A method of predicting the number of smart meter replacements

Abstract

The present application relates to a kind of method for predicting the replacement quantity of intelligent electric meter, belong to electric meter replacement quantity prediction technical field.The present application is first by the improved grey model of the improved data based on weighted average weakening buffer operator correction to the replacement quantity of intelligent electric meter is predicted, further optimization is carried out to predicted value using improved Markov model, the improved grey-Markov prediction value after improvement is obtained, the weight of the influence factor of the replacement quantity of intelligent electric meter is calculated by entropy weight method to obtain final coefficient, finally final coefficient is added to the improved grey-Markov model, and the final predicted value is obtained.The present application uses the improved grey-Markov model, solves the problem of low precision of small sample prediction caused by insufficient sample, data statistics etc., improves the prediction precision of the replacement quantity model of intelligent electric meter, effectively reduces the operation and maintenance cost of intelligent electric meter, greatly improves the operation safety of smart grid.

Description

Technical Field

[0001] This invention relates to a method for predicting the number of smart meters to be replaced, belonging to the technical field of meter replacement quantity prediction. Background Technology

[0002] With the rapid development of smart grid technology, smart meters have become a new technological favorite. Smart meters, with a microprocessor at their core and composed of hardware and software, can collect and store large amounts of data in real time. As a core device for power distribution management, compared to conventional meters, they integrate a microprocessor, input and output interfaces, communication devices, and a display screen. They have automatic metering and billing functions, data transmission capabilities, and are increasingly serving as the control center for intelligent residences. Power distribution companies can use smart meters to detect faults more quickly and respond promptly to various measures for the safe control and effective management of the power network. Therefore, accurately predicting the number of smart meters to be replaced is of great significance for effectively reducing the operating and maintenance costs of smart meters and further improving the operational safety of the smart grid.

[0003] Currently in China, the main method for identifying and addressing meter malfunctions in smart meters is through regular inspections. However, these regular inspections are extremely costly in terms of manpower, resources, and time, and require inspectors to possess solid professional knowledge; otherwise, they are prone to overlooking or misjudging problems. Therefore, while regular inspections can accurately determine the number of smart meters to be replaced, they are inefficient and susceptible to subjective judgment, leading to significant errors in the estimated replacement numbers. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting the number of smart meters to be replaced, so as to solve the problems of low efficiency and large error in the current process of determining the number of smart meters to be replaced.

[0005] To solve the above-mentioned technical problems, the present invention provides a method for predicting the number of smart meters to be replaced. The prediction method includes the following steps:

[0006] 1) Obtain the historical number of smart meters replaced and the environmental data on the impact of the same period on the lifespan of the meters, and determine the score of each environmental data as an influencing factor. The environmental data includes at least two of temperature, humidity, electrical stress and salt spray. The historical smart meter replacement data is processed by a gray model to obtain the predicted value of the number of smart meters replaced.

[0007] 2) Optimize the predicted values ​​using a Markov model;

[0008] 3) The entropy weight method is used to process the scores of each influencing factor in the historical environmental data of the same period, calculate the influence weight of the environmental factors of smart meters on the number of replacements, and then obtain the weight coefficient.

[0009] 4) Adjust the predicted values ​​after Markov model optimization using weighting coefficients to obtain the final prediction results.

[0010] This invention first uses a grey model to predict the number of smart meters to be replaced, then optimizes the prediction results using a Markov model, next calculates the weights of environmental factors on smart meter components using the entropy weight method, and finally adjusts the grey-Markov model using weight coefficients to make the prediction results more accurate. This invention improves the accuracy of smart meter replacement predictions by incorporating the entropy weight method into the grey-Markov model to calculate the weights of environmental factors on the number of smart meters to be replaced.

[0011] Further, the process of determining the weight coefficient in step 3) is as follows: the obtained influencing factor scores are used as the initial matrix, and the obtained initial matrix is ​​standardized; the entropy value of each influencing factor is calculated using the standardized initial matrix, the difference coefficient of each influencing factor is calculated based on the entropy value, and then the weight of each influencing factor is calculated based on the difference coefficient; according to the obtained weight of each influencing factor, the scores of each influencing factor are weighted and summed, and the calculated result is the weight coefficient σ.

