Wind power plant video monitoring management system and method
Through the intelligent analysis of real-time video data and the construction of fault identification models of the wind farm video surveillance management system, the problem that traditional wind turbine monitoring methods cannot achieve real-time monitoring and fault identification are solved, and the operation efficiency and reliability of wind farms have been significantly improved. .
Patent Information
- Application Number
- CN202411477656.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-22
- Publication Date
- 2025-02-11
AI Technical Summary
Traditional wind turbine monitoring methods rely on manual maintenance and simple sensor monitoring, and cannot achieve real-time monitoring, resulting in insufficient timely fault identification and high false alarm rates and missed alarm rates, which cannot meet the needs of high efficiency and reliability operations of modern wind farms.
Design a wind farm video surveillance management system, build a wind turbine fault identification model through intelligent analysis of real-time video data, and use convolutional neural network to predict and output the processing scheme to achieve intelligent monitoring.
It greatly improves the sensitivity to identify potential faults, significantly improves the operational efficiency and reliability of wind farms, and reduces the cycle and cost of manual maintenance.
Smart Images
Figure CN119418273A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of wind farm intelligent monitoring, and in particular to a wind farm video monitoring management system and method. Background Art
[0002] With the rapid development of the wind power industry, higher requirements are placed on the intelligent monitoring and fault diagnosis of wind turbines. At present, traditional wind turbine monitoring methods mainly rely on regular manual maintenance and simple sensor monitoring. The manual maintenance cycle is long and the cost is high. It is impossible to achieve real-time monitoring, which easily leads to untimely fault identification. In addition, the simple sensor monitoring method only analyzes specific data and cannot fully capture the subtle changes in the operating status of the wind turbine, making the identification of potential faults not sensitive enough. Since sensor data usually exists in isolation and is independent of each other, its false alarm rate and missed alarm rate are high, which increases the difficulty of wind turbine maintenance and cannot meet the needs of high-efficiency and reliable operation of modern wind farms. Based on this, in response to the above problems, we designed a wind farm video monitoring management system and method. Summary of the invention
[0003] The purpose of the present invention is to provide a wind farm video monitoring management system and method, which can better capture subtle changes in the operating status of wind turbines through real-time intelligent analysis of video data, greatly improve the sensitivity of identifying potential faults, and significantly improve the operating efficiency and reliability of wind farms.
[0004] The embodiments of the present invention are implemented by the following technical solutions:
[0005] A wind farm video monitoring management system, comprising:
[0006] Collection unit: obtains real-time video data and historical video data of wind farms;
[0007] Preprocessing unit: preprocessing the real-time video data and historical video data of the wind farm respectively, and dividing the preprocessed historical video data of the wind farm into a training set and a test set;
[0008] Computation unit: construct a wind turbine fault identification model, input the training set into the wind turbine fault identification model for calculation, obtain the wind turbine fault prediction result, calculate the loss value between the test set and the training set through the loss function, iteratively update the parameters of the wind turbine fault identification model through the loss value, until the maximum number of iterations is reached, and the training of the wind turbine fault identification model is completed;
[0009] Execution unit: Input the real-time video data of the wind farm into the trained wind turbine fault identification model for calculation, output the corresponding processing plan according to the prediction results, and complete the intelligent monitoring of the wind farm.
[0010] Optionally, the real-time video data and the historical video data of the wind farm are preprocessed respectively, which specifically comprises: frame extraction, image scaling, grayscale conversion and data enhancement of the real-time video data and the historical video data of the wind farm.
