A fish-light complementary photovoltaic power station disease data generation method

Abstract

The present application relates to a kind of fish light complementary photovoltaic power station disease data generation method, it is related to disease data generation technical field, it includes the following steps: step S1: obtaining existing disease data set;Step S2: based on existing disease data set, constructs disease diffusion model;Step S3: input initial disease data set and environmental condition, runs disease diffusion model, outputs simulation result, generates extended disease data set;Step S4: generating formula is against network optimization extended disease data set;Step S5: obtains optimized disease data set.The present application has the effect of improving the accuracy and reliability of disease identification.

Description

Technical Field

[0001] This invention relates to the field of disease data generation technology, and in particular to a method for generating disease data in a solar-fishery hybrid photovoltaic power station. Background Technology

[0002] In solar-fishery complementary photovoltaic power plants, accurate identification of defects is crucial to operation and maintenance efficiency and power plant safety. The accuracy of defect identification and the richness of the dataset directly affect the operation and maintenance results.

[0003] The existing disease datasets for solar-aquaculture hybrid power plants are generally small in scale. Due to limitations in actual shooting conditions and high data annotation costs, it is difficult to obtain disease data, resulting in insufficient samples. This limits the training effect of disease identification models and affects the accuracy and reliability of disease identification. Summary of the Invention

[0004] To improve the accuracy and reliability of disease identification, this application provides a method for generating disease data in a solar-fishery hybrid photovoltaic power station.

[0005] The method for generating defect data of a solar-fishery hybrid photovoltaic power station provided in this application adopts the following technical solution: A method for generating data on defects in a solar-fishery hybrid photovoltaic power station includes the following steps: Step S1: Obtain the existing disease dataset; Step S2: Construct a disease spread model based on the existing disease dataset; Step S3: Input the initial disease dataset and environmental conditions, run the disease spread model, output the simulation results, and generate an extended disease dataset; Step S4: Generative adversarial network optimizes and expands the disease dataset; Step S5: Obtain the optimized disease dataset.

[0006] By adopting the above technical solution, existing disease data is first collected to provide basic data support for the construction of the disease spread model. The constructed disease spread model simulates the spatial and temporal propagation patterns of diseases. The initial disease dataset and environmental conditions are input into the disease spread model, and the simulation results are obtained through the disease spread model calculation, thereby generating an extended disease dataset to achieve the initial expansion of the dataset. Generative adversarial networks are introduced, and the extended disease dataset is optimized through adversarial training between the generator and the discriminator to obtain a high-quality and diversified optimized disease dataset. The dataset is first initially expanded using the disease spread model, and then optimized using a generative adversarial network. This solves the problem of small scale and insufficient sample size of the existing disease dataset of fishery-solar hybrid power stations, improves the training effect of the disease identification model, and enhances the accuracy and reliability of disease identification.

[0007] Preferably, the disease spread model in step S2 is: ,in, Image of the disease at the next moment. This is an image of the disease at the current moment. It is the time step for feature learning. It is a factor influencing spatial distribution. These are factors influencing the ontological characteristics.

[0008] Preferably, in step S2 Using an average distribution, in step S2 The RGB three-color grayscale matrix of the disease image is used.

[0009] By adopting the above technical solutions, the disease spread model can capture the distribution patterns and ontological characteristics of diseases at different time and space scales through multi-scale modeling, simulate the spread of diseases in time and space, and generate an extended disease dataset that conforms to the real spread patterns.

[0010] Preferably, step S4 includes the following steps: Step S41: The generator takes the expanded disease dataset as the initial input and obtains the generated disease dataset through the generator's multi-layer convolutional structure. The discriminator takes the generated disease dataset and the existing disease dataset as the initial input and obtains the true probability judgment value through the discriminator's multi-layer convolutional structure. Step S42: Calculate the generator loss value ; Calculate the discriminator loss value ; Step S43: If the generator loss value is not lower than the generator loss threshold or the discriminator loss value is not lower than the discriminator loss threshold, proceed to step S44; if the generator loss value is lower than the generator loss threshold and the discriminator loss value is lower than the discriminator loss threshold, generate a disease dataset as an optimized disease dataset and proceed to step S5. Step S44: If the generator loss value is not lower than the generator loss threshold, then correct the generator multi-layer convolutional structure; if the discriminator loss value is not lower than the discriminator loss threshold, then correct the discriminator multi-layer convolutional structure. Step S45; Proceed to step S41.

[0011] Preferably, the multi-layer convolutional structure of the generator in step S41 is as follows: ,in, It is the first The generator output of the layer, It is the first The generator output of the layer, It is a convolution kernel. It's a bias. The ReLU activation function is used.

