Photovoltaic monitoring system and method based on Internet of Things

Abstract

The invention discloses a photovoltaic monitoring system and method based on the Internet of Things, and relates to the technical field of photovoltaic monitoring, and the system comprises a data analysis unit, a parameter primary screening unit, a parameter repeated screening unit, and a photovoltaic module cleaning unit. According to the invention, the correlation value between the multi-source data and the output power of the photovoltaic module in the dust-deposition-free state is analyzed through the data analysis unit, so that various variables most related to the output power of the photovoltaic module can be effectively screened out subsequently, and the space occupied by the data is saved; the parameter initial screening unit and the parameter repeated screening unit can identify parameters which are good in performance on different data subsets instead of optimal parameters on a single data set through multiple rounds of verification, the performance of the model is effectively improved, meanwhile, all sample data are fully utilized in the mode, and the efficiency is improved. And the photovoltaic module cleaning unit sends prompt information to the user regularly, so that the condition of excessive dust accumulation on the photovoltaic module is avoided, and the electric quantity loss of the photovoltaic module is further reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic monitoring, and particularly to a photovoltaic monitoring system and method based on the Internet of Things. Background Art

[0002] The photovoltaic monitoring system based on the Internet of Things collects data such as the power generation, component temperature, and irradiance of a power station in real time through sensors, and transmits them to a cloud platform via a wireless network. It can help users optimize power generation efficiency, reduce operation and maintenance costs, and improve the reliability and economic benefits of photovoltaic power stations. The patent with the application number 201810339116.6 discloses "a photovoltaic array monitoring system based on the Internet of Things and its working method. This photovoltaic array monitoring system includes: measuring and controlling devices distributed in each photovoltaic array, and each measuring and controlling device is connected to an Internet of Things node to upload the collected photovoltaic array data to a base station; the present invention is used to assist in the formation of a network, and according to the environmental information of the network and the distribution characteristics of the nodes, a networking strategy is constructed. Moreover, during the networking process, the base station mainly communicates with the nodes at the first level, reducing the pressure on the base station; moreover, the networking strategy is mainly sent to specific nodes rather than each node, reducing the amount of data transmission in the network and improving the networking efficiency, thereby meeting the operation and maintenance of large-scale photovoltaic power stations."

[0003] The above-mentioned prior art solves problems such as the ineffective transmission of real-time data. However, during the operation of the system, since the correlation value between multi-source data and the output power of photovoltaic components without dust accumulation cannot be calculated, the variable screening process lacks data support, which may affect the accuracy and efficiency of the prediction model. Moreover, the multi-round verification mechanism is not adopted in the model-related parameter screening link, and only a single data set is relied on to evaluate the accuracy, resulting in the risk of model overfitting. At the same time, the system does not have a cleaning reminder function, making it impossible for maintenance personnel to clean in time, leading to a continuous increase in power generation loss. Summary of the Invention

[0004] The purpose of the present invention is to provide a photovoltaic monitoring system and method based on the Internet of Things to solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A photovoltaic monitoring system based on the Internet of Things includes a photovoltaic component cleaning unit.

[0006] A data analysis unit. The data analysis unit collects solar radiation, temperature, humidity, wind speed, wind direction, air pressure, and the output power of photovoltaic components without dust accumulation through multiple sensors, stores them in a database, counts the offset values corresponding to variables such as solar radiation, temperature, humidity, wind speed, wind direction, air pressure, and the output power of photovoltaic components without dust accumulation, and analyzes the correlation values between variables according to the offset values.

[0007] The initial parameter screening unit uses solar irradiance, temperature, and humidity as optimal variables, constructs corresponding sample data, determines the initial parameters of the power prediction model and sets the search range. It divides all sample data into five parts, rotates to select the test set, and the rest are the training set. After each training, it evaluates the model accuracy, and the average accuracy of five rounds of evaluation is used as the actual accuracy of the current parameters;

[0008] The parameter repeated screening unit counts the parameter values corresponding to the highest actual accuracy, screens the optimal range according to the parameter values and calculates the new search interval, generates parameters according to the optimal range and the new search interval, deletes the original parameters, calculates the actual accuracy of the current parameters, and continues to iterate if the accuracy is lower than the threshold, otherwise outputs the best parameters and completes the training of the power prediction model.

