Data-driven Identification Method and Device for Optimizing the Cleaning Time of a Photovoltaic Array

Abstract

The present invention provides a data-driven identification method and device for optimizing the cleaning time of a photovoltaic array, belonging to the field of optimizing the cleaning time of a photovoltaic array. The method of the present invention includes the following steps: collecting relevant data of the dust accumulation density and output power of the photovoltaic array under preset conditions; based on the relevant data, conducting a principle analysis and process description of the model identification of the stag hunt optimization algorithm; based on the model parameter identification method of the stag hunt optimization algorithm, respectively identifying the dust accumulation attenuation model and the peak hour number prediction model of the photovoltaic system; conducting a rationality evaluation and quantitative analysis of the predicted cleaning time of the photovoltaic array. The present invention can indirectly obtain the optimal cleaning time interval through the established dust accumulation attenuation model and peak hour number prediction model of the photovoltaic system, providing a reference for the dynamic optimization of the cleaning time. By comparing the changes in the power generation benefits of the photovoltaic system before and after cleaning through benefit quantification, the rationality of the obtained cleaning time plan can be effectively evaluated.

Description

Technical Field

[0001] The present invention belongs to the field of optimizing the cleaning time of a photovoltaic array, and particularly relates to a data-driven identification method and device for optimizing the cleaning time of a photovoltaic array. Background Art

[0002] With the energy transition worldwide, solar energy is actively used in multiple fields such as power generation and heating, and its application in the power generation process shows an increasingly prosperous trend, and countries' attention to it is gradually increasing. The research focus of a photovoltaic system is how to improve the power generation efficiency of a photovoltaic array or restore the efficiency loss caused by various internal and external factors. In this context, as one of the main reasons affecting the energy efficiency of a photovoltaic system, the dust accumulation phenomenon on the surface of the glass cover of a photovoltaic module or the pollution of flat glass has become an issue that cannot be ignored. It is necessary to evaluate the economic losses caused by problems such as dust accumulation on a photovoltaic array over time, and then determine a reasonable cleaning time interval. However, existing research has limited exploration of this problem. Its importance in improving the power generation efficiency and economy of photovoltaic power generation remains to be further explored.

[0003] The performance of a photovoltaic module is affected by factors such as conversion efficiency, weather patterns, soil, photovoltaic tilt angle, photovoltaic cell temperature, environmental temperature, and humidity. To mitigate the impact of problems such as dust accumulation or pollution on the surface of the module during its operation in an outdoor environment, cleaning and maintenance are the key means to ensure its safe and efficient operation. The choice of cleaning time interval is an important factor determining its operation and maintenance costs. Although frequent cleaning operations can effectively increase power generation, the resulting costs of water, electricity, equipment, and labor will directly reduce the benefits or even cause the situation of not increasing but decreasing. On the contrary, if the cleaning interval is too long, it will directly affect the power generation efficiency of the photovoltaic system, thereby causing a decline in benefits. Therefore, the cleaning of a photovoltaic array is essentially an economic benefit optimization problem considering power generation and cleaning costs. Although different methods are used to optimize the cleaning time or cycle of a photovoltaic array, most of them achieve the prediction of cleaning time and benefit optimization through the modeling of dust accumulation speed and cleaning costs.

[0004] Therefore, considering the rapid development of big data technology, the attempt and application of data-driven identification methods in optimizing the cleaning time of a photovoltaic array have great potential. The present invention proposes a data-driven identification method and device for optimizing the cleaning time of a photovoltaic array. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art, and provides a data-driven identification method and device for optimizing the cleaning time of a photovoltaic array.

[0006] On the one hand, the present invention provides a data-driven identification method for optimizing the cleaning time of a photovoltaic array, including the following specific steps:

[0007] Collect relevant data on the dust accumulation density and output power of a photovoltaic array under preset conditions;

[0008] Based on the relevant data, conduct a principle analysis and process description of the model identification of the stag hunting optimization algorithm;

[0009] Based on the model parameter identification method of the stag hunting optimization algorithm, identify the dust accumulation attenuation model and the peak hour prediction model of the photovoltaic system respectively;

[0010] Conduct a rationality evaluation and quantitative analysis of the dust cleaning prediction time of the photovoltaic array.

[0011] Optionally, the collection of relevant data on the dust accumulation density and output power of the photovoltaic array under preset conditions includes:

[0012] Collect relevant data on the dust accumulation density and output power of the photovoltaic system based on climate, environment, and terrain factors;

[0013] Select multiple combinations of climate, environment, and terrain conditions for data collection.

[0014] Optionally, the principle analysis and process description of the model identification of the stag hunting optimization algorithm based on the relevant data includes:

[0015] Set a preset form for the model to be identified;

[0016] Regard the best position of the stag hunting as the optimal model parameter vector θ * , and optimize the model identification of the stag hunting optimization algorithm;

[0017] Obtain the optimal hunting position H according to the model identification of the stag hunting optimization algorithm best , and θ * =H best .

[0018] Optionally, regarding the best position of the stag hunting as the optimal model parameter vector θ * , and optimizing the model identification of the stag hunting optimization algorithm includes:

[0019] Take s hunters as a population H, and randomly initialize the hunter population as H = {H 1 , H 2 , …, H s};

[0020] Determine the initial wind direction angle and position angle for hunting stags;

[0021] Calculate the position closest to the optimal solution according to the fitness function;

[0022] When the optimal hunting position is determined, the hunters stop updating their positions.

