Charging pile equipment fault prediction method and system, and computer readable storage medium

By using a lightweight 7-layer separable convolutional neural network prediction model, the problems of low efficiency and high cost in charging pile fault detection are solved. It achieves accurate fault prediction with limited data, reduces equipment costs and reduces safety hazards.

CN116010853BActive Publication Date: 2024-11-29SHENZHEN TIANKE INFORMATION TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310106138.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-02-13
Publication Date
2024-11-29
Estimated Expiration
2043-02-13

AI Technical Summary

Technical Problem

Existing charging pile fault detection is inefficient, and prediction models that rely on large-scale data training from external devices are costly. Furthermore, charging pile operators have difficulty obtaining charging pile data, leading to inaccurate fault prediction and potential safety hazards.

Method used

A lightweight 7-layer separable convolutional neural network prediction model was established. By acquiring the fault element information of charging pile equipment, the model was encoded and generated into tensors, which were divided into training and testing datasets. The model was trained and optimized using a Few-shot training strategy to generate the optimal prediction model and predict the equipment fault index in real time.

Benefits of technology

Accurate fault prediction of charging pile equipment was achieved with limited data, reducing costs and CPU memory requirements, and enabling real-time prediction and reduction of safety hazards.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116010853B_ABST
    Figure CN116010853B_ABST
Patent Text Reader

Abstract

The present invention belongs to the technical field of new energy vehicle charging operation management, and specifically relates to a charging pile equipment fault prediction method and system, and a computer-readable storage medium. A charging pile equipment fault prediction method specifically includes the following steps: S1: establishing a lightweight prediction model; S2: obtaining the element information of the charging pile equipment fault; S3: encoding the element information, and stacking the encoded element information into a tensor V of Z×m×n dimensions, where Z is the number of types of element information, m is the length of the m×n matrix, and n is the width of the m×n matrix; S4: dividing the tensor V into a training data set and a test data set in a ratio of 8:2. The present invention can predict a relatively accurate charging pile equipment fault prediction result under the support of limited element information, can read the charging pile data in real time and give the equipment fault index, and can effectively reduce the safety hazards of the charging pile.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy vehicle charging operation management, and relates to charging pile equipment fault prediction technology, and specifically to a charging pile equipment fault prediction system and a computer-readable storage medium. Background Art

[0002] With the implementation of global energy conservation and emission reduction policies, the number of new energy electric vehicles continues to increase. In order to meet the electricity needs of car owners, electric charging piles are also deployed in various places in towns. In addition to the initial equipment cost, the daily operation and maintenance expenses of the deployment of charging piles are also a considerable expense. Therefore, whether the charging pile can achieve profitability, in addition to considering the simple charging charges, the operating loss of the equipment should also be considered, such as the aging of the original parts, the repair and replacement of parts, etc. Moreover, if the failure of the charging pile equipment is not predicted in time for troubleshooting, there will be great safety hazards, which can easily cause irreparable losses to personnel or new energy vehicles when users use it. However, in the prior art, the fault detection of charging piles is mostly carried out by employees at fixed points and at regular intervals. There is no specific difference in quality inspection for the maintenance of charging piles of different types, different environments, and different years. This method has low efficiency in troubleshooting of charging piles and greatly wastes human resources. Or some charging pile operators do not conduct regular quality inspections at all, which is not only not conducive to the normal profitability of charging piles, but also poses a considerable safety hazard to charging cars. Most of the equipment failure prediction models in the existing technology rely on external devices to collect a large amount of existing charging pile data for large-scale data training. The prediction model is large in size and needs to rely on external high-performance CPU or even GPU to achieve real-time prediction. The cost of model deployment is very high and there are inconveniences.

[0003] Moreover, for most charging pile operators, the effective charging pile data available is very limited, and it is also very challenging to accurately predict equipment failures based on small data. Summary of the invention

[0004] In order to solve the above-mentioned problems in the prior art, the present invention provides a charging pile equipment fault prediction system and a computer-readable storage medium.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for predicting faults of a charging pile device is provided, which is characterized by comprising the following steps:

[0007] S1: Establish a lightweight prediction model;

[0008] S2: Obtaining the element information of the charging pile equipment failure;

[0009] S3: Encode the feature information and stack the encoded feature information into a tensor V of Z×m×n dimensions, where Z is the number of types of feature information, m is the length of the m×n matrix, and n is the width of the m×n matrix;

[0010] S4: Divide the tensor V into a training data set and a test data set in a ratio of 8:2, and group the training data set and the test data set respectively. For each group of training data sets, continuously stack the features to generate V1 and then reduce its dimension to generate a few-shot training tensor Vf1. For each group of test data sets, continuously stack the features to generate V2 and then reduce its dimension to generate a few-shot test tensor Vf2. The number of samples in each training data set is the same as that in each test data set.