[0012] Preferably, the multi-layer convolutional structure of the discriminator in step S41 is as follows: ,in, It is the first The discriminator output of the layer, It is the first The discriminator output of the layer, It is a convolution kernel. It's a bias. The ReLU activation function is used.

[0013] Preferably, the generator loss value in step S42 ,in, Generator adversarial loss value, It is the generator spatial distribution loss weight value. Generator spatial distribution loss value, These are the generator ontology feature loss weight values. Generator ontology feature loss value.

[0014] Preferably, the generator is adversarial to the loss value ,in To expand the input noise of the disease dataset, For generator, The discriminator; the generator spatial distribution loss value ,in, To generate the spatial distribution mean vector of the sample set, The mean vector of the spatial distribution of the real sample set; the generator ontology feature loss value. ,in These are real disease image samples.

[0015] Preferably, the discriminator loss value in step S42 ,in, Discriminator adversarial loss value, These are the discriminator spatial distribution loss weight values. Discriminator spatial distribution loss value, These are the discriminator's ontological feature loss weight values. Discriminator ontological feature loss value.

[0016] Preferably, the discriminator is effective against loss values. ,in These are samples from a real disease dataset. To expand the input of the disease dataset, For generator, The discriminator; the spatial distribution loss value of the discriminator ,in The mean of the discriminant's judgment results for the generated samples is denoted as . The mean of the true sample distribution; the discriminator ontology feature loss value. ,in This represents the discriminator's judgment output on the generated samples. These are real disease image samples.

[0017] By adopting the above technical solutions, the multi-layer convolutional structures of the generator and discriminator can more effectively extract image features. Through layer-by-layer processing, richer spatial information and details are captured. The generator's multi-layer convolutional structure is corrected based on the generator's loss value, and the discriminator's multi-layer convolutional structure is corrected based on the discriminator's loss value, thereby improving the realism and consistency of the generated disease dataset. The introduction of spatial distribution loss values ​​for the generator and discriminator facilitates the matching of the overall structure of the generated image with the real sample. The introduction of ontological feature loss values ​​for the generator and discriminator enhances the similarity of the generated samples in the feature space. The generator and discriminator are trained adversarially and dynamically corrected, which facilitates the generator to generate diverse and high-quality generated disease datasets based on the extended disease dataset, improving the realism and detailed features of the generated samples, and achieving the effect of iterative optimization of the extended disease dataset.

[0018] In summary, this application includes at least one of the following beneficial technical effects: 1. By using multi-scale modeling, the disease spread model can capture the distribution patterns and ontological characteristics of diseases at different time and space scales, simulate the spread of diseases in time and space, generate an extended disease dataset that conforms to the real spread patterns, and optimize the extended disease dataset by combining generative adversarial networks, thereby enhancing the realism of the generated samples and making them closer to the actual situation, thus obtaining a high-quality and diverse optimized disease dataset. 2. Introducing multi-layer convolutional structures for the generator and discriminator in generative adversarial networks can more effectively extract image features and capture richer spatial information and details through layer-by-layer processing; 3. Introducing generator spatial distribution loss and discriminator spatial distribution loss facilitates the matching of the overall structure of the generated image with the real sample. Introducing generator ontology feature loss and discriminator ontology feature loss enhances the similarity of the generated samples in the feature space, making it easier for the generator to generate diverse and high-quality generated disease datasets based on the extended disease dataset, improving the realism and detailed features of the generated samples, and achieving the effect of iterative optimization of the extended disease dataset. Attached Figure Description

[0019] Figure 1This is a flowchart of a method for generating defect data in a solar-fishery complementary photovoltaic power station according to an embodiment of this application.

[0020] Figure 2 This is a flowchart illustrating the generative adversarial network optimization and expansion of the disease dataset in this application embodiment. Detailed Implementation

[0021] The following is in conjunction with the appendix Figure 1-2 This application will be described in further detail.

[0022] This application discloses a method for generating defect data in a solar-fishery hybrid photovoltaic power station. (Refer to...) Figure 1 and Figure 2 This includes the following steps.

[0023] Step S1: Collect and organize existing disease data, which includes disease types, distribution patterns, ontological characteristics, and environmental conditions, to obtain an existing disease dataset.

[0024] Step S2: Based on the existing disease dataset, a mathematical model combining random walk and convolutional neural networks is selected to describe the disease propagation mechanism, thus constructing a disease diffusion model. Multi-scale modeling enables the disease diffusion model to capture the distribution patterns and ontological characteristics of diseases at different temporal and spatial scales, simulating the spatial distribution patterns and ontological characteristics of diseases in a solar-fishery hybrid power station. The disease diffusion model is as follows: .in, Image of the disease at the next moment. This is an image of the disease at the current moment. It is the time step for feature learning. It is a factor influencing spatial distribution. Using a uniform distribution, It is an influencing factor of ontological characteristics. The RGB three-color grayscale matrix of the disease image is used.