[0023] Optionally, the initial wind direction angle and position angle for determining the hunting of stags are obtained using the following formula:

[0024] φ k = 2πl

[0025]

[0026] where φ represents the wind direction angle, represents the position angle; l is a random number in the range [0, 1], and k represents the current iteration step.

[0027] Optionally, calculating the position closest to the optimal solution according to the fitness function includes:

[0028] Based on the propagation process of the leader position, when the best position is determined, each hunter in the population attempts to reach the best position, and at this time, the position update process is triggered. Among them, the hunter surrounding behavior can be expressed by the following formula:

[0029] H k+1 = H lead - M·c|R×H lead - H k |

[0030] where: H lead represents the leader position, c is a random number considering the wind direction angle, and its value range is (0, 2], and both M and R refer to parameter vectors;

[0031] Based on the propagation process of the position angle, the search space is improved by considering the position angle during the hunter's position update process. During the deer hunting process, assuming that the hunting process is effective at a certain position angle, a new parameter ds k is introduced to update the position angle, and this parameter is developed based on the variance between the wind angles. Its visible angle is as follows:

[0032] ds k = φ k - vs k

[0033] and, the visible angle vs of the prey k is as follows:

[0034]

[0035] where rad is a random number taken from [0, 1];

[0036] And, the position angle is updated using the following relationship:

[0037]

[0038] The position of the hunter called update is carried out using the following relational expressions:

[0039] H k+1 = H lead - c|cos(w) × H lead - H k |;

[0040] Where: H k+1 is the position vector of the hunter at time k, H k is the position of the hunter at time k + 1 after update, and w represents the position angle;

[0041] Based on the propagation process of the successor position: Then assume that the value of the vector R is less than 1, and the update process of the hunter position depends on the position of the successor, as shown in the following relational expressions:

[0042] H k+1 = H successor - M·c|R × H successor - H k |

[0043] Where, H successor represents the position of the successor.

[0044] Optionally, for the model parameter identification method based on the stag hunting optimization algorithm, the dust accumulation attenuation model and the peak hour prediction model of the photovoltaic system are respectively identified, including:

[0045] Select the time when the photovoltaic module has just been cleaned as the sampling starting point under different seasons and weather conditions, and use the change of the rated peak power of the photovoltaic module over time after cleaning as a reference to establish a power attenuation trend model that can accurately describe the photovoltaic module with the dust accumulation duration;

[0046] Select the environmental temperature, air humidity, cloud thickness and air density data from the sampling data as the modeling input, and the solar irradiance as the output. Based on the preset model form, the unknown parameters of the model are identified through the DHO algorithm, and the peak hours are calculated.

[0047] Optionally, the peak hours are calculated using the following relational expression:

[0048]

[0049] Where, the peak hours H refer to the hours within a certain period of time when the total solar irradiance is converted into standard test conditions, that is, the irradiance is 1000 W / m 2 , and the temperature is 25 °C;

[0050] t 1 and t 2They are the start time point and the end time point of the calculated time period respectively, and R(t) is the irradiance of the photovoltaic array at time t.

[0051] Optionally, the rationality evaluation and quantitative analysis of the predicted time for cleaning the dust on the photovoltaic array include:

[0052] Plot the rated peak power change curve of the following photovoltaic system before and after dust accumulation and cleaning, and obtain the power generation increment ΔQ generated by the photovoltaic system after cleaning;

[0053] Quantitatively calculate the increment of power station revenue ΔE brought by the cleaning operation to photovoltaic power generation through the following formula:

[0054] ΔE = E clean - E dust - C

[0055] E clean = r × Q clean

[0056] E dust = r × Q dust

[0057] Among them, E clean and E dust respectively represent the power generation revenues with and without cleaning the photovoltaic array, C is the cleaning cost, Q clean represents the cumulative power generation when the photovoltaic array is cleaned at time t within the time period Tc, Q 1 represents the cumulative power generation when the photovoltaic array is not cleaned within the time period Tc, and r represents the grid-connected electricity price of photovoltaic power generation; dust And, judge whether ΔE is greater than 0 to obtain the rationality of the cleaning time and seek the optimal cleaning time interval.

[0058] On the other hand, the present invention provides a data-driven identification device for optimizing the cleaning time of a photovoltaic array, including: a collection module, an analysis module, an identification module, and an evaluation module; among them,

[0059] The collection module is used to collect relevant data on the dust accumulation density and output power of the photovoltaic array under preset conditions;

[0060] The analysis module is used to conduct principle analysis and process description on the model identification of the stag hunt optimization algorithm based on the relevant data;

[0061] The identification module is used to identify the dust accumulation attenuation model and the peak hour number prediction model of the photovoltaic system respectively based on the model parameter identification method of the stag hunt optimization algorithm;

[0062]

[0063] ​The evaluation module is used to evaluate the rationality and quantitatively analyze the predicted time for cleaning the dust accumulation on the photovoltaic array.

