Intelligent cleaning system of centralized photovoltaic power station

Abstract

The embodiment of the invention relates to the technical field of photovoltaic power station cleaning, and provides an intelligent cleaning system of a centralized photovoltaic power station, which can perceive environmental conditions by using a deep learning algorithm based on environmental parameters and surface state images of solar cell panels and judge the pollution degree of the solar cell panels according to the environmental conditions, so that the cleaning efficiency of the solar cell panels is improved. Therefore, the cleaning speed of the cleaning robot is adaptively adjusted, the cleaning efficiency and the performance of the solar cell panel are effectively improved, and the cleaning and maintenance cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic power station cleaning technology, and in particular to an intelligent cleaning system for a centralized photovoltaic power station. Background Art

[0002] Photovoltaic power generation is a technology that converts solar energy into electrical energy. Solar panels can be used to directly convert sunlight into electrical energy. This type of electricity is a clean, renewable form of energy with many advantages. It is widely considered to be one of the important directions for future energy development. The working principle of photovoltaic power generation is based on the photoelectric effect. When sunlight shines on solar panels, the photons in them interact with the semiconductor atoms in the material, stimulating electrons to jump out of the atoms and forming an electric current. This current flows through the wires on the panels and can be used to supply power or stored for later use.

[0003] Photovoltaic power generation does not produce greenhouse gases such as carbon dioxide and other pollutants, and has almost no negative impact on the environment. Compared with traditional fossil fuel power generation, photovoltaic power generation can significantly reduce greenhouse gas emissions and help address climate change. Sunlight is an inexhaustible energy resource. Photovoltaic power generation uses solar energy, which is renewable. Globally, solar energy resources are very abundant and can meet human energy needs as long as they are used reasonably. Photovoltaic power generation can be carried out in a distributed manner, that is, photovoltaic power stations can be built in various places, and power generation equipment can be installed on the roofs of buildings, solar farms, deserts and other places, making the generation of electricity more flexible and more adaptable. This distributed power generation method can reduce energy transmission losses and improve energy utilization efficiency. Once the photovoltaic power generation system is installed, its maintenance cost is relatively low. Solar panels have no moving parts and do not require frequent maintenance and replacement. They only need regular cleaning and inspection, which reduces the operating cost of energy production.

[0004] However, photovoltaic power generation also faces some challenges in practical applications. Solar panels are exposed to outdoor environments for a long time and are susceptible to pollutants such as dust, leaves, and bird droppings. These pollutants will accumulate on the surface of the panels, causing corrosion and damage to the surface of the solar panels, reducing the light absorption rate of the surface of the panels, and thus reducing the efficiency of photovoltaic power generation. Regular cleaning of the panels can remove these pollutants, prevent long-term damage to the panels by pollutants, restore the light absorption capacity of the panels, extend the service life of the panels, and improve the power generation efficiency. Keeping the panels clean can also reduce the absorption and adsorption of pollutants on the panels, reduce chemical reactions and heat accumulation on the panels, and help maintain the stability and reliability of the panels. Studies have shown that after cleaning, the power generation efficiency of solar panels can be increased by 5% to 30%. Therefore, regular cleaning of solar panels is one of the necessary measures to ensure the performance of photovoltaic power generation.

[0005] In addition, regular cleaning of solar panels can also reduce maintenance costs. If the panels are not cleaned for a long time, the accumulation of pollutants may cause more serious damage, requiring more expensive repairs or replacements. Regular cleaning can detect and deal with problems in a timely manner, reducing maintenance costs. Solar panels are one of the core components of photovoltaic power generation systems. Cleaning the panels can ensure the stable operation of the system and reduce failures and downtime. Especially in arid areas or dusty environments, regular cleaning is particularly important to maintain the reliability and stability of the system. The energy output of a photovoltaic power generation system is directly related to the power generation efficiency of the solar panels. By cleaning the solar panels regularly, the energy output can be maximized and the economic benefits of the photovoltaic power generation system can be increased.

[0006] At present, the cleaning method of centralized photovoltaic power stations mainly adopts manual cleaning. Manual cleaning requires a lot of manpower input, the cleaning process is time-consuming and relatively inefficient. For large-scale centralized photovoltaic power stations, the number of panels is huge, and a lot of time and human resources are needed to complete the cleaning work, which may increase the downtime of the power station, affect the continuity and efficiency of photovoltaic power generation, and the cleaning and maintenance costs may be high. In addition, the manual cleaning method requires the employment of special cleaning personnel, the payment of labor costs and related management costs, and the use of cleaning tools, water sources, cleaning agents and other materials, which increases the operating costs. For large photovoltaic power stations, the cost of manual cleaning may be very high. In addition, there are certain safety risks in manually cleaning photovoltaic power stations. Cleaning personnel need to operate at high altitudes and may face the risk of falling from high altitudes, electric shock and other accidental injuries. Especially under adverse weather conditions, such as strong winds or slippery surfaces of photovoltaic panels, the safety risks are more prominent. In this case, in order to ensure the safety of staff, additional safety measures and training are required, which increases the complexity of management and operation. Summary of the invention

[0007] The present disclosure aims to solve at least one of the problems existing in the prior art and provide an intelligent cleaning system for a centralized photovoltaic power station.

