Health Status Evaluation Method for Charging Terminal Devices Based on Wide Learning and K-means

Abstract

Health status evaluation method for charging terminal devices based on width learning and K-means. 1) Use the electric energy collection terminal to configure the charging terminal network; 2) Establish a health status evaluation model for the charging terminal device for the battery being charged; 3) Dynamically self-learn and adjust the model according to device data; 4) Establish a health evaluation model for the charging terminal device; 5) Calculate the health level of the charging terminal device according to the K-means evaluation model, and find out the charging terminal node corresponding to this health level through the configuration model in step 1, and give different treatment suggestions according to different health levels. Through the electric energy collection terminal configuration model, the number of electric energy collection terminal configurations is reduced, the cost is saved, the health status of the battery being charged is evaluated through width learning, and at the same time, a weighted network is introduced to realize the real-time update of the width learning model. Finally, a health evaluation model for the charging terminal device is established through the K-means model.

Description

Technical Field

[0001] The present invention relates to the field of charging terminal device detection, and particularly to a method for evaluating the health status of charging terminal devices based on width learning and K-means. Background Art

[0002] In recent years, with the continuous development of electric vehicles, the market share of electric vehicles in China has been increasing. However, with the increase in the number of electric vehicles, the charging technology of electric vehicles has also become a key point of concern. As an efficient management platform for the charging and discharging of electric vehicles, the electric vehicle charging pile network undertakes functions such as energy management, demand-side management, and implementation of orderly charging and discharging plans, providing support for the grid-connected management of electric vehicles.

[0003] The primary condition for efficient repair and maintenance of charging piles is to quickly locate the fault point. Therefore, an effective method to ensure the reliability of charging piles is to improve the fault diagnosis technology. Since the current research on charging pile fault diagnosis is still in its infancy, maintenance personnel often rely on past experience for diagnosis and repair. However, inexperienced maintenance personnel often cannot quickly locate the cause of the fault and lack the ability to monitor transient power quality indicators. Due to the lack of a background database as support, the data storage and management capabilities are weak, and the data of each monitoring point in the whole network cannot be stored in a structured manner. Moreover, the data dump is not flexible, the interfaces and data formats of equipment from different manufacturers are not unified, and there is no data sharing ability. The number of monitoring points is small, the real-time performance is poor, and the data is often affected by the current system operation state. The test results cannot reflect the long-term usage situation of the charging network and cannot provide comprehensive reference data for the management of charging piles, resulting in relatively low diagnostic efficiency. Summary of the Invention

[0004] To solve the above problems, the present invention proposes a method for evaluating the health status of charging terminal devices based on width learning and K-means. By configuring the power acquisition terminal model, the number of power acquisition terminals is reduced, saving costs. The health status of the charging battery is evaluated through width learning. At the same time, a weighted network is introduced to realize the real-time update of the width learning model. Finally, a health assessment model for charging terminal devices is established through the K-means model. According to the calculated health level of the charging terminal devices, different treatment suggestions are given, and targeted maintenance can be carried out on the charging terminal devices. To achieve this purpose, the present invention provides a method for evaluating the health status of charging terminal devices based on width learning and K-means, and the specific steps are as follows, characterized in that:

[0005] Step 1, configure the charging terminal network using the power acquisition terminal to reduce the number of power acquisition terminal configurations;

[0006] Step 2: Use width learning to establish a health status evaluation model for the charging terminal device to the battery being charged, and evaluate the status of the charging battery;

[0007] Step 3: Introduce a weighted network into the width learning model, and perform dynamic self-learning adjustment of the model according to device data;

[0008] Step 4: Establish a health assessment model for the charging terminal device through the current, voltage, power change curve, utilization rate, temperature sensor data, smoke sensor data, and order distribution time of the device's historical orders;

[0009] Step 5: Calculate the health level of the charging terminal device according to the K-means evaluation model, and find the corresponding charging terminal node through the configuration model in Step 1, and give different treatment suggestions according to different health levels.

