AI algorithm-based photovoltaic module temperature field deviation adaptive system

Abstract

The application provides an AI algorithm-based photovoltaic module temperature field deviation adaptive system, which comprises a data acquisition module, a data preprocessing module, an AI algorithm processing module and a control module, wherein the AI algorithm processing module comprises a temperature field deviation identification model, a regulation parameter correction model, a model self-iteration optimization module and an abnormal early warning and emergency decision submodule. The application realizes accurate capture of local subtle deviation, adaptive control of complex scenes, intelligent production control and stable control of temperature field uniformity by fine partitioning and laying temperature measuring points and collecting full-dimensional data in combination with the AI algorithm, so as to solve the problems of poor adaptability, low accuracy and dependence on manual parameter adjustment in the prior art.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic module manufacturing technology, specifically to an AI algorithm-based adaptive system for temperature field deviation in photovoltaic modules, which is used to solve the problem of temperature field uniformity control in the lamination process of photovoltaic modules and improve the production quality and stability of modules. Background Technology

[0002] The lamination process is the core process in photovoltaic module encapsulation. It uses high temperature and pressure to tightly bond glass, encapsulant film, and solar cells together, and the uniformity of the temperature field directly determines the quality of the module. Temperature deviations can easily lead to defects such as bubbles, delamination, and microcracks, even resulting in batch scrapping. As the photovoltaic industry upgrades to larger sizes and higher power, the heat dissipation difference between the edges and center of large-size modules such as 210mm is significant. New types of encapsulant films, such as POE, require higher temperature accuracy, with a deviation of ≤±1℃. Furthermore, module specifications, environmental conditions, and equipment aging all increase the difficulty of temperature control.

[0003] Currently, the industry generally relies on traditional temperature control technology, which can meet basic production needs, but lacks adaptability and accuracy, and cannot cope with the control requirements of multiple scenarios. Component defect rates remain high, requiring frequent manual parameter adjustments, which is inefficient and costly. There is an urgent need for a self-adaptive, self-learning, high-precision control solution to break through these bottlenecks. Summary of the Invention

[0004] The purpose of this invention is to provide an adaptive temperature field deviation system for photovoltaic modules based on AI algorithms. By finely zoning and deploying temperature measurement points and collecting data from all dimensions, combined with AI algorithms, it can accurately capture subtle local deviations, adaptively manage complex scenarios, intelligently manage production, and stably control the uniformity of the temperature field. This solves the problems of poor adaptability, low accuracy, and reliance on manual parameter adjustment in existing technologies.

[0005] The present invention achieves the above objectives through the following technical solutions: An AI-based adaptive system for temperature field deviation in photovoltaic modules includes: The data acquisition module is configured to collect multi-dimensional data during the lamination process; The data preprocessing module is used to perform noise reduction, outlier removal, and data range standardization on the collected raw data to generate preprocessed data. The AI ​​algorithm processing module includes: Temperature field deviation identification model: A convolutional neural network is used to process the preprocessed temperature data, construct a temperature field distribution feature map, identify temperature deviations in each zone, and classify them according to the severity of the deviations; Control parameter correction model: Based on long short-term memory network, the control parameters are adaptively corrected according to the deviation identification results, current component specifications, environmental conditions and equipment operating status. Model self-iterative optimization module: adopts reinforcement learning Q-learning mechanism, and dynamically iteratively optimizes temperature field deviation identification model and control parameter correction model based on defect data and control effect data fed back from subsequent processes; Anomaly warning and emergency decision-making submodule: Based on logistic regression algorithm, it identifies sudden anomalies in the temperature field, determines the cause of the anomaly, triggers graded warnings, and executes emergency control measures; Control module: Based on the control instructions output by the AI ​​algorithm processing module, the heating module and pressure regulating valve of the laminator are driven to make adjustments. At the same time, the execution results are fed back to the AI ​​algorithm processing module in real time to form a closed-loop control, so as to realize the adaptive control of temperature field deviation during the lamination process of photovoltaic modules.

[0006] According to the present invention, a photovoltaic module temperature field deviation adaptive system based on an AI algorithm includes multi-dimensional data such as: Temperature data includes the temperatures of at least 12 independently controlled temperature zones of the laminator heating plate, each zone having at least 3 temperature measuring points; the temperatures of the center and at least 4 edge temperature measuring points on the component surface; and the ambient temperatures of the upper and lower chambers of the laminator. Equipment operating data includes real-time power, current, and voltage of each zone heating module, real-time pressure of the laminator chamber, operating time of the heating modules, and cumulative operating mileage of the equipment; Component specifications; Workshop environmental data, including real-time temperature, humidity, and dust concentration.

