Fire early warning method for photovoltaic power generation device

Abstract

The present invention relates to the technical field of photovoltaic power generation device early warning, and in particular to a fire early warning method for a photovoltaic power generation device. The method comprises the following steps: acquiring historical image data of a photovoltaic power generation device; constructing a photovoltaic power generation device fire early warning model, inputting the historical image data of the photovoltaic power generation device into the photovoltaic power generation device fire early warning model, iteratively performing forward propagation, loss calculation, and backward propagation, and when the number of iterations reaches a threshold, outputting the trained photovoltaic power generation device fire early warning model; and deploying the trained photovoltaic power generation device fire early warning model into a monitoring system, processing real-time image data of the photovoltaic power generation device in real time by means of the monitoring system, and determining whether a fire occurs in the photovoltaic power generation device, if not, repeatedly processing the real-time image data of the photovoltaic power generation device in real time by means of the monitoring system, and if yes, triggering an early warning mechanism.

Description

A method for early warning of fires in photovoltaic power generation equipment Technical Field

[0001] This invention relates to the field of early warning technology for photovoltaic power generation equipment, and more specifically, to a method for early warning of fires in photovoltaic power generation equipment. Background Technology

[0002] Currently, traditional fire detection methods for photovoltaic (PV) power generation equipment typically employ ground fault detection, DC arc detection, and fire hazard analysis. Existing methods, such as ground fault detection, often fail to detect problems promptly, while DC arc detection and fire hazard analysis require comprehensive consideration of risk factors. Consequently, the accuracy and timeliness of fire detection are relatively low, potentially impacting the safety and stability of PV power generation equipment and affecting its normal operation. Therefore, to address these issues, we have designed a fire early warning method for PV power generation equipment.

[0003] Summary of the Invention

[0004] The purpose of this invention is to provide a method for early warning of fires in photovoltaic power generation equipment. This method constructs a fire early warning model for photovoltaic power generation equipment using historical data. By setting the loss calculation process of the fire early warning model and optimizing it with an optimization algorithm, it can not only improve the accuracy and timeliness of fire detection, but also enable the monitoring system based on the fire early warning model to achieve intelligent early warning and emergency response, thus helping to reduce the losses caused by fires.

[0005] The embodiments of the present invention are achieved through the following technical solutions:

[0006] A method for early warning of fires in photovoltaic power generation equipment, the method comprising the following steps:

[0007] Acquire historical image data of photovoltaic power generation equipment;

[0008] A fire early warning model for photovoltaic power generation equipment is constructed. Historical image data of photovoltaic power generation equipment is input into the fire early warning model for photovoltaic power generation equipment. After iteratively executing forward propagation, loss calculation, and back propagation until the number of iterations reaches the threshold, the trained fire early warning model for photovoltaic power generation equipment is output.

[0009] The trained photovoltaic power generation equipment fire early warning model is deployed to the monitoring system. The monitoring system processes the real-time image data of the photovoltaic power generation equipment in real time and determines whether a fire has occurred. If not, the monitoring system repeats the real-time processing of the real-time image data of the photovoltaic power generation equipment; if so, the early warning mechanism is triggered.

[0010] Optionally, the photovoltaic power generation equipment fire early warning model is specifically a YOLOv3 model, which includes an input layer, a convolutional layer, and a multi-scale detection layer; the forward propagation specifically involves inputting historical image data of the photovoltaic power generation equipment into the photovoltaic power generation equipment fire early warning model, extracting and outputting the potential fire boundary box and related confidence scores of the photovoltaic power generation equipment.

[0011] Optionally, the loss calculation is specifically used to assess the difference between the output value and the true value of the photovoltaic power generation equipment fire early warning model, including: coordinate loss, confidence loss and category loss.

[0012] Optionally, the formula for calculating the coordinate loss is:

[0013] Among them, S 2 B represents the number of grid cells, and B represents the number of bounding boxes in each grid cell. To determine whether the j-th bounding box in the i-th grid contains an object, x i y i w i h i These are the coordinates and size of the predicted bounding box, respectively. These are the coordinates and size of the actual bounding box, respectively.

[0014] Optionally, the formula for calculating the confidence loss is:

[0015] Among them, C i The confidence level of the prediction. For true confidence level, λ is the indicator function for meshes without objects. noobj The weight of the bounding box that does not contain objects.

[0016] Optionally, the formula for calculating the category loss is:

[0017] Where, p i (c) represents the probability that the object predicted by the photovoltaic power generation equipment fire early warning model belongs to category c. This represents the true probability.

