Comprehensive evaluation method and device for performance of power utilization information collection equipment

Abstract

The application discloses a kind of comprehensive evaluation method and device for the performance of electric information collection equipment, and the method steps include: model training stage, the operating state data set of different electric information collection equipment is obtained and feature vector extraction is carried out, and training data set is formed;Training data set is input into neural network classifier and is trained, in the training process, each electric information collection equipment is formed multiple sample pairs, the feature vector of two electric information collection equipment in each sample pair is input into two identical neural networks respectively, and the difference between the output results of two neural networks is used to calculate loss function, and the sorting model is obtained after training;Evaluation stage, the operating state data of the electric information collection equipment to be evaluated is obtained and feature vector extraction is carried out, and is input into sorting model, and the sorting result output of comprehensive evaluation is obtained.The application has the advantages of simple implementation method, low cost, accurate evaluation result and high efficiency.

Description

Technical Field

[0001] This invention relates to the field of electrical information acquisition equipment technology, and in particular to a comprehensive evaluation method and apparatus for the performance of electrical information acquisition equipment. Background Technology

[0002] Traditional equipment reliability assurance primarily relies on on-site inspections and post-failure replacements. On-site inspections, based on accumulated experience, involve personnel periodically checking and verifying metering and data acquisition equipment according to prescribed maintenance schedules. However, due to the lack of integrated diagnostic equipment, this method often fails to pinpoint the cause of anomalies and is highly inefficient, depending heavily on expert expertise. Post-failure replacement, on the other hand, forces the replacement of metering and data acquisition equipment after a failure, but this can have immeasurable impacts on power supply reliability and the integrity of electricity data. If the performance of each electricity data acquisition device could be accurately evaluated and ranked in real time—that is, if the performance of different devices could be differentiated in order of priority—then each device could be monitored according to its performance ranking. This would allow for timely prediction of fault conditions before equipment failure occurs, preventing equipment malfunctions. Therefore, comprehensive performance evaluation and ranking of electricity data acquisition equipment is of great significance for fault prediction, operational status management, and real-time monitoring of online data acquisition equipment.

[0003] For performance monitoring of electricity consumption information collection equipment, existing technologies typically employ on-site debugging tools or systems. Regarding transmission channels, this usually involves assessing noise and attenuation based on transmission rates and comparisons with normal communication metrics. However, these tools and systems can only monitor the functionality of individual communication modules; they cannot accurately analyze the carrier frequency of wired or wireless signals used for electricity consumption information collection in real time, nor can they monitor electromagnetic radiation and attenuation changes on power line or wireless channels. Therefore, a comprehensive evaluation of the electricity consumption information collection equipment's performance is impossible. If traditional classifiers are used directly for performance evaluation, the large volume and high dimensionality of the operational monitoring data make it difficult to directly input this data into a traditional classifier, hindering comprehensive evaluation and ranking. Summary of the Invention

[0004] The technical problem to be solved by this invention is: in view of the technical problems existing in the prior art, this invention provides a comprehensive evaluation method and device for the performance of electricity information collection equipment that is simple to implement, low in cost, accurate in evaluation results and highly efficient.

[0005] To solve the above-mentioned technical problems, the technical solution proposed by this invention is as follows:

[0006] Compared with the prior art, the advantages of the present invention are as follows:

[0007] A comprehensive evaluation method for the performance of electricity consumption information collection equipment, comprising the following steps:

[0008] During the model training phase, the operating status datasets of different electricity consumption information collection devices are acquired and feature vectors are extracted to form a training dataset. The training dataset is then input into a neural network classifier for training. During the training process, each electricity consumption information collection device is paired up to form multiple sample pairs. The feature vectors of the two electricity consumption information collection devices in each sample pair are respectively input into two identical neural networks, and the loss function is calculated using the difference between the outputs of the two neural networks. After training, a ranking model is obtained.

[0009] During the evaluation phase, the operating status data of the electricity consumption information collection equipment to be evaluated is acquired and feature vectors are extracted. The feature vectors of the electricity consumption information collection equipment to be evaluated are input into the ranking model to obtain the ranking result of the comprehensive evaluation.

