, R(TP,FN) represents recall, TP represents true positive examples, FP represents false positive examples, and FN represents false negative examples.

[0057] Calculate the F1 value: The F1 value is the harmonic mean of precision and recall, which is used to comprehensively evaluate the performance of the model. Its calculation formula is:

[0058]

[0059] According to the calculated F1 value, the classification performance of the model is evaluated. A higher F1 value indicates that the model has achieved a better balance between precision and recall. The larger the F1 score, the smaller the classification error and the better the prediction effect of the model. If the performance of the model does not meet the requirements, the model can be optimized by adjusting the parameters of the random forest, performing feature selection, or trying other feature engineering techniques. Cross-validation can also be used to more accurately evaluate model performance and prevent overfitting.

[0060] Preferably, the random forest modeling process mainly includes the following steps:

[0061] Step 1: Extract samples from the original data set using bootstrap sampling to form multiple sub-data sets;

[0062] Step 2: For each sub-dataset, build a decision tree. At each node, randomly select a portion of features for splitting;

[0063] Step 3: Repeat the above steps until the specified number of decision trees are generated.

[0064] When selecting the optimal features for partitioning, the randomly selected feature subsets are different each time, which introduces nonlinearity and enables the formation of a more effective decision boundary in the feature space;

[0065] The representation ability of a single decision tree is limited. Random forest improves the prediction ability of new data by reducing the variance of a single decision tree. The prediction results of multiple decision trees are combined by voting or averaging, which can effectively reduce the variance of the model and improve the stability and accuracy of the model.

[0066] Preferably, the feature importance of the random forest model is usually calculated in the following two ways:

[0067] Average impurity reduction: When building each tree of a random forest, each feature split results in some degree of impurity reduction (e.g., Gini impurity or information gain), and the importance of a feature can be defined as the average of its impurity reduction across all trees;

[0068] Average Drop in Precision: Another way to calculate feature importance is by randomly permuting the feature values ​​and measuring the effect of this permutation on the model accuracy. The importance of a feature can be defined as the average drop in model accuracy when it is randomly permuted.

[0069] Preferably, the confidence interval calculation method for large samples: when n>=30, it is a large sample that conforms to the normal distribution, and the z-statistic is used to calculate, the confidence interval = sample mean +- z*standard error

[0070] The standard error: SE calculation method:

[0071]

[0072] Preferably, the method for calculating confidence intervals with a small sample size: a sample size n<30 is considered a small sample and belongs to the t distribution;

[0073] Use the t-statistic;

[0074] Degrees of freedom: df = n-1 The amount of information that can be freely varied without affecting the given restrictions. Or the amount of independent information that can be obtained when estimating other information;

[0075] The t distribution can be one-sided or two-sided.

[0076] Preferably, the upper limit (or lower limit, depending on the direction of the error) of the confidence interval does not exceed the 5% error range allowed by the project (i.e. |UB|≤0.05 or |LB|≥-0.05, note that the positive and negative nature of the error needs to be considered here), then it can be considered that the error meets the project requirements statistically. If there are strict requirements on the error and the confidence interval is wide, it may be necessary to increase the number of measurements n to reduce the width of the confidence interval, thereby improving the accuracy of the estimate;

[0077] This process requires a high reliance on high-precision measuring equipment to obtain accurate meter values ​​and line loss values, and to combine appropriate calculation methods to accurately calculate the current error value of the meter and its confidence interval; these calculation methods should be based on statistical principles and reliable mathematical models to ensure the accuracy and reliability of error assessment. The error calculation results, confidence intervals and meter age are stored in the database for subsequent replacement decision analysis.