[0010] Furthermore, the process of the power acquisition terminal configuration model in Step 1 can be expressed as:

[0011] The power acquisition terminal configuration model is expressed by the following formula:

[0012]

[0013] In the formula, C is the network configuration cost of the power acquisition terminal, x i is the configuration coefficient of the i-th charging node. If the power acquisition terminal is configured at this node, then x i is equal to 1, otherwise x i is equal to 0, c i is the cost coefficient of configuring the power acquisition terminal at the i-th charging node, n is the number of nodes, u i equal to 1 indicates that the voltage of the i-th node is measurable, I ij equal to 1 indicates that the current between the i-th node and the j-th node is measurable, r ij represents the connection coefficient between the i-th node and the j-th node.

[0014] Furthermore, the establishment process of the state evaluation model of the charging battery in Step 2 can be expressed as:

[0015] For electric vehicle batteries of different models, use width learning to establish a theoretical health status evaluation model respectively. The width learning model can be expressed as:

[0016] Y N×Q =[Z N×b |H N×d ·W1 (B+b)×Q (2)

[0017] Among them, Z is the feature node layer, H is the enhanced node layer, Y is the output of the output layer, W1 is the connection weight matrix, N is the number of input samples, b is the number of feature nodes, d is the number of enhanced nodes, and Q is the dimension of the output layer; in the formula, Z N×b can be expressed as:

[0018] Z N×b =X N×M ·W2 M+b (3)

[0019] Among them, X is the input data, W2 is the optimal input weight matrix obtained by sparse auto-encoding, M is the feature dimension of the input data, and N is the number of input data; H in formula (1) N×d can be expressed as:

[0020] H N×b =φ(Z N×b W3 b×d +β N×d ) (4)

[0021] In the formula, W3 is a random matrix, and β is a bias; the feature nodes are subjected to feature extraction by a sparse auto-encoder to remove redundant features, and then enhanced into enhanced nodes through a non-linear activation function to form an enhanced node layer H; then the feature layer and the enhanced layer are merged, and the network is expanded horizontally, and finally the connection weight matrix W1 is obtained by using the ridge regression algorithm;

[0022] The voltage, current, charging time, impedance, and temperature data during battery charging are combined into a multi-dimensional matrix as the input data of the input layer of the width learning, and the SOH value of the battery is used as the output of the output layer of the width learning network to establish a width learning model;

[0023] Furthermore, the process of dynamic self-learning adjustment of the model in step 3 can be expressed as:

[0024] A weighted network is added to the width learning model, and the width learning model is adjusted according to the weights in the weighted network to obtain a width learning model that can dynamically self-learn;

[0025] The weighted network first reduces the dimension of the input data set X of the input layer through the convolutional layer, then obtains the feature map R of N×M through the activation function, and then normalizes R through the convolutional layer and the activation function to generate the attention weight σ of 0 to 1, and its calculation formula is:

[0026] M=tanh(comv1X) (5)

[0027] σ=sigmoid(conv2M) (6)

[0028] Among them, conv1 and conv2 represent convolution operations, tanh and sigmoid represent the tanh function and sigmoid function respectively, and the feature output of the final feature weighting is X':

[0029] X′ = σ × X (7)

[0030] At the same time, the weighted network normalizes the weighted feature output into the feature node layer through a fully connected layer. The self - learned feature node layer is represented as follows:

[0031] Z N×b = X′ N×M ·W2 M×b (8)

[0032] Furthermore, the process of establishing the charging terminal device health assessment model in step 4 can be expressed as:

[0033] Establish a K - means evaluation model to evaluate the health level of the charging terminal device. As an unsupervised clustering method, K - means is usually used to automatically divide samples into k clusters. The purpose of the K - means clustering method is to assign all N samples to k clusters by minimizing the sum of the distances from points to the centroids. K - means minimizes the squared error J for the cluster partition C obtained by clustering as follows:

[0034]