[0008] The present disclosure provides an intelligent cleaning system for a centralized photovoltaic power station, the intelligent cleaning system comprising a cleaning robot and a charging station; the charging station is used to provide electrical energy for the cleaning robot;

[0009] The cleaning robot is equipped with a camera, a light sensor, a temperature sensor and a humidity sensor, and the cleaning robot also includes a cleaning and regulating module;

[0010] The cleaning and regulating module comprises:

[0011] A surface state image acquisition unit, used to acquire the surface state image of the solar panel to be cleaned in the centralized photovoltaic power station collected by the camera;

[0012] a data acquisition unit, used to acquire light intensity values, temperature values ​​and humidity values ​​corresponding to the solar cell panel to be cleaned, respectively collected by the photosensor, the temperature sensor and the humidity sensor at a plurality of predetermined time points within a predetermined time period;

[0013] An environmental state feature extraction unit, used to extract environmental state features from the light intensity value, the temperature value and the humidity value to obtain an environmental state multi-parameter associated time series feature vector;

[0014] A feature integration unit, used to integrate the solar panel surface state features of the surface state image into the environmental state multi-parameter associated time series feature vector to obtain an environmental state channel embedded solar panel surface state feature map;

[0015] The cleaning speed control unit is used to determine whether to increase the cleaning speed of the cleaning robot based on the surface state characteristic map of the solar panel embedded in the environmental state channel.

[0016] Optionally, the environmental state feature extraction unit includes:

[0017] A data preprocessing subunit, used for performing data preprocessing on the light intensity value, the temperature value and the humidity value to obtain an environmental multi-parameter time series input matrix;

[0018] The feature extraction subunit is used to use a deep learning network model to perform feature extraction on the environmental multi-parameter time series input matrix to obtain the environmental state multi-parameter associated time series feature vector.

[0019] Optionally, the data preprocessing subunit is specifically used to:

[0020] The light intensity value, the temperature value and the humidity value are arranged according to the time dimension into the environmental multi-parameter time series input matrix.

[0021] Optionally, the deep learning network model is an environmental state temporal feature extractor based on a convolutional neural network model;

[0022] The feature extraction subunit is specifically used for:

[0023] The environment multi-parameter time series input matrix is ​​passed through the environment state time series feature extractor based on the convolutional neural network model to obtain the environment state multi-parameter associated time series feature vector.

[0024] Optionally, the environmental state temporal feature extractor based on the convolutional neural network model includes: an input layer, a convolution layer, a pooling layer, an activation layer and an output layer.

[0025] Optionally, the feature integration unit includes:

[0026] A solar panel surface state feature extraction subunit is used to extract the solar panel surface state features of the surface state image to obtain a solar panel surface state feature map;

[0027] A feature distribution correction subunit is used to perform feature distribution correction on the environmental state multi-parameter associated time series feature vector to obtain a corrected environmental state multi-parameter associated time series feature vector;

[0028] The cross-modal feature fusion subunit is used to integrate the solar panel surface state feature map and the corrected environmental state multi-parameter associated time series feature vector through a cross-modal feature fusion module based on MetaNet to obtain the environmental state channel embedded in the solar panel surface state feature map.

[0029] Optionally, the solar panel surface state feature extraction subunit is specifically used for:

[0030] The surface state image is passed through a surface state feature extractor based on a convolutional neural network model to obtain a surface state feature map of the solar panel.

[0031] Optionally, the cross-modal feature fusion subunit is specifically used to:

[0032] The solar panel surface state feature map is passed through the convolutional neural network of the MetaNet-based cross-modal feature fusion module to obtain a solar panel surface state feature vector;

[0033] Passing the corrected environmental state multi-parameter associated time series feature vector through the one-dimensional convolution layer of the MetaNet-based cross-modal feature fusion module to obtain an environmental state multi-parameter feature vector;

[0034] Channel weighting is performed on the solar panel surface state feature vector using the environmental state multi-parameter feature vector to obtain a solar panel surface state feature map in which the environmental state channel is embedded.