[0009] Preferably, the data analysis unit includes a data acquisition module and a power extraction module. The data acquisition module uses a pyranometer to collect solar irradiance at different time periods, collects temperature data of the photovoltaic module at different time periods through a temperature sensor, uses a humidity sensor to collect humidity data of the photovoltaic module at different time periods, uses an anemometer to collect wind speed data and wind direction around the photovoltaic module at different time periods, and collects atmospheric pressure around the photovoltaic module at different time periods through a barometer. It transmits solar irradiance, temperature data, humidity data, wind speed data, wind direction, atmospheric pressure, and the corresponding recording time to the database for storage. The power extraction module uses a smart meter to collect the output power of the photovoltaic module without dust accumulation at different time periods, and transmits the photovoltaic output power and the corresponding recording time to the database for storage.

[0010] Preferably, the data analysis unit further includes an offset value calculation module and a correlation value determination module. The offset value calculation module counts the corresponding solar irradiance F1, F2,..., F k ,..., F n , temperature data T1, T2,..., T k ,…, T n , humidity data H1, H2,…, H k ,…, H n , wind speed data V1, V2,..., V k ,..., V n , wind direction data W1, W2,..., W k ,..., W n , atmospheric pressure P1, P2,..., P k ,..., P n at different recording times t1, t2,..., t k ,..., P nand photovoltaic output powers OP1, OP2, …, OP k , …, OP n After that, according to F1, F2, …, F k , …, F n calculate the k-th solar radiation offset value F′ k , where Utilize T1, T2, …, T k , …, T n calculate the k-th temperature offset value T′ k , where Utilize H1, H2, …, H k , …, H n calculate the k-th humidity offset value H′ k , where Through V1, V2, …, V k , …, V n calculate the k-th wind speed offset value V′ k , where According to W1, W2, …, W k , …, W n calculate the k-th wind direction offset value W′ k , where Utilize P1, P2, …, P k , …, P n calculate the k-th atmospheric pressure offset value P′ k , where Through OP1, OP2, …, OP k , …, OP n calculate the k-th photovoltaic output power offset value OP′ k , where k represents a parameter, n represents the amount of data. The correlation value determination module repeats the operation until all solar radiation offset values, temperature offset values, humidity offset values, wind speed offset values, wind direction offset values, atmospheric pressure offset values, and photovoltaic output power offset values are calculated, and then analyzes the solar radiation offset values, temperature offset values, humidity offset values, wind speed offset values, wind direction offset values, atmospheric pressure offset values, and photovoltaic output power offset values to obtain the correlation values among solar radiation, temperature, humidity, wind speed, wind direction, atmospheric pressure, and photovoltaic output power.

[0011] Preferably, the parameter primary screening unit includes a sample generation module and a parameter setting module. The sample generation module selects the three optimal variables most relevant to the photovoltaic output power according to the magnitude of the correlation value. The three optimal variables are solar irradiance, temperature, and humidity respectively. After extracting the solar irradiance, temperature data, humidity data, and photovoltaic output power in the database at different time periods, multiple sample data are constructed based on the recording time, solar irradiance, temperature data, humidity data, and photovoltaic output power, and stored in the original dataset. After the parameter setting module constructs a power prediction model using SVM technology, it sets the initial values, ranges, and maximum search intervals of the adjustment coefficient and kernel function parameters in the power prediction model, and obtains multiple adjustment coefficients and kernel function parameters by taking values of the adjustment coefficient and kernel function parameters according to the maximum search interval.

[0012] Preferably, the parameter primary screening unit further includes a sample division module, an accuracy analysis module, and an average value calculation module. The sample division module divides the sample data in the original dataset into five subsets. After randomly selecting an unlabeled subset as the test set and labeling it, the remaining subsets are used as the training set. After the accuracy analysis module selects the adjustment coefficient and kernel function parameters, it transmits the sample data in the training set to the current power prediction model for analysis. After completing the model training, it uses the sample data in the test set to determine the accuracy of the current model. The average value calculation module repeats the operation until each subset is labeled, and then calculates the corresponding average value using the five accuracies, and takes this average value as the actual accuracy corresponding to the current adjustment coefficient and kernel function parameters. The power prediction model is specifically:

[0013]

[0014] Among them, represents the vector output function, λ1 represents the first adjustment coefficient, λ2 represents the second adjustment coefficient, λ m represents the mth adjustment coefficient, represents the first input column vector, represents the second input column vector, represents the mth input column vector, represents the input vector parameter, represents the first kernel function, represents the second kernel function, represents the mth kernel function, and c represents the participation coefficient.