[0011] Optionally, the wind turbine fault identification model is specifically a convolutional neural network model, including: an input layer, a convolution layer, a pooling layer and a fully connected layer; wherein the calculation formula of the convolution layer is:
[0012]
[0013] Where I is the input feature map, K is the convolution kernel, (I*K)(i,j) is the output value of the convolution operation at position (i,j), i and j are the pixel position indexes on the feature map, m and n are the position indexes on the convolution kernel, which are used to traverse the convolution kernel size;
[0014] The calculation formula of the pooling layer is:
[0015] y(i,j)=max{x(i+m,j+n)|0≤m<M,0≤n<N
[0016] Where y(i,j) is the output value of position (i,j) after the pooling operation, and M×N is the pooling window size;
[0017] The calculation formula of the fully connected layer is:
[0018]
[0019] Among them, z is the linear output of the fully connected layer, Prediction results of wind turbine fault identification model.
[0020] Optionally, the wind turbine fault identification model is further provided with a Re lu activation function, the calculation formula of which is:
[0021]
[0022] Among them, f(x) is the output of the activation function.
[0023] Optionally, the loss value between the test set and the training set is calculated by a loss function, and the specific calculation formula is:
[0024]
[0025] Among them, min(L) is the loss function, N is the number of samples, K is the total number of categories, yik is the actual probability that the i-th sample belongs to the k-th category, is the probability that the i-th sample belongs to the k-th class predicted by the wind turbine fault recognition model, L is the number of network layers, λ1 and λ2 are the hyperparameters of the regularization strength, Jl is the number of neurons in the l-th layer, and w lj is the weight of the jth neuron in the lth layer, α is the weight of the variational regularization term, and KL(q(z|x)||p(z)) is the KL divergence.
[0026] Optionally, the parameters of the wind turbine fault identification model are updated iteratively through the loss value, and the calculation formula is:
[0027]
[0028] Among them, w' lj is the updated weight, η is the learning rate, is the cross entropy loss L CE For weight w lj The partial derivative of sign(w lj ) is w lj The symbol function of is the KL divergence term for weight w lj The partial derivative of .
[0029] Optionally, the processing scheme corresponding to the output of the prediction result is specifically:
[0030] Define the fault threshold as T;
[0031] Determine whether the prediction result is greater than or equal to T;
[0032] If yes, it is determined that there is a wind farm equipment failure, the alarm system is activated, and the emergency automation mechanism is executed;
[0033] If not, it is determined that the wind farm equipment is normal, and real-time video data of the wind farm is continuously acquired.
[0034] A wind farm video monitoring management method, the method comprising the following steps:
[0035] Obtain real-time and historical video data of wind farms;
[0036] Preprocess the real-time video data and historical video data of the wind farm respectively, and divide the preprocessed historical video data of the wind farm into a training set and a test set;
[0037] Construct a wind turbine fault identification model, input the training set into the wind turbine fault identification model for calculation, obtain the wind turbine fault prediction result, calculate the loss value between the test set and the training set through the loss function, iteratively update the parameters of the wind turbine fault identification model through the loss value, until the maximum number of iterations is reached, and complete the training of the wind turbine fault identification model;
[0038] The real-time video data of the wind farm is input into the trained wind turbine fault identification model for calculation, and the corresponding processing plan is output according to the prediction results to complete the intelligent monitoring of the wind farm.
[0039] The technical solution of the embodiment of the present invention has at least the following advantages and beneficial effects:
[0040] The embodiments of the present invention can better capture subtle changes in the operating status of wind turbines through intelligent analysis of real-time video data, greatly improve the sensitivity of identifying potential faults, and significantly improve the operating efficiency and reliability of wind farms. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 A logical schematic diagram of a wind farm video monitoring management system provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0042] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.
[0043] like Figure 1 As shown, the present invention provides one embodiment: a wind farm video monitoring management system, comprising:
[0044] Collection unit: obtains real-time video data and historical video data of wind farms;
[0045] Preprocessing unit: preprocessing the real-time video data and historical video data of the wind farm respectively, and dividing the preprocessed historical video data of the wind farm into a training set and a test set;
[0046] Computation unit: construct a wind turbine fault identification model, input the training set into the wind turbine fault identification model for calculation, obtain the wind turbine fault prediction result, calculate the loss value between the test set and the training set through the loss function, iteratively update the parameters of the wind turbine fault identification model through the loss value, until the maximum number of iterations is reached, and the training of the wind turbine fault identification model is completed;
[0047] Execution unit: Input the real-time video data of the wind farm into the trained wind turbine fault identification model for calculation, output the corresponding processing plan according to the prediction results, and complete the intelligent monitoring of the wind farm.