[0012] Entropy weighting is an objective method for determining weights. Compared with subjective methods such as the analytic hierarchy process (AHP), it has a certain degree of accuracy. Furthermore, the weights determined by this method can be modified, which determines its high adaptability. This invention uses entropy weighting to calculate the weights of the environment's influence on smart meter components, making the prediction results more accurate.

[0013] Furthermore, the gray model is an improved model based on the weighted average weakened buffer operator. The processing procedure of this model is as follows: the number of historical smart meter replacements is used as the original data sequence, and a level ratio test is performed on the original data sequence. For sequences that do not meet the level ratio test, the weighted average weakened buffer operator is used to correct the original data sequence until the corrected sequence meets the level ratio test. The gray model is then used to predict the sequences that meet the level ratio test.

[0014] Furthermore, the process of prediction using the grey model is as follows: the sequences that satisfy the level ratio test are accumulated to obtain the accumulated generated sequence. The whitening differential equation of the grey model is constructed using the accumulated generated sequence. After solving the equation, the predicted value of the accumulated generated sequence is obtained. Finally, the predicted value of the original data sequence is obtained by restoring the original data sequence.

[0015] Grey prediction has the advantages of simple operation, easy verification, and no consideration of distribution patterns and trends. Traditional grey prediction model GM(1,1) rarely performs level ratio verification on the original data sequence during prediction, resulting in a large deviation between the prediction results and the actual results. This invention uses a weighted average weakened buffer operator to correct the data, correct the original data and perform level ratio verification, and correct the sequences that do not meet the level ratio verification, so that the prediction results are closer to the actual results.

[0016] Furthermore, the Markov model is an improved model based on the sum of the probabilities of multiple predicted states and taking the maximum value. The processing procedure of this model is as follows: select the known states of the n consecutive times closest to the time to be predicted, and use the corresponding k-step state transition probability matrix to obtain the state probability vector of the time to be predicted, and form its probability matrix; sum the probabilities corresponding to multiple states, and determine the state of the time to be predicted by the maximum probability criterion; then take the predicted state as the known state, and use this state as the basis to predict the number of smart meters to be removed and replaced in the next step.

[0017] Furthermore, the optimization results in step 2) for:

[0018]

[0019] In the formula, Let e ​​represent the estimated number of smart meters to be removed and replaced in the kth month as predicted by the grey model. 1j and e 2j State E j The lower and upper limits.

[0020] In the actual prediction of the number of smart meters to be disassembled and replaced, it was found that the state transition probability matrix calculated by the frequency of a certain state is often unstable. This invention adopts a method of accumulating the state probabilities of multiple predictions and taking the maximum value to find the state at the time of the test, thereby reducing the error generated by the traditional Markov method for transition prediction and making the prediction results more accurate.

[0021] Furthermore, the final prediction result is:

[0022]

[0023] In the formula, σ is the weighting coefficient. To represent the estimated number of smart meters to be removed and replaced in the kth month as predicted by the grey model, e 1j and e 2j State E j The lower and upper limits. Attached Figure Description

[0024] Figure 1This is a flowchart of the method for predicting the number of smart meters to be replaced according to the present invention;

[0025] Figure 2 This is a flowchart of the modeling process for the improved gray model in this invention;

[0026] Figure 3 This is a comparison diagram of the error results before and after the improvement of the gray model in the embodiment of the present invention;

[0027] Figure 4 This is a comparison of the prediction errors of the improved grey model and the improved grey-Markov model in this embodiment of the invention. Detailed Implementation

[0028] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0029] Implementation examples of methods for predicting the number of smart meters to be replaced

[0030] This invention first obtains the raw data sequence of historical smart meter replacement numbers, and then uses an improved grey model to predict the replacement number of smart meters through a series ratio test. Next, the predicted value is tested for error. Then, an improved Markov model is used to further optimize the predicted value, resulting in an improved grey-Markov predicted value. The weights of factors influencing the smart meter replacement number are calculated using the entropy weight method, and the weights of these factors are fused and processed to obtain the final coefficients. Finally, these final coefficients are added to the improved grey-Markov model to obtain the final predicted value. The implementation process of this method is as follows: Figure 1 As shown, the following will illustrate this with specific examples.