[0015] Preferably, the parameter repeated screening unit includes a search interval adjustment module, a parameter output module, and a model training module. After the search interval adjustment module calculates the actual accuracy of all current adjustment parameters and kernel function parameters, it selects the adjustment coefficient and kernel function parameter with the highest actual accuracy, filters out the optimal range according to the values of the adjustment coefficient and kernel function parameter, calculates a new search interval according to the size of the optimal range, re-analyzes multiple adjustment coefficients and kernel function parameters using the optimal range and the search interval, and deletes the original adjustment coefficients and kernel function parameters. The parameter output module calculates the actual accuracy corresponding to all adjustment coefficients and kernel function parameters. If the actual accuracy is lower than the threshold, it continues to filter out the optimal range and generate new adjustment coefficients and kernel function parameters. If the actual accuracy is higher than or equal to the threshold, it stops filtering out the optimal range and outputs the adjustment coefficients and kernel function parameters corresponding to the current maximum actual accuracy. The model training module transmits the adjustment coefficients and kernel function parameters to the power prediction model, randomly extracts multiple sample data from the original data set and stores them in the random test set, trains the current power prediction model using the remaining sample data in the original data set, and calculates the accuracy of the current power prediction model through the sample data in the random test set.

[0016] Preferably, the photovoltaic module cleaning unit includes an output power prediction module, a power loss analysis module, a loss value calculation module, and a prompt message sending module. After the output power prediction module obtains the solar irradiance, temperature data, and humidity data in the current time period, it transmits them to the power prediction model for analysis to obtain the output power of the photovoltaic module in a dust-free state. The power loss analysis module uses the inverter to collect the actual output power of the current photovoltaic module, and calculates the daily power loss value through the output power of the photovoltaic module in a dust-free state and the actual output power. The loss value calculation module obtains the daily electricity price and the most recent cleaning time, and analyzes the electricity price and the daily power loss value using the power loss determination algorithm to obtain the actual loss value of the current photovoltaic module. If the actual loss value is lower than the preset value, the prompt message sending module does not perform any operation. Otherwise, it retrieves the meteorological data within the next 48 hours. If the rainfall in the meteorological data is not zero, it does not perform any operation. If the rainfall in the meteorological data is zero, it sends a prompt message to clean the photovoltaic module.

[0017] The photovoltaic monitoring method based on the Internet of Things includes the following steps:

[0018] S1. Sensor data collection: Collect solar irradiance, temperature, humidity, wind speed, wind direction, air pressure, and the output power of the photovoltaic module in a dust-free state through multiple sensors, store them in the database, calculate the offset values corresponding to these variables of solar irradiance, temperature, humidity, wind speed, wind direction, air pressure, and the output power of the photovoltaic module in a dust-free state, and analyze the correlation values between the variables according to the offset values;

[0019] S2. Divide the sample data: Take solar irradiance, temperature, and humidity as the optimal variables, construct the corresponding sample data, determine the initial parameters of the power prediction model and set the search range. Divide all the sample data into five parts, rotate to select the test set, and the rest are the training set. Evaluate the model accuracy after each training, and take the average accuracy of the five rounds of evaluation as the actual accuracy of the current parameters;

[0020] S3. Screen the best parameters: Statistically analyze the parameter values corresponding to the highest actual accuracy, screen the optimal range according to the parameter values and calculate the new search interval. Generate parameters based on the optimal range and the new search interval, and delete the original parameters. Calculate the actual accuracy of the current parameters. If the accuracy is lower than the threshold, continue the iteration; otherwise, output the best parameters and complete the training of the power prediction model;

[0021] S4. Output a cleaning reminder: Obtain the real-time solar irradiance and temperature and humidity data, analyze the output power under the dust-free state of the photovoltaic module through the power prediction model, and compare it with the actual power collected by the inverter, so as to calculate the daily power loss value. Combine the electricity price and the cleaning time to analyze the actual loss value of the photovoltaic module. If the loss value exceeds the preset value and the meteorological data shows no rain in the next 48 hours, send a cleaning reminder.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] The present invention analyzes the correlation value between multi-source data and the output power of the photovoltaic module under the dust-free state through the data analysis unit, which is convenient for effectively screening out various variables most relevant to the output power of the photovoltaic module subsequently. This method not only improves the accuracy of the prediction model, but also saves the space occupied by the data. Moreover, the initial parameter screening unit and the repeated parameter screening unit can identify the parameters that perform well on different data subsets through multiple rounds of verification, rather than the optimal parameters on a single data set, effectively improving the performance of the model. At the same time, this method makes full use of all sample data. The photovoltaic module cleaning unit regularly sends reminder information to users to avoid excessive dust accumulation on the photovoltaic module, further reducing the power loss of the photovoltaic module. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the overall system flow provided by the embodiment of the present invention;