[0011] S5: All the few-shot training tensors Vf1 and all the few-shot test tensors Vf2 are input into the prediction model for training, and the prediction model parameters are generated and tuned to generate the optimal prediction model parameters and obtain the optimal prediction model;

[0012] S6: Input the factor information collected in real time into the optimal prediction model to predict the equipment failure index.

[0013] Preferably, in step S1,

[0014] The prediction model is a 7-layer separable convolutional neural network. The input of the prediction model is the Few-shot training tensor Vf1 and the Few-shot test tensor Vf2 of 4M×am×bn. The first layer of the prediction model is a deep separable convolutional layer with kernel sizes of 1×am / 256×bn / 256 and 4M×1×1 respectively. The second, third and fourth layers are deep separable convolutional layers with kernel sizes of 1×am / 256×bn / 256 and M×1×1 respectively. The fifth and sixth layers of the prediction model are the upsampling layer and downsampling layer respectively, and the seventh layer is the Softmax layer.

[0015] Preferably, in step S2,

[0016] The element information of charging pile equipment failure includes: the flashing frequency I of the charging indicator light per unit time, the working status W of the charging gun per unit time, the temperature T of the charging gun per unit time, and the difference Us between the voltage of the charging gun and the rated voltage per unit time.

[0017] Preferably, in step S3,

[0018] One-Hot is used to encode non-numeric features, and numeric features are encoded as vectors. The encoded features per unit time are stacked to generate a 4×m×n dimensional tensor V.

[0019] Preferably, in step S5,

[0020] First, initialize the preset optimal loss value λ to 0, and compare the loss value λ(k) of the current iteration with the loss value λ(k-1) of the previous iteration.

[0021] If λ(k-1)>λ(k), the model parameter p of the current iteration is saved and the current actual optimal loss value λt=λ(k) is recorded. If λ=λ(k) occurs during the training process, the model training is terminated and the optimal parameter pbest=p of the prediction model is obtained. If λ=λ(k) does not occur during the training process, the optimal parameter pbest=p(λt) of the prediction model is obtained. p(λt) is the model parameter at the iteration with the actual optimal loss value.

[0022] Preferably, in step S5,

[0023] The loss function Loss for calculating the loss value is as follows:

[0024] Loss = 1 / 4 [-logP (ans | Vf1, Vf2; p)] (Formula 1)

[0025] Among them, ans represents the prediction result of the prediction model, and P(ans|Vf1,Vf2; p) represents the probability that the prediction model predicts ans when the few-shot training tensor Vf1 and the few-shot test tensor Vf2 are input into the prediction model under the condition that the model parameter is p.

[0026] Preferably, in step S6,

[0027] The calculation method of the prediction result ans is as follows:

[0028] ans=Q1I+Q2W+Q3T+Q4Us(Formula 2)

[0029] Among them, Q1 represents the weight of the flashing frequency I of the charging indicator light per unit time, Q2 represents the weight of the working state W of the charging gun per unit time, Q3 represents the weight of the temperature T of the charging gun per unit time, and Q4 represents the weight of the difference Us between the voltage of the charging gun and the rated voltage per unit time.

[0030] Preferably, in step S6,

[0031] The equipment failure index is divided into risk levels, including level 1 risk, level 2 risk, level 3 risk and level 4 risk;

[0032] The equipment failure index range for level one risk is 0.8-1, the equipment failure index range for level two risk is 0.6-0.8, the equipment failure index range for level three risk is 0.4-0.6, and the equipment failure index range for level four risk is 0-0.4.