[0031] 1. Obtain the original data sequence by acquiring the historical number of smart meters replaced, and then use a grey model to obtain the predicted value of the number of smart meters replaced.

[0032] Grey prediction GM(1,1) performs well in predicting simple systems, mainly because: 1) it generally only requires four data points, solving the problems of limited historical data, low sequence integrity, and low reliability; 2) it can fully exploit the essence of the system using differential equations with high accuracy; and 3) it can generate sequences with strong regularity from irregular original data. Therefore, grey prediction is characterized by its ease of computation, ease of verification, and lack of consideration for distribution patterns and trends. Traditional grey prediction models like GM(1,1) rarely perform level ratio tests on the original data sequences during prediction, leading to significant deviations between predicted and actual results. This invention proposes a method using a weighted average weakened buffer operator to correct the data, perform level ratio tests on the original data, and correct sequences that do not meet the level ratio test until the corrected sequence meets the test. The improved grey model modeling process is as follows: Figure 2 As shown.

[0033] (1) Obtain the raw data sequence X of the number of smart meters removed and replaced each month. (0) ,Right now

[0034] X (0) ={x (0) (1),x (0) (2),...,x (0) (n)}

[0035] In the formula, x (0) (k) represents the number of smart meters that were removed and replaced in the kth month.

[0036] This embodiment uses the monthly number of smart meters removed and replaced by a company as the raw data for analysis, and establishes a grey model GM(1,1) for smart meter replacement. The number of smart meters removed and replaced from January to December 2021 is shown in Table 1. To facilitate model calculation and verification, the number of smart meter replacements is treated as a continuous time series, with the first 9 months as sample data and the last 3 numbers as test values. An improved grey model is used to predict the sample data, and the accuracy and effectiveness of the model are verified.

[0037] Table 1. Statistics on the Number of Smart Meters Removed and Replaced from January to December 2021

[0038]

[0039]

[0040] (2) For the original data sequence X (0) The expression for the grade ratio test is as follows:

[0041]

[0042] for If σ(k)∈(e -2 / (n+1) ,e 2 / (n+1) If the result is positive, it means that the original data sequence of the number of smart meters disassembled and replaced has passed the level ratio test.

[0043] (3) If the original data sequence fails the level ratio test, the weighted average weakening buffer operator D is introduced to revise the original data sequence, and X is set to 0. (0) D={x (0) (1)d,x (0) (2)d,…,x (0) (n)d}, to obtain the corrected generated sequence

[0044]

[0045] In the formula, This represents the number of smart meters removed and replaced in month k after correction using the weighted average weakening buffer operator D, and its expression is:

[0046]

[0047] (4) The corrected generated sequence Repeat the grade ratio test. If it still fails, continue making corrections until the grade ratio test passes.

[0048] In this embodiment, the first nine data points of the monthly number of smart meters to be disassembled and replaced are subjected to a grade ratio test. The test results show that the sample does not conform to the grey prediction model. Then, the disassembly and replacement numbers for the first nine months are corrected, and the corrected data meets the grade ratio test.

[0049] (5) The corrected generated sequence Perform an accumulation once to obtain the accumulated sequence X. (1) Its expression is:

[0050] X (1) ={x (1) (1),x (1) (2),...,x (1) (n)}

[0051] In the formula, x (1) (k) represents the number of smart meters to be removed and replaced in the kth month after the initial accumulation, and its expression is:

[0052]

[0053] (6) Construct the cumulative generation sequence X (1) The mean generation sequence Z (1) Its expression is:

[0054] Z (1) ={z (1) (2),z (1) (3),...,z (1) (n)}

[0055] In the formula, z (1) (k) represents the number of smart meters removed and replaced in the k-th month of the average generation sequence, and its expression is:

[0056] z (1) (k)=0.5x (1) (k)+0.5x (1) (k-1)(k=2,3,…n)

[0057] (7) Accumulate to generate sequence X (1)The whitening differential equation for the grey model GM(1,1) is constructed as follows:

[0058]

[0059] In the formula, a is the development gray level of the gray model GM(1,1), and b is the endogenous control gray level of the gray model GM(1,1).