[0064] The present invention aims to provide a data-driven identification method for optimizing the cleaning time of a photovoltaic array to predict the optimal cleaning time of a photovoltaic system and improve its power generation efficiency and economy. This method constructs a data-driven model identification method through a stag hunt optimization algorithm and applies it to the modeling of a dust accumulation attenuation model and a peak hour number prediction model of a photovoltaic system. At the same time, cost constraints are considered in this process, and the feasibility and rationality of the cleaning strategy are evaluated with economic benefits as the performance index. It provides an effective reference for obtaining the optimal cleaning time of the photovoltaic array and plays a great role in promoting the economic, efficient, safe and stable operation of the photovoltaic power generation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a schematic diagram of a data-driven identification method for optimizing the cleaning time of a photovoltaic array according to an embodiment of the present invention;

[0066] Figure 2 It is a flowchart of a data-driven identification method for optimizing the cleaning time of a photovoltaic array according to another embodiment of the present invention;

[0067] Figure 3 It is a curve of the rated peak power change of a photovoltaic system after dust accumulation and cleaning according to another embodiment of the present invention;

[0068] Figure 4 It is a schematic diagram of a data-driven identification device for optimizing the cleaning time of a photovoltaic array according to another embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0069] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0070] As Figure 1 and Figure 2 shown, on the one hand, the present invention provides a data-driven identification method S100 for optimizing the cleaning time of a photovoltaic array, including the following specific steps S110 to S140:

[0071] S110. Collect relevant data on the dust accumulation density and output power of the photovoltaic array under preset conditions.

[0072] It should be noted that different climate, environment, and terrain factors can cause significant differences in air humidity, oxygen content, and dust density, which in turn affect the dust accumulation situation of the photovoltaic array. For example, in seasons with more rainfall, the dust on the surface of the photovoltaic array can be effectively removed through the washing of rainwater, achieving a cleaning effect. At this time, the power generation efficiency of the photovoltaic system will be effectively improved. Based on this, step S110 can be specifically refined into the following S1101 to S1102.

[0073] S1101. Collect data related to the dust accumulation density and output power of the photovoltaic system based on climate, environment, and terrain factors.

[0074] S1102. To ensure the comprehensiveness of the sampled data, select multiple combinations of climate, environment, and terrain conditions for data collection. If there are variables in the dust accumulation density and output power of the photovoltaic system that cannot be directly measured, then sample the data related to them and perform calculations to obtain the indirectly measured target. Let the sampling step size be T, the number of combinations of climate, environment, and terrain conditions be n, the number of data vectors sampled under each combination be N, and the dimension of the data vector be m.

[0075] Based on the data related to the dust accumulation density and output power of the photovoltaic system obtained in step S110, a specific description of the model identification based on the Deer Hunting Optimization (DHO) algorithm is given in step S120.

[0076] S120. Based on the relevant data, conduct a principle analysis and process description of the model identification of the deer hunting optimization algorithm.

[0077] Specifically, S1201. Set a preset form for the model to be identified, such as the transfer function or subspace model form, and set the parameter vector to be identified as θ, and its optimal value as θ * 。

[0078] S1202. The hunting method of deer by humans provides inspiration for developing the DHO algorithm. The key objective of the DHO algorithm is to determine the best or effective position for hunting deer. Some specific characteristics of male deer (i.e., deer) can protect themselves from predators or hunters, such as excellent vision, extraordinary smell, and vigilance against ultra-high-frequency sounds. Based on the above analysis, regard the best position for hunting deer as the optimal model parameter vector θ * , and the model identification based on the DHO algorithm is mainly divided into four stages.

[0079] First, population initialization. Consider s hunters as a population H, and randomly initialize the hunter population as H = {H 1 ,H 2 ,…,H s}:

[0080] Second, initialize the wind and the position angle. The initialization of the position and the wind direction angle is considered a key process for the hunter to determine the best ideal position for hunting the stag. The wind direction angle φ and the position angle are given by the following formulas respectively:

[0081] φ k = 2πl (1)

[0082]

[0083] where: l is a random number in the range of [0,1], and k represents the current iteration step.

[0084] Third, position propagation. In the initial stage, the position of the optimal or ideal space is uncertain. Therefore, the position closest to the optimal solution calculated according to the fitness function is regarded as the optimal space. Usually, two position update schemes are considered, namely the successor position and the leader position.

[0085] Propagation process based on the leader position: Once the best position is determined, each hunter in the population tries to reach the best position, and at this time, the position update process is triggered. The hunter surrounding behavior can be expressed by the following formula:

[0086] H k+1 = H lead - M·c|R×H lead - H k | (3)

[0087] where: represents the leader position, c is a random number considering the wind direction angle, and its value range is (0,2], and both M and R refer to the parameter vectors, and the calculation formulas are as follows:

[0088]

[0089] R = 2rd (5)

[0090] where: rd is a random number taken from [0,1], k max is the maximum number of iterations, and the parameter a is used to measure the coefficient vector in the equation.

[0091] Propagation process based on the position angle: In this case, the search space is improved by considering the position angle during the position update process of the hunter. During the deer hunting process, it is assumed that the hunting process is effective at a certain position angle. In addition, a new parameter ds k is introduced for the position angle update and this parameter is developed based on the variance between the wind angles, and its visible angle is expressed in formula (6). At the same time, the visible angle vs k of the prey is defined by formula (7).