[0033] A charging pile equipment fault prediction system, characterized in that it includes:

[0034] Feature acquisition module;

[0035] Information processing module;

[0036] Model training module;

[0037] Risk prediction module;

[0038] The element acquisition module is used to obtain the element information of the charging pile equipment failure, and the information processing module is used to encode the element information and stack the encoded element information to generate a 4×m×n dimensional tensor V, where m is the length of the m×n matrix and n is the width of the m×n matrix;

[0039] The model training module is used to divide the tensor V into a training data set and a test data set in a ratio of 8:2, and group the training data set and the test data set respectively, stack the features of each group of training data sets continuously to generate V1 and then reduce its dimension to generate a few-shot training tensor Vf1, stack the features of each group of test data sets continuously to generate V2 and then reduce its dimension to generate a few-shot test tensor Vf2, the number of samples in each training data set is the same as that in each test data set, and then input all the few-shot training tensors Vf1 and all the few-shot test tensors Vf2 into the prediction model for training, generate and tune the prediction model parameters, and generate the optimal prediction model;

[0040] The risk prediction module is used to input the element information collected in real time into the optimal prediction model to predict the equipment failure index.

[0041] A computer-readable storage medium, comprising:

[0042] Used to store an executable program, and the executable program is used to execute the charging pile equipment fault prediction method.

[0043] The present invention provides a charging pile equipment fault prediction system and a computer-readable storage medium, and the beneficial effects are embodied in: the charging pile equipment fault prediction method provided by the present invention can predict a relatively accurate charging pile equipment fault prediction result under the support of limited element information, and has low requirements for COU running memory, which can greatly reduce the cost of charging pile equipment fault prediction, and can achieve a faster prediction speed on the charging pile main control CPU. The prediction method can read the charging pile data in real time and give the equipment fault index, which can effectively reduce the safety hazards of the charging pile. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1It is a flow chart of a method for predicting faults of charging pile equipment;

[0045] Figure 2 Schematic diagram of the prediction model;

[0046] Figure 3 This is a flow chart for predicting faults of charging pile equipment;

[0047] Figure 4 This is a block diagram of a charging pile equipment fault prediction system. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] See also Figure 1-Figure 4 As shown, the specific embodiments provided by the present invention are as follows:

[0050] Embodiment 1:

[0051] like Figure 1 As shown, a charging pile equipment fault prediction method is characterized in that it specifically includes the following steps:

[0052] A method for predicting faults of charging pile equipment, characterized in that it specifically comprises the following steps:

[0053] S1: Establish a lightweight prediction model;

[0054] S2: Obtaining the element information of the charging pile equipment failure;

[0055] S3: Encode the feature information and stack the encoded feature information into a tensor V of Z×m×n dimensions, where Z is the number of types of feature information, m is the length of the m×n matrix, and n is the width of the m×n matrix;

[0056] S4: Divide the tensor V into a training data set and a test data set in a ratio of 8:2, and group the training data set and the test data set respectively. For each group of training data sets, continuously stack the features to generate V1 and then reduce its dimension to generate a few-shot training tensor Vf1. For each group of test data sets, continuously stack the features to generate V2 and then reduce its dimension to generate a few-shot test tensor Vf2. The number of samples in each training data set is the same as that in each test data set.

[0057] S5: All the few-shot training tensors Vf1 and all the few-shot test tensors Vf2 are input into the prediction model for training, and the prediction model parameters are generated and tuned to generate the optimal prediction model parameters and obtain the optimal prediction model;

[0058] S6: Input the factor information collected in real time into the optimal prediction model to predict the equipment failure index.

[0059] With the implementation of global energy conservation and emission reduction policies, the number of new energy electric vehicles is increasing. In the daily use of charging piles, due to the installation in different places, the loss of charging piles also has many irregularities. For example, the aging of original parts and the maintenance of parts have a great impact on whether the charging piles are profitable. However, in the existing technology, if the equipment failure of the charging pile cannot be detected in time, there will be huge safety hazards. Once a safety accident occurs in the charging pile, it is very easy to cause dangerous situations such as fire and explosion. In addition, the regular quality inspection of the staff is greatly interfered by human factors, and it is very easy to miss inspections or fail to inspect. Therefore, it is very valuable to provide a charging pile equipment failure prediction system.

[0060] The existing charging pile equipment failure prediction models also rely on charging pile data collected by external devices for large-scale data training. The prediction model is large in size and needs to rely on external high-performance CPUs or even GPUs to achieve real-time operation, which is inconvenient in model deployment. In addition, due to industry and market competition, it is difficult for operators to obtain a large amount of charging pile application data, so it is very challenging to accurately predict charging pile equipment failures.