[0060] Least squares estimation ensures that the sum of squared residuals in solving the differential equation is minimized, thus allowing us to find the values ​​of parameters a and b in the differential equation, which are:

[0061] [ab] T = (B T B) -1 B T Y

[0062] In the formula, the expressions for matrix B and matrix Y are as follows:

[0063]

[0064] The solution to the whitening differential equation of the grey model GM(1,1) can be obtained as follows:

[0065]

[0066] In the formula, and These are the estimated values ​​of parameters a and b, respectively.

[0067] (8) After obtaining the predicted values ​​of the accumulated generated sequence, the predicted value sequence of the original data can be obtained by restoring it. for:

[0068]

[0069] In the formula, The estimated number of smart meters to be removed and replaced in the kth month is expressed as:

[0070]

[0071] This is the predicted number of smart meters to be removed and replaced.

[0072] To demonstrate the effectiveness of the improved grey model, an error test was performed on the improved grey model, and it was compared with the traditional grey model to verify its accuracy.

[0073] Predicted number of smart meters to be removed and replaced x its actual number of disassemblies (0) The relative error sequence of (k) is denoted as δ(k), and its expression is:

[0074]

[0075] Based on engineering practice, the relative error of gray prediction is taken to be within 20%, that is, when the relative error sequence satisfies |δ(k)|<0.2, the predicted value of the number of smart meters to be disassembled and replaced can be considered to meet the prediction requirements; otherwise, the predicted value is considered not to meet the prediction requirements.

[0076] To comprehensively improve the prediction accuracy of the grey model GM(1,1), the mean absolute percentage error (MAPE) is used to compare with the predictions of the traditional grey model GM(1,1). The expression for the mean absolute percentage error is:

[0077]

[0078] In this embodiment, the traditional GM(1,1) prediction model and the improved GM(1,1) prediction model are used to predict the number of disassemblies and replacements in the first nine months, and the relative error of the prediction is calculated to facilitate the verification of model accuracy. A comparison of the prediction results before and after the improvement of the grey model is shown in Table 2:

[0079] Table 2 Comparison of prediction results before and after the improvement of the grey model from January to September

[0080]

[0081]

[0082] The mean absolute percentage error (MAPE1) of the traditional grey model GM(1,1) was calculated to be 4.99%, while the mean absolute percentage error (MAPE2) of the improved grey model GM(1,1) was 4.44%. The traditional and improved GM(1,1) prediction models were used to predict two test data points, and the prediction results for October-December before and after the improvement are shown in Table 3.

[0083] Table 3. Comparison of prediction results before and after the improvement of the grey model from October to December.

[0084]

[0085] Comparative analysis of the table above shows that the improved GM(1,1) prediction results are closer to the actual values ​​than the traditional GM(1,1) prediction results. The comparison results of the errors before and after the grey model improvement are as follows: Figure 3 As shown. A comparative analysis of the prediction results before and after the grey model improvement for March-September and October-December is conducted, combining their mean absolute percentage error and... Figure 3As shown by the curve, the improved grey model GM(1,1) has a better prediction effect. Therefore, this invention adopts the improved grey model GM(1,1) as the grey prediction model for the number of smart meters to be disassembled and replaced.

[0086] 2. Optimize the predicted values ​​using a Markov model.

[0087] The transition probability matrix represents the process of transitioning from one state to another in a Markov process. Generally, the transition probabilities between states are uncertain; the frequency of state transitions between samples is usually used to represent the transition probabilities. Let the samples be in state E. i The number of N i , using N ij (k) represents state E i After k steps, the state transitions to E. j The number of transitions, and using Let represent the transition probability at step k. The state transition probability matrix at step k can be expressed as:

[0088]

[0089] In the formula, p ij ≥0, and

[0090] Assume the object to be predicted is at position E. g Given states (g = 1, 2, ..., n) and an initial state probability vector, predicting the state at the next moment only requires observing the transition probabilities in the g-th row of the state transition probability matrix. If the j-th column is the maximum probability in the g-th row, then the next state is highly likely to be from state E. g Transition to state E j .