[0092] dsk = φ k - vs k (6)

[0093]

[0094] Update the position angle using Equation (8), and update the position of the hunter based on the position angle shown in Equation (9):

[0095]

[0096] H k+1 = H lead - c|cos(w) × H lead - H k | (9)

[0097] Where: H k+1 is the position of the hunter at time k + 1 updated using the hunter position vector H k at time k, and w represents the position angle.

[0098] Propagation process based on the successor position: The surrounding mechanism is used to change the vector R in the exploration stage. However, in the initial state, random search is first considered, and then it is assumed that the value of the vector R is less than 1. The update process of the hunter's position depends on the position of the successor, as shown in Equation (10). If the value of the vector R is less than 1, a search agent is randomly selected; otherwise, the optimal solution is selected to update the position of the search agent.

[0099] H k+1 = H successor - M·c|R × H successor - H k | (10)

[0100] Fourth, the termination process. When the optimal hunting position is determined, the hunter stops updating the position, and the above optimization process ends.

[0101] S1203. For the data-driven modeling of this embodiment, the optimal hunting position H best obtained at this time is the optimal solution of the model parameter vector, that is, θ * = H best .

[0102] According to the model parameter identification method based on DHO designed in step S120, the identification of the dust accumulation attenuation model and the peak hour prediction model of the photovoltaic system is carried out separately in step S130.

[0103] S130. Based on the model parameter identification method of the deer hunting optimization algorithm, the dust accumulation attenuation model and the peak hour prediction model of the photovoltaic system are identified separately.

[0104] Specifically, S1301, identification of the dust accumulation attenuation model of the photovoltaic system. Select the time when the photovoltaic modules have just been cleaned as the sampling starting point under different seasons and weather conditions, and take the change of the rated peak power of the photovoltaic modules over time after cleaning as a reference, so as to establish a power attenuation trend model that can accurately describe the photovoltaic modules with the dust accumulation duration, providing a reference for reasonably planning the cleaning time of the photovoltaic modules.

[0105] The dust accumulation attenuation characteristics of the photovoltaic system are related to solar irradiance, photovoltaic module temperature and output power. Therefore, in the modeling process, corresponding variables should be selected from the collected data first, then the form of the attenuation model is preset, and then the unknown parameters of the model are optimized and identified through the above-mentioned DHO-based modeling method.

[0106] S1302, identification of the peak hour number prediction model. The peak hour number H refers to the number of hours when the total solar irradiance in a certain period is converted into standard test conditions, that is, the irradiance is 1000W / m 2 , and the temperature is 25°C. Generally, it can be calculated by the following formula:

[0107]

[0108] where: t 1 and t 2 are the starting time point and the ending time point of the calculated time period respectively, and R(t) is the irradiance of the photovoltaic array at time t.

[0109] As can be seen from the above formula, the peak hour number is directly related to solar irradiance. Therefore, the influencing factors of solar irradiance change need to be considered. Through the weather change situation during the time period from t 1 to t 2 , it can be known that the change of solar irradiance is closely related to the weather clarity, cloud thickness, oxygen content and humidity of the air, etc.

[0110] Therefore, based on five weather types: sunny, cloudy, overcast, light rain and heavy rain to storm, select data such as ambient temperature, air humidity, cloud thickness and air density from the sampling data as the modeling input, and solar irradiance as the output. Based on the preset model form, the unknown parameters of the model are identified through the DHO algorithm. Then, the peak hour number is calculated using formula (11).

[0111] S1303, according to the established dust accumulation attenuation model and peak hour number prediction model of the photovoltaic system, combined with the economy of the photovoltaic system, obtain the optimal interval of the cleaning time of the photovoltaic array.

[0112] S140, conduct a rationality evaluation and quantitative analysis on the predicted cleaning time of the dust on the photovoltaic array.

[0113] Specifically, in step S140, a quantitative analysis of the cleaning benefits of the photovoltaic array is performed to dynamically evaluate the rationality of the obtained dust cleaning time. First, draw the rated peak power change curves of the following photovoltaic system before and after dust accumulation and cleaning, as Figure 3 shown, where P 0 is the rated peak power before dust cleaning, Tc represents the cleaning time interval, and t 1 represents the specific time of a certain cleaning. It can be obtained from Figure 3 that after cleaning, the photovoltaic system can generate an additional power generation of ΔQ.

[0114] Based on this Figure 3 , the change in the power station revenue caused by the cleaning operation is quantitatively calculated through a formula. First, give the power generation formula of the photovoltaic system:

[0115] Q(t) = P(t) × H(t) (12)

[0116] where: H(t) represents the peak hours of the t-th day calculated by formula (11).

[0117] If the photovoltaic array is not cleaned within the Tc time period, the cumulative power generation can be expressed as:

[0118]

[0119] If the photovoltaic array is cleaned at time t 1 within the Tc time period, the cumulative power generation at this time becomes:

[0120]

[0121] Let r represent the grid-connected electricity price of photovoltaic power generation, then the revenue E can be obtained as:

[0122]

[0123] where: E clean and E dust respectively represent the power generation revenues with and without cleaning the photovoltaic array, and C is the cleaning cost. At this time, it can be judged whether the obtained cleaning time is reasonable by whether ΔE is greater than 0 and the optimal cleaning time interval can be sought.