[0007] According to the present invention, a photovoltaic module temperature field deviation adaptive system based on AI algorithm includes the following noise reduction steps: Data prediction: Based on the accurate data after noise reduction processing at the previous moment. Control quantity at the previous moment Estimate the reasonable data value for the current moment. , represented as:

[0008] Among them, coefficient A The value is 1, and the coefficient is... B The value is 0. This is the control quantity of the laminator at the previous moment; Predictive uncertainty estimation: based on the predictive uncertainty of the previous moment Calculate the uncertainty of the predicted data at the current moment. , is represented as;

[0009] in, Q For process noise covariance; Weight calculation: based on the uncertainty of prediction at the current moment Calculate the data weight at the current moment. , represented as:

[0010] in, H For the observation matrix, R To determine the value of the noise covariance; Data correction: Combine the predicted reasonable data values ​​at the current moment. The raw data directly collected by the sensor at the current moment and data weights Output the accurate data after noise reduction at the current moment. ， Represented as:

[0011] in, These are the raw data collected by sensors such as temperature, power, and pressure at the current moment. The observed values ​​are for predicting the data.

[0012] According to the adaptive system for temperature field deviation of photovoltaic modules based on AI algorithm provided by the present invention, the noise reduction process further includes an uncertainty update step: Data weights based on the current moment Observation matrix H and the uncertainty of the prediction at the previous moment Calculate the uncertainty of the data after noise reduction at the current moment. , represented as:

[0013] in, I It is the identity matrix. This uncertainty value is used to quantify the reliability of the data after noise reduction at the current moment, and will be used as the input parameter for the uncertainty estimation step at the next moment, so as to realize the dynamic transmission and iterative update of uncertainty during the noise reduction process.

[0014] According to the adaptive system for temperature field deviation of photovoltaic modules based on AI algorithm provided by the present invention, the abnormal data removal includes the following steps: Data distribution characteristic calculation: For each type of denoised dataset, calculate the average value of the data. m and degree of dispersion s ,in: average value m Calculated using the following formula:

[0015] in, For a single denoised data point in this type of dataset, n The number of samples in this type of dataset. m Used to reflect the overall level of data; Dispersion s Calculated using the following formula:

[0016] in, s Used to reflect the range of data fluctuations; Outlier detection and removal: Set the criterion for extreme erroneous data as a single data entry. Compared with the average m The absolute value of the difference is greater than 3 times the degree of dispersion s When a piece of data meets the data discrimination criteria, it is directly judged as an extreme error and removed.

[0017] According to the present invention, a photovoltaic module temperature field deviation adaptive system based on AI algorithm includes the following steps in unifying data range processing: For each type of valid data after noise reduction and removal of outlier extreme data, determine the minimum value for that type of data. and maximum value ;in, This represents the statistical minimum value of this type of valid data. This represents the maximum statistical value of this type of valid data. For each valid data x The standard data after unification is calculated using the following formula. :

[0018] Standard data output: As a standardized data output with a unified scope, it is directly used as the input feature for the AI ​​algorithm processing module.

[0019] According to the present invention, a photovoltaic module temperature field deviation adaptive system based on AI algorithm is provided, wherein the temperature field deviation identification model is configured to perform the following operations: Based on the preprocessed temperature data output by the data preprocessing module, a temperature field distribution characteristic map of each independent temperature control zone of the laminator heating plate is constructed. A convolutional neural network is used to extract and analyze the features of the temperature field distribution map, and to identify the temperature deviation value of each independent temperature control zone. The temperature deviation value is calculated as the difference between the real-time temperature of the corresponding zone and the standard temperature of the zone. The standard temperature is preset based on the specifications of the photovoltaic module and the type of lamination film; Based on the absolute value of the temperature deviation, the temperature deviation of each zone is divided into three severity levels: slight deviation, moderate deviation, and severe deviation. The final output includes the temperature deviation value, deviation level, and deviation location information for each independent temperature control zone, which is the location identifier of the corresponding independent temperature control zone.