[0010] Furthermore, the feature parameters in the feature vector include the mean time between failures (MTBF) T. m Average operating current value I w Average bit error rate C r Average frequency offset of data transmission D t Stray radiation maximum R l Modulation center frequency M f Phase deviation maximum value D l , frequency deviation maximum value f l Average output power P m The number of terminals X whose connection success rate with the electrical information collection equipment exceeds a preset threshold within a specified time period can be any one or more of the following:

[0011] Furthermore, the construction of the training dataset includes:

[0012] Calculate the operating status score of each electricity consumption information collection device based on the operating status data of each device;

[0013] Each electricity consumption information collection device is paired up to form multiple sample pairs. A label value is set for each sample pair based on the difference between the operating status scores of the two electricity consumption information collection devices in each sample pair, so as to mark the performance ranking relationship between the two electricity consumption information collection devices.

[0014] Furthermore, the operating status score of each electricity consumption information collection device is calculated using a linear function, which is:

[0015] g(S h )=U(Sh ) T

[0016] Where g() represents a linear function, U represents the coefficient vector of the linear function, and S h Indicates the running status.

[0017] Furthermore, inputting the training dataset into the neural network classifier for training includes:

[0018] Traverse the sample pairs (Q) formed by the pairwise pairing of each electricity consumption information collection device. i Q j ), where Q i Q j These represent two different electricity consumption information collection devices;

[0019] Electricity consumption information collection device Q i and various electricity information collection devices Q j eigenvector S i and S j The inputs are fed into two identical neural networks, and the outputs of the two neural networks are obtained respectively;

[0020] Calculate the difference O between the outputs of the two neural networks. i -O j ;

[0021] According to the difference O i -O j Calculate the loss function L ij And establish the training optimization objective function based on the calculated loss functions;

[0022] Two neural networks are trained based on the optimization objective function to obtain the final ranking model.

[0023] Furthermore, the loss function during training is calculated according to the following formula:

[0024]

[0025] Among them, O i -O j To integrate electricity consumption information collection equipment O i Electricity consumption information collection equipment O j The difference between the outputs of two identical neural networks obtained after inputting the feature vectors into them, L ij This indicates that the electricity consumption information collection device O i Electricity consumption information collection equipment O j The feature vectors correspond to the calculated loss function.

[0026] Furthermore, the optimization objective function is established according to the following formula:

[0027] Min∑L ij

[0028] Here, Min represents the operation of finding the minimum value.

[0029] Furthermore, when training the two neural networks according to the optimized objective function, the gradient descent method is used to train the weight parameters w of the two neural networks, where the learning rate λ ij The formula for adaptive adjustment is:

[0030]

[0031] The gradient of the loss function with respect to the weight parameter w in the ranking is calculated as follows:

[0032]

[0033] Among them, L ij This indicates that the electricity consumption information collection device O i Electricity consumption information collection equipment O j The feature vectors correspond to the calculated loss function.

[0034] A computer device includes a processor and a memory, the memory being used to store a computer program, and the processor being used to execute the computer program to perform the method described above.

[0035] A computer-readable storage medium storing a computer program that, when executed, implements the method described above.

[0036] 1. This invention combines a neural network classifier to achieve a comprehensive performance evaluation of electricity information collection equipment. At the same time, it calculates the difference between the outputs of two identical neural networks as the evaluation ranking score. The loss function used in the training process is calculated based on the difference between the outputs of the two identical neural networks, so that the ranking result and the loss function have consistent characteristics. This effectively ensures the accuracy of the comprehensive performance evaluation of electricity information collection equipment, while greatly reducing the complexity of the evaluation implementation and the amount of data processing required, and effectively improving the evaluation efficiency.

[0037] 2. In the training process of the ranking model, the present invention further employs gradient descent to train two neural networks. This results in the ranking score of samples ranked higher being improved and the ranking score of samples ranked lower being reduced, based on the calculation results of the loss function. This further improves the evaluation accuracy and makes the results as close as possible to the actual working conditions of the various electro-brainwashing data collection devices. Attached Figure Description

[0038] Figure 1 This is a schematic diagram illustrating the implementation process of the comprehensive evaluation method for the performance of the electricity information collection equipment in this embodiment.

[0039] Figure 2 This is a detailed flowchart illustrating the comprehensive evaluation of the performance of the electricity information collection equipment in a specific application embodiment of the present invention.

[0040] Figure 3 This is a schematic diagram of the Epoch loss curve of the model training obtained in a specific application embodiment of the present invention. Detailed Implementation

[0041] The present invention will be further described below with reference to the accompanying drawings and specific preferred embodiments, but this does not limit the scope of protection of the present invention.

