Intelligent Scheduling Method and System for Photovoltaic Cleaning Device

Abstract

The present invention discloses an intelligent scheduling method and system for a photovoltaic cleaning device, belonging to the technical field of intelligent scheduling. Based on the historical photovoltaic panel images and the corresponding dirt states of each historical photovoltaic panel image, an integrated self-monitoring model is determined. Then, the integrated self-monitoring model is used to identify and analyze the current photovoltaic panel image to determine the probability value of the real-time dirt state corresponding to the current photovoltaic panel image. When the probability value of the real-time dirt state corresponding to the current photovoltaic panel image is greater than a preset threshold, the photovoltaic cleaning device is turned on, thereby enabling intelligent scheduling. Compared with the prior art, it can not only effectively ensure the cleanliness of the photovoltaic panel, but also only clean when cleaning is needed, effectively avoiding ineffective cleaning and improving the cleaning efficiency of the photovoltaic panel.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent scheduling, and particularly relates to an intelligent scheduling method and system for a photovoltaic cleaning device. Background Art

[0002] With the rapid development of the photovoltaic industry, the scale of photovoltaic power stations has been continuously expanding. The surface of photovoltaic modules is prone to dust accumulation, which affects the power generation efficiency. A photovoltaic cleaning device is a device specifically used to clean the surface of photovoltaic panels (solar panels), and its purpose is to maintain the high-efficiency power generation capacity of the photovoltaic panels. The photovoltaic panels are easily contaminated with dust, bird droppings, leaves and other sundries after being exposed to the outdoor environment for a long time. These dirt will reduce the light transmittance of the photovoltaic panels, thereby affecting the power generation efficiency. At present, photovoltaic cleaning devices are more and more widely used in photovoltaic power stations, but the scheduling methods of existing cleaning devices are relatively simple, such as setting timed tasks for cleaning, and cannot perform intelligent scheduling according to the actual situation, resulting in poor cleaning effects and affecting the overall operation efficiency of photovoltaic power stations. Summary of the Invention

[0003] The present invention provides an intelligent scheduling method and system for a photovoltaic cleaning device to solve the technical problems existing in the prior art.

[0004] On the one hand, the present invention provides an intelligent scheduling method for a photovoltaic cleaning device, including:

[0005] Obtaining historical photovoltaic panel images and the corresponding dirt states of each historical photovoltaic panel image; wherein, the dirt state includes that there is dirt on the photovoltaic panel or there is no dirt on the photovoltaic panel;

[0006] Based on the historical photovoltaic panel images and the corresponding dirt states of each historical photovoltaic panel image, determining an integrated self-monitoring model;

[0007] Collecting the current photovoltaic panel image, and using the integrated self-monitoring model to identify and analyze the current photovoltaic panel image to determine the probability value of the real-time dirt state corresponding to the current photovoltaic panel image;

[0008] When the probability value of the real-time dirt state corresponding to the current photovoltaic panel image is greater than a preset threshold, the photovoltaic cleaning device is turned on for intelligent scheduling.

[0009] Further, when a timed cleaning task comes, it is detected whether the power of the photovoltaic panel is greater than a preset threshold. If so, the start of the photovoltaic cleaning device is delayed until the power of the photovoltaic panel is less than the preset photovoltaic threshold, otherwise the photovoltaic cleaning device is turned on.

[0010] Further, based on the historical photovoltaic panel images and the corresponding dirt states of each historical photovoltaic panel image, determining an integrated self-monitoring model, including:

[0011] Determine multiple machine learning models;

[0012] Use the historical photovoltaic panel images as the input of the machine learning models, and use the corresponding soiling states of the historical photovoltaic panel images as the corresponding expected labels to obtain multiple optimized machine learning models;

[0013] Initialize the weighting coefficients corresponding to each neural network model, and use an intelligent optimization algorithm to optimize the weighting coefficients to determine the optimal weighting coefficients;

[0014] Weight the multiple optimized machine learning models with the optimal weighting coefficients to obtain an integrated self-monitoring model.

[0015] Further, the machine learning models include multiple of convolutional neural network, multi-layer perceptron, support vector machine, logistic regression, Bayesian model, and random forest model.