[0020] According to the adaptive system for temperature field deviation of photovoltaic modules based on AI algorithm provided by the present invention, the model self-iterative optimization module is configured to perform the following operations: Obtain defect data and control effect data from subsequent processes. Using the current temperature field deviation characteristics, component specification parameters, and environmental operating condition data as the state space, the adjustment amount of the convolution kernel parameters of the temperature field deviation identification model and the adjustment amount of the hidden layer weights of the control parameter correction model as the action space, and the reduction in defect rate and the degree of temperature deviation reduction after control as the reward function, construct the Q-table or deep Q-network of the reinforcement learning Q-learning algorithm. By iteratively updating the Q value, dynamically adjust the convolution kernel parameters of the temperature field deviation identification model and the hidden layer weights of the control parameter correction model to achieve adaptive optimization of the model for the lamination scenario. Dynamic updates of model parameters are achieved using a dual-mode approach: real-time iteration and batch iteration. After a single control operation is completed, based on the real-time effect data of that control, the local convolution kernel parameters of the temperature field deviation identification model or the short-term memory weights of the control parameter correction model are fine-tuned to achieve a rapid response of the model to the effect of a single control. The system automatically summarizes the control effect data, defect data, environmental condition data, and component specification data of all lamination processes during the preset time period each day. It then uses the Q-learning algorithm to batch optimize the global convolution kernel parameters of the temperature field deviation identification model and the long-term memory weights of the control parameter correction model.

[0021] According to the present invention, a photovoltaic module temperature field deviation adaptive system based on AI algorithm is provided. In the model self-iterative optimization module, the combined index of temperature field deviation correction rate and lamination defect rate is used as the reward function of reinforcement learning Q-learning algorithm. The temperature field deviation correction rate is the proportion of independent temperature control zones whose temperature deviation value is reduced to the target range after regulation; the lamination defect rate is the ratio of the number of defects in the module after lamination to the total number of modules. When the temperature field deviation correction rate under a certain operating condition is found to be less than the first preset value or the lamination defect rate is found to be greater than the second preset value, the following operation will be automatically triggered: The full-dimensional dataset under this working condition is extracted as a new training sample. The full-dimensional dataset includes: preprocessed temperature field distribution data, component specification parameters, environmental working condition data, control parameter instructions and corresponding subsequent process defect data. Add the new training samples to the training sets of the CNN bias recognition model and the LSTM parameter adjustment model, re-execute the iterative optimization process of the Q-learning algorithm, and update the convolution kernel parameters of the CNN bias recognition model and the hidden layer weights of the LSTM parameter adjustment model.

[0022] According to the present invention, a photovoltaic module temperature field deviation adaptive system based on AI algorithm is provided, the control module comprising: The execution equipment group includes 12 independent zone heating modules for the laminator, pressure regulating valves, backup sensors, audible and visual early warning devices, and operation and maintenance terminals; The execution control unit includes an edge controller and a PLC controller. The edge controller is configured to receive control commands output by the AI ​​algorithm processing module. The control commands include the power adjustment values ​​and temperature field deviation compensation coefficients of each independent zone heating module. The PLC controller is communicatively connected to the edge controller and is configured to drive the corresponding independent zone heating module to adjust its power according to the control commands, and synchronously adjust the pressure regulating valve of the corresponding area to achieve temperature and pressure coordinated control of heating power and chamber pressure. The data feedback unit is configured to collect temperature data and equipment operation data in real time during the execution of control commands, and feed the collected data back to the AI ​​algorithm processing module for the verification of deviation correction effect and the self-iterative optimization of temperature field deviation identification model and control parameter correction model.

[0023] Therefore, compared with the existing technology, the photovoltaic module temperature field deviation adaptive system based on AI algorithm proposed in this invention has the following beneficial effects: 1. This invention uses 12 zones and multiple measurement points for data acquisition and preprocessing, combined with a CNN algorithm to achieve deviation classification identification and positioning, resulting in more accurate identification and meeting the high-precision control requirements of POE film at ±1℃.

[0024] 2. This invention achieves adaptive power adjustment in partitioned areas by coupling multiple influencing factors through LSTM, which can be adapted to scenarios such as large-size components, complex environments, and equipment aging.

[0025] 3. This invention achieves dual-mode self-iteration of the model based on Q-learning, and completes hierarchical early warning and automatic emergency response with logistic regression, which can be continuously optimized without manual intervention.