[0048] Specifically, the real-time video data and the historical video data of the wind farm are preprocessed respectively, which specifically includes: frame extraction, image scaling, grayscale conversion and data enhancement of the real-time video data and the historical video data of the wind farm.
[0049] In the implementation, frame extraction is specifically to extract static image frames from the real-time video data and historical video data of the wind farm; image scaling is specifically to adjust the extracted image frames to a uniform size. Interpolation methods (such as bilinear interpolation, nearest neighbor interpolation, etc.) are used to scale the image to the input size required by the model. Grayscale conversion is specifically to convert the color image into a grayscale image through a weighted RGB channel conversion formula to reduce the amount of calculation and complexity. While retaining the brightness information, it is also suitable for many wind farm image analysis tasks. Data enhancement is specifically to expand the data set and increase its diversity, thereby improving the generalization ability of the model.
[0050] In this embodiment, the wind turbine fault identification model is specifically a convolutional neural network model, including: an input layer, a convolution layer, a pooling layer and a fully connected layer; wherein the input layer is W×H×C:, representing the data dimension of the input image, wherein W and H are the width and height of the input image respectively, and C is the number of channels of the image.
[0051] The calculation formula of the convolutional layer is:
[0052]
[0053] Among them, I is the input feature map, K is the convolution kernel, (I*K)(i,j) is the output value of the convolution operation at position (i,j), i and j are the pixel position indexes on the feature map, m and n are the position indexes on the convolution kernel, which are used to traverse the convolution kernel size.
[0054] Rel u activation function, its calculation formula is:
[0055] f(x)=max(0,x)
[0056] Among them, f(x) is the output of the activation function.
[0057] The calculation formula of the pooling layer is:
[0058] y(i,j)=max{x(i+m,j+n)|0≤m<M,0≤n<N
[0059] Among them, y(i,j) is the output value of position (i,j) after the pooling operation, and M×N is the pooling window size.
[0060] The calculation formula of the fully connected layer is:
[0061]
[0062] Among them, z is the linear output of the fully connected layer, Prediction results of wind turbine fault identification model.
[0063] In the training process of the wind turbine fault identification model, the loss value between the test set and the training set is calculated by the loss function, and the specific calculation formula is:
[0064]
[0065] Among them, min(L) is the loss function, N is the number of samples, K is the total number of categories, yik is the actual probability that the i-th sample belongs to the k-th category, is the probability that the i-th sample belongs to the k-th class predicted by the wind turbine fault recognition model, L is the number of network layers, λ1 and λ2 are the hyperparameters of the regularization strength, Jl is the number of neurons in the l-th layer, and w lj is the weight of the jth neuron in the lth layer, α is the weight of the variational regularization term, and KL(q(z|x)||p(z)) is the KL divergence.
[0066] In implementation, this embodiment aims to minimize the loss function, while taking into account the performance and generalization ability of the model on the training data, and is able to handle a variety of complex fault conditions in wind farms. L1 regularization helps to automatically select important features, and the combination of multiple regularization methods improves the stability of the model. By analyzing the weights, the importance of different features can be understood.
[0067] Based on an implementation case: a training set is set to include labeled video clips of normal operation and various fault states, and training is performed using the loss function set in this embodiment, including cross entropy loss and multiple regularization terms, and the model parameters are iteratively updated through back propagation and gradient descent.