[0016] Further, using the historical photovoltaic panel images as the input of the machine learning models, and using the corresponding soiling states of the historical photovoltaic panel images as the corresponding expected labels to obtain multiple optimized machine learning models includes:

[0017] Encode and initialize the parameters of the machine learning models, and randomly generate M encoded individuals; where each encoded individual includes multiple parameters of the machine learning model;

[0018] Apply the parameters corresponding to each encoded individual to the machine learning models, and use the historical photovoltaic panel images as the input of the machine learning models, and use the corresponding soiling states of the historical photovoltaic panel images as the corresponding expected labels to obtain the error function values corresponding to the encoded individuals;

[0019] According to the error function values corresponding to the encoded individuals, eliminate N encoded individuals with larger error function values;

[0020] Randomly add N encoded individuals to the existing population;

[0021] Set the crossover probability, randomly select two machine learning models from each encoded individual in the existing population as crossover points, and perform crossover operations according to the crossover probability;

[0022] Set the mutation probability, and perform mutation operations on each encoded individual in the existing population according to the mutation probability;

[0023] Repeat the above elimination, addition, crossover, and mutation operations multiple times, and use the encoded individual with the smallest error function value as the final parameter of the machine learning model to obtain multiple optimized machine learning models.

[0024] Further, initialize the weighting coefficients corresponding to each neural network model, and use an intelligent optimization algorithm to optimize the weighting coefficients to determine the optimal weighting coefficients, including:

[0025] Initialize the weighting coefficients corresponding to each neural network model to determine the positions corresponding to multiple particles and the velocities corresponding to the particles;

[0026] Calculate the fitness function value of the particle. If the fitness function value is greater than its historical optimal value, update the historical optimal value corresponding to the particle;

[0027] According to the fitness function value corresponding to each particle, update the global optimal value based on the fitness function value;

[0028] Update the velocity and position of the particle according to the historical optimal value and the global optimal value to obtain the updated particle;

[0029] Determine whether the number of particle updates is equal to the preset iteration number threshold. If so, use the output global optimal value as the optimal weighting coefficient, otherwise return to calculate the fitness function value of the particle.

[0030] Further, calculate the fitness function value of the particle, including:

[0031] For any particle, apply the weighting coefficients included in the particle to the output of the corresponding optimized machine learning model, and use the historical photovoltaic panel image as the input of each optimized machine learning model to obtain multiple weighted outputs;

[0032] Add the multiple weighted outputs to obtain the comprehensive actual output. Use the soiling state corresponding to the historical photovoltaic panel image as the corresponding expected label, and obtain the error function value based on the comprehensive actual output and the expected label;

[0033] Add the error function value obtained using the particle to a preset constant term and take the reciprocal to obtain the fitness function value corresponding to the particle.

[0034] Further, update the velocity and position of the particle according to the historical optimal value and the global optimal value to obtain the updated particle, including:

[0035] Update the velocity of the particle using differential improvement and the double optimal leadership strategy to obtain the updated velocity, and use the updated velocity to update the position of the particle once to obtain the particle after one update;

[0036] For the particle after one update, construct an adaptively changing mutation decision probability, and determine the mutation action based on the mutation decision probability; where the mutation action includes the need to mutate or not to mutate;

[0037] When the mutation action requires mutation, a trigonometric function and a perturbation function that combines the adaptation function value are used to perform a secondary update on the particles after the first update, resulting in the particles after the secondary update.

[0038] Furthermore, an optimal weighting coefficient is used to weight multiple optimized machine learning models to obtain an integrated self-monitoring model, including:

[0039] For each optimized machine learning model, the output of the machine learning model is weighted using the weighting coefficient corresponding to the machine learning model in the optimal weighting coefficient to obtain the output corresponding to the weighted machine learning model;

[0040] The outputs corresponding to the weighted machine learning models are summed to determine the integrated self-monitoring model.

[0041] On the other hand, the present invention provides an intelligent scheduling system for a photovoltaic cleaning device, including: a data acquisition module, a model acquisition module, a dirt prediction module, and an intelligent scheduling module;

[0042] The data acquisition module is used to acquire historical photovoltaic panel images and the dirt status corresponding to each historical photovoltaic panel image; wherein, the dirt status includes that the photovoltaic panel is dirty or the photovoltaic panel is not dirty;

[0043] The model acquisition module is used to determine an integrated self-monitoring model based on the historical photovoltaic panel images and the dirt status corresponding to each historical photovoltaic panel image;

[0044] The dirt prediction module is used to collect the current photovoltaic panel image and use the integrated self-monitoring model to perform identification and analysis on the current photovoltaic panel image to determine the probability value of the real-time dirt status corresponding to the current photovoltaic panel image;

[0045] The integrated self-monitoring model is used to turn on the photovoltaic cleaning device for intelligent scheduling when the probability value of the real-time dirt status corresponding to the current photovoltaic panel image is greater than a preset threshold.