[0025] Step S3: Input the initial disease dataset and environmental conditions, run the disease spread model, simulate the spread of the disease in space and time, and output the simulation results to generate an extended disease dataset that conforms to the real spread law.

[0026] Step S4: Generative adversarial network optimization expands the disease dataset, including the following steps.

[0027] Step S41: The generator's multi-layer convolutional structure is as follows: ,in, It is the first The generator output of the layer, It is the first The generator output of the layer, It is a convolution kernel. It is a biased value with the same dimension as the convolution kernel. The ReLU activation function is used. The discriminator has a multi-layer convolutional structure. ,in, It is the first The discriminator output of the layer, It is the first The discriminator output of the layer, It is a convolution kernel. It is a biased value with the same dimension as the convolution kernel. The ReLU activation function is employed. The generator takes an expanded disease dataset as its initial input and obtains a generated disease dataset through a multi-layer convolutional structure. The discriminator takes the generated disease dataset and the existing disease dataset as its initial input and obtains the true probability values ​​through a multi-layer convolutional structure. The multi-layer convolutional structures of the generator and discriminator can more effectively extract image features, capturing richer spatial information and details through layer-by-layer processing.

[0028] Step S42: Generator Loss Value ,in, It is the generator's adversarial loss value. It is the generator spatial distribution loss weight value. It is the generator spatial distribution loss value and The spatial distribution is a mean distribution. These are the generator ontology feature loss weight values. This is the generator ontology feature loss value. Generator adversarial loss value. The adversarial loss function in standard generative adversarial networks is used, and the calculation formula is as follows: ,in To expand the input noise of the disease dataset, For generator, The generator serves as the discriminator. The generator spatial distribution loss value is used to constrain the consistency between generated samples and real samples in spatial distribution; the calculation formula is as follows: ,in and These are the mean vectors of the spatial distributions of the generated sample set and the real sample set, respectively. The generator ontology feature loss value is calculated by extracting features through a pre-trained feature extraction network and calculating the feature difference, using the following formula: ,in For feature extraction function, These are real disease image samples. Weighting coefficients. and The contribution of each loss adjustment was determined through experimental optimization, with an initial value of [value missing]. and .

[0029] Discriminator loss value ,in, It is the discriminator adversarial loss value. These are the discriminator spatial distribution loss weight values. It is the spatial distribution loss value of the discriminator and The spatial distribution is a mean distribution. These are the discriminator's ontological feature loss weight values. This is the discriminator's ontological feature loss value. The discriminator's adversarial loss value. The adversarial loss function in standard generative adversarial networks is used, and the calculation formula is as follows: ,in These are samples from a real disease dataset. To expand the disease dataset input. Discriminator spatial distribution loss value. The formula used to constrain the discriminator's consistency in judging the spatial distribution of generated samples and real samples is as follows: ,in The mean of the discriminant's judgment results for the generated samples is denoted as . The mean of the true sample distribution. Discriminator ontology feature loss value. The discriminant's judgment result is extracted from the feature difference between the pre-trained feature extraction network and the real sample. The calculation formula is as follows: ,in This represents the discriminator's judgment output on the generated samples. These are real disease image samples. Weighting coefficients. and The contribution of each loss adjustment was determined through experimental optimization, with an initial value of [value missing]. and .

[0030] Calculate generator loss value ; Calculate the discriminator loss value Introducing generator spatial distribution loss and discriminator spatial distribution loss facilitates the matching of the overall structure of the generated image with the real sample. Introducing generator ontology feature loss and discriminator ontology feature loss enhances the similarity of the generated samples in the feature space.

[0031] Step S43: If the generator loss value is not lower than the generator loss threshold or the discriminator loss value is not lower than the discriminator loss threshold, proceed to step S44; if the generator loss value is lower than the generator loss threshold and the discriminator loss value is lower than the discriminator loss threshold, generate a disease dataset as an optimized disease dataset and proceed to step S5.

[0032] Step S44: If the generator loss value is not lower than the generator loss threshold, then correct the generator's multi-layer convolutional structure; if the discriminator loss value is not lower than the discriminator loss threshold, then correct the discriminator's multi-layer convolutional structure. Correcting the generator's multi-layer convolutional structure based on the generator loss value and the discriminator's multi-layer convolutional structure based on the discriminator loss value improves the realism and consistency of the generated disease dataset.

[0033] Step S45; Proceed to step S41.

[0034] Step S5: Obtain the optimized disease dataset.