[0042] like Figure 1 As shown, the steps of the comprehensive evaluation method for the performance of electricity information collection equipment in this embodiment include:

[0043] S01. During the model training phase, the operating status datasets of different electricity consumption information collection devices are obtained and feature vectors are extracted to construct a training dataset. The training dataset is then input into a neural network classifier for training. During the training process, each electricity consumption information collection device is paired up to form multiple sample pairs. The feature vectors of the two electricity consumption information collection devices in each sample pair are input into two identical neural networks one-to-one, and the loss function is calculated using the difference between the outputs of the two neural networks. After training, the ranking model is obtained.

[0044] S02. During the evaluation phase, the operating status data of the electricity consumption information collection equipment to be evaluated is obtained and feature vectors are extracted. The feature vectors of the electricity consumption information collection equipment to be evaluated are input into the ranking model to obtain the ranking result of the comprehensive evaluation.

[0045] In a specific application embodiment, the model training stage adopts an offline method. That is, firstly, information on the electricity consumption information collection devices used for training is collected, and the feature vectors of the electricity consumption information collection devices are calculated. Then, a comprehensive evaluation ranking training dataset of the performance of the electricity consumption information collection devices is constructed, and a neural network classifier is used to train the ranking model on the training dataset. When the information of the electricity consumption information collection devices to be evaluated is obtained, the feature vector of the electricity consumption information collection devices to be evaluated is calculated, and it is input into the trained ranking model, so as to quickly and accurately obtain the comprehensive evaluation ranking result of the performance of the electricity consumption information collection devices to be evaluated.

[0046] In this embodiment, the feature parameters in the feature vector include the mean time between failures (MTBF) T. m Average operating current value I w Average bit error rate C rAverage frequency offset of data transmission D t Stray radiation maximum R l Modulation center frequency M f Phase deviation maximum value D l , frequency deviation maximum value f l Average output power P m The number of terminals X that successfully connect to the electricity information collection device within a specified time period exceeds a preset threshold, etc., that is, the feature vector S of the electricity information collection device = [T m ,I w C r D t ,R l M f D l ,f l ,P m [X]. The feature vectors mentioned above are constructed by selecting 10 representative operational information quantities to build the feature vectors of the electricity information collection equipment. This not only effectively characterizes the performance of the electricity information collection equipment but also reduces the data dimensionality. Of course, the specific selection of feature parameters can be made according to actual needs.

[0047] In this embodiment, the construction of the training dataset includes:

[0048] S101. Calculate the operating status score of each electricity consumption information collection device based on the operating status data of each device;

[0049] S102. Pair each electricity consumption information collection device into multiple sample pairs, and set a label value for each sample pair according to the difference between the operating status scores of the two electricity consumption information collection devices in each sample pair, so as to mark the performance ranking relationship between the two electricity consumption information collection devices.

[0050] In a specific application embodiment, Q i and Q j All represent the electricity consumption information collected by the equipment used for training. The tag value setting rule is: when Q i Its performance is better than Q j Q i The overall evaluation ranking in Q j Previously, the label value of this sample set was +1; when Q... i Performance and Q j Same, i.e., Q i The overall evaluation ranking and Q j When they are the same, the label value of the sample set is 0; when Q... j Its performance is better than Q i Q j The overall evaluation ranking in Q iPreviously, the label value for this sample set was -1. Specifically, this was achieved by comparing Q... i and Q j The running status score, if g(S) Qi )-g(S Qj If the sample pair (Q) is greater than or equal to the first preset threshold (e.g., 0.05), then the sample pair (Q) is considered valid. i Q j The label value of g(S) is +1; if g(S) Qi )-g(S Qj If the sample pair (Q) is less than or equal to the second preset threshold (e.g., -0.05), then the sample pair (Q) is considered valid. i Q j The label value of ) is -1; if the second preset threshold (e.g., -0.05) ≤ g(S) Qi )-g(S Qj If the sample pair (Q) is less than or equal to the first preset threshold (e.g., 0.05), then the sample pair (Q) is considered valid. i Q j The label value is 0. The specific values ​​of the above label values ​​and the first and second preset thresholds can be configured according to actual needs.

[0051] In this embodiment, a linear function is specifically used to calculate the operating status score of each electricity consumption information collection device, which can be expressed as:

[0052] g(S h )=U(S h ) T (1)

[0053] Where g() represents the coefficient vector of the linear function, S h This indicates the running status. The above-mentioned linear function g() can specifically take the form of Y = a*x1 + b*x2 + c*x3....