[0035] Optionally, the cleaning speed control unit is specifically used to:

[0036] After the environmental state channel is embedded in the solar panel surface state characteristic map, a corresponding classification result is obtained through a classifier; the classification result is used to indicate whether to increase the cleaning speed of the cleaning robot.

[0037] Optionally, the intelligent cleaning system also includes a control center, which is used to manage the cleaning robot.

[0038] Compared with the prior art, the present invention can use a deep learning algorithm to perceive environmental conditions based on environmental parameters and surface state images of solar panels, and use this to judge the degree of contamination of solar panels, thereby adaptively adjusting the cleaning speed of the cleaning robot, effectively improving the cleaning efficiency and performance of solar panels, and reducing cleaning and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplifications do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0040] Figure 1 A schematic diagram of the structure of an intelligent cleaning system for a centralized photovoltaic power station provided in one embodiment of the present disclosure;

[0041] Figure 2 A flowchart of a cleaning adjustment module included in a cleaning robot in an intelligent cleaning system of a centralized photovoltaic power station provided by another embodiment of the present disclosure;

[0042] Figure 3 A schematic diagram of data flow of a cleaning adjustment module included in a cleaning robot in an intelligent cleaning system of a centralized photovoltaic power station provided by another embodiment of the present disclosure;

[0043] Figure 4 This is an application scenario diagram of an intelligent cleaning system for a centralized photovoltaic power station provided in another embodiment of the present disclosure. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical scheme and advantages of the embodiments of the present disclosure clearer, the embodiments of the present disclosure will be described in detail below in conjunction with the accompanying drawings. However, it can be understood by those skilled in the art that in each embodiment of the present disclosure, many technical details are proposed in order to enable readers to better understand the present disclosure. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed for protection in the present disclosure can also be implemented. The division of the following embodiments is for the convenience of description and should not constitute any limitation on the specific implementation of the present disclosure. The various embodiments can be combined and referenced with each other without contradiction.

[0045] Unless otherwise specified, all technical and scientific terms used in the embodiments of the present disclosure have the same meaning as those commonly understood by those skilled in the art to which the present disclosure belongs. The terms used in the present disclosure are only for the purpose of describing specific embodiments and are not intended to limit the scope of the present disclosure.

[0046] In the description of the embodiments of the present disclosure, it should be noted that, unless otherwise specified and limited, the term "connection" should be understood in a broad sense. For example, it can be an electrical connection or a connection between two elements. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0047] It should be noted that the terms "first\second\third" involved in the embodiments of the present disclosure are only used to distinguish similar objects, and do not represent a specific order for the objects. It is understandable that the specific order or sequence of "first\second\third" can be interchanged where permitted. It should be understood that the objects distinguished by "first\second\third" can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein.

[0048] In order to solve the problems existing in the cleaning methods of centralized photovoltaic power stations in the prior art, the present disclosure provides an intelligent cleaning system for a centralized photovoltaic power station, the technical concept of which is as follows: based on the surface state images of the solar panels to be cleaned in the centralized photovoltaic power station and their corresponding environmental parameters, a deep learning algorithm is used to perceive the environmental conditions of the solar panels to be cleaned, and the degree of contamination of the solar panels to be cleaned is judged thereby, so as to adaptively adjust the cleaning speed of the cleaning robot.

[0049] like Figure 1 As shown, an intelligent cleaning system 100 of a centralized photovoltaic power station provided by an embodiment of the present disclosure includes a cleaning robot 1 and a charging station 2. The charging station 2 is used to provide power to the cleaning robot 1. Among them, a plurality of charging piles can be arranged in the charging station 2, and each charging pile can be located at a different position in the centralized photovoltaic power station, so that the cleaning robot 1 can be charged nearby when the power is insufficient. Figure 1 As shown, the intelligent cleaning system 100 of the centralized photovoltaic power station also includes a control center 3, which is used to manage the cleaning robot 1. For example, when the remaining power of the cleaning robot 1 reaches the minimum power threshold, the control center 3 can control the cleaning robot 1 to go to the charging station 2 for charging. Alternatively, the control center 3 can also control the cleaning robot 1 to go to the location of the solar panel to be cleaned to clean the solar panel to be cleaned. Of course, the control center 3 can also manage the cleaning robot 1 in other ways, and this embodiment does not limit this.

[0050] like Figure 1 As shown, a camera 11 , a photosensor 12 , a temperature sensor 13 and a humidity sensor 14 are deployed on the cleaning robot 1 , and the cleaning robot 1 also includes a cleaning and regulating module 15 .