[0035] This study utilizes a disease diffusion model to simulate the spatial distribution and ontological characteristics of diseases in solar-aquaculture hybrid power plants, generating diverse disease samples to form an expanded disease dataset. Based on this expanded dataset, a generative adversarial network (GAN) is introduced to further optimize it. By training the generator and discriminator, the generator produces higher-quality and more diverse samples, improving realism and detail. Iterative optimization of the expanded disease dataset yields an optimized dataset. This addresses the issues of small size and insufficient samples in existing solar-aquaculture hybrid power plant disease datasets, improving the training effect of the disease identification model and enhancing the accuracy and reliability of disease identification.

[0036] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for generating data on defects in a solar-fishery hybrid photovoltaic power station, characterized in that: Includes the following steps: Step S1: Obtain the existing disease dataset; Step S2: Construct a disease spread model based on the existing disease dataset; Step S3: Input the initial disease dataset and environmental conditions, run the disease spread model, output the simulation results, and generate an extended disease dataset; Step S4: Generative adversarial network optimizes and expands the disease dataset; Step S5: Obtain the optimized disease dataset.

2. The method for generating data on defects in a solar-fishery hybrid photovoltaic power station according to claim 1, characterized in that: The disease spread model in step S2 is: ,in, Image of the disease at the next moment. This is an image of the disease at the current moment. It is the time step for feature learning. It is a factor influencing spatial distribution. These are factors that influence the ontological characteristics.

3. The method for generating defect data of a solar-fishery complementary photovoltaic power station according to claim 2, characterized in that: In step S2 Using an average distribution, in step S2 The RGB three-color grayscale matrix of the disease image is used.

4. The method for generating defect data of a solar-fishery hybrid photovoltaic power station according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: The generator takes the expanded disease dataset as the initial input and obtains the generated disease dataset through the generator's multi-layer convolutional structure. The discriminator takes the generated disease dataset and the existing disease dataset as the initial input and obtains the true probability judgment value through the discriminator's multi-layer convolutional structure. Step S42: Calculate the generator loss value ; Calculate the discriminator loss value ; Step S43: If the generator loss value is not lower than the generator loss threshold or the discriminator loss value is not lower than the discriminator loss threshold, proceed to step S44; if the generator loss value is lower than the generator loss threshold and the discriminator loss value is lower than the discriminator loss threshold, generate a disease dataset as an optimized disease dataset and proceed to step S5. Step S44: If the generator loss value is not lower than the generator loss threshold, then correct the generator multi-layer convolutional structure; if the discriminator loss value is not lower than the discriminator loss threshold, then correct the discriminator multi-layer convolutional structure. Step S45; Proceed to step S41.

5. The method for generating defect data of a solar-fishery hybrid photovoltaic power station according to claim 4, characterized in that: The generator multi-layer convolutional structure in step S41 is as follows: ,in, It is the first The generator output of the layer, It is the first The generator output of the layer, It is a convolution kernel. It's a bias. The ReLU activation function is used.

6. The method for generating defect data of a solar-fishery hybrid photovoltaic power station according to claim 4, characterized in that: The multi-layer convolutional structure of the discriminator in step S41 is as follows: ,in, It is the first The discriminator output of the layer, It is the first The discriminator output of the layer, It is a convolution kernel. It's a bias. The ReLU activation function is used.

7. The method for generating defect data of a solar-fishery hybrid photovoltaic power station according to claim 4, characterized in that: generator loss value in step S42 ,in, Generator adversarial loss value, It is the generator spatial distribution loss weight value. Generator spatial distribution loss value, These are the generator ontology feature loss weight values. Generator ontology feature loss value.

8. The method for generating defect data of a solar-fishery hybrid photovoltaic power station according to claim 7, characterized in that: The generator resists loss values ,in To expand the input noise of the disease dataset, For generator, The discriminator; the generator spatial distribution loss value ,in, To generate the spatial distribution mean vector of the sample set, The mean vector of the spatial distribution of the real sample set; the generator ontology feature loss value. ,in These are real disease image samples.

9. The method for generating defect data of a solar-fishery complementary photovoltaic power station according to claim 4, characterized in that: The discriminator loss value in step S42 ,in, Discriminator adversarial loss value, These are the discriminator spatial distribution loss weight values. Discriminator spatial distribution loss value, These are the discriminator's ontological feature loss weight values. Discriminator ontological feature loss value.

10. A method for generating defect data in a solar-fishery hybrid photovoltaic power station according to claim 9, characterized in that: The discriminator opposes the loss value ,in These are samples from a real disease dataset. To expand the input of the disease dataset, For generator, The discriminator; the spatial distribution loss value of the discriminator ,in The mean of the discriminant's judgment results for the generated samples is denoted as . The mean of the true sample distribution; the discriminator ontology feature loss value. ,in This represents the discriminator's judgment output on the generated samples. These are real disease image samples.

Images