[0054] In this embodiment, inputting the training dataset into the neural network classifier for training includes:

[0055] S111. Traverse the sample pairs (Q) consisting of pairs of each electricity consumption information collection device. i Q j ), where Q i Q j These represent two different electricity consumption information collection devices;

[0056] S112. Install the electricity information collection device Q i and various electricity information collection devices Q j eigenvector S i and S j The inputs are fed into two identical neural networks, and the outputs of the two neural networks are obtained respectively;

[0057] S113. Calculate the difference O between the outputs of the two neural networks. i -O j ;

[0058] S114. Based on the difference O i -O j Calculate the loss function L ij And establish the training optimization objective function based on the calculated loss functions;

[0059] S115. Train two neural networks according to the optimization objective function to obtain the final ranking model.

[0060] The loss function during the above training process is calculated according to the following formula:

[0061]

[0062] Among them, O i -O j To integrate electricity consumption information collection equipment O i Electricity consumption information collection equipment O j The difference between the outputs of two identical neural networks obtained after inputting the feature vectors into them, L ij This indicates that the electricity consumption information collection device O i Electricity consumption information collection equipment O j The feature vectors correspond to the calculated loss function.

[0063] The above optimization objective function is specifically established according to the following formula:

[0064] Min∑L ij (3)

[0065] Here, Min represents the operation of finding the minimum value.

[0066] In step S105 above, when training the two neural networks according to the objective function, the gradient descent method is specifically used to train the weight parameters w of the two neural networks, where the learning rate λ ij The specific formula for adaptive adjustment is as follows:

[0067]

[0068] In this embodiment, the gradient calculation formula of the loss function with respect to the weight parameter w in the ranking is as follows:

[0069]

[0070] Among them, L ij This indicates that the electricity consumption information collection device O i Electricity consumption information collection equipment O jThe feature vectors correspond to the calculated loss function.

[0071] This embodiment automatically adjusts the value of the weight parameter w during the training process based on the gradient calculation result of the loss function, thereby increasing the ranking score of samples ranked higher and decreasing the ranking score of samples ranked lower, which can further improve the ranking evaluation accuracy.

[0072] In a specific application embodiment, a comprehensive evaluation ranking result of the performance of the electricity consumption information collection equipment to be evaluated is obtained. When the result value is +1, it indicates that the electricity consumption information collection equipment Q is being evaluated. i Its performance is better than Q j When the result value is -1, it indicates that the electricity consumption information collection device Q to be evaluated is... j Its performance is better than Q i When the result value is 0, it indicates that the electricity consumption information collection device Q to be evaluated is... i With Q j The performance is the same.

[0073] Electricity consumption information collection equipment generates a large amount of complex data, typically high-dimensional arrays, which cannot be directly used as input vectors for clustering models or classifiers. This invention combines a neural network classifier to achieve a comprehensive performance evaluation of electricity consumption information collection equipment. It uses a neural network classifier to train a ranking model and calculates the difference between the outputs of two identical neural networks as the ranking score. The loss function used during training is calculated based on this difference, ensuring consistency between the ranking result and the loss function. When O... i -O j When >0, L ij The value of the loss function is very small and close to 0. That is, when the ranking of the i-th sample is higher than that of the j-th sample, the score of the i-th sample is also higher than that of the j-th sample, and the value of the loss function is very small. Conversely, if the evaluation ranking score does not match the sample ranking, the larger the score difference, the larger the value of the loss function. This can effectively ensure the accuracy of the comprehensive performance evaluation of the electricity information collection equipment, while greatly reducing the complexity of the evaluation implementation and the amount of data processing required, and effectively improving the evaluation efficiency.

[0074] like Figure 2 As shown, the detailed steps for implementing a comprehensive performance evaluation of electricity information collection equipment using the above method in a specific application embodiment of the present invention are as follows:

[0075] Step S01. Data Acquisition

[0076] First, the operational status information of 500 sets of electricity information collection devices used for training was collected using wide-range time-domain and frequency-domain online monitoring technology. This involved coupled monitoring of power line signals and spectral characteristics of wireless signals without interrupting service transmission, as well as testing and statistical analysis of frequency band utilization and occupancy. Performance monitoring of communication characteristics was achieved, including output signal level, out-of-band interference level, power spectral density, radiated interference limits, carrier power, data transmission frequency offset, spurious radiation limits, modulation spectrum, switching spectrum, phase deviation, frequency deviation, output power, output carrier power, carrier frequency error, transmission frequency offset, spurious radio frequency components, and startup time. Time-domain sampling detection technology was used to monitor the status of weak current interfaces, AC analog quantities, RS-485 interface load capacity, interface driving capability, and signal level establishment time. Optical performance monitoring technology was used to monitor the physical layer transmission status of EPON optical power and wavelength. A power analyzer was used to monitor the average operating current and average output power; and the number of terminals with a daily connection success rate exceeding 80% with the electricity information collection devices was counted.