[0025] Figure 2 It is an internal module block diagram of the data analysis unit provided by the embodiment of the present invention;

[0026] Figure 3 It is an internal module block diagram of the initial parameter screening unit provided by the embodiment of the present invention;

[0027] Figure 4 It is the internal module block diagram of the parameter repeated screening unit provided by the embodiment of the present invention;

[0028] Figure 5 It is the internal module block diagram of the photovoltaic module cleaning unit provided by the embodiment of the present invention.

[0029] In the figure: 1. Data analysis unit; 101. Data acquisition module; 102. Power extraction module; 103. Offset value calculation module; 104. Correlation value determination module; 2. Parameter primary screening unit; 201. Sample generation module; 202. Parameter setting module; 203. Sample division module; 204. Accuracy analysis module; 205. Average value calculation module; 3. Parameter repeated screening unit; 301. Search interval adjustment module; 302. Parameter output module; 303. Model training module; 4. Photovoltaic module cleaning unit; 401. Output power prediction module; 402. Power loss analysis module; 403. Loss value calculation module; 404. Prompt information sending module. Specific implementation manners

[0030] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0031] Please refer to Figures 1-5 , the present invention provides a technical solution: an Internet of Things-based photovoltaic monitoring system, including a photovoltaic module cleaning unit 4;

[0032] A data analysis unit 1, the data analysis unit 1 collects solar irradiance, temperature, humidity, wind speed, wind direction, air pressure and the output power of the photovoltaic module under the condition of no dust accumulation through multiple sensors, stores them in the database, statistically analyzes the offset values corresponding to these variables of solar irradiance, temperature, humidity, wind speed, wind direction, air pressure and the output power of the photovoltaic module under the condition of no dust accumulation, and analyzes the correlation values between the variables according to the offset values;

[0033] A parameter primary screening unit 2, the parameter primary screening unit 2 takes solar irradiance, temperature and humidity as the optimal variables, constructs corresponding sample data, determines the initialization parameters of the power prediction model and sets the search range, divides all sample data into five parts, rotates to select the test set, and the rest are the training set. After each training, the accuracy of the model is evaluated, and the average value of the accuracies evaluated in five rounds is used as the actual accuracy of the current parameters;

[0034] The parameter repeated screening unit 3 counts the parameter values corresponding to the highest actual accuracy, screens the optimal range according to the parameter values and calculates a new search interval, generates parameters based on the optimal range and the new search interval, deletes the original parameters, calculates the actual accuracy of the current parameters, and continues to iterate if the accuracy is lower than the threshold, otherwise outputs the optimal parameters and completes the training of the power prediction model.

[0035] The data analysis unit 1 includes a data acquisition module 101 and a power extraction module 102. The data acquisition module 101 uses a pyranometer to collect the solar irradiance at different time periods, collects the temperature data of the photovoltaic modules at different time periods through a temperature sensor, uses a humidity sensor to collect the humidity data of the photovoltaic modules at different time periods, uses an anemometer to collect the wind speed data and wind direction around the photovoltaic modules at different time periods, and collects the atmospheric pressure around the photovoltaic modules at different time periods through a barometer. It transmits the solar irradiance, temperature data, humidity data, wind speed data, wind direction, atmospheric pressure, and the corresponding recording time to the database for storage. The power extraction module 102 uses a smart meter to collect the output power of the photovoltaic modules in a non-dusty state at different time periods, and transmits the photovoltaic output power and the corresponding recording time to the database for storage;