[0091] In the process of researching the prediction of the number of smart meters to be removed and replaced, it was found that the state transition probability matrix calculated by the frequency of a certain state is often unstable. To reduce the error caused by this method, this invention adopts a method of accumulating the state probabilities of multiple predictions and taking the maximum value to determine the state at the time to be predicted. The known states of the five consecutive times closest to the time to be predicted are selected, and the state probability vector of the time to be predicted is obtained using the corresponding k-step state transition probability matrix, forming its probability matrix. The probabilities corresponding to multiple states are accumulated, and the state at the time to be predicted is determined by the maximum probability criterion. This predicted state is then used as the known state, and the number of smart meters to be removed and replaced is predicted based on this state.

[0092] The relative error sequence δ(k) obtained from the improved grey model prediction is divided into n states, and any state interval is: E i =(e1i ,e 2i ), where e 1i e 2i State E i The lower and upper limits. State division is usually based on the number of samples and the range of prediction error, dividing the relative error into 3 to 5 states.

[0093] Assume the states of the object to be predicted at times 1 to k are E1, E2, ..., E k The steps to predict the state at time k+1 and time k+2 are as follows:

[0094] (1) Using the known states (E) of the 5 consecutive times before time k. k E k-1 E k-2 E k-3 E k-4 By combining this with the state transition probability matrix, the state probability matrix P at time k+1 can be obtained. k+1 Its expression is:

[0095]

[0096] (2) Calculate the state probability matrix P at time k+1. k+1 For each column of probabilities, the state corresponding to the maximum value of the probability sum is the state at time k+1.

[0097] (3) When predicting the state at time k+2, the predicted state at time k+1 is taken as the known state. When predicting the state of the five consecutive times before time k+1, the state at time k+1 is replaced with the state at time k-4. This process is repeated to predict the state sequence of subsequent times.

[0098] (4) The predicted value of the grey-Markov model is related to the state of its residuals. Generally, the midpoint of the state interval is taken as the correction value. Let the predicted object move to E in the next step. j The state then determines the predicted value of the improved grey-Markov model. for:

[0099]

[0100] In the formula, e 1j and e 2j State E j The lower and upper limits.

[0101] In this embodiment, the predicted number of smart meter replacements for nine months is obtained by improving the grey model GM(1,1), and the states are divided according to the relative error to obtain the states required for Markov prediction. Since the data for April fluctuates significantly, resulting in a much higher prediction deviation than other months, the error for April is replaced by the maximum error of other months when dividing the Markov states. The prediction data error is divided into equal intervals, and the relative error sequence predicted by the improved GM(1,1) is divided into four states, as shown in Table 4.

[0102] Table 4 Relative Error Sequence Status Table

[0103]

[0104] The number of smart meters replaced from March to October was divided into different status intervals, as shown in Table 5:

[0105] Table 5. Classification of Smart Meter Replacement Quantities from March to October

[0106] state <![CDATA[E4]]> <![CDATA[E4]]> <![CDATA[E3]]> <![CDATA[E1]]> <![CDATA[E1]]> <![CDATA[E2]]> <![CDATA[E2]]>

[0107] Calculate the state transition probability matrix P from step 1 to step 5. (1) P (2) P (3) P (4) P (5) Based on the improved grey-Markov model, the predicted values ​​for times 10 to 12 were obtained, and the model accuracy verification index was calculated. The results are summarized in Table 6:

[0108] Table 6. Accuracy test index values ​​of the improved grey-Markov model

[0109]

[0110] The comparison of prediction errors between the two models (i.e., the improved grey model and the improved grey-Markov model) is plotted as follows: Figure 4 As shown. Analysis Figure 4 Furthermore, the mean absolute percentage error (MAPE2) of the improved grey-Markov model was calculated to be 2.11%. Comparing the calculation results of the two models, it was found that the mean absolute percentage error of the improved grey model was smaller than that of the improved grey model, indicating that the improved grey-Markov model had a better prediction effect.