[0051] Specifically, the camera 11 is used to capture the surface condition of the solar panel to be cleaned, collect the surface condition image of the solar panel to be cleaned, and provide a clear image for subsequent processing. The photosensitive sensor 12, the temperature sensor 13, and the humidity sensor 14 are important devices for sensing environmental conditions, which can reflect the cleaning requirements of the solar panel to be cleaned. For example, the photosensitive sensor 12 can detect changes in light intensity to determine the photoelectric absorption rate and power generation efficiency of the solar panel to be cleaned. The temperature sensor 13 can detect temperature changes to determine the aging risk of the solar panel to be cleaned. The humidity sensor 14 can detect humidity changes to determine the corrosion risk of the solar panel to be cleaned.

[0052] like Figure 1 As shown, the cleaning adjustment module 15 includes a surface state image acquisition unit 151 , a data acquisition unit 152 , an environment state feature extraction unit 153 , a feature integration unit 154 , and a cleaning speed control unit 155 .

[0053] The surface state image acquisition unit 151 is used to acquire the surface state image of the solar panel to be cleaned in the centralized photovoltaic power station collected by the camera 11. Specifically, when the surface state image of the solar panel to be cleaned is acquired by the camera 11, it is necessary to ensure that the camera can accurately capture the surface condition of the solar panel to be cleaned to provide a clear image for subsequent processing. In this way, the surface state image of the solar panel to be cleaned provides visual information of the surface state of the solar panel to be cleaned, which provides a basis for subsequent cleaning decisions and analysis.

[0054] The data acquisition unit 152 is used to acquire the light intensity value, temperature value and humidity value corresponding to the solar panel to be cleaned, which are respectively acquired by the photosensitive sensor 12, the temperature sensor 13 and the humidity sensor 14 at multiple predetermined time points within a predetermined time period. Specifically, during the acquisition process of the photosensitive sensor 12, the temperature sensor 13 and the humidity sensor 14, the accuracy and stability of each sensor must be ensured to acquire reliable environmental data, namely, light intensity value, temperature value and humidity value. In this way, the light intensity value, temperature value and humidity value respectively provide time series data of environmental parameters such as light, temperature and humidity, which can be used to analyze the impact of the environment on the solar panel to be cleaned.

[0055] The environmental state feature extraction unit 153 is used to extract environmental state features from the light intensity value, the temperature value and the humidity value to obtain an environmental state multi-parameter associated time series feature vector. Specifically, the environmental state feature extraction unit 153 can select an appropriate feature extraction method, such as statistical features, frequency domain features or time domain features, to capture the key features of the environmental state and obtain an environmental state multi-parameter associated time series feature vector. In this way, the environmental state multi-parameter associated time series feature vector provides a comprehensive description of the environmental state corresponding to the solar panel to be cleaned, providing a basis for subsequent decision-making.

[0056] The feature fusion unit 154 is used to fuse the surface state features of the solar panel of the surface state image into the multi-parameter associated time series feature vector of the environmental state, and obtain the environmental state channel embedded solar panel surface state feature map. Specifically, the feature fusion unit 154 can reasonably fuse the surface state features and the environmental state features to maintain the integrity and accuracy of the information, and obtain the environmental state channel embedded solar panel surface state feature map. In this way, the environmental state channel embedded solar panel surface state feature map provides a comprehensive state map of the solar panel to be cleaned, which combines the surface condition and environmental information of the solar panel to be cleaned, and can be used to more comprehensively understand the cleaning requirements of the solar panel to be cleaned.

[0057] The cleaning speed control unit 155 is used to determine whether to increase the cleaning speed of the cleaning robot 1 based on the environmental state channel embedded solar panel surface state characteristic map. Specifically, the cleaning speed control unit 155 can flexibly adjust the cleaning speed of the cleaning robot 1 according to the comprehensive evaluation of the environmental state and the surface state of the solar panel to be cleaned, so as to improve the cleaning efficiency and effect. In this way, the cleaning speed control unit 155 realizes intelligent control of the cleaning speed of the cleaning robot according to the real-time environmental information and the state information of the solar panel to be cleaned, thereby improving the cleaning efficiency and reducing the cleaning time.

[0058] In particular, during the actual operation of the intelligent cleaning system 100 of a centralized photovoltaic power station, it should be considered that the accumulation of dust and pollutants on the surface of the solar panel will be affected by environmental conditions. That is, the accumulation of dust and pollutants on the surface of the solar panel will be different under different time periods and environmental conditions. Therefore, this embodiment can accurately clean the solar panel to be cleaned according to actual conditions, that is, during the cleaning process, if the degree of contamination on the surface of the solar panel to be cleaned is relatively light, the cleaning speed of the cleaning robot 1 can be appropriately increased to reduce energy consumption; on the contrary, if the degree of contamination on the surface of the solar panel to be cleaned is relatively heavy, the cleaning speed of the cleaning robot 1 can be reduced to ensure adequate cleaning.