[0124] To predict the optimal cleaning time of a photovoltaic system to improve its power generation efficiency and economy, this embodiment provides a data-driven identification method based on the stag hunt optimization algorithm. This method takes into account the effects of different climate, environment, and terrain factors on the dust accumulation density and output power of the photovoltaic array, and collects relevant data. Then, based on the rapid development of big data technology and swarm intelligence optimization algorithms, a data-driven model identification method integrating the stag hunt optimization algorithm is proposed and applied to the modeling of the dust accumulation attenuation model and peak hour prediction model of the photovoltaic system. The optimal cleaning time interval can be indirectly obtained through the established dust accumulation attenuation model and peak hour prediction model of the photovoltaic system, providing a reference for the dynamic optimization of the cleaning time. Finally, by comparing the changes in the power generation benefits of the photovoltaic system before and after cleaning through benefit quantification, the rationality of the obtained cleaning time plan can be effectively evaluated.

[0125] As Figure 4 shown, on the other hand, the present invention provides a data-driven identification device 200 for optimizing the cleaning time of a photovoltaic array, including: a collection module 210, an analysis module 220, an identification module 230, and an evaluation module 240; wherein, the collection module 210 is used to collect relevant data on the dust accumulation density and output power of the photovoltaic array under preset conditions; the analysis module 220 is used to perform principle analysis and process description on the model identification of the stag hunt optimization algorithm based on the relevant data; the identification module 230 is used to identify the dust accumulation attenuation model and peak hour prediction model of the photovoltaic system respectively based on the model parameter identification method of the stag hunt optimization algorithm; the evaluation module 240 is used to perform rationality evaluation and quantitative analysis on the predicted cleaning time of the dust accumulation of the photovoltaic array.

[0126] It should be noted that the optimization process of the cleaning time of the photovoltaic array by the device in this embodiment is based on the method described above, and will not be elaborated here.

[0127] The data-driven identification device for optimizing the cleaning time of the photovoltaic array provided in this embodiment is used to predict the optimal cleaning time of the photovoltaic system to improve its power generation efficiency and economy. The device constructs a data-driven model identification method through the stag hunt optimization algorithm and applies it to the modeling of the dust accumulation attenuation model and peak hour prediction model of the photovoltaic system. At the same time, cost constraints are considered in this process, and the feasibility and rationality of the cleaning strategy are evaluated with economic benefits as the performance index. It provides an effective reference for obtaining the optimal cleaning time of the photovoltaic array, and plays a great role in promoting the economic, efficient, safe and stable operation of the photovoltaic power generation system.

[0128] The following will specifically illustrate the data-driven identification method for optimizing the cleaning time of the photovoltaic array with specific embodiments:

[0129] Please refer to Figure 1 ,Figure 1 This is the schematic diagram of the data-driven identification method for optimizing the cleaning time of a photovoltaic array provided in this embodiment. This embodiment is completed relying on the Matlab software platform and specifically consists of the following 4 steps:

[0130] S1: Collect relevant data on the dust accumulation density and output power of the photovoltaic array under different climate, environment, and terrain factors;

[0131] S2: Analyze the model identification principle and process description based on the deer hunting optimization algorithm;

[0132] S3: Identify the dust accumulation attenuation model and peak hour prediction model of the photovoltaic system;

[0133] S4: Evaluate the rationality and conduct quantitative analysis of the predicted cleaning time of the dust accumulation.

[0134] Different climate, environment, and terrain factors will cause significant differences in air humidity, oxygen content, and dust density, thereby affecting the dust accumulation situation of the photovoltaic array. For example, in seasons with more rain, the dust on the surface of the photovoltaic array can be effectively removed through the scouring of rain to achieve the cleaning effect, and the power generation efficiency of the photovoltaic system will be effectively improved at this time. Based on this, step S1 can be specifically divided into:

[0135] S1.1: Collect relevant data on the dust accumulation density and output power of the photovoltaic system based on climate, environment, and terrain factors.

[0136] S1.2: To ensure the comprehensiveness of the sampled data, select multiple combinations of climate, environment, and terrain conditions for data collection. If there are variables in the dust accumulation density and output power of the photovoltaic system that cannot be directly measured, then sample the relevant data for calculation to obtain the indirectly measured target. Let the sampling step be T = 3 min, the number of combinations of climate, environment, and terrain conditions be n = 8, and the number of data vectors sampled under each combination be N = 8000.

[0137] Based on the relevant data of the dust accumulation density and output power of the photovoltaic system obtained in step S1, in step S2, a specific description of the model identification based on the deer hunting optimization (DHO) algorithm is made.

[0138] S2.1: Set a preset form for the model to be identified, such as the transfer function or subspace model form, and set the parameter vector to be identified as θ, and its optimal value as θ * 。

[0139] S2.2: The hunting method of humans for deer provides inspiration for the development of the DHO algorithm. The key objective of the DHO algorithm is to determine the optimal or effective position for hunting deer. Certain specific characteristics of stags (i.e., deer) can protect themselves from predators or hunters, such as excellent vision, extraordinary sense of smell, and vigilance against ultra-high-frequency sounds. Based on the above analysis, the optimal position for hunting deer is regarded as the optimal model parameter vector θ * , and the model identification based on the DHO algorithm is mainly divided into four stages.