[0111] Compare the prediction results of the grey model before and after improvement from March to September with the prediction results of the improved grey-Markov model from March to December, and analyze their mean absolute percentage error and... Figure 4The curves shown indicate that the improved grey-Markov model has a smaller relative prediction error than the improved grey model, meaning the prediction results are more accurate. Therefore, the improved grey-Markov model can be used to accurately predict the number of smart meters to be replaced.

[0112] 3. Obtain the impact of historical environmental data on meter failures, use the entropy weight method to calculate the weight of environmental factors on the number of replacements for smart meters, and perform data fusion processing on the weights of influencing factors to obtain weight coefficients.

[0113] Entropy weighting, also known as objective weighting, is a mathematical method for judging the dispersion of an indicator. It is a quantitative evaluation method, more objective than some qualitative methods, calculating the weight of each indicator based on the amount of information it reflects. In information theory, entropy is a measure of uncertainty; the greater the information content, the lower the degree of disorder, and therefore the less uncertainty and the lower the entropy value. Conversely, the smaller the information content, the higher the degree of disorder, the greater the uncertainty and the greater the entropy value. In practical applications, entropy weighting can yield relatively objective indicator weights. This invention incorporates entropy weighting into a grey-Markov model to calculate the weight of the environment's influence on smart meter components, thereby making the prediction results more accurate. The main calculation steps of entropy weighting are as follows:

[0114] (1) Suppose a comprehensive evaluation problem involves n indicators, which are composed of n samples. The collected data are processed to obtain the initial matrix A as follows:

[0115]

[0116] That is, X = |X ij | m×n (0≤i≤m,0≤j≤n), where X ij This represents the value of the j-th evaluation index for the i-th sample.

[0117] (2) Standardize the data:

[0118]

[0119] In the formula, i = 1, 2, ..., n; j = 1, 2, ..., m.

[0120] (3) Calculate the ratio of each indicator under each scheme, which is the proportion of the j-th indicator in the i-th scheme. This is essentially to calculate the magnitude of the variation of that indicator.

[0121]

[0122] (4) Calculate the entropy value of the j-th factor:

[0123]

[0124] In the formula, k>0, E i is the natural logarithm, E j ≥0; the constant k is related to the number of samples m, generally k=1 / lnm, so 0≤E i ≤1.

[0125] (5) Calculate the difference coefficient G of the j-th factor j :

[0126] G j =1-E j

[0127] (6) Determine the weight w j :

[0128]

[0129] (7) After performing data fusion processing on the weights of influencing factors, the final coefficient σ is obtained.

[0130] In this embodiment, based on years of collection of damaged electric meters and data from accelerated aging experiments, 6 types of components that are relatively prone to failure are selected in a targeted manner, and a set of data is compiled by combining statistics of the working environment of electric meters. Four environmental factors that have the most obvious impact on component life are extracted, and the impact of each environmental factor on component failure rate is evaluated. The evaluation process is as follows:

[0131] (1) Count the damage probability of different components in smart meters in the same region and the same environment.

[0132] (2) Calculate the differences of temperature, humidity, ultraviolet light (light intensity received by the electric meter), and salt fog in different regional environments, which are x1,x2,…,x n .

[0133] (3) Calculate the differences of failure probability of the same component in different regional environments, which are y1,y2,…,y n .

[0134] (4) Calculate the sample correlation coefficient:

[0135]

[0136] In the formula, X i % is a statistic representing the linear correlation strength and direction between two random variables, it is unitless and -1<X i %<1. Finally, the life impact score table of different environments on different components is obtained as shown in Table 7:

[0137] Table 7 Score Table of Impact of Environment on Component Life

[0138]

[0139] Next, based on the principle of entropy weighting and related mathematical formulas, the weights of each influencing factor were calculated; then, reliability and validity analyses were performed on the data. The calculation results show that the data is reasonable, more practical, and meets the research requirements.