[0059] Compared with the prior art, the intelligent cleaning system of the centralized photovoltaic power station provided by the embodiment of the present disclosure can adaptively adjust the cleaning speed of the cleaning robot, effectively improve the cleaning efficiency and the performance of the solar cell panel, and reduce the cleaning and maintenance costs.

[0060] Exemplarily, the environment state feature extraction unit 153 includes a data preprocessing subunit and a feature extraction subunit.

[0061] The data preprocessing subunit is used to perform data preprocessing on the light intensity value, the temperature value and the humidity value to obtain an environmental multi-parameter time series input matrix.

[0062] The feature extraction subunit is used to extract features from the environmental multi-parameter time series input matrix using a deep learning network model to obtain the environmental state multi-parameter associated time series feature vector, thereby capturing the synergistic correlation feature distribution between light intensity, temperature value and humidity value.

[0063] The data preprocessing subunit is specifically used to arrange the light intensity value, the temperature value and the humidity value according to the time dimension into an environmental multi-parameter time series input matrix.

[0064] Specifically, by arranging the light intensity values, temperature values, and humidity values ​​of the environmental parameters according to the time dimension, the light, temperature, and humidity data at a series of time points can be obtained. This time series arrangement can be used for time series analysis, such as detecting periodic changes, trends, and seasonal patterns of environmental parameters, which helps to have a deeper understanding and prediction of changes in environmental conditions. The time series input matrix can be used to monitor changes in environmental parameters in real time and provide immediate feedback control. Arranging the light intensity values, temperature values, and humidity values ​​at multiple predetermined time points according to the time dimension into an environmental multi-parameter time series input matrix can provide more comprehensive and accurate environmental status information, provide a basis for subsequent analysis, modeling, and control, and help optimize the operation strategy of the cleaning robot, improve cleaning efficiency, reduce costs, and ensure the optimal performance of solar panels.

[0065] Exemplarily, the deep learning network model is an environment state temporal feature extractor based on a convolutional neural network model. The feature extraction subunit is specifically used to: pass the environment multi-parameter temporal input matrix through the environment state temporal feature extractor based on the convolutional neural network model to obtain the environment state multi-parameter associated temporal feature vector.

[0066] Among them, the environmental state temporal feature extractor based on the convolutional neural network model includes: an input layer, a convolution layer, a pooling layer, an activation layer and an output layer.

[0067] Specifically, a convolutional neural network is a powerful feature extractor that can learn features with good characterization capabilities from input data. By inputting the environmental multi-parameter time series input matrix into the environmental state time series feature extractor based on the convolutional neural network model, the environmental state time series feature extractor can automatically learn the time series characteristics of the environmental state to capture the correlation and change pattern between environmental parameters, and provide a more comprehensive and accurate description of the environmental state. The time series feature extraction of environmental parameters can capture the time series correlation between parameters. For example, during the cleaning process of the solar panel to be cleaned, parameters such as light intensity, temperature and humidity may change over time and affect each other. By using the environmental state time series feature extractor based on the convolutional neural network model to extract time series features from the environmental multi-parameter time series input matrix, the environmental state time series feature extractor can learn the time series correlation between parameters, thereby providing a more accurate and comprehensive environmental state feature vector. Through the feature extraction process of the environmental state time series feature extractor, the high-dimensional environmental multi-parameter time series input matrix can be converted into a low-dimensional environmental state multi-parameter associated time series feature vector. This dimensionality compression and representation learning can reduce the redundancy of features, extract key environmental state information, and provide more efficient input for subsequent decision-making and control.

[0068] By using the environmental state temporal feature extractor based on the convolutional neural network model for feature extraction, useful features can be extracted from the environmental multi-parameter temporal input matrix, which can better describe the correlation and temporal changes of the environmental state, provide more accurate and comprehensive information for subsequent decision-making and control, and help optimize the operation strategy of the cleaning robot and improve the cleaning efficiency and the performance of the solar panel.

[0069] Exemplarily, the feature fusion unit 154 includes a solar panel surface state feature extraction subunit, a feature distribution correction subunit, and a cross-modal feature fusion subunit.

[0070] The solar panel surface state feature extraction subunit is used to extract the solar panel surface state features of the surface state image to obtain the solar panel surface state feature map.

[0071] The characteristic distribution correction subunit is used to perform characteristic distribution correction on the environmental state multi-parameter associated time series characteristic vector to obtain the corrected environmental state multi-parameter associated time series characteristic vector.

[0072] The cross-modal feature fusion subunit is used to embed the surface state feature map of the solar panel and the multi-parameter correlation time series feature vector of the corrected environmental state through a cross-modal feature fusion module based on MetaNet to obtain the environmental state channel embedded in the surface state feature map of the solar panel.