[0078] Compared with the prior art, the present invention has the following beneficial effects:

[0079] 1. The model of the present invention is based on the data and alarm events collected by the smart meter, relying on the big data processing and analysis capabilities of the electric energy data platform, to achieve real-time monitoring of the meter status, abnormal diagnosis and decision replacement. By setting the meter abnormality query rules and error calculation formula, the present invention can automatically perform meter abnormality diagnosis and meter error data calculation, and perform statistical analysis and display, providing powerful decision support for grassroots metering personnel;

[0080] 2. The present invention also supports the generation of meter replacement plans, realizing the automation and intelligence of meter replacement management. The realization of these functional points will greatly improve the efficiency and accuracy of meter management, and provide a strong guarantee for the refined management of power resources;

[0081] 3. In the smart grid management, this technical solution shows significant advantages over the existing technology through remote real-time data collection, accurate data processing, intelligent abnormality diagnosis, precise error calculation and scientific decision analysis;

[0082] 4. The present invention realizes real-time and comprehensive acquisition of data, ensuring the accuracy and consistency of data; improves data processing efficiency through complex algorithms, quickly identifies and solves the problem of inaccuracy of electric energy meters, introduces error formulas to calculate error values, and provides a basis for replacement decisions. Finally, a comprehensive decision-making model is constructed using big data and intelligent algorithms to realize intelligent meter replacement decisions.

[0083] 5. The present invention improves the intelligence level of power grid management, reduces costs, speeds up response, and enhances system reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0085] In the attached picture:

[0086] Figure 1 It is a smart meter status evaluation and replacement model of the present invention;

[0087] Figure 2 This is a diagram of the abnormality diagnosis rules of the electric meter of the present invention;

[0088] Figure 3 This is a random forest model diagram of the present invention. DETAILED DESCRIPTION

[0089] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0090] See also Figure 1-3 , a smart meter status evaluation and replacement model, this system realizes the automation and intelligence of the entire chain from data collection to decision output, providing strong support for the refined management of the power industry.

[0091] First, at the data collection level, remote communication technology is used to obtain key operating data including current, voltage, power, meter code, clock, etc. from smart energy meters in real time and continuously. These data form the basis for subsequent analysis.

[0092] Next, the data processing layer uses a series of complex algorithms to clean, fuse and preprocess the raw data to ensure the accuracy, completeness and consistency of the data. This step is the key to ensure the accuracy of subsequent analysis.

[0093] In the abnormal diagnosis layer, according to the preset abnormal diagnosis rules, such as meter code flying away, sudden change in power, current imbalance, etc., the processed data is deeply analyzed to quickly identify the possible inaccuracy problems of the electricity meter and classify and mark the abnormal types.

[0094] The error calculation layer further introduces the error formula to calculate the current error value of the electricity meter and its confidence interval. This step provides a key basis for subsequent replacement decisions.

[0095] Finally, at the decision model level, a comprehensive decision support system for electricity meter replacement was built. It comprehensively considers multi-dimensional information such as meter aging, abnormal conditions, metering accuracy, and prediction robustness, and uses the random forest algorithm to achieve efficient classification and prediction. This model not only significantly improves the intelligence level and accuracy of electricity meter replacement decisions, but also ensures the scientificity and effectiveness of the decision.

[0096] Data collection level: Using remote communication technology, key operating data including current, voltage, power, meter code, clock, etc. are obtained from smart electric energy meters in real time and continuously;

[0097] Data processing layer: Through a series of algorithms, the raw data is cleaned, integrated and preprocessed to ensure the accuracy, completeness and consistency of the data;

[0098] Abnormal diagnosis layer: Based on the preset abnormal diagnosis rules, such as meter code flying away, power sudden change, current imbalance, in-depth analysis of the processed data, rapid identification of possible inaccuracy problems of the energy meter, and classification and marking of abnormal types;

[0099] Error calculation layer: further introduce the error formula to calculate the current error value of the electric energy meter and its confidence interval;

[0100] Decision model layer: A comprehensive decision support system for electricity meter replacement was built, which comprehensively considered multi-dimensional information such as meter aging, abnormal conditions, metering accuracy, and prediction robustness, and used the random forest algorithm to achieve efficient classification and prediction. This model not only significantly improved the intelligence level and accuracy of electricity meter replacement decisions, but also ensured the scientificity and effectiveness of the decision;