[0036] The data analysis unit 1 further includes an offset value calculation module 103 and a correlation value determination module 104. The offset value calculation module 103 counts the solar irradiance F1, F2, …, F k , …, F n corresponding to different recording times t1, t2, …, t k , …, F n , the temperature data T1, T2, …, T k , …, T n , the humidity data H1, H2, …, H k , …, H n , the wind speed data V1, V2, …, V k , …, V n , the wind direction data W1, W2, …, W k , …, W n , the atmospheric pressure P1, P2, …, P k , …, P n , and the photovoltaic output power OP1, OP2, …, OP k , …, OP n . After that, according to F1, F2, …, F k , …, F n , it calculates the k-th solar radiation offset value F′ k , where Using T1, T2, …, T k , …, Tn Calculate the k-th temperature offset value T′ k , where Use H1, H2,..., H k ,..., H n Calculate the k-th humidity offset value H′ k , where Through V1, V2,..., V k ,..., V n Calculate the k-th wind speed offset value V′ k , where According to W1, W2,…, W k ,..., W n Calculate the k-th wind direction offset value W′ k , where Use P1, P2,..., P k ,..., P n Calculate the k-th atmospheric pressure offset value P′ k , where Through OP1, OP2,…, OP k ,…, OP n Calculate the k-th photovoltaic output power offset value OP′ k , where k represents a parameter, n represents the amount of data. The correlation value determination module 104 repeats the operation until all solar radiation offset values, temperature offset values, humidity offset values, wind speed offset values, wind direction offset values, atmospheric pressure offset values, and photovoltaic output power offset values are calculated, and then analyzes the solar radiation offset values, temperature offset values, humidity offset values, wind speed offset values, wind direction offset values, atmospheric pressure offset values, and photovoltaic output power offset values to obtain the correlation values between solar radiation, temperature, humidity, wind speed, wind direction, atmospheric pressure, and photovoltaic output power;

[0037] The parameter initial screening unit 2 includes a sample generation module 201 and a parameter setting module 202. The sample generation module 201 selects the three optimal variables most relevant to the photovoltaic output power according to the magnitude of the correlation value. The three optimal variables are solar radiation, temperature, and humidity respectively. After extracting the solar radiation, temperature data, humidity data, and photovoltaic output power at different time periods from the database, multiple sample data are constructed according to the recording time, solar radiation, temperature data, humidity data, and photovoltaic output power, and stored in the original data set. The parameter setting module 202 uses the SVM technology to construct a power prediction model, and then sets the initial values, ranges, and maximum search intervals of the adjustment coefficient and kernel function parameters in the power prediction model, and obtains multiple adjustment coefficients and kernel function parameters by taking values for the adjustment coefficient and kernel function parameters according to the maximum search interval;

[0038] The parameter initial screening unit 2 further includes a sample division module 203, an accuracy analysis module 204, and an average value calculation module 205. The sample division module 203 divides the sample data in the original dataset into five subsets. After randomly selecting an unlabeled subset as the test set and labeling it, the remaining subsets are used as the training set. After the accuracy analysis module 204 selects the adjustment coefficient and kernel function parameters, it transmits the sample data in the training set to the current power prediction model for analysis. After completing the model training, it uses the sample data in the test set to determine the accuracy of the current model. The average value calculation module 205 repeats the operation until each subset is labeled, and then calculates the corresponding average value using the five accuracies. This average value is used as the actual accuracy corresponding to the current adjustment coefficient and kernel function parameters. The power prediction model is specifically:

[0039]

[0040] Among them, represents the vector output function, λ1 represents the first adjustment coefficient, λ2 represents the second adjustment coefficient, λ m represents the m-th adjustment coefficient, represents the first input column vector, represents the second input column vector, represents the m-th input column vector, represents the input vector parameter, represents the first kernel function, represents the second kernel function, represents the m-th kernel function, and c represents the participation coefficient;

[0041] The parameter repeated screening unit 3 includes a search interval adjustment module 301, a parameter output module 302, and a model training module 303. After the search interval adjustment module 301 counts the actual accuracies of all current adjustment parameters and kernel function parameters, it selects the adjustment coefficient and kernel function parameter with the highest actual accuracy, screens out the optimal range according to the values of the adjustment coefficient and kernel function parameter, calculates a new search interval according to the size of the optimal range, re-analyzes multiple adjustment coefficients and kernel function parameters using the optimal range and the search interval, and deletes the original adjustment coefficients and kernel function parameters. The parameter output module 302 counts the actual accuracies corresponding to all adjustment coefficients and kernel function parameters. If the actual accuracy is lower than the threshold, it continues to screen the optimal range and generate new adjustment coefficients and kernel function parameters. If the actual accuracy is higher than or equal to the threshold, it stops screening the optimal range and outputs the adjustment coefficients and kernel function parameters corresponding to the current maximum actual accuracy. The model training module 303 transmits the adjustment coefficients and kernel function parameters to the power prediction model, randomly extracts multiple sample data from the original data set and stores them in the random test set, and uses the remaining sample data in the original data set to train the current power prediction model, and calculates the accuracy of the current power prediction model through the sample data in the random test set;