[0018] Optionally, the photovoltaic power generation equipment fire early warning model is further optimized through an optimization algorithm during the backpropagation process. The calculation formula of the optimization algorithm is as follows:

[0019] Where, θ t Let η be the parameter at time step t, and η be the learning rate. These are the first and second moment estimates after bias correction, respectively, where α is a constant.

[0020] Optionally, the objective function of the trained photovoltaic power generation equipment fire early warning model is: L total =λ coord L coord +L conf +λ class L class

[0021] Among them, L total For the total loss, λ coord , λ class These are the weights for coordinate loss and category loss, respectively.

[0022] The technical solutions of the embodiments of the present invention have at least the following advantages and beneficial effects:

[0023] This invention constructs a fire early warning model for photovoltaic power generation equipment using historical data. By setting the loss calculation process of the fire early warning model and optimizing it with an optimization algorithm, not only can the accuracy and timeliness of fire detection be improved, but the monitoring system based on the fire early warning model can also realize intelligent early warning and emergency response, helping to reduce the losses caused by fire. Attached Figure Description

[0024] Figure 1 is a flowchart illustrating a method for early warning of fires in photovoltaic power generation equipment according to an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0026] As shown in Figure 1, the present invention provides one embodiment: a method for early warning of fires in photovoltaic power generation equipment, the method comprising the following steps:

[0027] Acquire historical image data of photovoltaic power generation equipment;

[0028] A fire early warning model for photovoltaic power generation equipment is constructed. Historical image data of photovoltaic power generation equipment is input into the fire early warning model for photovoltaic power generation equipment. After iteratively executing forward propagation, loss calculation, and back propagation until the number of iterations reaches the threshold, the trained fire early warning model for photovoltaic power generation equipment is output.

[0029] The trained photovoltaic power generation equipment fire early warning model is deployed to the monitoring system. The monitoring system processes the real-time image data of the photovoltaic power generation equipment in real time and determines whether a fire has occurred. If not, the monitoring system repeats the real-time processing of the real-time image data of the photovoltaic power generation equipment; if so, the early warning mechanism is triggered.

[0030] In this embodiment, the photovoltaic power generation equipment fire early warning model is specifically a YOLOv3 model, which includes an input layer, a convolutional layer, and a multi-scale detection layer. The forward propagation specifically involves inputting historical image data of the photovoltaic power generation equipment into the photovoltaic power generation equipment fire early warning model, extracting and outputting the potential fire boundary boxes and related confidence scores of the photovoltaic power generation equipment.

[0031] In implementation, during the forward propagation phase, since this embodiment uses the YOLOv3 model, the input layer of this embodiment sets an appropriate input resolution based on the size of the photovoltaic power generation equipment image; the convolutional layer specifically acts as a feature extractor; and the multi-scale detection layer specifically performs detection at three different scales to detect fire features of different sizes. The YOLOv3 model input image passes through the network, calculating the output of each convolutional layer until the final output layer. This process can be viewed as a function f(x; θ'), where x is the input image and θ' is the network parameter. Furthermore, it is understood that the early warning mechanism in this embodiment can be any method, such as audible or visual alarms.

[0032] Furthermore, the loss calculation set in this embodiment consists of multiple parts, used to evaluate the difference between the output value and the true value of the photovoltaic power generation equipment fire early warning model, including: coordinate loss, confidence loss and category loss.

[0033] Specifically, the coordinate loss is used to calculate the difference between the position and size of the bounding box predicted by the model and the position and size of the true bounding box, and its calculation formula is as follows:

[0034] Among them, s 2 B represents the number of grid cells, and B represents the number of bounding boxes in each grid cell. To determine whether the j-th bounding box in the i-th grid contains an object, x i y i w i h i These are the coordinates and size of the predicted bounding box, respectively. These are the coordinates and size of the actual bounding box, respectively.

[0035] Specifically, the confidence loss is used to evaluate the difference between the confidence level of the model predicting the presence of objects in the bounding box and the true confidence level, and its calculation formula is as follows:

[0036] Among them, C i The confidence level of the prediction. For true confidence level, λ is the indicator function for meshes without objects. noobj The weight of the bounding box that does not contain objects.

[0037] Specifically, for the photovoltaic power generation equipment fire early warning model, the category loss is used to calculate the difference between the probability that the object in the model's predicted image is a fire and the actual probability. The calculation formula is as follows:

[0038] Where, p i (c) represents the probability that the object predicted by the photovoltaic power generation equipment fire early warning model belongs to category c. This represents the true probability.