[0101] First, list all unique category values ​​of "meter abnormality type" and "meter abnormal event". Assume that "meter abnormality type" has n category values, recorded as C1, C2, ..., C n, “Abnormal events of electric meters” has m category values, denoted as E1, E2, ..., E n , create a coding matrix for each feature; for "meter abnormality type", create an N×n matrix (N is the number of samples), where each row corresponds to a sample and each column corresponds to a category value; for "meter abnormal event", create an N×m matrix; traverse each sample in the data set, and according to the category value of its "meter abnormality type" and "meter abnormal event", set the corresponding position in the corresponding coding matrix to 1, and the remaining positions to 0, that is: if the "meter abnormality type" of the i-th sample is C j , then the i-th row and j-th column of the encoding matrix are set to 1. Similarly, similar operations are performed for "electricity meter abnormal events". The two encoding matrices are merged with other numerical features in the data set to form a new data set. Since then, "electricity meter abnormal type" and "electricity meter abnormal events" have been quantified by one-hot encoding. In addition, the lower limit and upper limit of the confidence interval are taken as two independent features as features.

[0102] 1. Implementation Method of Data Collection Layer

[0103] In order to obtain the key operating data of smart energy meters in real time and uninterruptedly, remote communication technologies such as LTE-M, NB-IoT or 5G are used. A remote communication module is installed on each smart energy meter. The module can send key data such as current, voltage, power, meter code, clock, etc. to the data center through the wireless network at a fixed time or according to preset conditions (such as data changes exceeding the threshold). The data center deploys a corresponding receiving system to ensure the complete reception and storage of data, providing a solid data foundation for subsequent analysis;

[0104] 2. Data Processing Layer Implementation

[0105] At the data processing layer, a distributed computing architecture is adopted, and multiple high-performance servers are used to process the collected raw data in parallel. First, the data is cleaned. The data cleaning strategy needs to be particularly cautious, because some seemingly "abnormal" data may be a direct reflection of meter failure or error. Basic data cleaning can be performed, such as processing obvious format errors and non-numeric characters, but it does not involve complex denoising or error data correction. For obvious erroneous data (such as readings that obviously exceed the meter range), reasonable default values ​​can be used to replace them, but these replacement operations should be recorded so that they can be identified and processed in subsequent analysis. Then, data fusion technology is used to integrate data from different sources to form a unified data format to improve the efficiency and accuracy of subsequent analysis. The entire processing process uses automated process control to ensure the quality and consistency of data processing;

[0106] 3. Abnormal Diagnosis Layer Implementation Method

[0107] In the meter abnormality diagnosis layer, a series of meter abnormality diagnosis rules are pre-set, such as meter code flying away (meter code changes abnormally quickly), power mutation (power increases or decreases significantly in a short period of time), current imbalance (three-phase current deviation exceeds the set threshold). These rules are formulated based on professional knowledge and practical experience in the power industry and have high accuracy and reliability. Figure 2 The processed data is input into the abnormal diagnosis system. The system conducts in-depth analysis of the data according to preset rules, quickly identifies possible inaccuracy problems of the electric energy meter, and classifies and marks the abnormal types. At the same time, the system supports the abnormal alarm function. Once an abnormality is found, an alarm message will be sent to the management personnel immediately so that timely measures can be taken;

[0108] 4. Error Calculation Layer Implementation

[0109] At the error calculation layer, a professional error calculation formula is introduced:

[0110]

[0111] The current error value of the electric energy meter is calculated based on the total meter power value, sub-meter power value and total line loss power value of the electric energy meter. In order to evaluate the reliability of the error, the confidence interval of the error is also calculated. The confidence interval can help understand the credibility of the error estimate. For example, if the confidence interval of a meter error is very narrow, it means that the error estimate is very accurate, and you can rely on this error value more confidently when making decisions. On the contrary, if the confidence interval is very wide, it means that there is a large uncertainty in the error estimate, and you need to be more cautious when making decisions to consider the impact of this uncertainty on the results;

[0112] By measuring the total meter electricity, the total sub-meter electricity and the total line loss electricity independently for multiple times (for example, measuring in different time periods or under different conditions), multiple sample values ​​of electricity errors can be obtained. Based on these sample values, the sample mean and sample standard deviation can be calculated, and the confidence interval of the electricity error can be further calculated. In order to calculate the confidence interval of the electricity meter error, and considering that the project allows an electricity error within 5%, the following steps can be followed:

[0113] Perform multiple (assuming n) independent measurements, and record the total meter power (T i ), total electricity consumption of each meter (S i ) and total line loss power (L i ), where i represents the number of measurements;

[0114] For each measurement, the meter error (E i ):

[0115]

[0116] This will result in a series of error sample values ​​E1, E2, ..., E n ;

[0117] Sample mean

[0118]

[0119] The standard equation of the sample (S E ):

[0120]

[0121] Select the confidence level, usually 95% (or other level selected according to needs); at the 95% confidence level, the corresponding z value is z a / 2 =1.96 (from the standard normal distribution table);

[0122] Confidence interval width (ME):

[0123]

[0124] Confidence interval upper and lower limits (LB and UB):

[0125]

[0126]

[0127] 5. Decision Model Layer Implementation Method

[0128] At the decision model level, the present invention deeply integrates big data analysis and intelligent optimization algorithms to construct a comprehensive decision model to scientifically guide the replacement decision of electric energy meters; the model comprehensively considers the age of the meter (the number of years of use is used as an indicator of equipment aging), abnormal types and events (identifying common abnormal types and abnormal events through abnormal diagnosis rules), meter errors (evaluating measurement accuracy) and confidence interval data (ensuring the robustness of prediction results). To achieve this goal, the random forest algorithm is used as the basic classifier because it can effectively process multi-dimensional features and has good generalization ability. After the model training is completed, the F1 score evaluation index is used to comprehensively evaluate the performance and decision accuracy of the model. This process not only improves the intelligence and accuracy of the replacement decision, but also provides solid data support and decision-making basis for the long-term stable operation of the power system;

[0129] If the upper limit (or lower limit, depending on the direction of the error) of the confidence interval does not exceed the 5% error range allowed by the project (i.e. |UB|≤0.05 or |LB|≥-0.05, note that the positive and negative nature of the error needs to be considered here), then it can be considered that the error meets the project requirements statistically. If there are strict requirements on the error and the confidence interval is wide, it may be necessary to increase the number of measurements n to reduce the width of the confidence interval, thereby improving the accuracy of the estimate;

[0130] This process requires a high degree of reliance on high-precision measuring equipment to obtain accurate meter values ​​and line loss values, and to combine appropriate calculation methods to accurately calculate the current error value of the meter and its confidence interval; these calculation methods should be based on statistical principles and reliable mathematical models to ensure the accuracy and reliability of error assessment. The error calculation results, confidence intervals and meter age are stored in the database for subsequent replacement decision analysis.

[0131] Small sample confidence interval calculation method: Sample size n < 30 is a small sample and belongs to t distribution;

[0132] Use the t-statistic;

[0133] Degrees of freedom: df = n-1 The amount of information that can be freely varied without affecting the given restrictions. Or the amount of independent information that can be obtained when estimating other information;

[0134] The t distribution can be one-sided or two-sided.

[0135] For data sets containing categorical features, such as "electricity meter abnormality type" and "electricity meter abnormality event", one-hot encoding is usually used for quantization. One-hot encoding is a technique for converting categorical variables into numerical variables. Simply put, it is achieved by creating a new binary column for each category value. These columns are 1 at the category value of the corresponding sample and 0 at the rest. It can avoid simply comparing category values ​​as numerical values, thereby preventing potential misunderstandings and incorrect analysis.

[0136] The dataset is further divided into training set, validation set and test set. This is usually done by random sampling to ensure that the distribution of the training set, validation set and test set on the target variable is similar. The algorithm of the model is random forest, such as Figure 3 As shown, it is an ensemble learning method that improves prediction performance by building multiple decision trees and outputting their mode (for classification problems) or average (for regression problems);

[0137] Decision trees are the basic building blocks of random forests. When building decision trees, some criteria are usually used to select the best split point to minimize the loss function. For classification problems, commonly used criteria include Gini impurity and information gain;

[0138] Gini impurity: For a set T, its Gini impurity G(T) is defined as:

[0139]

[0140] Where C is the number of categories, p i is the proportion of samples belonging to the i-th class in the set T;

[0141] When splitting nodes, select features and split points that minimize Gini impurity.