[0046] An intelligent scheduling method and system for a photovoltaic cleaning device provided by the present invention, based on the historical photovoltaic panel images and the corresponding dirt states of each historical photovoltaic panel image, determines an integrated self-monitoring model, and then uses the integrated self-monitoring model to identify and analyze the current photovoltaic panel image to determine the probability value of the real-time dirt state corresponding to the current photovoltaic panel image. When the probability value of the real-time dirt state corresponding to the current photovoltaic panel image is greater than a preset threshold, the photovoltaic cleaning device is turned on, so as to achieve intelligent scheduling. Compared with the prior art, it can not only effectively ensure the cleanliness of the photovoltaic panel, but also only clean when cleaning is needed, effectively avoiding ineffective cleaning and improving the cleaning efficiency of the photovoltaic panel. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings are incorporated herein and form a part of this specification, showing embodiments consistent with the present invention and, together with the specification, are used to explain the principles of the present invention.

[0048] Figure 1 It is a schematic flow chart of an intelligent scheduling method for a photovoltaic cleaning device provided by an embodiment of the present invention.

[0049] Figure 2 It is a schematic structural diagram of an intelligent scheduling system for a photovoltaic cleaning device provided by an embodiment of the present invention.

[0050] Among them, 201 - data acquisition module, 202 - model acquisition module, 203 - dirt prediction module, 204 - intelligent scheduling module.

[0051] Through the above accompanying drawings, specific embodiments of the present invention have been shown, and there will be more detailed descriptions hereinafter. These drawings and written descriptions are not intended to limit the scope of the inventive concept in any way, but to illustrate the concept of the present invention to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] Here, the exemplary embodiments will be described in detail, and the examples are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0053] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0054] As Figure 1 shown, an embodiment of the present invention provides an intelligent scheduling method for a photovoltaic cleaning device, including:

[0055] S101. Obtain historical photovoltaic panel images and the corresponding dirt states of each historical photovoltaic panel image; wherein, the dirt state includes that the photovoltaic panel has dirt or the photovoltaic panel has no dirt;

[0056] The dirt state corresponding to each historical photovoltaic panel image is marked by the staff whether there is dirt. When there is dirt, it can be determined that cleaning is required.

[0057] S102. Based on the historical photovoltaic panel images and the corresponding dirt states of each historical photovoltaic panel image, determine an integrated self-monitoring model;

[0058] Learn the correlation between the historical photovoltaic panel images and the corresponding dirt states of each historical photovoltaic panel image through a machine learning algorithm, so that when collecting photovoltaic panel images subsequently, image analysis can be performed through the integrated self-monitoring model to determine the probability value of the photovoltaic panel being in a dirty state.

[0059] S103. Collect the current photovoltaic panel image, and use the integrated self-monitoring model to identify and analyze the current photovoltaic panel image to determine the probability value of the real-time dirt state corresponding to the current photovoltaic panel image;

[0060] Since the previous integrated self-monitoring model is trained with historical data, when there is a large amount of historical data, it can theoretically accurately identify the dirt state. Therefore, the current photovoltaic panel image can be collected for identification.

[0061] S104. When the probability value of the real-time dirt state corresponding to the current photovoltaic panel image is greater than a preset threshold, turn on the photovoltaic cleaning device for intelligent scheduling.

[0062] Optionally, when the probability value of the real-time dirt state corresponding to the current photovoltaic panel image is greater than a preset threshold, it can also be determined whether the current time is less than a preset time threshold from the sunset time. If so, set a scheduled task to execute the task of turning on the photovoltaic cleaning device after sunset, otherwise directly turn on the photovoltaic cleaning device.

[0063] When the probability value of the real-time dirt state corresponding to the current photovoltaic panel image is less than a preset threshold, it proves that the dirt is not serious or there is no dirt, so there is no need to turn on the photovoltaic cleaning device.