[0042] The photovoltaic module cleaning unit 4 includes an output power prediction module 401, a power loss analysis module 402, a loss value calculation module 403, and a prompt information sending module 404. After the output power prediction module 401 obtains the solar irradiance, temperature data, and humidity data in the current time period, it transmits them to the power prediction model for analysis to obtain the output power of the photovoltaic module in a dust-free state. The power loss analysis module 402 uses the inverter to collect the actual output power of the current photovoltaic module, and calculates the daily power loss value through the output power of the photovoltaic module in a dust-free state and the actual output power. The loss value calculation module 403 obtains the daily electricity price and the most recent cleaning time, and analyzes the electricity price and the daily power loss value using the power loss determination algorithm to obtain the actual loss value of the current photovoltaic module. If the actual loss value of the prompt information sending module 404 is lower than the preset value, it does not perform any operation. Otherwise, it retrieves the meteorological data within the next 48 hours. If the rainfall in the meteorological data is not zero, it does not perform any operation. If the rainfall in the meteorological data is zero, it sends a prompt message to clean the photovoltaic module. The power loss determination algorithm is specifically:

[0043]

[0044] Among them, G represents the actual loss value, s x represents the electricity price on the x-th day, q xy represents the power loss value of the y-th photovoltaic module within the x-th day, M represents the number of days between the current time and the most recent cleaning time, N represents the number of photovoltaic modules, and x, y represent parameters;

[0045] The photovoltaic monitoring method based on the Internet of Things includes the following steps:

[0046] S1. Sensor data collection: Collect solar irradiance, temperature, humidity, wind speed, wind direction, air pressure, and the output power of the photovoltaic module under the condition of no dust accumulation through multiple sensors, store them in the database, count the offset values corresponding to these variables of solar irradiance, temperature, humidity, wind speed, wind direction, air pressure, and the output power of the photovoltaic module under the condition of no dust accumulation, and analyze the correlation values between the variables according to the offset values;

[0047] S2. Sample data division: Take solar irradiance, temperature, and humidity as the optimal variables, construct the corresponding sample data, determine the initial parameters of the power prediction model and set the search range, divide all sample data into five parts, rotate and select the test set, and the rest are the training set. Evaluate the model accuracy after each training, and take the average value of the accuracies evaluated in five rounds as the actual accuracy of the current parameters;

[0048] S3. Optimal parameter screening: Count the parameter values corresponding to the highest actual accuracy, screen the optimal range according to the parameter values and calculate the new search interval, generate parameters according to the optimal range and the new search interval, delete the original parameters, calculate the actual accuracy of the current parameters, and continue to iterate if the accuracy is lower than the threshold, otherwise output the optimal parameters and complete the training of the power prediction model;

[0049] S4. Output cleaning prompt: Obtain the real-time solar irradiance and temperature and humidity data, analyze the output power under the condition of no dust accumulation through the power prediction model, compare it with the actual power collected by the inverter, so as to calculate the daily power loss value, combine the electricity price and the cleaning time to analyze the actual loss value of the photovoltaic module. If the loss value exceeds the preset value and the meteorological data shows that there is no rain in the next 48 hours, a cleaning prompt is issued.

[0050] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0051] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A photovoltaic monitoring system based on the Internet of Things, including a photovoltaic module cleaning unit (4), characterized in that: A data analysis unit (1), which collects solar irradiance, temperature, humidity, wind speed, wind direction, air pressure, and the output power of photovoltaic modules without dust accumulation through multiple sensors, stores them in a database, statistically analyzes the offset values corresponding to these variables of solar irradiance, temperature, humidity, wind speed, wind direction, air pressure, and the output power of photovoltaic modules without dust accumulation, and analyzes the correlation values between variables based on the offset values; A parameter initial screening unit (2), which takes solar irradiance, temperature, and humidity as the optimal variables, constructs corresponding sample data, determines the initial parameters of the power prediction model and sets the search range, divides all sample data into five parts, rotates to select the test set, and the rest are the training set. After each training, the model accuracy is evaluated, and the average value of the accuracies evaluated in five rounds is used as the actual accuracy of the current parameters; A parameter repeated screening unit (3), which statistically analyzes the parameter values corresponding to the highest actual accuracy, screens the optimal range according to the parameter values and calculates a new search interval, generates parameters according to the optimal range and the new search interval, deletes the original parameters, calculates the actual accuracy of the current parameters, and if the accuracy is lower than the threshold, continues to iterate, otherwise outputs the best parameters and completes the training of the power prediction model.