[0026] 4. This invention achieves precise temperature and pressure control through the construction of a closed-loop management system throughout the entire process, effectively reducing defects such as bubbles and debonding, significantly reducing the defect rate, and improving production yield and stability.

[0027] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0028] Figure 1 This is a schematic diagram illustrating the implementation principle of an embodiment of a photovoltaic module temperature field deviation adaptive system based on AI algorithm according to the present invention.

[0029] Figure 2 This is a schematic diagram illustrating the implementation principle of the temperature field deviation identification model in an embodiment of an AI-based adaptive system for photovoltaic module temperature field deviation.

[0030] Figure 3 This is a schematic diagram illustrating the implementation principle of the model self-iterative optimization module in an embodiment of a photovoltaic module temperature field deviation adaptive system based on AI algorithm according to the present invention.

[0031] Figure 4 This is a schematic diagram illustrating the implementation principle of the control module in an embodiment of an AI-based adaptive system for temperature field deviation in photovoltaic modules according to the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0033] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0034] See Figure 1 to Figure 4 This embodiment provides an adaptive system for temperature field deviation of photovoltaic modules based on AI algorithms, including: The data acquisition module is configured to collect multi-dimensional data during the lamination process; The data preprocessing module is used to perform noise reduction, outlier removal, and data range standardization on the collected raw data to generate preprocessed data. The AI ​​algorithm processing module includes: Temperature field deviation identification model: A convolutional neural network is used to process the preprocessed temperature data, construct a temperature field distribution feature map, identify temperature deviations in each zone, and classify them according to the severity of the deviations; Control parameter correction model: Based on long short-term memory network, the control parameters are adaptively corrected according to the deviation identification results, current component specifications, environmental conditions and equipment operating status. Model self-iterative optimization module: adopts reinforcement learning Q-learning mechanism, and dynamically iteratively optimizes temperature field deviation identification model and control parameter correction model based on defect data and control effect data fed back from subsequent processes; Anomaly warning and emergency decision-making submodule: Based on logistic regression algorithm, it identifies sudden anomalies in the temperature field, determines the cause of the anomaly, triggers graded warnings, and executes emergency control measures; Control module: Based on the control instructions output by the AI ​​algorithm processing module, the heating module and pressure regulating valve of the laminator are driven to make adjustments. At the same time, the execution results are fed back to the AI ​​algorithm processing module in real time to form a closed-loop control, so as to realize the adaptive control of temperature field deviation during the lamination process of photovoltaic modules.

[0035] In this embodiment, the multi-dimensional data includes: Temperature data includes the temperatures of at least 12 independently controlled temperature zones of the laminator heating plate, each zone having at least 3 temperature measuring points; the temperatures of the center and at least 4 edge temperature measuring points on the component surface; and the ambient temperatures of the upper and lower chambers of the laminator. Equipment operating data includes real-time power, current, and voltage of each zone heating module, real-time pressure of the laminator chamber, operating time of the heating modules, and cumulative operating mileage of the equipment; Component specifications; Workshop environmental data, including real-time temperature, humidity, and dust concentration.

[0036] Specifically, the laminator heating plate is divided into 12 independent temperature control zones, with 3 PT100 temperature measuring points (T1, T2, T3) in each zone; a high-temperature resistant flexible test piece is attached to the component surface, with a central temperature measuring point (T1, T2, T3). C ) and 4 edge temperature measurement points (T E1 ~T E4 ), synchronously collect data from the upper chamber (T U ), lower chamber (T) D The ambient temperature. The collected equipment operating data includes the real-time power of each zone's heating module (0-100%, P1~P...). 12 ), current (0-10A, I1~I 12 Voltage (220V±10V, U1~U) 12 Real-time pressure of the laminator chamber (0.1-0.3MPa, P) chamber Heating module running time t run Accumulated operating mileage of equipment S total It is used to determine the aging status of equipment.