[0061] In this embodiment, a charging pile equipment fault prediction method is provided, which is characterized by comprising the following steps:

[0062] S1: Establish a lightweight prediction model;

[0063] S2: Obtaining the element information of the charging pile equipment failure;

[0064] S3: Encode the feature information and stack the encoded feature information into a tensor V of Z×m×n dimensions, where Z is the number of types of feature information, m is the length of the m×n matrix, and n is the width of the m×n matrix;

[0065] S4: Divide the tensor V into a training data set and a test data set in a ratio of 8:2, and group the training data set and the test data set respectively. For each group of training data sets, continuously stack the features to generate V1 and then reduce its dimension to generate a few-shot training tensor Vf1. For each group of test data sets, continuously stack the features to generate V2 and then reduce its dimension to generate a few-shot test tensor Vf2. The number of samples in each training data set is the same as that in each test data set.

[0066] S5: All the few-shot training tensors Vf1 and all the few-shot test tensors Vf2 are input into the prediction model for training, and the prediction model parameters are generated and tuned to generate the optimal prediction model parameters and obtain the optimal prediction model;

[0067] S6: Input the factor information collected in real time into the optimal prediction model to predict the equipment failure index.

[0068] The present invention predicts the equipment failure index through a few-shot training strategy. First, a lightweight prediction model is established. When establishing the model, the type of element information of the faulty equipment is simultaneously referred to. Then, the element information of the equipment failure is obtained, and the element information is encoded in sequence. A variety of element information is fused into a tensor V in the form of information stacking. Then, the tensor V is divided into a training data set and a training data set in a ratio of 8:2, and the training data set and the test data set are grouped respectively. For each group of training data sets, continuous feature stacking is performed to generate V1, and then dimension reduction is performed to generate a few-shot training tensor Vf1. For each group of test data sets, continuous feature stacking is performed to generate V2, and then dimension reduction is performed to generate a few-shot test tensor Vf2. Dimension reduction is to screen out useless information after encoding by allocating different weights. This method is beneficial to improving the performance of the prediction model and accelerating the model convergence speed during the training process.

[0069] The charging pile equipment fault prediction method provided by the present invention can predict a relatively accurate charging pile equipment fault prediction result under the support of limited factor information, has low requirements for CPU running memory, can greatly reduce the cost of charging pile equipment fault prediction, and can achieve a faster prediction speed on the charging pile main control CPU. The prediction method can read the charging pile data in real time and give the equipment fault index, which can effectively reduce the safety hazards of the charging pile.

[0070] Embodiment 2:

[0071] In step S1,

[0072] The prediction model is a 7-layer separable convolutional neural network. The input of the prediction model is the Few-shot training tensor Vf1 and the Few-shot test tensor Vf2 of 4M×am×bn. The first layer of the prediction model is a deep separable convolutional layer with kernel sizes of 1×am / 256×bn / 256 and 4M×1×1 respectively. The second, third and fourth layers are deep separable convolutional layers with kernel sizes of 1×am / 256×bn / 256 and M×1×1 respectively. The fifth and sixth layers of the prediction model are the upsampling layer and downsampling layer respectively, and the seventh layer is the Softmax layer.

[0073] In this embodiment, there are four kinds of element information of equipment failure, and the prediction model is a 7-layer separable convolutional neural network. The input of the prediction model is a training tensor Vf1 and a test tensor Vf2 of 4M×am×bn. A deep convolution layer and a point convolution layer constitute a deep separable convolution layer. The first layer of the prediction model is a deep separable convolution layer with kernel sizes of 1×am / 256×bn / 256 and 4M×1×1, respectively. In other words, the kernel size of the first deep convolution layer of the prediction model is 1×am / 256×bn / 256, and the prediction The kernel size of the first point convolution layer of the model is 4M×1×1, and the second, third and fourth layers are all depth-separable convolution layers with kernel sizes of 1×am / 256×bn / 256 and M×1×1, respectively. They are not described here. The fifth and sixth layers of the prediction model are the upsampling layer and the downsampling layer, respectively. The two layers are combined to form a bottleneck structure. Without introducing additional parameters and computational complexity, the present invention is easier to extract key information of features. The seventh layer is the Softmax layer, which is used to output the device failure index.