2. The photovoltaic monitoring system based on the Internet of Things according to claim 1, characterized in that: The data analysis unit (1) includes a data acquisition module (101) and a power extraction module (102). The data acquisition module (101) uses a pyranometer to collect solar irradiance at different time periods, collects temperature data of photovoltaic modules at different time periods through a temperature sensor, collects humidity data of photovoltaic modules at different time periods using a humidity sensor, collects wind speed data and wind direction around photovoltaic modules at different time periods using a wind speed sensor, and collects the atmospheric pressure around photovoltaic modules at different time periods through a barometric pressure sensor, and transmits solar irradiance, temperature data, humidity data, wind speed data, wind direction, atmospheric pressure, and the corresponding recording time to the database for storage. The power extraction module (102) uses a smart meter to collect the output power of photovoltaic modules without dust accumulation at different time periods, and transmits the photovoltaic output power and the corresponding recording time to the database for storage.

3. The photovoltaic monitoring system based on the Internet of Things according to claim 2, characterized in that: The data analysis unit (1) further comprises an offset value calculation module (103) and a correlation value determination module (104). The offset value calculation module (103) counts different recording times t1, t2, ..., t k ,…,t n The corresponding solar radiation F1, F2, ..., F k ,…,F n 、Temperature data T1, T2,…, T k ,…,T n 、Humidity data H1, H2,…, H k ,...,H n 、Wind speed data V1, V2, ..., V k ,...,V n 、wind direction data W1, W2, ..., W k ,...,W n 、Atmospheric pressure P1, P2, ..., P k ,...,P n And photovoltaic output power OP1, OP2, ..., OP k ,...,OP n Then, according to F1, F2, ..., F k ,...,F n Calculate the kth solar radiation offset value F′ k ,in Using T1, T2, ..., T k ,...,T n Calculate the kth temperature offset value T′ k ,in Using H1,H2,...,H k ,...,H n Calculate the kth temperature offset value H′ k ,in Through V1, V2, ..., V k ,...,V n Calculate the kth wind speed offset value V′ k ,in According to W1,W2,...,W k ,...,W n Calculate the kth wind direction offset value W′ k ,in Using P1, P2, ..., P k ,...,P n Calculate the kth atmospheric pressure offset value P′ k ,in Through OP1, OP2, ..., OP k ,…,OP n Calculate the k-th photovoltaic output power offset value OP′ k , where k represents a parameter, n represents the amount of data. The correlation value determination module (104) repeats the operation until all the solar radiation offset values, temperature offset values, humidity offset values, wind speed offset values, wind direction offset values, atmospheric pressure offset values, and photovoltaic output power offset values are calculated, and then analyzes the solar radiation offset values, temperature offset values, humidity offset values, wind speed offset values, wind direction offset values, atmospheric pressure offset values, and photovoltaic output power offset values to obtain the correlation values among the solar radiation degree, temperature, humidity, wind speed, wind direction, atmospheric pressure, and photovoltaic output power.

4. The photovoltaic monitoring system based on the Internet of Things according to claim 1, wherein: The parameter primary screening unit (2) includes a sample generation module (201) and a parameter setting module (202). The sample generation module (201) selects the three optimal variables most relevant to the photovoltaic output power according to the magnitude of the correlation value. The three optimal variables are solar irradiance, temperature, and humidity respectively. After extracting the solar irradiance, temperature data, humidity data, and photovoltaic output power in the database at different time periods, multiple sample data are constructed based on the recording time, solar irradiance, temperature data, humidity data, and photovoltaic output power, and stored in the original dataset. After the parameter setting module (202) constructs a power prediction model using the SVM technology, it sets the initial values, ranges, and maximum search intervals of the adjustment coefficient and kernel function parameters in the power prediction model, and takes values for the adjustment coefficient and kernel function parameters according to the maximum search interval, obtaining multiple adjustment coefficients and kernel function parameters.

5. The photovoltaic monitoring system based on the Internet of Things according to claim 4, characterized in that: The parameter primary screening unit (2) further includes a sample division module (203), an accuracy analysis module (204), and an average value calculation module (205). The sample division module (203) divides the sample data in the original dataset into five subsets. After randomly selecting an unlabeled subset as the test set and labeling it, the remaining subsets are used as the training set. After the accuracy analysis module (204) selects the adjustment coefficient and kernel function parameters, it transmits the sample data in the training set to the current power prediction model for analysis. After completing the model training, it uses the sample data in the test set to determine the accuracy of the current model. The average value calculation module (205) repeats the operation until each subset is labeled, and then calculates the corresponding average value using the five accuracies, and takes this average value as the actual accuracy corresponding to the current adjustment coefficient and kernel function parameters.