[0077] Furthermore, the feature vector of the electricity consumption information collection equipment is calculated:

[0078] S = [T] m ,I w C r D t ,R l M f D l ,f l ,P m [X]

[0079] Among them, T m Mean time between failures (MTBF); I w The average operating current value; C r D is the average bit error rate; t The mean frequency offset of the number of transmissions; R l M is the maximum value of stray radiation. f For modulation center frequency; D l f is the maximum value of the phase deviation; l P represents the maximum frequency deviation. m X represents the average output power; X is the number of terminals with a daily connection success rate of over 80% to the electricity information acquisition equipment.

[0080] Then, a comprehensive evaluation and ranking training dataset for the performance of electricity information collection equipment is constructed:

[0081] Step S101: Calculate the operating status score of the collected electricity consumption information acquisition equipment used for training using a linear function:

[0082] g(Sh )=(3,0.5,0.5,0.5,1,0.5,1,2,0.5,2)(S h ) T

[0083] The running status score calculated using a linear function is used only as a label for training and may not reflect the actual situation.

[0084] Step 102: Construct the collected electricity consumption information acquisition devices used for training into a paired sample set, i.e., (Q i Q j ), where Q i and Q j All of these represent the electricity consumption information collected by the equipment used for training.

[0085] Step 103: Set the labels to three categories, with a label value of +1 representing Q. i Its performance is better than Q j Q i The overall evaluation ranking in Q j Previously; a label value of 0 indicates Q. i Performance and Q j Same, i.e., Q i The overall evaluation ranking and Q j Same; a label value of -1 indicates Q. j Its performance is better than Q i Q j The overall evaluation ranking in Q i Before;

[0086] Step 104: Compare Q i and Q j The running status score, if g(S) Qi )-g(S Qj )≥0.05, sample pairs (Q i Q j The label value of g(S) is +1; if g(S) Qi )-g(S Qj )≤-0.05, sample pair (Q i Q j The label value of ) is -1; if -0.05 ≤ g(S) Qi )-g(S Qj )≤-0.05, sample pair (Q i Q j The label value for ) is 0. The percentages of the three labels are shown in Table 1.

[0087] Table 1: Proportion of Labels

[0088]

[0089] Step S105. Train the ranking model using a neural network classifier:

[0090] Step S151: Use two identical three-layer fully connected neural networks, with 10 neurons in the input layer (consistent with the dimension of the feature vector), 128 neurons in the first hidden layer, and 32 neurons in the second hidden layer, to establish a sorting and training model.

[0091] Step S152, iterate through all device sample pairs (Q) i Q j ), will device Q i and equipment Q j eigenvector S i and S j The inputs are respectively fed into the two identical three-layer fully connected neural networks mentioned above;

[0092] Step S153, put device Q i and equipment Q j eigenvector S i and S j The inputs are fed into two identical neural networks respectively;

[0093] Step S154: Calculate the difference between the outputs of two identical neural networks as the evaluation ranking score, i.e., O. i -O j ;

[0094] Step S155, establish the loss function for the training process:

[0095]

[0096] Step S:156, Establish the optimization objective function for training:

[0097] Min∑L ij

[0098] Here, Min represents the operation of finding the minimum value.

[0099] Step S157: Train two neural networks using gradient descent, where the learning rate λ ij Adaptive adjustment, the calculation formula is:

[0100]

[0101] The gradient of the loss function with respect to the parameter w in the ranking is calculated as follows:

[0102]

[0103] Through the above steps, the value of parameter w in the ranking process is automatically adjusted based on the gradient calculation result of the loss function during training. The ranking score of samples ranked higher is increased, while the ranking score of samples ranked lower is decreased. The Epoch loss function value during model training is as follows: Figure 3 As shown.

[0104] Finally, the feature vectors of the electricity consumption information collection devices to be evaluated are input into the trained ranking model to obtain the comprehensive evaluation ranking results of the performance of the electricity consumption information collection devices. The statistical results using the ranking model are shown in Table 2. The number of correct results in the table indicates that the model's judgment result is consistent with the label. If they are inconsistent, they are considered as errors. For sample pairs with a label of 0, they are all included in the number of incorrect sample pairs.