[0074] Compared with traditional convolutional layers, the depthwise separable convolutional layer has a deeper network but fewer parameters, and can achieve higher accuracy even with a smaller data set.

[0075] In this embodiment, for the same input 1×28×28, the model parameters of Letnet are 4310800, and the value after the number of parameters is converted into storage units is 1.64MB. The value after the number of parameters of the network without depthwise separable convolution is converted into storage units is 1.53MB, while the value after the number of parameters of the present invention is converted into storage units is 0.96MB. It can be seen that the prediction model of the present invention is more lightweight, has less requirements for charging pile storage, and has a greater probability of being popularized to various charging piles.

[0076] Embodiment 3:

[0077] In step S2,

[0078] The element information of charging pile equipment failure includes: the flashing frequency I of the charging indicator light per unit time, the working status W of the charging gun per unit time, the temperature T of the charging gun per unit time, and the difference Us between the voltage of the charging gun and the rated voltage per unit time.

[0079] Most of the existing technologies directly detect faults on charging piles. At this time, the detected faults are existing faults, while some charging devices that are at the critical point of failure but have not yet failed will not be included in the fault list. These charging pile devices that are about to fail pose a great safety hazard to users, just like a time bomb that may go wrong at any time. At this time, it is very valuable to provide a method or system to predict the equipment fault information of charging piles.

[0080] In this embodiment, the flashing frequency I of the charging indicator light per unit time can reflect the number of abnormalities that have occurred in the device. The higher the flashing frequency I of the charging indicator light per unit time, the higher the device failure index. The lower the flashing frequency I of the charging indicator light per unit time, the lower the device failure index.

[0081] The working status W of the charging gun per unit time can reflect the usage of the charging pile. W has two states: charging and idle. The charging pile equipment fault index can be further judged by obtaining the relationship between the number of idle times and the number of charging times of W. The more idle times, the higher the loss index, and the more charging times, the lower the loss index.

[0082] When the temperature of the charging gun remains high for a fixed period of time, it is considered that the charging pile equipment is in poor operating condition. Excessive temperature of the charging gun and the charging cable will cause accelerated aging of the wires. At this time, the equipment failure index of the charging pile equipment will increase.

[0083] In this embodiment, the charging gun temperature T per unit time may represent the maximum temperature of the charging pile within a fixed time period. The maximum temperature value may be fine-tuned according to the environment in which the charging pile is located. For example, the maximum temperature value range is 50-65 degrees Celsius.

[0084] In another embodiment, the charging gun temperature T per unit time may represent the average temperature of the charging pile when charging within a fixed time period. The value of the average temperature may be fine-tuned according to the environment in which the charging pile is located. For example, the value range of the average temperature is 40-45 degrees Celsius.

[0085] When the difference between the charging gun voltage and the rated voltage is large, and the working state W of the charging gun is "idle", that is, the voltage difference is large and the charging gun is not working, it is considered that the charging pile has leakage or component aging, and the risk level of the device will increase. The present invention uses the difference between the charging gun voltage and the rated voltage Us per unit time and the working state W of the charging gun per unit time to perform model training and model prediction, which can more accurately predict the equipment failure index.

[0086] Among them, the working state W of the charging gun per unit time, the temperature T of the charging gun per unit time, and the difference Us between the voltage of the charging gun and the rated voltage per unit time characterize the depreciation of the charging pile. The present invention predicts future equipment failures in advance, and can automatically monitor the element information of the charging pile, timely predict the quality of equipment failures, and timely repair and replace the charging pile equipment according to the equipment failure index, which greatly improves the efficiency and safety of the use of the charging pile.

[0087] Embodiment 4:

[0088] In step S3,

[0089] One-Hot is used to encode non-numeric features, and numeric features are encoded as vectors. The encoded features per unit time are stacked to generate a 4×m×n dimensional tensor V.

[0090] The purpose of using One-Hot to encode non-numeric elements is to facilitate the concatenation and stacking of vectors. Stacking the encoded elements can fuse features of different dimensions, largely avoiding overfitting and thus improving the accuracy of the prediction model.

[0091] Embodiment 5:

[0092] In step S5,

[0093] First, initialize the preset optimal loss value λ to 0, and compare the loss value λ(k) of the current iteration with the loss value λ(k-1) of the previous iteration.