6. The photovoltaic monitoring system based on the Internet of Things according to claim 1, characterized in that: The parameter repeated screening unit (3) includes a search interval adjustment module (301), a parameter output module (302), and a model training module (303). After the search interval adjustment module (301) statistically analyzes the actual accuracies of all current adjustment parameters and kernel function parameters, it selects the adjustment coefficient and kernel function parameter with the highest actual accuracy, screens out the optimal range according to the values of the adjustment coefficient and kernel function parameter, calculates a new search interval according to the size of the optimal range, re-analyzes multiple adjustment coefficients and kernel function parameters using the optimal range and the search interval, and deletes the original adjustment coefficients and kernel function parameters. The parameter output module (302) statistically analyzes the actual accuracies corresponding to all adjustment coefficients and kernel function parameters. If the actual accuracy is lower than the threshold, it continues to screen the optimal range and generates new adjustment coefficients and kernel function parameters. If the actual accuracy is higher than or equal to the threshold, it stops screening the optimal range and outputs the adjustment coefficients and kernel function parameters corresponding to the current maximum actual accuracy. The model training module (303) transmits the adjustment coefficients and kernel function parameters to the power prediction model, randomly extracts multiple sample data from the original data set and stores them in the random test set, trains the current power prediction model using the remaining sample data in the original data set, and calculates the accuracy of the current power prediction model through the sample data in the random test set.

7. The photovoltaic monitoring system based on the Internet of Things according to claim 1, characterized in that: The photovoltaic module cleaning unit (4) includes an output power prediction module (401), a power loss analysis module (402), a loss value calculation module (403), and a prompt message sending module (404). After the output power prediction module (401) obtains the solar irradiance, temperature data, and humidity data in the current time period, it transmits them to the power prediction model for analysis to obtain the output power of the photovoltaic module in a dust-free state. The power loss analysis module (402) uses the inverter to collect the actual output power of the current photovoltaic module, and calculates the daily power loss value through the output power of the photovoltaic module in a dust-free state and the actual output power. The loss value calculation module (403) obtains the daily electricity price and the most recent cleaning time, and analyzes the electricity price and the daily power loss value using the power loss determination algorithm to obtain the actual loss value of the current photovoltaic module. If the actual loss value is lower than the preset value, the prompt message sending module (404) does not perform any operation. Otherwise, it retrieves the meteorological data within the next 48 hours. If the rainfall in the meteorological data is not zero, it does not perform any operation. If the rainfall in the meteorological data is zero, it sends a prompt message to clean the photovoltaic module.

8. Photovoltaic monitoring method based on the Internet of Things, characterized in that, The photovoltaic monitoring method is applicable to the Internet of Things-based photovoltaic monitoring system according to any one of claims 1-7, and includes the following steps: S1. Sensor data acquisition: Collect solar irradiance, temperature, humidity, wind speed, wind direction, air pressure, and the output power of the photovoltaic module under the condition of no dust accumulation through multiple sensors, store them in the database, calculate the offset values corresponding to these variables of solar irradiance, temperature, humidity, wind speed, wind direction, air pressure, and the output power of the photovoltaic module under the condition of no dust accumulation, and analyze the correlation values between the variables according to the offset values; S2. Sample data division: Take solar irradiance, temperature, and humidity as the optimal variables, construct the corresponding sample data, determine the initial parameters of the power prediction model and set the search range. Divide all sample data into five parts, rotate to select the test set, and the rest are the training set. Evaluate the model accuracy after each training, and take the average accuracy of the five rounds of evaluation as the actual accuracy of the current parameters; S3. Select the best parameters: Calculate the parameter values corresponding to the highest actual accuracy, screen the optimal range according to the parameter values and calculate the new search interval, generate parameters according to the optimal range and the new search interval, delete the original parameters, calculate the actual accuracy of the current parameters. If the accuracy is lower than the threshold, continue to iterate, otherwise output the best parameters and complete the training of the power prediction model; S4. Output cleaning prompt: Obtain real-time solar irradiance and temperature and humidity data, analyze the output power under the condition of no dust accumulation through the power prediction model, and compare it with the actual power collected by the inverter, so as to calculate the daily power loss value. Combine the electricity price and cleaning time to analyze the actual loss value of the photovoltaic module. If the loss value exceeds the preset value and the meteorological data shows that there is no rain in the next 48 hours, issue a cleaning prompt.

Images