[0105] Table 2: Statistical results using the ranking model

[0106]

[0107] This embodiment also provides a computer device, including a processor and a memory, the memory being used to store a computer program, characterized in that the processor is used to execute the computer program to perform the method as described above.

[0108] This embodiment also provides a computer-readable storage medium storing a computer program, which, when executed, implements the method described above.

[0109] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should fall within the protection scope of the present invention.

Claims

1. A comprehensive evaluation method for the performance of electricity information collection equipment, characterized by the following steps: include: During the model training phase, the operating status datasets of different electricity consumption information collection devices are acquired and feature vectors are extracted to construct the training dataset. The training dataset is input into a neural network classifier for training. During the training process, each electricity consumption information collection device is paired up to form multiple sample pairs. The feature vectors of the two electricity consumption information collection devices in each sample pair are respectively input into two identical neural networks. The loss function is calculated using the difference between the output results of the two neural networks. After training, a ranking model is obtained. During the evaluation phase, the operating status data of the electricity consumption information collection equipment to be evaluated is obtained and feature vectors are extracted. The feature vectors of the electricity consumption information collection equipment to be evaluated are input into the ranking model to obtain the ranking result of the comprehensive evaluation. The construction of the training dataset includes: Calculate the operating status score of each electricity consumption information collection device based on the operating status data of each device; Each electricity consumption information collection device is paired up to form multiple sample pairs. A label value is set for each sample pair according to the difference between the operating status score values ​​of the two electricity consumption information collection devices in each sample pair, so as to mark the performance ranking relationship between the two electricity consumption information collection devices. The operating status score of each electricity consumption information collection device is calculated using a linear function, which is: in, This represents the coefficient vector of a linear function. Indicates the running status.

2. The comprehensive evaluation method for the performance of electricity information collection equipment according to claim 1, characterized in that, The feature vector includes feature parameters such as mean time between failures (MTBF). Average operating current value Average bit error rate Average frequency deviation of data transmission Stray radiation maximum Modulation center frequency Maximum phase deviation Maximum frequency deviation Average output power The number of terminals whose connection success rate with the electrical information collection equipment exceeds a preset threshold within a specified time period. Any one or more of the following.

3. The comprehensive evaluation method for the performance of electricity information collection equipment according to any one of claims 1 to 2, characterized in that, The process of inputting the training dataset into a neural network classifier for training includes: Traverse the sample pairs consisting of each pair of electricity consumption information collection device. Q i , Q j ),in Q i , Q j These represent two different electricity consumption information collection devices; Electricity consumption information collection equipment Q i and various electricity consumption information collection devices Q j eigenvectors i and j The inputs are fed into two identical neural networks, and the outputs of the two neural networks are obtained respectively; Calculate the difference between the outputs of two neural networks i - j ; Based on the difference i - j Calculate the loss function And establish the training optimization objective function based on the calculated loss functions; Two neural networks are trained based on the optimization objective function to obtain the final ranking model.

4. The comprehensive evaluation method for the performance of electricity information collection equipment according to claim 3, characterized in that, The loss function during training is calculated according to the following formula: in, i - j To collect electricity information equipment i Electricity consumption information collection equipment j The difference between the outputs of two identical neural networks obtained after inputting the feature vectors of the network. This indicates that the electricity consumption information collection equipment i Electricity consumption information collection equipment j The feature vectors correspond to the calculated loss function.

5. The comprehensive evaluation method for the performance of electricity information collection equipment according to claim 3, characterized in that, The optimization objective function is obtained according to the following formula: Here, Min represents the operation of finding the minimum value.

6. The comprehensive evaluation method for the performance of electricity information collection equipment according to claim 3, characterized in that, When training the two neural networks according to the optimized objective function, the gradient descent method is used to train the weight parameters of the two neural networks. w The learning rate The formula for adaptive adjustment is: The loss function affects the weight parameters in the ranking. w The gradient calculation formula is: in, This indicates that the electricity consumption information collection equipment i Electricity consumption information collection equipment j The feature vectors correspond to the calculated loss function.

7. A computer device comprising a processor and a memory, the memory being used to store a computer program, characterized in that, The processor is used to execute the computer program to perform the method as described in any one of claims 1 to 6.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed, it implements the method as described in any one of claims 1 to 6.

Images