[0140] Continue to calculate X according to the formula. j Entropy value E j Coefficient of difference G j Sum of weights W j After organizing the calculation results, the weight data of the four influencing factors were tabulated to obtain the weights and ranking results of the influencing factors, as shown in Table 8:

[0141] Table 8. Weights and Ranking of Influencing Factors

[0142]

[0143] Since climate data varies from month to month, to make the predictions more accurate, it is necessary to use weather data from the past three years for each month, combined with the impact of different weather conditions on the lifespan of smart meters, to obtain a correction factor σ for the number of smart meters to be replaced each month. i Where i represents different months, and the values ​​of σ are shown in Table 9:

[0144] Table 9. Values ​​of σ for different months

[0145]

[0146]

[0147] 4. The weighting coefficients are added to the improved grey-Markov model to obtain the final prediction result. The final improved grey-Markov model after incorporating the entropy weighting method is as follows:

[0148]

[0149] In the formula, σ is the weighting coefficient.

Claims

1. A method for predicting the number of smart meters to be replaced, characterized in that, The method includes the following steps: 1) Obtain the historical number of smart meter replacements and the environmental data on the impact of the historical same period on the lifespan of the meters, and determine the score of each environmental data as an influencing factor. The environmental data includes at least two of temperature, humidity, electrical stress and salt spray. The historical smart meter replacement data is processed by a grey model to obtain the predicted value of the number of smart meter replacements. 2) Optimize the predicted values ​​using a Markov model; 3) Use the obtained influencing factor scores as the initial matrix, standardize the initial matrix, calculate the entropy value of each influencing factor using the standardized initial matrix, calculate the difference coefficient of each influencing factor based on the entropy value, and then calculate the weight of each influencing factor based on the difference coefficient; calculate the weighted sum of the scores of each influencing factor according to the obtained weights of each influencing factor, and the calculated result is the weight coefficient. 4) Adjust the predicted values ​​after the Markov model is optimized using the weighting coefficients to obtain the final prediction results.

2. The method for predicting the number of smart meters to be replaced according to claim 1, characterized in that, The grey model in step 1) is an improved model based on the weighted average weakened buffer operator. The processing procedure of this model is as follows: the number of historical smart meter replacements is used as the original data sequence. The original data sequence is subjected to a level ratio test. For sequences that do not meet the level ratio test, the weighted average weakened buffer operator is used to correct the original data sequence until the corrected sequence meets the level ratio test. The grey model is used to predict the sequences that meet the level ratio test.

3. The method for predicting the number of smart meters to be replaced according to claim 2, characterized in that, The process of prediction using the grey model is as follows: the sequences that satisfy the level ratio test are accumulated to obtain the accumulated generated sequence. The whitening differential equation of the grey model is constructed using the accumulated generated sequence. After solving the equation, the predicted value of the accumulated generated sequence is obtained. Finally, the predicted value of the original data sequence is obtained by restoring the original data sequence.

4. The method for predicting the number of smart meters to be replaced according to claim 3, characterized in that, The whitening differential equation is solved using the least squares method.

5. The method for predicting the number of smart meters to be replaced according to claim 1, characterized in that, The Markov model described is an improved model based on the sum of the probabilities of multiple predicted states and taking the maximum value. The processing procedure of this model is as follows: select the known states of the n consecutive times closest to the time to be predicted, and use the corresponding k-step state transition probability matrix to obtain the state probability vector of the time to be predicted, and form its probability matrix; sum the probabilities corresponding to multiple states, and determine the state of the time to be predicted by the maximum probability criterion; then take the predicted state as the known state, and use this state as the basis to predict the number of smart meters to be removed and replaced in the next step.

6. The method for predicting the number of smart meters to be replaced according to claim 1, characterized in that, The optimization result obtained in step 2) for: In the formula, Let e ​​represent the estimated number of smart meters to be removed and replaced in the kth month as predicted by the grey model. 1j and e 2j State E j The lower and upper limits.

7. The method for predicting the number of smart meters to be replaced according to claim 1, characterized in that, The final prediction result is as follows: In the formula, These are the weighting coefficients. To represent the estimated number of smart meters to be removed and replaced in the kth month as predicted by the grey model, e 1j and e 2j State E j The lower and upper limits.

Images