[0142] Information gain: For a set T, its entropy H(T) is defined as:

[0143]

[0144] Information gain is the difference between the entropy of the parent node and the weighted average entropy of all child nodes.

[0145] Random forest improves performance by building multiple decision trees and aggregating their prediction results. When building each tree, random forest randomly extracts samples from the original data set with replacement to build a training set for each tree. When selecting the split feature for each node, instead of considering all features, a subset is randomly selected from all features and the best split feature is selected from this subset.

[0146] The output of a random forest is usually the support of each category. These probabilities are calculated by aggregating the predictions of all decision trees. Specifically, for each category, the proportion of predictions of that category in all decision trees is calculated, and the category with the highest proportion is selected as the final prediction result.

[0147] The F1 value is used to evaluate the model because the main focus is on the performance of classification tasks, especially on unbalanced datasets. The F1 value is the harmonic mean of precision and recall, which provides a trade-off between the two to more comprehensively evaluate the classification ability of the model.

[0148] Precision: The proportion of samples that are actually positive among all samples predicted by the model to be positive.

[0149]

[0150] Recall rate: The proportion of samples correctly predicted as positive by the model among all samples that are actually positive.

[0151]

[0152] Among them, P(TP,FP) represents precision, R(TP,FN) represents recall, TP represents true positive examples, FP represents false positive examples, and FN represents false negative examples.

[0153] Calculate the F1 value: The F1 value is the harmonic mean of precision and recall, which is used to comprehensively evaluate the performance of the model. Its calculation formula is:

[0154]

[0155] According to the calculated F1 value, the classification performance of the model is evaluated. A higher F1 value indicates that the model has achieved a better balance between precision and recall. The larger the F1 score, the smaller the classification error and the better the prediction effect of the model. If the performance of the model does not meet the requirements, the model can be optimized by adjusting the parameters of the random forest, performing feature selection, or trying other feature engineering techniques. Cross-validation can also be used to more accurately evaluate model performance and prevent overfitting.

[0156] The modeling process of random forest mainly includes the following steps:

[0157] Step 1: Extract samples from the original data set using bootstrap sampling to form multiple sub-data sets;

[0158] Step 2: For each sub-dataset, build a decision tree. At each node, randomly select a portion of features for splitting;

[0159] Step 3: Repeat the above steps until the specified number of decision trees are generated.

[0160] When selecting the optimal features for partitioning, the randomly selected feature subsets are different each time, which introduces nonlinearity and enables the formation of a more effective decision boundary in the feature space;

[0161] The representation ability of a single decision tree is limited. Random forest improves the prediction ability of new data by reducing the variance of a single decision tree. The prediction results of multiple decision trees are combined by voting or averaging, which can effectively reduce the variance of the model and improve the stability and accuracy of the model.

[0162] The feature importance of a random forest model is usually calculated in two ways:

[0163] Average impurity reduction: When building each tree of a random forest, each feature split results in some degree of impurity reduction (e.g., Gini impurity or information gain), and the importance of a feature can be defined as the average of its impurity reduction across all trees;

[0164] Average Drop in Precision: Another way to calculate feature importance is by randomly permuting the feature values ​​and measuring the effect of this permutation on the model accuracy. The importance of a feature can be defined as the average drop in model accuracy when it is randomly permuted.

[0165] Method for calculating confidence intervals for large samples: When n>=30, it is a large sample that conforms to the normal distribution. Use the z-statistic to calculate, confidence interval = sample mean +-z*standard error.

[0166] The standard error: SE calculation method:

[0167]