[0064] An intelligent scheduling method for a photovoltaic cleaning device provided by the present invention determines an integrated self-monitoring model based on the historical photovoltaic panel images and the corresponding dirt states of each historical photovoltaic panel image, and then uses the integrated self-monitoring model to identify and analyze the current photovoltaic panel image to determine the probability value of the real-time dirt state corresponding to the current photovoltaic panel image. When the probability value of the real-time dirt state corresponding to the current photovoltaic panel image is greater than a preset threshold, the photovoltaic cleaning device is turned on, so that intelligent scheduling can be realized. Compared with the prior art, it can not only effectively ensure the cleanliness of the photovoltaic panel, but also clean only when cleaning is needed, effectively avoiding ineffective cleaning and improving the cleaning efficiency of the photovoltaic panel.

[0065] In an embodiment of the present invention, when a timed cleaning task comes, it is detected whether the power of the photovoltaic panel is greater than a preset threshold. If so, the start of the photovoltaic cleaning device is delayed until the power of the photovoltaic panel is less than the preset photovoltaic threshold, otherwise the photovoltaic cleaning device is turned on.

[0066] In an embodiment of the present invention, determining an integrated self-monitoring model based on the historical photovoltaic panel images and the corresponding dirt states of each historical photovoltaic panel image includes:

[0067] Determining a plurality of machine learning models;

[0068] Using the historical photovoltaic panel images as the input of the machine learning model and the corresponding dirt states of the historical photovoltaic panel images as the corresponding expected labels to obtain a plurality of optimized machine learning models;

[0069] Initializing the weighting coefficients corresponding to each neural network model and using an intelligent optimization algorithm to optimize the weighting coefficients to determine the optimal weighting coefficients;

[0070] Using the optimal weighting coefficients to weight a plurality of optimized machine learning models to obtain an integrated self-monitoring model.

[0071] In the prior art, a single machine learning model is often used for data recognition, and there may be misdetection. If multiple machine learning models are directly combined and used, the computational overhead brought by the combined use cannot be most effectively utilized, and the data recognition accuracy is not maximally improved.

[0072] In an embodiment of the present invention, the machine learning models include multiple of a convolutional neural network, a multi-layer perceptron, a support vector machine, a logistic regression, a Bayesian model, and a random forest model.

[0073] It should be noted that in addition to the above machine learning models, other machine learning models can also be used to realize the joint recognition of multiple machine learning models.

[0074] In an embodiment of the present invention, using the historical photovoltaic panel image as the input of a machine learning model and the corresponding soiling state of the historical photovoltaic panel image as the corresponding expected label, a plurality of optimized machine learning models are obtained, including:

[0075] Encode and initialize the parameters of the machine learning model, and randomly generate M encoded individuals; where each encoded individual includes a plurality of parameters of the machine learning model;

[0076] Apply the parameters corresponding to each encoded individual to the machine learning model, use the historical photovoltaic panel image as the input of the machine learning model, and use the corresponding soiling state of the historical photovoltaic panel image as the corresponding expected label to obtain the error function value corresponding to the encoded individual;

[0077] According to the error function value corresponding to the encoded individual, eliminate N encoded individuals with larger error function values;

[0078] Randomly add N encoded individuals to the existing population;

[0079] Set the crossover probability, randomly select two machine learning models from each encoded individual in the existing population as the crossover points, and perform crossover operations according to the crossover probability;

[0080] Set the mutation probability and perform mutation operations on each encoded individual in the existing population according to the mutation probability;

[0081] Repeat the above elimination, addition, crossover, and mutation operations multiple times, and use the encoded individual with the smallest error function value as the final parameter of the machine learning model to obtain a plurality of optimized machine learning models.

[0082] In an embodiment of the present invention, initialize the weighting coefficients corresponding to each neural network model, and use an intelligent optimization algorithm to optimize the weighting coefficients to determine the optimal weighting coefficients, including:

[0083] Initialize the weighting coefficients corresponding to each neural network model to determine the positions of a plurality of particles and the velocities of the particles;

[0084] Calculate the fitness function value of the particle. If the fitness function value is greater than its historical optimal value, update the historical optimal value corresponding to the particle;

[0085] According to the fitness function value corresponding to each particle, update the global optimal value according to the fitness function value;

[0086] Update the velocity and position of the particle according to the historical optimal value and the global optimal value to obtain the updated particle;

[0087] Determine whether the number of particle updates is equal to the preset iteration number threshold. If so, use the output global optimal value as the optimal weighting coefficient. Otherwise, return the calculated fitness function value of the particle.