[0073] Specifically, the surface state image of the solar panel to be cleaned can reflect information such as the distribution and degree of pollutants on the solar panel to be cleaned. The multi-parameter associated time series feature vector of the environmental state can reflect the environmental conditions of the solar panel to be cleaned, such as light intensity, temperature and humidity. These environmental conditions will affect the light absorption rate and power generation efficiency of the solar panel to be cleaned, and will also affect the adhesion of pollutants and the difficulty of cleaning. By cross-modal feature fusion of the surface state image and environmental conditions, the fused environmental state channel can be embedded in the surface state feature map of the solar panel with richer feature expression.

[0074] Exemplarily, the solar panel surface state feature extraction subunit is specifically used to: pass the surface state image through a surface state feature extractor based on a convolutional neural network model to obtain a solar panel surface state feature map.

[0075] Exemplarily, the cross-modal feature fusion subunit is specifically used for: passing the solar panel surface state feature map through a convolutional neural network of a cross-modal feature fusion module based on MetaNet to obtain a solar panel surface state feature vector; passing the corrected environmental state multi-parameter associated time series feature vector through a one-dimensional convolutional layer of a cross-modal feature fusion module based on MetaNet to obtain an environmental state multi-parameter feature vector; channel-weighting the solar panel surface state feature vector with the environmental state multi-parameter feature vector to obtain an environmental state channel-embedded solar panel surface state feature map.

[0076] Among them, the cross-modal feature fusion module based on MetaNet enables the multi-parameter associated time series feature vector of the environmental state as one-dimensional data to directly interact with the visual features represented by the surface state feature map of the solar panel, directly control the relevant characteristics of each feature channel, and help the network focus on the specific part of each feature channel.

[0077] When the solar panel surface state feature map and the multi-parameter associated time series feature vector of the environmental state are passed through a cross-modal feature fusion module based on MetaNet, each feature matrix of the solar panel surface state feature map expresses the image semantic features of the surface state image of the solar panel to be cleaned, and conforms to the channel distribution of the convolutional neural network in the channel dimension.

[0078] Here, when the image semantic feature space distribution weight constraint along the channel dimension is applied to the solar panel surface state feature map by using the environment state multi-parameter associated time series feature vector, considering that the feature expression intensity of the solar panel surface state feature map in the semantic space is much higher than that of the environment state multi-parameter associated time series feature vector, it may cause the target distribution information loss of the environment state multi-parameter associated time series feature vector, thus affecting the weight constraint effect of the environment state multi-parameter associated time series feature vector on the solar panel surface state feature map. Therefore, this embodiment first calculates the global mean of each feature matrix of the solar panel surface state feature map to obtain the solar panel surface state feature vector V 1 Then, the surface state feature vector of the solar panel is used to correlate the multi-parameter time series feature vector of the environmental state to V 2 Optimize and obtain the corrected environmental state multi-parameter correlation time series feature vector V 2 ′.

[0079] In other words, the feature integration unit 154 can be used to calculate the global mean of each feature matrix of the solar panel surface state feature map to obtain the solar panel surface state feature vector V 1 ; Based on the following optimization formula, the surface state characteristic vector V of the solar panel 1 The multi-parameter correlation time series feature vector V of the environmental state 2 Optimize and obtain the corrected environmental state multi-parameter correlation time series feature vector V 2 ′; the optimization formula is:

[0080]

[0081] Among them, V 1 is the surface state characteristic vector of the solar panel, V 2 is the multi-parameter associated time series feature vector of the environmental state, and Represent the surface state feature vector V of the solar panel 1 and the multi-parameter correlation time series feature vector V of the environmental state 2 The inverse of the global mean, I is a unit vector, V 2 ′ is the time series feature vector of multi-parameter correlation of environmental state after correction, represents positional addition, represents positional subtraction, and ⊙ represents positional multiplication.

[0082] That is to say, considering the uneven distribution of features, if the surface state feature vector V 1 Considered as the multi-parameter associated time series feature vector V of the environmental state 2The feature distribution of the enhanced input, considering the multi-parameter correlation time series feature vector V of the environment state 2 The target distribution information loss of the target features in the constraint space may lead to the loss of the constraint purpose. Therefore, by cross-penalizing the outlier distribution of the feature distribution relative to each other, the self-supervised balance of feature enhancement and constraint robustness can be achieved through feature interpolation fusion to improve the multi-parameter correlation time series feature vector V of the environment state. 2 Compared with the robust constraint effect of strong feature expression, the multi-parameter correlation time series feature vector V of the corrected environmental state is used. 2 ′ is used to perform channel weighting on the solar panel surface state feature map, so as to improve the feature expression effect of the obtained environmental state channel embedded in the solar panel surface state feature map, thereby improving the accuracy of the classification result obtained by the classifier.