Claims

1. A smart meter status evaluation and replacement model, characterized by: Establish a smart meter status evaluation system. The system realizes automation and intelligence of the entire chain from data collection to decision output, providing strong support for the refined management of the power industry; Data collection level: Using remote communication technology, key operating data including current, voltage, power, meter code, clock, etc. are obtained from smart electric energy meters in real time and continuously; Data processing layer: Through a series of algorithms, the raw data is cleaned, integrated and preprocessed to ensure the accuracy, completeness and consistency of the data; Abnormal diagnosis layer: Based on the preset abnormal diagnosis rules, such as meter code flying away, power sudden change, current imbalance, in-depth analysis of the processed data, rapid identification of possible inaccuracy problems of the energy meter, and classification and marking of abnormal types; Error calculation layer: further introduce the error formula to calculate the current error value of the electric energy meter and its confidence interval; Decision model layer: A comprehensive decision support system for electricity meter replacement was built, which comprehensively considered multi-dimensional information such as meter aging, abnormal conditions, metering accuracy, and prediction robustness, and used the random forest algorithm to achieve efficient classification and prediction. This model not only significantly improved the intelligence level and accuracy of electricity meter replacement decisions, but also ensured the scientificity and effectiveness of the decision; First, list all unique category values ​​of "meter abnormal type" and "meter abnormal event". Assume that "meter abnormal type" has n category values, recorded as C1, C2, ..., C n , "electricity meter abnormal event" has m category values, recorded as E1, E2, ..., E n , create a coding matrix for each feature; for "electricity meter abnormality type", create an N×n matrix (N is the number of samples), where each row corresponds to a sample and each column corresponds to a category value; for "electricity meter abnormality event", create an N×m matrix; traverse each sample in the data set, and according to the category value of its "electricity meter abnormality type" and "electricity meter abnormality event", set the corresponding position in the corresponding coding matrix to 1, and the remaining positions to 0, that is: if the "electricity meter abnormality type" of the i-th sample is C j , then the i-th row and j-th column of the encoding matrix are set to 1. Similarly, similar operations are performed for "electricity meter abnormal events". The two encoding matrices are merged with other numerical features in the data set to form a new data set. Since then, "electricity meter abnormal type" and "electricity meter abnormal events" have been quantified by one-hot encoding. In addition, the lower limit and upper limit of the confidence interval are taken as two independent features as features.

2. The smart meter status evaluation and replacement model according to claim 1 is characterized in that: The following steps are involved:

1. Implementation Method of Data Collection Layer In order to obtain the key operating data of smart energy meters in real time and uninterruptedly, remote communication technologies such as LTE-M, NB-IoT or 5G are used. A remote communication module is installed on each smart energy meter. The module can send key data such as current, voltage, power, meter code, clock, etc. to the data center through the wireless network at a fixed time or according to preset conditions (such as data changes exceeding the threshold). The data center deploys a corresponding receiving system to ensure the complete reception and storage of data, providing a solid data foundation for subsequent analysis; 2. Data Processing Layer Implementation At the data processing layer, a distributed computing architecture is adopted, and multiple high-performance servers are used to process the collected raw data in parallel. First, the data is cleaned. The data cleaning strategy needs to be particularly cautious, because some seemingly "abnormal" data may be a direct reflection of meter failure or error. Basic data cleaning can be performed, such as processing obvious format errors and non-numeric characters, but it does not involve complex denoising or error data correction. For obvious erroneous data (such as readings that obviously exceed the meter range), reasonable default values ​​can be used to replace them, but these replacement operations should be recorded so that they can be identified and processed in subsequent analysis. Then, data fusion technology is used to integrate data from different sources to form a unified data format to improve the efficiency and accuracy of subsequent analysis. The entire processing process uses automated process control to ensure the quality and consistency of data processing; 3. Abnormal Diagnosis Layer Implementation Method In the meter abnormality diagnosis layer, a series of meter abnormality diagnosis rules are pre-set, such as meter code flying away (meter code changes abnormally quickly), power mutation (power increases or decreases significantly in a short period of time), current imbalance (three-phase current deviation exceeds the set threshold). These rules are formulated based on the professional knowledge and practical experience of the power industry, with high accuracy and reliability. The processed data is input into the abnormality diagnosis system. The system conducts in-depth analysis of the data according to the preset rules, quickly identifies possible misalignment problems of the energy meter, and classifies and marks the abnormal types. At the same time, the system supports abnormal alarm function. Once an abnormality is found, an alarm message will be sent to the management personnel immediately so that timely measures can be taken; 4. Error Calculation Layer Implementation At the error calculation layer, a professional error calculation formula is introduced: The current error value of the electric energy meter is calculated based on the total meter power value, sub-meter power value and total line loss power value of the electric energy meter. In order to evaluate the reliability of the error, the confidence interval of the error is also calculated. The confidence interval can help understand the credibility of the error estimate. For example, if the confidence interval of a meter error is very narrow, it means that the error estimate is very accurate, and you can rely on this error value more confidently when making decisions. On the contrary, if the confidence interval is very wide, it means that there is a large uncertainty in the error estimate, and you need to be more cautious when making decisions to consider the impact of this uncertainty on the results; By measuring the total meter electricity, the total sub-meter electricity and the total line loss electricity independently for multiple times (for example, measuring in different time periods or under different conditions), multiple sample values ​​of electricity errors can be obtained. Based on these sample values, the sample mean and sample standard deviation can be calculated, and the confidence interval of the electricity error can be further calculated. In order to calculate the confidence interval of the electricity meter error, and considering that the project allows an electricity error within 5%, the following steps can be followed: Perform multiple (assuming n) independent measurements, and record the total meter power (T i ), total electricity consumption of each meter (S i ) and total line loss power (L i ), where i represents the number of measurements; For each measurement, the meter error (E i ): This will result in a series of error sample values ​​E1, E2, ..., E n ; Sample mean The standard equation of the sample (S E ): Select the confidence level, usually 95% (or other level selected according to needs); at the 95% confidence level, the corresponding z value is z a / 2 =1.96 (from the standard normal distribution table); Confidence interval width (ME): Confidence interval upper and lower limits (LB and UB):