[0039] In practice, by calculating coordinate loss, confidence loss and category loss, the current performance of the photovoltaic power generation equipment fire early warning model can be quantified, and necessary information can be provided for subsequent backpropagation.

[0040] In backpropagation, this embodiment calculates the gradient of the network parameters based on the results of the aforementioned loss function, and updates the parameters accordingly to improve the performance of the photovoltaic power generation equipment fire early warning model. At this stage, the photovoltaic power generation equipment fire early warning model determines the direction and magnitude of parameter updates by calculating the derivative of the loss function with respect to the network parameters. An optimizer, i.e., the optimization algorithm described in this embodiment, is used to adjust the network weights to reduce the value of the loss function. The calculation formula for the optimization algorithm is as follows:

[0041] Where, θ t Let η be the parameter at time step t, and η be the learning rate. These are the first and second moment estimates after bias correction, respectively, with α being a constant. The process of forward propagation, loss function calculation, and backpropagation is repeated continuously, updating the parameters of the photovoltaic power generation equipment fire early warning model in each iteration, until a threshold number of iterations is reached and the result is output.

[0042] In this embodiment, the ultimate goal is to minimize the weighted sum of the loss functions, including the coordinate loss, confidence loss, and class loss mentioned above. These three together constitute the overall loss, representing the goal of the photovoltaic power generation equipment fire early warning model to minimize the loss function during training. The calculation formula is: L total =λ coord L coord +L conf +λ class L class

[0043] Among them, L total For the total loss, λ coord , λ class These are the weights for coordinate loss and category loss, respectively.

[0044] The above are merely preferred embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A photovoltaic power plant fire early warning method, characterized in that, The method comprises the following steps: acquiring historical image data of a photovoltaic power generation device; constructing a photovoltaic power generation device fire warning model, inputting the historical image data of the photovoltaic power generation device into the photovoltaic power generation device fire warning model, performing forward propagation, loss calculation and backward propagation iteratively until the number of iterations reaches a threshold, and outputting the trained photovoltaic power generation device fire warning model; deploying the trained photovoltaic power generation device fire warning model to a monitoring system, processing real-time image data of the photovoltaic power generation device in real time through the monitoring system, determining whether a fire occurs in the photovoltaic power generation device, if not, repeatedly processing the real-time image data of the photovoltaic power generation device in real time through the monitoring system, and if yes, triggering a warning mechanism.

2. The photovoltaic power plant fire warning method according to claim 1, characterized in that, The photovoltaic power generation device fire warning model is specifically a YOLOV3 model, which comprises an input layer, a convolution layer and a multi-scale detection layer; the forward propagation is specifically as follows: inputting the historical image data of the photovoltaic power generation device into the photovoltaic power generation device fire warning model, extracting and outputting potential fire bounding boxes and related confidence of the photovoltaic power generation device.

3. The photovoltaic power plant fire warning method according to claim 2, characterized in that, The loss calculation is specifically used for evaluating the difference between the output value and the true value of the photovoltaic power generation device fire warning model, and comprises coordinate loss, confidence loss and category loss.

4. The photovoltaic power plant fire warning method according to claim 3, characterized in that, The calculation formula of the coordinate loss is: where S 2 is the number of grids, B is the number of bounding boxes in each grid, For whether the jth bounding box in the ith grid contains an object, x i , y i , w i , h i are the coordinates and size of the predicted bounding box, respectively, The real bounding box coordinates and size are respectively.

5. The photovoltaic power plant fire early warning method of claim 4, wherein, The calculation formula of the confidence loss is: wherein C i is a confidence of the prediction, for real trustworthiness, for the indicator function of the grid without objects, l noobj is the weight for the bounding box not containing an object.

6. The photovoltaic power plant fire warning method according to claim 5, characterized in that, The calculation formula of the category loss is: wherein p i (c) a probability that the object predicted by the fire warning model for the photovoltaic power generation device belongs to the category c, The real probability is.

7. The photovoltaic power plant fire early warning method of claim 6, wherein, The photovoltaic power generation equipment fire warning model is further optimized by an optimization algorithm in the process of back propagation, and a calculation formula of the optimization algorithm is: where θ t is the parameter at time step t, η is the learning rate, The first-order and second-order moment estimates after bias correction are respectively, and a is a constant.

8. The photovoltaic power plant fire early warning method of claim 7, wherein, The trained photovoltaic power generation equipment fire warning model has a target function as follows: L total = λ coord L coord + L conf + λ class L class where L total is the total loss, λ coord and λ class are the weights of the coordinate loss and the class loss, respectively.

Images