[0083] Exemplarily, the cleaning speed control unit 155 is specifically used to: embed the environmental state channel into the solar panel surface state characteristic map and then pass it through a classifier to obtain a corresponding classification result; the classification result is used to indicate whether to increase the cleaning speed of the cleaning robot.

[0084] Specifically, by embedding the environmental state channel into the solar panel surface state feature map and using a classifier to classify it, the cleaning speed of the cleaning robot can be adaptively adjusted according to different environmental states. For example, when the environmental state indicates that the surface of the solar panel to be cleaned is relatively dirty, the classification result can indicate that the cleaning speed should be increased to improve the cleaning efficiency. Conversely, when the environmental state indicates that the surface of the solar panel to be cleaned is relatively clean, the classification result can indicate that the cleaning speed should be reduced to save energy and extend the service life of the cleaning robot. This adaptive cleaning speed adjustment helps to improve the cleaning efficiency and the performance of the solar panel.

[0085] By embedding the environmental state channel into the solar panel surface state feature map and using a classifier for classification, real-time monitoring and feedback control of the environmental state can also be achieved. The classification results can timely reflect the current environmental state, and the cleaning robot can make corresponding adjustments to the cleaning speed based on the classification results. This real-time monitoring and feedback control can make the cleaning robot more adaptable to environmental changes, respond to different cleaning needs in a timely manner, and improve cleaning efficiency and the performance of solar panels.

[0086] Figure 2 The working process of the cleaning speed control unit 155 is shown. Figure 3 1 shows the data flow of the cleaning speed control unit 155. Figure 2 As shown, combined Figure 3, the working process of the cleaning speed control unit 155 includes the following steps:

[0087] Step 210: Acquire a surface state image of a solar panel to be cleaned in a centralized photovoltaic power station captured by a camera.

[0088] Step 220, obtaining light intensity values, temperature values ​​and humidity values ​​corresponding to the solar panel to be cleaned collected by the light sensor, the temperature sensor and the humidity sensor at multiple predetermined time points within a predetermined time period.

[0089] Step 230 , extracting environmental state features from the light intensity value, the temperature value, and the humidity value to obtain an environmental state multi-parameter associated time series feature vector.

[0090] Step 240 , integrating the solar panel surface state features of the surface state image into the environment state multi-parameter associated time series feature vector to obtain an environment state channel embedded solar panel surface state feature map.

[0091] Step 250, based on the environmental state channel embedded in the solar panel surface state characteristic map, determine whether to increase the cleaning speed of the cleaning robot.

[0092] The specific implementation method of each step in the work flow of the cleaning speed control unit 155 can be found in the description of the intelligent cleaning system for a centralized photovoltaic power station provided in the embodiment of the present disclosure, and will not be repeated here.

[0093] Exemplarily, the intelligent cleaning system 100 of the centralized photovoltaic power station can be implemented in various terminal devices, such as a server for intelligent cleaning of the centralized photovoltaic power station. In one example, the intelligent cleaning system 100 of the centralized photovoltaic power station can be integrated into the terminal device as a software module and / or a hardware module. For example, the intelligent cleaning system 100 of the centralized photovoltaic power station can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the intelligent cleaning system 100 of the centralized photovoltaic power station can also be one of the many hardware modules of the terminal device.

[0094] Alternatively, in another example, the intelligent cleaning system 100 of the centralized photovoltaic power station and the terminal device may also be separate devices, and the intelligent cleaning system 100 of the centralized photovoltaic power station may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0095] Figure 4 This is an application scenario diagram of the intelligent cleaning system of a centralized photovoltaic power station provided by the embodiment of the present disclosure. Figure 4As shown, in this application scenario, the cleaning adjustment module in the cleaning robot first obtains the surface state image of the solar panel to be cleaned collected by the camera deployed in the cleaning robot through the surface state image acquisition unit. The surface state image can be recorded as Figure 4 C1 in; then, the cleaning adjustment module obtains the light intensity value, temperature value and humidity value at multiple predetermined time points within a predetermined time period collected by the photosensitive sensor, temperature sensor and humidity sensor deployed on the cleaning robot through the data acquisition unit, wherein the light intensity value can be recorded as Figure 4 The temperature value of C2 in can be recorded as Figure 4 The humidity value of C3 can be Figure 4 C4 in; then, the cleaning adjustment module uses the environmental state feature extraction unit, the feature fusion unit, and the cleaning speed control unit to process the surface state image, the light intensity value, the temperature value, and the humidity value to determine whether to increase the cleaning speed of the cleaning robot.