[0035] where C = {C1, C2, …, C k} represents the k types of charging terminal device health level clusters, is a feature matrix of the current charging terminal device's current, voltage, power change curve, utilization rate, temperature sensor data, smoke sensor data, and order distribution time data, represents the centroid of the i - th cluster;

[0036] Step 4.1, randomly select k cluster centroids of the charging terminal device health level feature samples;

[0037] Step 4.2, measure the distance between the sample and each centroid, and assign each sample to the nearest cluster centroid, and iterate n times in this way;

[0038] Step 4.3, in each iteration process, update the centroids of each cluster using the mean value;

[0039] Step 4.4, for the k cluster centroids, use steps 4.2 and 4.3 for iterative update until the cluster centroids are stable or the function converges.

[0040] Health Status Assessment Method of Charging Terminal Equipment Based on Wide Learning and K-means, Beneficial Effects: The technical effects of the present invention are as follows:

[0041] 1. The present invention reduces the number of power collection terminal configurations through the power collection terminal configuration model, saving costs;

[0042] 2. The present invention evaluates the health status of the charging battery through wide learning. The model has a simple structure and fast training speed, and a weighted network is introduced to realize real-time update of the model;

[0043] 3. The present invention establishes a health assessment model for charging terminal equipment through the K-means model, and gives different treatment suggestions according to the calculated health level of the charging terminal equipment, so as to carry out targeted maintenance on the charging terminal equipment. Description of the Drawings

[0044] Figure 1 is the flowchart of the present invention;

[0045] Figure 2 is the state assessment model of the charging battery of the present invention;

[0046] Figure 3 is the flowchart of the health level processing of the charging terminal equipment of the present invention. Detailed Embodiment

[0047] The present invention will be further described in detail below in conjunction with the drawings and specific embodiments:

[0048] The present invention proposes a health status assessment method for charging terminal equipment based on wide learning and K-means. The flowchart of the present invention is as Figure 1 shown. The steps of the present invention will be described in detail below in conjunction with the flowchart.

[0049] Step 1: Use the power collection terminal to configure the charging terminal network to reduce the number of power collection terminal configurations;

[0050] The power collection terminal configuration model is expressed by the following formula:

[0051]

[0052] In the formula, C is the power collection terminal network configuration cost, x i is the configuration coefficient of the i-th charging node. If the power collection terminal is configured at this node, then x i is equal to 1, otherwise x i is equal to 0, c i is the cost coefficient of configuring the power collection terminal at the i-th charging node, n is the number of nodes, u iEqual to 1 indicates that the voltage of the i-th node is measurable, I ij Equal to 1 indicates that the current between the i-th node and the j-th node is measurable, r ij Indicates the connection coefficient between the i-th node and the j-th node.

[0053] Step 2: Use width learning to establish a health state evaluation model for the charging terminal device to the battery being charged as Figure 2 shown, and evaluate the state of the charging battery;

[0054] For electric vehicle batteries of different models, use width learning to establish a health state theoretical evaluation model respectively. The width learning model can be expressed as:

[0055] Y N×Q =[Z N×b |H N×d ·W1 (b+d)×Q (2)

[0056] where Z is the feature node layer, H is the enhanced node layer, Y is the output of the output layer, W1 is the connection weight matrix, N is the number of input samples, b is the number of feature nodes, d is the number of enhanced nodes, and Q is the dimension of the output layer; in the formula, Z N×b can be expressed as:

[0057] Z N×b =X N×M ·W2 M×b (3)

[0058] where X is the input data, W2 is the optimal input weight matrix obtained by sparse autoencoding, M is the feature dimension of the input data, and N is the number of input data; in formula 1, H N×d can be expressed as:

[0059] H N×b =φ(Z N×b W3 b×d +β N×d ) (4)

[0060] In the formula, W3 is a random matrix and β is a bias; the feature nodes are subjected to feature extraction by a sparse autoencoder to remove redundant features, and then enhanced by a non-linear activation function to form an enhanced node layer H; then the feature layer and the enhanced layer are merged, and the network is extended horizontally. Finally, the connection weight matrix W1 is obtained by using the ridge regression algorithm;