[0140] S2.2.1: Population initialization. Take s hunters as a population H, and randomly initialize the hunter population as H = {H 1 , H 2 , …, H s}:

[0141] S2.2.2: Initialize wind and position angle. The initialization of position and wind direction angle is considered a key process for hunters to determine the best ideal position for hunting stags. The wind direction angle φ and the position angle are respectively given by the following formulas:

[0142] φ k = 2πl (1)

[0143]

[0144] where: l is a random number in the range of [0, 1], and k represents the current iteration step.

[0145] S2.2.3: Position propagation. In the initial stage, the position of the optimal or ideal space is uncertain. Therefore, the position closest to the optimal solution calculated according to the fitness function is regarded as the optimal space. Usually, two position update schemes are considered, namely the successor position and the leader position.

[0146] Propagation process based on the leader position: Once the best position is determined, each hunter in the population tries to reach the best position, and at this time, the position update process is triggered. The hunter surrounding behavior can be expressed by the following formula:

[0147] H k+1 = H lead - M·c|R × H lead - H k | (3)

[0148] where: represents the leader position, c is a random number considering the wind direction angle, and its value range is (0, 2], and both M and R refer to the parameter vectors, and the calculation formulas are as follows:

[0149]

[0150] R = 2rd (5)

[0151] where: rd is a random number taken from [0, 1], k max = 50 is the maximum number of iterations, and the parameter a is used to measure the coefficient vector in the equation, and here it is taken as 0.1.

[0152] Propagation process based on the position angle: In this case, the search space is improved by considering the position angle in the process of updating the hunter's position during the deer hunting process. It is assumed that the hunting process is effective at a certain position angle. In addition, a new parameter ds k is introduced to update the position angle and this parameter is developed based on the variance between the wind angles, and its visible angle is expressed in formula (6). At the same time, the visible angle vs of the prey k is defined by equation (7).

[0153] ds k = φ k - vs k (6)

[0154]

[0155] The position angle is updated using equation (8), and the hunter's position is updated based on the position angle shown in equation (9):

[0156]

[0157] H k+1 = H lead - c|cos(w) × H lead - H k | (9)

[0158] Propagation process based on the successor position: The surrounding mechanism is used to change the vector R in the exploration stage. However, in the initial state, random search is first considered, and then it is assumed that the value of the vector R is less than 1. The update process of the hunter's position depends on the position of the successor, as shown in equation (10). If the value of the vector R is less than 1, a search agent is randomly selected, otherwise the optimal solution is selected to update the position of the search agent.

[0159] H k+1 = H successor - M·c|R × H successor - H k | (10)

[0160] S2.2.4: Termination process. When the optimal hunting position is determined, the hunter stops updating the position, and the above optimization process ends.

[0161] S2.3: For the data-driven modeling of the present invention, the optimal hunting position H obtained at this timebest is the optimal solution of the model parameter vector, i.e., θ * = H best .

[0162] According to the DHO-based model parameter identification method designed in step S2, the identification of the photovoltaic system dust accumulation attenuation model and the peak hour number prediction model is carried out respectively in step S3.

[0163] S3.1: Identification of the photovoltaic system dust accumulation attenuation model. Select the time when the photovoltaic module has just been cleaned as the sampling starting point under different seasons and weather conditions, and take the change of the rated peak power of the photovoltaic module over time after cleaning as a reference, so as to establish a model that can accurately describe the power attenuation trend of the photovoltaic module with the dust accumulation duration, and provide a reference for reasonably planning the cleaning time of the photovoltaic module.

[0164] The dust accumulation attenuation characteristics of the photovoltaic system are related to solar irradiance, photovoltaic module temperature and output power. Therefore, in the modeling process, the corresponding variables should be selected from the collected data first, then the form of the attenuation model is preset, and then the unknown parameters of the model are optimized and identified through the above DHO-based modeling method.

[0165] S3.2: Identification of the peak hour number prediction model. The peak hour number H refers to the number of hours when the total solar irradiance within a certain period of time is converted into the standard test conditions, that is, the irradiance is 1000 W / m 2 , and the temperature is 25 °C. Generally, it can be calculated by the following formula:

[0166]

[0167] Where: t 1 and t 2 are the starting time point and the ending time point of the calculated time period respectively, and R(t) is the irradiance of the photovoltaic array at time t.

[0168] As can be seen from the above formula, the peak hour number is directly related to solar irradiance. Therefore, the influencing factors of solar irradiance change need to be considered. Through the weather change situation within the time period from t 1 to t 2 , it can be seen that the change of solar irradiance is closely related to the weather clarity, cloud thickness, oxygen content and humidity of the air, etc.

[0169] Therefore, based on five weather types of clear, cloudy, overcast, light rain and heavy rain to rainstorm, select data such as ambient temperature, air humidity, cloud thickness and air density from the sampling data as the modeling input, and solar irradiance as the output, and identify the unknown parameters of the model through the DHO algorithm based on the preset model form. Then, the peak hour number is calculated using formula (11).