[0068] In a certain place, in windy weather, the Internet of Things or other detection equipment detected that a certain wind turbine might be abnormal. After model recognition, it was determined that the wind turbine was operating normally: 25%; blade cracks: 65%. The judgment was based on: abnormal texture changes on the blade surface were detected in continuous video frames, and the wind turbine blades had a small but abnormal bending pattern during rotation. The system immediately alerted the maintenance team and provided video clips and analysis results. The maintenance staff confirmed the abnormality through remote video and then sent a team for on-site inspection. A crack about 30 cm long was found and repaired in time to avoid more serious damage.
[0069] In the back propagation of the wind turbine fault identification model, the parameters of the wind turbine fault identification model are iteratively updated through the loss value, and the calculation formula is:
[0070]
[0071] Among them, w' lj is the updated weight, η is the learning rate, is the cross entropy loss L CE For weight w lj The partial derivative of sign(w lj ) is w lj The symbol function of is the KL divergence term for weight w lj The partial derivative of .
[0072] In the specific application of this embodiment, the processing scheme corresponding to the output according to the prediction result is specifically:
[0073] Define the fault threshold as T;
[0074] Determine whether the prediction result is greater than or equal to T;
[0075] If yes, it is determined that there is a wind farm equipment failure, the alarm system is activated, and the emergency automation mechanism is executed;
[0076] If not, it is determined that the wind farm equipment is normal, and real-time video data of the wind farm is continuously acquired.
[0077] In specific applications, if it is determined that the wind farm equipment is faulty, the following actions are performed: 1. Immediately notify the on-site operation and maintenance personnel or relevant persons in charge through the alarm system. 2. Start the emergency automation mechanism, including deceleration or shutdown. 3. Record the fault information, time, possible fault type and preliminary diagnosis in the system log for subsequent analysis. 4. Determine the specific fault type and location. If it is determined that the wind farm equipment is normal, continue to monitor the video data in real time.
[0078] This embodiment optimizes the unit operation efficiency through intelligent control during high winds and low temperatures. The specific process is: obtain environmental data of the wind farm such as the current wind speed and temperature through meteorological sensors; collect real-time operation data of the wind turbine, including: wind turbine speed, power output, unit status, etc. Determine whether the current wind speed exceeds the rated wind speed of the wind turbine; determine whether the current temperature is lower than the optimal operating temperature range; calculate the optimal load mode at the current wind speed and temperature, and solve the ideal power generation at the current wind speed through the following calculation formula: ρ is air density, A is swept area, λ' and β are speed ratio and yaw angle respectively, V is wind speed, and Cp(λ',β) is power coefficient. The angle and speed of wind turbine blades are automatically adjusted according to power output and environmental conditions.
[0079] In another embodiment, the present invention provides a wind farm video monitoring management method, the method comprising the steps of:
[0080] Obtain real-time and historical video data of wind farms;
[0081] Preprocess the real-time video data and historical video data of the wind farm respectively, and divide the preprocessed historical video data of the wind farm into a training set and a test set;
[0082] Construct a wind turbine fault identification model, input the training set into the wind turbine fault identification model for calculation, obtain the wind turbine fault prediction result, calculate the loss value between the test set and the training set through the loss function, iteratively update the parameters of the wind turbine fault identification model through the loss value, until the maximum number of iterations is reached, and complete the training of the wind turbine fault identification model;
[0083] The real-time video data of the wind farm is input into the trained wind turbine fault identification model for calculation, and the corresponding processing plan is output according to the prediction results to complete the intelligent monitoring of the wind farm.
[0084] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A wind farm video monitoring management system, characterized in that: include: Collection unit: obtains real-time video data and historical video data of wind farms; Preprocessing unit: preprocessing the real-time video data and historical video data of the wind farm respectively, and dividing the preprocessed historical video data of the wind farm into a training set and a test set; Computation unit: construct a wind turbine fault identification model, input the training set into the wind turbine fault identification model for calculation, obtain the wind turbine fault prediction result, calculate the loss value between the test set and the training set through the loss function, iteratively update the parameters of the wind turbine fault identification model through the loss value, until the maximum number of iterations is reached, and the training of the wind turbine fault identification model is completed; Execution unit: Input the real-time video data of the wind farm into the trained wind turbine fault identification model for calculation, output the corresponding processing plan according to the prediction results, and complete the intelligent monitoring of the wind farm.