[0088] In the embodiments of the present invention, calculating the fitness function value of a particle includes:

[0089] For any one particle, apply the weighting coefficient included in the particle to the output of the corresponding optimized machine learning model, and use the historical photovoltaic panel image as the input of each optimized machine learning model to obtain multiple weighted outputs;

[0090] Add up the multiple weighted outputs to obtain a comprehensive actual output. Use the soiling state corresponding to the historical photovoltaic panel image as the corresponding expected label, and obtain the error function value according to the comprehensive actual output and the expected label;

[0091] Add the error function value obtained by the particle to the preset constant term and take the reciprocal to obtain the fitness function value corresponding to the particle.

[0092] In the embodiments of the present invention, by obtaining the best coefficients of each machine learning model, the best prediction combination can be found, so as to effectively use each model for joint prediction and improve the recognition accuracy of the soiling state.

[0093] Optionally, constraint conditions can also be set, that is, the sum of the weight coefficients corresponding to each optimized machine learning model is one. However, it should be noted that the constraint conditions can also not be set, and only an upper and lower limit of the weight coefficient can be set. When the weight coefficient exceeds the limit, boundary crossing processing can be performed. Then, when the best weight coefficient combination is determined, the maximum sum of outputs can be determined, and the staff can determine the probability threshold for turning on the photovoltaic cleaning device, so as to realize the control of the customized intelligent cleaning device.

[0094] For example, when there are two machine learning models, the weight coefficient corresponding to the first machine learning model is 2, and the weight coefficient corresponding to the second machine learning model is 3. Therefore, the maximum sum of outputs is 2×1 + 3×1 = 5. Therefore, the staff can set a value between 0 and 5 as the turning-on threshold (such as 4). When the predicted probabilities of the two machine learning models are weighted and summed, if it is greater than 4, the photovoltaic cleaning device can be turned on.

[0095] In the embodiments of the present invention, according to the historical optimal value and the global optimal value, update the velocity and position of the particle to obtain the updated particle, including:

[0096] A1. The velocity of the particle is updated by using differential improvement and the double-optimal leadership strategy to obtain the updated velocity, and the position of the particle is updated once with the updated velocity to obtain the particle after the first update.

[0097] In this embodiment, updating the velocity of the particle by using differential improvement and the double-optimal leadership strategy may include:

[0098]

[0099] Among them, represents the differential coefficient, represents the updated velocity, represents the first random particle, represents the second random particle, represents the individual learning factor, represents the population learning factor, represents the first random number between (0, 1), represents the second random number between (0, 1), represents the t historical optimal value corresponding to the i th particle in the th update process, i represents the global optimal particle, that is, the global optimal value; , , are all different.

[0100] Updating the position of the particle once with the updated velocity, the particle after the first update is:

[0101] ;

[0102] A2. For the particle after the first update, an adaptive mutation decision probability is constructed, and the mutation action is determined with this mutation decision probability; among them, the mutation action includes needing to mutate or not needing to mutate.

[0103] For example, the mutation decision probability can be set as: , represents the preset maximum mutation decision probability, represents the preset minimum mutation strategy probability, T represents the preset maximum number of updates, so that there can be more mutation possibilities in the early stage of the algorithm, avoiding falling into the local optimum, and the mutation probability can be reduced in the later stage of the algorithm, thus ensuring the convergence degree of the algorithm.

[0104] A third random number between (0, 1) can be generated, and it is determined whether the third random number is less than the mutation decision probability. If so, it is determined that the mutation action is required to mutate; otherwise, it is determined that the mutation action is not required to mutate.

[0105] A3. When the mutation action is required to mutate, the particles after the first update are secondarily updated by using trigonometric functions and a perturbation function that combines the fitness function values to obtain the particles after the secondary update.

[0106] The secondary update of the particles after the first update by using trigonometric functions and a perturbation function that combines the fitness function values is as follows:

[0107]

[0108]

[0109]

[0110]

[0111]

[0112] Among them, represents the d-th dimensional parameter in the m-th particle after the first update, where d = 1, 2, …, D, and D represents the total dimension of the parameters included in the particle. represents the perturbation function. represents the d-th dimensional parameter in the particle after the secondary update. represents a symbolic variable that is randomly -1 or 1. represents the pi. represents the arctangent function. represents an intermediate parameter. represents the perturbation amount affected by the fitness value. represents a constant term (such as set to 10). represents the d-th dimensional parameter in the historical best individual of the m-th particle. represents the d-th dimensional parameter of the global optimal value. represents the d-th dimensional parameter in the historical best individual of the n-th particle. represents the degree of fitness quality. represents the fitness value corresponding to the m-th particle. represents the fitness value corresponding to the global optimal value.