[0094] If λ(k-1)>λ(k), the model parameter p of the current iteration is saved and the current actual optimal loss value λt=λ(k) is recorded. If λ=λ(k) appears during the training process, the model training is terminated and the optimal parameter pbest=p of the prediction model is predicted. If λ=λ(k) does not appear during the training process, the optimal parameter pbest=p(λt) of the prediction model is predicted. p(λt) is the model parameter of the iteration with the actual optimal loss value. Judging the loss value at each iteration can largely avoid the influence of oscillation on model convergence.

[0095] The loss function Loss for calculating the loss value is as follows:

[0096] Loss = 1 / 4 [-logP (ans | Vf1, Vf2; p)] (Formula 1)

[0097] Among them, ans represents the prediction result of the prediction model, and P(ans|Vf1,Vf2; p) represents the probability that the prediction model predicts ans when the few-shot training tensor Vf1 and the few-shot test tensor Vf2 are input into the prediction model under the condition that the model parameter is p.

[0098] The calculation method of the prediction result ans is as follows:

[0099] ans=Q1I+Q2W+Q3T+Q4Us(Formula 2)

[0100] Among them, Q1 represents the weight of the flashing frequency I of the charging indicator light per unit time, Q2 represents the weight of the working state W of the charging gun per unit time, Q3 represents the weight of the temperature T of the charging gun per unit time, and Q4 represents the weight of the difference Us between the voltage of the charging gun and the rated voltage per unit time.

[0101] In actual prediction, Figure 3 As shown, the prediction specifically includes the following steps:

[0102] S61: Acquire prediction factor information of the charging device in real time;

[0103] S62: Encode the prediction factor information, and stack the encoded prediction factor information to generate a 4×m×n dimensional prediction tensor Vc;

[0104] S63: Input the prediction tensor Vc into the optimal prediction model, and the optimal prediction model outputs the equipment failure index.

[0105] Embodiment 6:

[0106] In step S6,

[0107] The equipment failure index is divided into risk levels, including level 1 risk, level 2 risk, level 3 risk and level 4 risk;

[0108] The equipment failure index range for level one risk is 0.8-1, the equipment failure index range for level two risk is 0.6-0.8, the equipment failure index range for level three risk is 0.4-0.6, and the equipment failure index range for level four risk is 0-0.4.

[0109] In this embodiment, different risk indices correspond to different risk levels. The higher the risk level, the less recommended it is, and the lower the risk level, the greater the probability of profit.

[0110] Embodiment 7:

[0111] like Figure 4 As shown, a charging pile equipment fault prediction system is characterized by comprising:

[0112] Feature acquisition module;

[0113] Information processing module;

[0114] Model training module;

[0115] Risk prediction module;

[0116] The element acquisition module is used to obtain the element information of the charging pile equipment failure, and the information processing module is used to encode the element information and stack the encoded element information to generate a 4×m×n dimensional tensor V, where m is the length of the m×n matrix and n is the width of the m×n matrix;

[0117] The model training module is used to divide the tensor V into a training data set and a test data set in a ratio of 8:2, and group the training data set and the test data set respectively, stack the features of each group of training data sets continuously to generate V1 and then reduce its dimension to generate a few-shot training tensor Vf1, stack the features of each group of test data sets continuously to generate V2 and then reduce its dimension to generate a few-shot test tensor Vf2, the number of samples in each training data set is the same as that in each test data set, and then input all the few-shot training tensors Vf1 and all the few-shot test tensors Vf2 into the prediction model for training, generate and tune the prediction model parameters, and generate the optimal prediction model;

[0118] The risk prediction module is used to input the element information collected in real time into the optimal prediction model to predict the equipment failure index.

[0119] A computer-readable storage medium, comprising:

[0120] Used to store an executable program, and the executable program is used to execute the charging pile equipment fault prediction method.

[0121] In this embodiment, a charging pile equipment fault prediction system is characterized by including an element acquisition module; an information processing module; a model training module; a risk prediction module; a storage module; and an element acquisition module data receiving end, wherein the data receiving end can receive element information related to the charging pile equipment fault from the charging pile lower computer interface, and the element information may include: the flashing frequency I of the charging indicator light per unit time, the working state W of the charging gun per unit time, the temperature T of the charging gun per unit time, and the difference Us between the voltage of the charging gun and the rated voltage per unit time. The data transmission protocol is the AT command protocol, and the data type is single floating point numerical data and character data. After the element information is acquired, it is stored in the storage module.