[0061] Form a multi-dimensional matrix with the voltage, current, charging time, impedance, and temperature data during battery charging as the input data of the input layer of width learning, and use the SOH value of the battery as the output of the output layer of the width learning network to establish a width learning model;

[0062] Step 3: Introduce a weighted network into the wide learning model and perform dynamic self-learning adjustment of the model according to device data;

[0063] Add a weighted network to the wide learning model and adjust the wide learning model according to the weights in the weighted network to obtain a wide learning model that can perform dynamic self-learning;

[0064] The weighted network first reduces the dimension of the input layer data set X through the convolutional layer, then obtains the feature map R of N×M through the activation function, and then performs a normalization operation on R through the convolutional layer and the activation function to generate the attention weight σ from 0 to 1. The calculation formula is as follows:

[0065] M = tanh(comv1X) (5)

[0066] σ = sigmoid(conv2M) (6)

[0067] Where conv1 and conv2 represent convolution operations, tanh and sigmoid represent the tanh function and the sigmoid function respectively, and the finally feature-weighted feature output is X':

[0068] X′ = σ × X (7)

[0069] At the same time, the weighted network normalizes the weighted feature output into the feature node layer through the fully connected layer. The self-learned feature node layer is represented as follows:

[0070] Z N×b = X′ N×M ·W2 M×b (8)

[0071] Step 4: Establish a health assessment model for the charging terminal device based on the current, voltage, power change curve, utilization rate, temperature sensor data, smoke sensor data, and order distribution time of the device's historical orders;

[0072] Establish a K-means evaluation model to evaluate the health level of the charging terminal device. As an unsupervised clustering method, K-means is usually used to automatically divide samples into k clusters. The purpose of the K-means clustering method is to assign all N samples to k clusters by minimizing the sum of the distances from points to the centroids. K-means minimizes the squared error J for the cluster partition C obtained by clustering as follows:

[0073]

[0074] Where C = {C1, C2, …, C k} represents the health level clusters of k types of charging terminal devices, It is a feature matrix of the current charging terminal device's current, voltage, power change curve, utilization rate, temperature sensor data, smoke sensor data, and order distribution time data. represents the cluster centroid of the i-th cluster;

[0075] Step 4.1: Randomly select k cluster centroids of the charging terminal device health level feature samples;

[0076] Step 4.2: Measure the distance between each sample and each centroid, and assign each sample to the nearest cluster centroid, and iterate n times in this way;

[0077] Step 4.3: In each iteration process, update the centroids of each cluster using the mean value;

[0078] Step 4.4: For the k cluster centroids, use Step 4.2 and Step 4.3 to perform iterative updates until the cluster centroids are stable or the function converges.

[0079] Step 5: Calculate the health level of the charging terminal device according to the K-means evaluation model, and find the charging terminal node corresponding to this health level through the configuration model in Step 1. Give different processing suggestions according to different health levels. The processing flow chart of the charging terminal device health level is as Figure 3 shown.

[0080] The above is only a preferred embodiment of the present invention, and it is not a limitation of the present invention in any other form. Any modification or equivalent change made according to the technical essence of the present invention still belongs to the scope protected by the present invention.