5. Decision Model Layer Implementation Method At the decision model level, the present invention deeply integrates big data analysis and intelligent optimization algorithms to construct a comprehensive decision model to scientifically guide the replacement decision of electric energy meters; the model comprehensively considers the age of the meter (the number of years of use is used as an indicator of equipment aging), abnormal types and events (identifying common abnormal types and abnormal events through abnormal diagnosis rules), meter errors (evaluating measurement accuracy) and confidence interval data (ensuring the robustness of prediction results). To achieve this goal, the random forest algorithm is used as the basic classifier because it can effectively process multi-dimensional features and has good generalization ability. After the model training is completed, the F1 score evaluation index is used to comprehensively evaluate the performance and decision accuracy of the model. This process not only improves the intelligence and accuracy of the replacement decision, but also provides solid data support and decision-making basis for the long-term stable operation of the power system; For data sets containing categorical features, such as "electricity meter abnormality type" and "electricity meter abnormality event", one-hot encoding is usually used for quantization. One-hot encoding is a technique for converting categorical variables into numerical variables. Simply put, it is achieved by creating a new binary column for each category value. These columns are 1 at the category value of the corresponding sample and 0 at the rest. It can avoid simply comparing category values ​​as numerical values, thereby preventing potential misunderstandings and incorrect analysis.

3. The smart meter status evaluation and replacement model according to claim 2 is characterized in that: The dataset is further divided into training, validation, and test sets. This is usually done by random sampling to ensure that the distribution of the training, validation, and test sets on the target variable is similar. The model algorithm, random forest, is an ensemble learning method that improves prediction performance by building multiple decision trees and outputting their mode (for classification problems) or average (for regression problems). Decision trees are the basic building blocks of random forests. When building decision trees, some criteria are usually used to select the best split point to minimize the loss function. For classification problems, commonly used criteria include Gini impurity and information gain; Gini impurity: For a set T, its Gini impurity G(T) is defined as: Where C is the number of categories, p i is the proportion of samples belonging to the i-th class in the set T; When splitting nodes, select features and split points that minimize Gini impurity. Information gain: For a set T, its entropy H(T) is defined as: Information gain is the difference between the entropy of the parent node and the weighted average entropy of all child nodes. Random forest improves performance by building multiple decision trees and aggregating their prediction results. When building each tree, random forest randomly extracts samples from the original data set with replacement to build a training set for each tree. When selecting the split feature for each node, instead of considering all features, a subset is randomly selected from all features and the best split feature is selected from this subset. The output of a random forest is usually the support of each category. These probabilities are calculated by aggregating the predictions of all decision trees. Specifically, for each category, the proportion of predictions of that category in all decision trees is calculated, and the category with the highest proportion is selected as the final prediction result. The F1 value is used to evaluate the model because the main focus is on the performance of classification tasks, especially on unbalanced datasets. The F1 value is the harmonic mean of precision and recall, which provides a trade-off between the two to more comprehensively evaluate the classification ability of the model. Precision: The proportion of samples that are actually positive among all samples predicted by the model to be positive. Recall rate: The proportion of samples correctly predicted as positive by the model among all samples that are actually positive. Among them, P(TP,FP) represents precision, R(TP,FN) represents recall, TP represents true positive examples, FP represents false positive examples, and FN represents false negative examples. Calculate the F1 value: The F1 value is the harmonic mean of precision and recall, which is used to comprehensively evaluate the performance of the model. Its calculation formula is: According to the calculated F1 value, the classification performance of the model is evaluated. A higher F1 value indicates that the model has achieved a better balance between precision and recall. The larger the F1 score, the smaller the classification error and the better the prediction effect of the model. If the performance of the model does not meet the requirements, the model can be optimized by adjusting the parameters of the random forest, performing feature selection, or trying other feature engineering techniques. Cross-validation can also be used to more accurately evaluate model performance and prevent overfitting.