[0170] S3.3: According to the established dust accumulation attenuation model and peak hour prediction model of the photovoltaic system, combined with the economy of the photovoltaic system, obtain the optimal interval of the cleaning time of the photovoltaic array.

[0171] Next, in step S4, quantitatively analyze the cleaning benefits of the photovoltaic array to realize the dynamic evaluation of the rationality of the obtained dust cleaning time.

[0172] First, draw the following rated peak power change curves of the photovoltaic system before and after dust accumulation and cleaning, as Figure 3 shown, P 0 is the rated peak power before dust accumulation cleaning, Tc represents the cleaning time interval, and t 1 represents the specific time of a certain cleaning. It can be seen from the figure that after cleaning, the photovoltaic system can generate an additional power generation of ΔQ.

[0173] For this Figure 3 , quantitatively calculate the change in the power station revenue caused by the cleaning operation through the formula. First, give the power generation formula of the photovoltaic system:

[0174] Q(t) = P(t) × H(t) (12)

[0175] Where: H(t) represents the peak hours of the t-th day calculated by formula (11).

[0176] If the photovoltaic array is not cleaned within the Tc time period, the cumulative power generation can be expressed as:

[0177]

[0178] If the photovoltaic array is cleaned at time t 1 within the Tc time period, the cumulative power generation at this time becomes:

[0179]

[0180] Let r represent the grid-connected electricity price of photovoltaic power generation, then the revenue E can be obtained as:

[0181] E = r × Q (15)

[0182] Then the revenue increment ΔE brought by the cleaning operation to photovoltaic power generation is:

[0183]

[0184] Where: E clean and E dust respectively represent the power generation revenues with and without cleaning the photovoltaic array, and C is the cleaning cost. At this time, it can be judged whether the obtained cleaning time is reasonable by whether ΔE is greater than 0 and seek the optimal cleaning time interval.

[0185] The present invention provides a data-driven identification method and device for optimizing the cleaning time of a photovoltaic array, which has the following beneficial effects compared with the prior art:

[0186] First, the present invention combines the stag hunt optimization algorithm to provide a data-driven identification method for optimizing the cleaning time of a photovoltaic array to predict the optimal cleaning time of the photovoltaic system and improve its power generation efficiency and economy.

[0187] Second, considering the rapid development of big data technology and swarm intelligence optimization algorithms, the present invention proposes a data-driven model identification method based on the stag hunt optimization algorithm to make the modeling process more intelligent and flexible, and applies it to the modeling of the dust accumulation attenuation model and peak hour prediction model of the photovoltaic system.

[0188] Third, based on the established dust accumulation attenuation model and peak hour prediction model of the photovoltaic system, the present invention compares the changes in the power generation income of the photovoltaic system before and after the cleaning operation through income quantification, thereby verifying the rationality of the obtained cleaning time and providing an effective reference for obtaining the optimal cleaning time of the photovoltaic array.

[0189] It can be understood that the above embodiments are merely exemplary embodiments adopted to illustrate the principle of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also regarded as the protection scope of the present invention.

Claims

1. A data-driven identification method for optimizing the cleaning time of a photovoltaic array, characterized in that, it includes the following specific steps: Collect relevant data on the dust accumulation density and output power of the photovoltaic array under preset conditions, including: collecting relevant data on the dust accumulation density and output power of the photovoltaic system based on climate, environment, and terrain factors; selecting multiple combinations of climate, environment, and terrain conditions for data collection; Based on the relevant data, conduct a principle analysis and process description of the model identification of the stag hunting optimization algorithm, including: Set a preset form for the model to be identified; Regard the best position for deer hunting as the optimal model parameter vector θ * , and optimize the model identification of the deer hunting optimization algorithm; Identify the optimal hunting position H according to the deer hunting optimization algorithm model best , and θ * =H best ; Based on the model parameter identification method of the stag hunting optimization algorithm, identify the dust accumulation attenuation model and peak hour prediction model of the photovoltaic system respectively, including: selecting the time when the photovoltaic module has just been cleaned as the sampling starting point in different seasons and weather conditions, and using the change of the rated peak power of the photovoltaic module over time after cleaning as a reference to establish a power attenuation trend model describing the photovoltaic module with the dust accumulation duration; Select environmental temperature, air humidity, cloud thickness, and air density data from the sampling data as the modeling input, and solar irradiance as the output. Identify the unknown parameters of the model through the stag hunting optimization algorithm based on the preset model form, and calculate the peak hours; Conduct a rationality evaluation and quantitative analysis of the predicted cleaning time of the photovoltaic array, including: drawing the following curve of the rated peak power change of the photovoltaic system before and after dust accumulation and cleaning, and obtaining the power generation increment ΔQ generated by the photovoltaic system after cleaning; Quantitatively calculate the increment of power station revenue ΔE brought by the cleaning operation to photovoltaic power generation through the following formula: ΔE = E clean - E dust - C E clean = r × Q clean E dust = r × Q dust Among them, E clean and E dust respectively represent the power generation income from whether to clean the photovoltaic array, C is the cleaning cost, and Q clean represents the cumulative power generation when the photovoltaic array is cleaned at time t 1 during the Tc time period, Q dust represents the cumulative power generation of the photovoltaic array without cleaning during the Tc time period, and r represents the grid-connected electricity price of photovoltaic power generation; And, determine the rationality of the cleaning time and seek the optimal cleaning time interval by judging whether ΔE is greater than 0.