[0122] For example, the storage module is a solid-state memory, and the storage module has a first storage unit, a second storage unit, a third storage unit, and a fourth storage unit. The first storage unit is used to store the flashing frequency I of the charging indicator light per unit time, the second storage unit is used to store the working state W of the charging gun per unit time, the third storage unit is used to store the temperature T of the charging gun per unit time, and the fourth storage unit is used to store the difference Us between the charging gun voltage and the rated voltage per unit time.

[0123] Among them, the model training module is used to divide the tensor V into a training data set and a test data set in a ratio of 8:2, and group the training data set and the test data set respectively, continuously stack the features of each group of training data sets to generate V1 and then reduce its dimension to generate a training tensor Vf1, continuously stack the features of each group of test data sets to generate V2 and then reduce its dimension to generate a test tensor Vf2. The number of samples in each training data set is the same as that in each test data set. By adjusting different weights, useless information after encoding is screened out. Dimensionality reduction is beneficial to improving the performance of the model and accelerating the convergence speed of the model during training.

[0124] In the description of the embodiments of the present invention, it needs to be understood that terms such as “up”, “down”, “front”, “back”, “left”, “right”, “vertical”, “horizontal”, “center”, “top”, “bottom”, “top”, “bottom”, “inside”, “outside”, “inside”, and “outside” indicate orientation or positional relationships.

[0125] In the description of the embodiments of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "install", "connect", "connect", and "assemble" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0126] In the description of the embodiments of the present invention, specific features, structures, materials or characteristics may be combined in a suitable manner in any one or more embodiments or examples.

[0127] In the description of the embodiments of the present invention, it is to be understood that "-" and "~" represent the range between two values, and the range includes the endpoints. For example: "AB" represents a range greater than or equal to A and less than or equal to B. "A~B" represents a range greater than or equal to A and less than or equal to B.

[0128] In the description of the embodiments of the present invention, the term "and / or" herein is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " herein generally indicates that the associated objects before and after are in an "or" relationship.

[0129] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A charging pile equipment fault prediction method, characterized in that: The specific steps include: S1: Establish a lightweight prediction model; S2: Obtaining the element information of the charging pile equipment failure; S3: Encode the feature information and stack the encoded feature information into a tensor V of Z×m×n dimensions, where Z is the number of types of feature information, m is the length of the m×n matrix, and n is the width of the m×n matrix; S4: Divide the tensor V into a training data set and a test data set in a ratio of 8:2, and group the training data set and the test data set respectively. For each group of training data sets, continuously stack the features to generate V1 and then reduce its dimension to generate a few-shot training tensor Vf1. For each group of test data sets, continuously stack the features to generate V2 and then reduce its dimension to generate a few-shot test tensor Vf2. The number of samples in each training data set is the same as that in each test data set. S5: All the few-shot training tensors Vf1 and all the few-shot test tensors Vf2 are input into the prediction model for training, and the prediction model parameters are generated and tuned to generate the optimal prediction model parameters and obtain the optimal prediction model; S6: Input the element information collected in real time into the optimal prediction model to predict the equipment failure index; In step S1, the prediction model is a 7-layer separable convolutional neural network. The input of the prediction model is a 4M×am×bn Few-shot training tensor Vf1 and a Few-shot test tensor Vf2. The first layer of the prediction model is a deep separable convolutional layer with kernel sizes of 1×am / 256×bn / 256 and 4M×1×1, respectively. The second, third, and fourth layers are all deep separable convolutional layers with kernel sizes of 1×am / 256×bn / 256 and M×1×1, respectively. The fifth and sixth layers of the prediction model are upsampling layers and downsampling layers, respectively, and the seventh layer is a Softmax layer.

2. A charging pile equipment fault prediction method according to claim 1, characterized in that: In step S2, The element information of charging pile equipment failure includes: the flashing frequency I of the charging indicator light per unit time, the working status W of the charging gun per unit time, the temperature T of the charging gun per unit time, and the difference Us between the voltage of the charging gun and the rated voltage per unit time.