4. The smart meter status evaluation and replacement model according to claim 1, characterized in that: The modeling process of random forest mainly includes the following steps: Step 1: Extract samples from the original data set using bootstrap sampling to form multiple sub-data sets; Step 2: For each sub-dataset, build a decision tree. At each node, randomly select a portion of features for splitting; Step 3: Repeat the above steps until the specified number of decision trees are generated. When selecting the optimal features for partitioning, the randomly selected feature subsets are different each time, which introduces nonlinearity and enables the formation of a more effective decision boundary in the feature space; The representation ability of a single decision tree is limited. Random forest improves the prediction ability of new data by reducing the variance of a single decision tree. The prediction results of multiple decision trees are combined by voting or averaging, which can effectively reduce the variance of the model and improve the stability and accuracy of the model.

5. The smart meter status evaluation and replacement model according to claim 1 is characterized in that: The feature importance of a random forest model is usually calculated in two ways: Average impurity reduction: When building each tree of a random forest, each feature split results in some degree of impurity reduction (e.g., Gini impurity or information gain), and the importance of a feature can be defined as the average of its impurity reduction across all trees; Average Drop in Precision: Another way to calculate feature importance is by randomly permuting the feature values ​​and measuring the effect of this permutation on the model accuracy. The importance of a feature can be defined as the average drop in model accuracy when it is randomly permuted.

6. The smart meter status evaluation and replacement model according to claim 1 is characterized by: Large sample confidence interval calculation method: When n>=30, it is a large sample that conforms to the normal distribution. Use the z-statistic to calculate. Confidence interval = sample mean +- z*standard error The standard error: SE calculation method:

7. The smart meter status evaluation and replacement model according to claim 1 is characterized by: Small sample confidence interval calculation method: Sample size n < 30 is a small sample and belongs to t distribution; Use the t-statistic; Degrees of freedom: df = n-1 The amount of information that can be freely varied without affecting the given restrictions. Or the amount of independent information that can be obtained when estimating other information; The t distribution can be one-sided or two-sided.

8. The smart meter status evaluation and replacement model according to claim 1 is characterized by: If the upper limit (or lower limit, depending on the direction of the error) of the confidence interval does not exceed the 5% error range allowed by the project (i.e. |UB|≤0.05 or |LB|≥-0.05, note that the positive and negative nature of the error needs to be considered here), then it can be considered that the error meets the project requirements statistically. If there are strict requirements on the error and the confidence interval is wide, it may be necessary to increase the number of measurements n to reduce the width of the confidence interval, thereby improving the accuracy of the estimate; This process requires a high degree of reliance on high-precision measuring equipment to obtain accurate energy meter values ​​and line loss values, and combines appropriate calculation methods to accurately calculate the current error value of the energy meter and its confidence interval; These calculation methods should be based on statistical principles and reliable mathematical models to ensure the accuracy and reliability of error assessment. The error calculation results, confidence intervals and meter age are stored in the database for subsequent replacement decision analysis.

Images