[0113] The mutation update provided by the embodiments of the present invention can provide stronger mutation ability when the particle is close to the optimal particle, and the mutation changes with the number of updates, having stronger adaptability, so as to effectively avoid the algorithm falling into local optimum. Compared with the prior art, it has stronger global search ability and solves the technical problems existing in the existing particle swarm algorithm.

[0114] In the embodiments of the present invention, the optimal weighted coefficients are used to weight multiple optimized machine learning models to obtain an integrated self-monitoring model, including:

[0115] For each optimized machine learning model, the output of the machine learning model is weighted by the weighted coefficient corresponding to the machine learning model in the optimal weighted coefficients to obtain the output corresponding to the weighted machine learning model;

[0116] The outputs corresponding to the weighted machine learning models are summed to determine the integrated self-monitoring model.

[0117] As Figure 2 shown, the present invention provides an intelligent scheduling system for a photovoltaic cleaning device, including: a data acquisition module 201, a model acquisition module 202, a dirt prediction module 203, and an intelligent scheduling module 204;

[0118] The data acquisition module 201 is used to acquire historical photovoltaic panel images and the dirt state corresponding to each historical photovoltaic panel image; wherein, the dirt state includes that the photovoltaic panel has dirt or the photovoltaic panel has no dirt;

[0119] The model acquisition module 202 is used to determine an integrated self-monitoring model based on the historical photovoltaic panel images and the dirt state corresponding to each historical photovoltaic panel image;

[0120] The dirt prediction module 203, the user collects the current photovoltaic panel image, and uses the integrated self-monitoring model to identify and analyze the current photovoltaic panel image to determine the probability value of the real-time dirt state corresponding to the current photovoltaic panel image;

[0121] The intelligent scheduling module 204 is used to turn on the photovoltaic cleaning device for intelligent scheduling when the probability value of the real-time dirt state corresponding to the current photovoltaic panel image is greater than a preset threshold.

[0122] The intelligent scheduling system for a photovoltaic cleaning device provided by the embodiments of the present invention can execute the above method technical solutions, and its principle and beneficial effects are similar, which will not be elaborated here.

[0123] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0124] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0125] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0126] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0127] Those of ordinary skill in the art can understand that all or part of the steps in implementing the above-mentioned methods and facts can be completed by instructing relevant hardware through a program. The involved program or the described program can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: At this time, the corresponding method steps are introduced. The storage medium can be ROM / RAM, magnetic disk, optical disk, etc.