Claims

1. A method for evaluating the health status of charging terminal devices based on width learning and K-means, the specific steps are as follows, and it is characterized in that: Step 1: Use the electric energy collection terminal to configure the charging terminal network to reduce the number of configured electric energy collection terminals; Step 2: Respectively establish a theoretical health status evaluation model for electric vehicle batteries of different models using width learning. The width learning model can be expressed as: Among them, Z is the feature node layer, H is the enhanced node layer, Y is the output of the output layer, W1 is the connection weight matrix, N is the number of input samples, b is the number of feature nodes, d is the number of enhanced nodes, and Q is the dimension of the output layer; in the formula, Z N×b can be expressed as: where X is the input data, W2 is the optimal input weight matrix obtained by sparse auto - encoding, M is the feature dimension of the input data, and N is the number of input data; H in Equation (2) N×d can be expressed as: In the formula, W3 is a random matrix, and β is a bias; the feature nodes are subjected to feature extraction through a sparse autoencoder to eliminate redundant features, and then enhanced into enhanced nodes through a non-linear activation function to form an enhanced node layer H; then the feature layer and the enhanced layer are merged, and the network is extended in a horizontal manner. Finally, the ridge regression algorithm is used to obtain the connection weight matrix W1; Add a weighted network to the width learning model, and adjust the width learning model according to the weights in the weighted network to obtain a width learning model that can dynamically self-learn; The weighted network first takes the data set of the input layer X to reduce the dimension through the convolutional layer, and then obtains the feature map of N×M through the activation function R , and then through the convolutional layer and the activation function to R perform a normalization operation to generate the attention weights from 0 to 1 σ , and the calculation formula is as follows: Among them, conv1 and conv2 represent convolution operations, tanh and sigmoid represent the tanh function and the sigmoid function respectively, and the finally feature-weighted feature output is X’ : At the same time, the weighted network outputs the weighted features into the feature node layer through full connection layer normalization. The self-learned feature node layer is expressed as follows: Form a multi-dimensional matrix with the voltage, current, charging time, impedance, and temperature data during battery charging as the input data of the input layer of width learning, and use the SOH value of the battery as the output of the output layer of the width learning network to establish a width learning model; Step 3: Establish a health evaluation model for the charging terminal device through the current, voltage, power change curve, utilization rate, temperature sensor data, smoke sensor data, and order distribution time of the device's historical orders; Step 4: Calculate the health level of the charging terminal device according to the K-means evaluation model, and find the charging terminal node corresponding to this health level through the configuration model in Step 1, and give different treatment suggestions according to different health levels.

2. The method for evaluating the health state of a charging terminal device based on width learning and K-means according to claim 1, wherein: The process of the electric energy collection terminal configuration model in Step 1 can be expressed as: The electric energy collection terminal configuration model is expressed by the following formula: Where C is the network configuration cost of the electric energy acquisition terminal, x i is the configuration coefficient of the i th charging node. If the electric energy acquisition terminal is configured for this node, then x i is equal to 1; otherwise x i is equal to 0. c i is the cost coefficient for configuring the electric energy acquisition terminal for the i th charging node. n is the number of nodes. u i being equal to 1 indicates that the voltage of the i th node is measurable. I ij being equal to 1 indicates that the current between the i th node and the j th node is measurable. r ij represents the connection coefficient between the i th node and the j th node.

3. The method for evaluating the health status of a charging terminal device based on width learning and K-means according to claim 1, wherein: The process of establishing the health evaluation model for the charging terminal device in Step 3 can be expressed as: Establish a K-means evaluation model to evaluate the health level of the charging terminal device. As an unsupervised clustering method, K-means is usually used to automatically divide samples into k clusters. The purpose of the K-means clustering method is to minimize the sum of the distances from points to the centroids and assign all N samples to k clusters. K-means minimizes the squared error J for the cluster partition C obtained by clustering as follows: where \(C = \{C_1, C_2, \ldots, C\) k \}\) represents the healthy level clusters of \(k\) types of charging terminal devices, is a feature matrix of the current charging terminal device's current, voltage, power change curve, utilization rate, temperature sensor data, smoke sensor data, and order distribution time data, represents the cluster centroid of the \(i\)-th cluster; Step 4.1: Randomly select k cluster centroids of the health level feature samples of the charging terminal device; Step 4.2: Measure the distance between the sample and each centroid, and assign each sample to the nearest cluster centroid, and iterate n times in this way; Step 4.3: During each iteration, update the centroids of each cluster using the mean value; Step 4.4: For the k cluster centroids, use Step 4.2 and Step 4.3 to perform iterative updates until the cluster centroids are stable or the function converges.

Images