[0096] Those skilled in the art will appreciate that the above-mentioned embodiments are specific embodiments for implementing the present disclosure, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present disclosure.

Claims

1. An intelligent cleaning system for a centralized photovoltaic power station, characterized in that: The intelligent cleaning system comprises a cleaning robot and a charging station; the charging station is used to provide electrical energy to the cleaning robot; The cleaning robot is equipped with a camera, a light sensor, a temperature sensor and a humidity sensor, and the cleaning robot also includes a cleaning and regulating module; The cleaning and regulating module comprises: A surface state image acquisition unit, used to acquire the surface state image of the solar panel to be cleaned in the centralized photovoltaic power station collected by the camera; a data acquisition unit, used to acquire light intensity values, temperature values ​​and humidity values ​​corresponding to the solar cell panel to be cleaned, respectively collected by the photosensor, the temperature sensor and the humidity sensor at a plurality of predetermined time points within a predetermined time period; An environmental state feature extraction unit, used to extract environmental state features from the light intensity value, the temperature value and the humidity value to obtain an environmental state multi-parameter associated time series feature vector; A feature integration unit, used to integrate the solar panel surface state features of the surface state image into the environmental state multi-parameter associated time series feature vector to obtain an environmental state channel embedded solar panel surface state feature map; A cleaning speed control unit is used to determine whether to increase the cleaning speed of the cleaning robot based on the surface state characteristic map of the solar panel embedded in the environmental state channel.

2. The intelligent cleaning system according to claim 1, characterized in that: The environmental state feature extraction unit comprises: A data preprocessing subunit, used for performing data preprocessing on the light intensity value, the temperature value and the humidity value to obtain an environmental multi-parameter time series input matrix; The feature extraction subunit is used to use a deep learning network model to perform feature extraction on the environmental multi-parameter time series input matrix to obtain the environmental state multi-parameter associated time series feature vector.

3. The intelligent cleaning system according to claim 2, characterized in that: The data preprocessing subunit is specifically used for: The light intensity value, the temperature value and the humidity value are arranged according to the time dimension into the environmental multi-parameter time series input matrix.

4. The intelligent cleaning system according to claim 2, characterized in that: The deep learning network model is an environmental state temporal feature extractor based on a convolutional neural network model; The feature extraction subunit is specifically used for: The environment multi-parameter time series input matrix is ​​passed through the environment state time series feature extractor based on the convolutional neural network model to obtain the environment state multi-parameter associated time series feature vector.

5. The intelligent cleaning system according to claim 4, characterized in that: The environmental state temporal feature extractor based on the convolutional neural network model includes: an input layer, a convolution layer, a pooling layer, an activation layer and an output layer.

6. The intelligent cleaning system according to claim 1, characterized in that: The feature integration unit includes: A solar panel surface state feature extraction subunit is used to extract the solar panel surface state features of the surface state image to obtain a solar panel surface state feature map; A feature distribution correction subunit is used to perform feature distribution correction on the environmental state multi-parameter associated time series feature vector to obtain a corrected environmental state multi-parameter associated time series feature vector; The cross-modal feature fusion subunit is used to integrate the solar panel surface state feature map and the corrected environmental state multi-parameter associated time series feature vector through a cross-modal feature fusion module based on MetaNet to obtain the environmental state channel embedded in the solar panel surface state feature map.

7. The intelligent cleaning system according to claim 6, characterized in that: The solar panel surface state feature extraction subunit is specifically used for: The surface state image is passed through a surface state feature extractor based on a convolutional neural network model to obtain a surface state feature map of the solar panel.

8. The intelligent cleaning system according to claim 6, characterized in that: The cross-modal feature fusion subunit is specifically used for: The solar panel surface state feature map is passed through the convolutional neural network of the MetaNet-based cross-modal feature fusion module to obtain a solar panel surface state feature vector; Passing the corrected environmental state multi-parameter associated time series feature vector through the one-dimensional convolution layer of the MetaNet-based cross-modal feature fusion module to obtain an environmental state multi-parameter feature vector; Channel weighting is performed on the solar panel surface state feature vector using the environmental state multi-parameter feature vector to obtain a solar panel surface state feature map in which the environmental state channel is embedded.

9. The intelligent cleaning system according to any one of claims 1 to 8, characterized in that: The cleaning speed control unit is specifically used for: After the environmental state channel is embedded in the solar panel surface state characteristic map, a corresponding classification result is obtained through a classifier; the classification result is used to indicate whether to increase the cleaning speed of the cleaning robot.

10. The intelligent cleaning system according to claim 9, characterized in that: The intelligent cleaning system also includes a control center, which is used to manage the cleaning robot.

Images