[0128] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is only for the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent scheduling method for a photovoltaic cleaning device, characterized in that Including: Obtaining historical photovoltaic panel images and the corresponding soiling states of each historical photovoltaic panel image; wherein, the soiling state includes that the photovoltaic panel is soiled or the photovoltaic panel is not soiled; Based on the historical photovoltaic panel images and the corresponding soiling states of each historical photovoltaic panel image, determining an integrated self-monitoring model, which includes: Determining a plurality of machine learning models, wherein the machine learning models include multiple of a convolutional neural network, a multi-layer perceptron, a support vector machine, a logistic regression, a Bayesian model, and a random forest model; Using the historical photovoltaic panel images as the input of the machine learning models and the soiling states corresponding to the historical photovoltaic panel images as the corresponding expected labels, obtaining a plurality of optimized machine learning models; Initializing the weighting coefficients corresponding to each neural network model and using an intelligent optimization algorithm to optimize the weighting coefficients, determining the optimal weighting coefficients, which includes: Initializing the weighting coefficients corresponding to each neural network model to determine the positions and velocities of a plurality of particles; Calculating the fitness function value of the particle. If the fitness function value is greater than its historical optimal value, then updating the historical optimal value corresponding to the particle; According to the fitness function value corresponding to each particle and updating the global optimal value according to the fitness function value; Updating the velocity and position of the particle according to the historical optimal value and the global optimal value to obtain the updated particle, which includes: A1. Updating the velocity of the particle by using differential improvement and a double-optimal leadership strategy to obtain the updated velocity, and using the updated velocity to update the position of the particle once to obtain the particle after one update; Updating the velocity of the particle by using differential improvement and a double-optimal leadership strategy, including: Among them, represents the difference coefficient, represents the speed after update, represents the first random particle, represents the second random particle, represents the individual learning factor, represents the population learning factor, represents the first random number between (0, 1), represents the second random number between (0, 1), represents the t th historical optimal value corresponding to the i th particle during the th update process, i represents the global optimal particle, i.e., the global optimal value; , , are all different; Updating the position of the particle once by using the updated velocity to obtain the particle after one update as: ； A2. For the particle after one update, constructing an adaptively changing mutation decision probability and determining a mutation action based on the mutation decision probability; wherein, the mutation action includes the need to mutate or the need not to mutate; The mutation decision probability is set to: , represents the preset maximum mutation decision probability, represents the preset minimum mutation strategy probability, and T represents the preset maximum number of updates, so that the algorithm has more mutation possibilities in the early stage, avoiding falling into local optima, and reducing the mutation probability in the later stage of the algorithm, thus ensuring the convergence degree of the algorithm; Generating a third random number between (0, 1) and determining whether the third random number is less than the mutation decision probability. If so, determining that the mutation action is the need to mutate, otherwise determining that the mutation action is the need not to mutate; A3. When the mutation action is the need to mutate, then performing a secondary update on the particle after one update by using a trigonometric function and a perturbation function that fuses the fitness function value to obtain the particle after the secondary update; Performing a secondary update on the particle after one update by using a trigonometric function and a perturbation function that fuses the fitness function value as: Among them, represents the d-th dimensional parameter in the m-th particle after the first update, where d = 1, 2, …, D, and D represents the total dimension of the parameters included in the particle, represents the perturbation function, represents the d-th dimensional parameter in the particle after the second update, represents a symbolic variable, randomly -1 or 1, represents pi, represents the arctangent function, represents an intermediate parameter, represents the perturbation amount affected by the fitness value, represents a constant term, represents the d-th dimensional parameter in the historical best individual of the m-th particle, represents the d-th dimensional parameter of the global optimal value, represents the d-th dimensional parameter in the historical best individual of the n-th particle, represents the degree of fitness quality, represents the fitness value corresponding to the m-th particle, represents the fitness value corresponding to the global optimal value; Judging whether the particle update times is equal to the preset iteration times threshold. If so, then using the output global optimal value as the optimal weighting coefficient, otherwise returning to calculate the fitness function value of the particle; Weighting the plurality of optimized machine learning models by using the optimal weighting coefficients to obtain the integrated self-monitoring model; Collect the current photovoltaic panel image, and use the integrated self-monitoring model to identify and analyze the current photovoltaic panel image to determine the probability value of the real-time dirt state corresponding to the current photovoltaic panel image; When the probability value of the real-time dirt state corresponding to the current photovoltaic panel image is greater than the preset threshold, turn on the photovoltaic cleaning device for intelligent scheduling.

2. The intelligent scheduling method for the photovoltaic cleaning device according to claim 1, wherein When the timed cleaning task arrives, detect whether the power of the photovoltaic panel is greater than the preset threshold. If so, delay the activation of the photovoltaic cleaning device until the power of the photovoltaic panel is less than the preset photovoltaic threshold; otherwise, turn on the photovoltaic cleaning device.

3. A smart scheduling system for a photovoltaic cleaning device, which is used to execute the smart scheduling method of the photovoltaic cleaning device according to any one of claims 1-2, characterized in that, It includes: A data acquisition module, a model acquisition module, a dirt prediction module, and an intelligent scheduling module; The data acquisition module is used to acquire historical photovoltaic panel images and the dirt state corresponding to each historical photovoltaic panel image; wherein, the dirt state includes that the photovoltaic panel has dirt or the photovoltaic panel has no dirt; The model acquisition module is used to determine the integrated self-monitoring model based on the historical photovoltaic panel images and the dirt state corresponding to each historical photovoltaic panel image; The dirt prediction module is used to collect the current photovoltaic panel image, and use the integrated self-monitoring model to identify and analyze the current photovoltaic panel image to determine the probability value of the real-time dirt state corresponding to the current photovoltaic panel image; The integrated self-monitoring model is used to turn on the photovoltaic cleaning device for intelligent scheduling when the probability value of the real-time dirt state corresponding to the current photovoltaic panel image is greater than the preset threshold.

Images