2. The wind farm video monitoring management system according to claim 1, characterized in that: The preprocessing of the wind farm real-time video data and the historical video data respectively specifically includes: frame extraction, image scaling, grayscale conversion and data enhancement of the wind farm real-time video data and the historical video data.
3. The wind farm video monitoring management system according to claim 2, characterized in that: The wind turbine fault identification model is specifically a convolutional neural network model, including: an input layer, a convolution layer, a pooling layer and a fully connected layer; wherein the calculation formula of the convolution layer is: Where I is the input feature map, K is the convolution kernel, (I*K)(i,j) is the output value of the convolution operation at position (i,j), i and j are the pixel position indexes on the feature map, m and n are the position indexes on the convolution kernel, which are used to traverse the convolution kernel size; The calculation formula of the pooling layer is: y(i,j)=max{x(i+m,j+n)|0≤m<M,0≤n<N Where y(i,j) is the output value of position (i,j) after the pooling operation, and M×N is the pooling window size; The calculation formula of the fully connected layer is: Among them, z is the linear output of the fully connected layer, Prediction results of wind turbine fault identification model.
4. The wind farm video monitoring management system according to claim 3, characterized in that: The wind turbine fault identification model is also provided with a Relu activation function, and its calculation formula is: f(x)=max(0,x) Among them, f(x) is the output of the activation function.
5. The wind farm video monitoring management system according to claim 4, characterized in that: The loss function is used to calculate the loss value between the test set and the training set, and the specific calculation formula is: Among them, min(L) is the loss function, N is the number of samples, K is the total number of categories, and y ik is the actual probability that the i-th sample belongs to the k-th class, is the probability that the i-th sample belongs to the k-th class predicted by the wind turbine fault identification model, L is the number of network layers, λ1 and λ2 are the hyperparameters of the regularization strength, and J l is the number of neurons in layer l, w lj is the weight of the jth neuron in the lth layer, α is the weight of the variational regularization term, and KL(q(z|x)||p(z)) is the KL divergence.
6. The wind farm video monitoring management system according to claim 5, characterized in that: The parameters of the wind turbine fault identification model are updated iteratively through the loss value, and the calculation formula is: Among them, w′ lj is the updated weight, η is the learning rate, is the cross entropy loss L CE For weight w lj The partial derivative of sign(w lj ) is w lj The symbol function of is the KL divergence term for weight w lj The partial derivative of .
7. The wind farm video monitoring management system according to claim 6, characterized in that: The processing scheme corresponding to the prediction result output is specifically as follows: Define the fault threshold as T; Determine whether the prediction result is greater than or equal to T; If yes, it is determined that there is a wind farm equipment failure, the alarm system is activated, and the emergency automation mechanism is executed; If not, it is determined that the wind farm equipment is normal, and real-time video data of the wind farm is continuously acquired.
8. A wind farm video monitoring management method, characterized in that: The steps of the method include: Obtain real-time and historical video data of wind farms; Preprocess the real-time video data and historical video data of the wind farm respectively, and divide the preprocessed historical video data of the wind farm into a training set and a test set; Construct a wind turbine fault identification model, input the training set into the wind turbine fault identification model for calculation, obtain the wind turbine fault prediction result, calculate the loss value between the test set and the training set through the loss function, iteratively update the parameters of the wind turbine fault identification model through the loss value, until the maximum number of iterations is reached, and complete the training of the wind turbine fault identification model; The real-time video data of the wind farm is input into the trained wind turbine fault identification model for calculation, and the corresponding processing plan is output according to the prediction results to complete the intelligent monitoring of the wind farm.