2. The method according to claim 1, characterized in that, Regarding the best location for deer hunting as the optimal model parameter vector θ * , and optimizing the model identification of the deer hunting optimization algorithm, including: Take s hunters as a population H, and randomly initialize the hunter population as H = {H 1 , H 2 , …, H s}; Determine the initial wind direction angle and position angle for hunting stags; Calculate the position closest to the optimal solution according to the fitness function; When the optimal hunting position is determined, the hunter stops updating the position.

3. The method according to claim 2, characterized in that, The initial wind direction angle and position angle for hunting stags are obtained by using the following formula: φ k = 2πl where φ represents the wind direction angle, represents the position angle; l is a random number within the range of [0, 1], and k represents the current iteration step.

4. The method according to claim 3, characterized in that, The calculation of the position closest to the optimal solution according to the fitness function includes: Based on the propagation process of the leader's position: when the best position is determined, each hunter in the population tries to reach the best position, and at this time, the position update process is triggered. Among them, the hunter's encirclement behavior can be expressed by the following formula: H k+1 = H lead -M·c|R×H lead -H k | Where: H lead represents the leader position, c is a random number considering the wind direction angle, and its value range is (0, 2], both M and R refer to parameter vectors; Position-angle-based propagation process: By considering the position angle in the hunter's position update process to improve the search space. During the deer hunting process, assume that the hunting process is effective at a certain position angle, and introduce a new parameter ds k to update the position angle, and this parameter is developed based on the variance between the wind angles. Its visual angle is as follows: ds k = φ k -vs k And, the visual angle vs of the prey k As follows: where rad is a random number taken from [0,1]; And, update the position angle using the following relational expression: Update the position of the hunter using the following relational expression: H k+1 = H lead - c|cos(w)×H lead - H k |; Where: H k+1 is the hunter position vector at time k, H k is the updated hunter position at time k + 1, and w represents the position angle; Based on the propagation process of the successor's position: assuming that the value of vector R is less than 1, the position update process of the hunter depends on the position of the successor, as shown in the following relational expression: H k+1 = H successor -M·c|R×H successor -H k | Among them, H successor represents the position of the successor.

5. The method according to claim 1, characterized in that, The peak hours are calculated using the following relational expression: Among them, the peak hour number H refers to the number of hours when the total solar irradiance within a certain period is converted under standard test conditions, that is, the irradiance is 1000W / m 2 , and the temperature is 25°C; t 1 and t 2 are respectively the starting time point and the ending time point of the calculated time period, and R(t) is the irradiance of the photovoltaic array at time t.

6. A data-driven identification device for optimizing the cleaning time of a photovoltaic array, characterized in that, it includes: A collection module, an analysis module, an identification module, and an evaluation module; among them, The acquisition module is used to acquire relevant data on the dust accumulation density and output power of a photovoltaic array under preset conditions, including: acquiring relevant data on the dust accumulation density and output power of a photovoltaic system based on climate, environment, and terrain factors, and selecting multiple combinations of climate, environment, and terrain conditions for data acquisition; The analysis module is used to conduct a principle analysis and process description of the model identification of the stag-hunting optimization algorithm based on the relevant data, including: Setting a preset form for the model to be identified; Regard the best position for hunting deer as the optimal model parameter vector θ * , and optimize the model identification of the deer hunting optimization algorithm; Identify the optimal hunting position H according to the deer hunting optimization algorithm model best , and θ * = H best ; The identification module is used to identify the dust accumulation attenuation model and the peak hour prediction model of the photovoltaic system respectively based on the model parameter identification method of the stag-hunting optimization algorithm, including: selecting the time when the photovoltaic module has just been cleaned as the sampling starting point in different seasons and weather conditions, and establishing a power attenuation trend model describing the change of the rated peak power of the photovoltaic module over time with the dust accumulation duration with reference to the change of the rated peak power of the photovoltaic module after cleaning; Selecting environmental temperature, air humidity, cloud thickness, and air density data from the sampling data as the modeling input, with solar irradiance as the output, identifying the unknown parameters of the model through the stag-hunting optimization algorithm based on the preset model form, and calculating the peak hours; The evaluation module is used to conduct a rationality evaluation and quantitative analysis of the predicted dust cleaning time of the photovoltaic array, including: plotting the change curve of the rated peak power of the following photovoltaic system before and after dust accumulation and cleaning, and obtaining the power generation increment ΔQ generated by the photovoltaic system after cleaning; Quantitatively calculating the increment of power station revenue ΔE brought by the cleaning operation to photovoltaic power generation by the formula as follows: ΔE = E clean - E dust - C E clean = r × Q clean E dust = r × Q dust Among them, E clean and E dust respectively represent the power generation income of whether to clean the photovoltaic array, C is the cleaning cost, and Q clean represents the cumulative power generation when the photovoltaic array is cleaned at time t 1 during the Tc time period, and Q dust represents the cumulative power generation of the photovoltaic array without cleaning during the Tc time period, and r represents the grid-connected electricity price of photovoltaic power generation; And, by judging whether ΔE is greater than 0 to obtain the rationality of the cleaning time and seek the optimal cleaning time interval.

Images