3. A charging pile equipment fault prediction method according to claim 2, characterized in that: In step S3, One-Hot is used to encode non-numeric features, and numeric features are encoded as vectors. The encoded features per unit time are stacked to generate a 4×m×n dimensional tensor V.

4. A charging pile equipment fault prediction method according to claim 3, characterized in that: In step S5, First, initialize the preset optimal loss value λ to 0, and compare the loss value λ(k) of the current iteration with the loss value λ(k-1) of the previous iteration. If λ(k-1)>λ(k), the model parameter p of the current iteration is saved and the current actual optimal loss value λt=λ(k) is recorded. If λ=λ(k) occurs during the training process, the model training is terminated and the optimal parameter pbest=p of the prediction model is obtained. If λ=λ(k) does not occur during the training process, the optimal parameter pbest=p(λt) of the prediction model is obtained. p(λt) is the model parameter at the iteration with the actual optimal loss value.

5. A charging pile equipment fault prediction method according to claim 4, characterized in that: In step S5, The loss function Loss for calculating the loss value is as follows: Loss = 1 / 4 [-logP (ans | Vf1, Vf2; p)] (Formula 1) Among them, ans represents the prediction result of the prediction model, and P(ans|Vf1,Vf2; p) represents the probability that the prediction model predicts ans when the few-shot training tensor Vf1 and the few-shot test tensor Vf2 are input into the prediction model under the condition that the model parameter is p.

6. A charging pile equipment fault prediction method according to claim 5, characterized in that: In step S6, The calculation method of the prediction result ans is as follows: ans=Q1I+Q2W+Q3T+Q4Us (Formula 2) Among them, Q1 represents the weight of the flashing frequency I of the charging indicator light per unit time, Q2 represents the weight of the working state W of the charging gun per unit time, Q3 represents the weight of the temperature T of the charging gun per unit time, and Q4 represents the weight of the difference Us between the voltage of the charging gun and the rated voltage per unit time.

7. A charging pile equipment fault prediction method according to claim 6, characterized in that: In step S6, The equipment failure index is divided into risk levels, including level 1 risk, level 2 risk, level 3 risk and level 4 risk; The equipment failure index range for level one risk is 0.8-1, the equipment failure index range for level two risk is 0.6-0.8, the equipment failure index range for level three risk is 0.4-0.6, and the equipment failure index range for level four risk is 0-0.

4.

8. A charging pile equipment fault prediction system, characterized in that: include, Feature acquisition module; Information processing module; Model training module; Risk prediction module; The element acquisition module is used to obtain the element information of the charging pile equipment failure, and the information processing module is used to encode the element information and stack the encoded element information to generate a 4×m×n dimensional tensor V, where m is the length of the m×n matrix and n is the width of the m×n matrix; The model training module is used to divide the tensor V into a training data set and a test data set in a ratio of 8:2, and group the training data set and the test data set respectively, stack the features of each group of training data sets continuously to generate V1 and then reduce its dimension to generate a few-shot training tensor Vf1, stack the features of each group of test data sets continuously to generate V2 and then reduce its dimension to generate a few-shot test tensor Vf2, the number of samples in each training data set is the same as that in each test data set, and then input all the few-shot training tensors Vf1 and all the few-shot test tensors Vf2 into the prediction model for training, generate and tune the prediction model parameters, and generate the optimal prediction model; Wherein, the risk prediction module is used to input the element information collected in real time into the optimal prediction model to predict the equipment failure index; The prediction model is a 7-layer separable convolutional neural network. The input of the prediction model is the Few-shot training tensor Vf1 and the Few-shot test tensor Vf2 of 4M×am×bn. The first layer of the prediction model is a deep separable convolutional layer with kernel sizes of 1×am / 256×bn / 256 and 4M×1×1 respectively. The second, third and fourth layers are deep separable convolutional layers with kernel sizes of 1×am / 256×bn / 256 and M×1×1 respectively. The fifth and sixth layers of the prediction model are the upsampling layer and downsampling layer respectively, and the seventh layer is the Softmax layer.

9. A computer-readable storage medium, characterized in that: include, Used to store an executable program, and the executable program is used to execute a charging pile equipment fault prediction method as described in any one of claims 1-7.

Citation Information

Patent Citations

  • Oil and gas gathering and transportation monitoring system and method

    CN115147655A

  • Charging pile state evaluation method and device based on CNN and random forest

    CN115619271A