[0037] In this embodiment, the noise reduction process includes the following steps: Data prediction: Based on the accurate data after noise reduction processing at the previous moment. Control quantity at the previous moment Estimate the reasonable data value for the current moment. , represented as:

[0038] in, To predict the data at the current moment (such as the predicted temperature at T1, the predicted power at P1), the coefficients... A The value is set to 1, which adapts to the linear continuity characteristics of conventionally collected data. The coefficient... B A value of 0 indicates simplified processing when there is no additional control input. This refers to the control parameters of the laminator at the previous moment, such as the heating power setpoint. Predictive uncertainty estimation: based on the predictive uncertainty of the previous moment Calculate the uncertainty of the predicted data at the current moment. , is represented as;

[0039] in, Q The process noise covariance is set to 0.01 (to adapt to the statistical characteristics of workshop environmental interference). Weight calculation: based on the uncertainty of prediction at the current moment Calculate the data weight at the current moment. (Values ​​range from 0 to 1, with values ​​closer to 1 indicating higher reliability of the original collected data), represented as:

[0040] in, H The observation matrix is ​​set to 1 (directly associated with the original acquired data), and the observation noise covariance is... R The value is 0.001 (matching the sensor's acquisition accuracy); Data correction: Combine the predicted reasonable data values ​​at the current moment. The raw data directly collected by the sensor at the current moment and data weights Output the accurate data after noise reduction at the current moment. ， Represented as:

[0041] in, This refers to the raw data collected by sensors at the current moment, such as the real-time temperature of T1 and the real-time current of I1. The observed values ​​are for predicting the data.

[0042] In this embodiment, the noise reduction process further includes an uncertainty update step: Data weights based on the current moment Observation matrix H and the uncertainty of the prediction at the previous moment Calculate the uncertainty of the data after noise reduction at the current moment. , represented as:

[0043] in, I It is the identity matrix. This uncertainty value is used to quantify the reliability of the data after noise reduction at the current moment, and will be used as the input parameter for the uncertainty estimation step at the next moment, so as to realize the dynamic transmission and iterative update of uncertainty during the noise reduction process.

[0044] In this embodiment, the abnormal data removal includes the following steps: Data distribution characteristic calculation: For each type of noise reduction dataset, including the noise reduction temperature dataset of all temperature measurement points, the noise reduction power dataset of all zone heating modules, and the noise reduction pressure dataset of the laminator chamber, calculate the average value of the data respectively. m and degree of dispersion s ,in: average value m Calculated using the following formula:

[0045] in, For a single denoised data point in this type of dataset, n This represents the number of samples in this type of dataset, calculated based on the sampling frequency. For example, if sampling is done at 10Hz, the sample size per minute is 600. m This is the average value of the data, used to reflect the overall level of the data; Dispersion s Calculated using the following formula:

[0046] in, s Used to reflect the range of data fluctuations Outlier detection and removal: Set the criterion for extreme erroneous data as a single data entry. Compared with the average m The absolute value of the difference is greater than 3 times the degree of dispersion s (i.e. |) xi m |>3 s When a piece of data meets this criterion, it is directly identified as an extreme error and removed. If three or more consecutive samples in the same dataset meet the above criteria, then perform the following operations simultaneously: Mark the corresponding data acquisition equipment (including temperature sensors, power acquisition lines, and pressure transmitters); The system pushes fault alerts to the system maintenance terminal, indicating that the acquisition device may have hardware failure or line interference. For example, if the temperature of three consecutive samples of T2 deviates from the normal range, it indicates that the temperature acquisition device may be faulty.

[0047] In this embodiment, the unified data range processing specifically includes the following steps: For each type of valid data after noise reduction and removal of outlier extreme data (including temperature measurement data, zone heating power data, laminator chamber pressure data, workshop environmental parameter data, etc.), determine the minimum value for that type of data. and maximum value ;in, This refers to the statistical minimum value of this type of valid data. For example, the minimum valid data for all temperature measurement points is -50℃, and the minimum valid data for all zone heating power is 0%. This refers to the statistical maximum value of this type of valid data. For example, the maximum value of valid data from all temperature measuring points is 200℃, and the maximum value of valid data from all zone heating power is 100%. For each valid data x That is, a single data point that has undergone noise reduction processing and has no abnormalities, such as T C core temperature, P _ chamber The chamber pressure is calculated using the following formula to obtain the standard data within a unified range. :

[0048] Standard data output: As a standard data output after unifying the scope The value range is strictly limited to 0~1, and it is directly used as the input feature of the AI ​​algorithm processing module (temperature field deviation identification submodule, control parameter correction submodule, etc.).

[0049] In this embodiment, as Figure 2 As shown, the temperature field deviation identification model is configured to perform the following operations: Based on the preprocessed temperature data output by the data preprocessing module, a temperature field distribution characteristic map of each independent temperature control zone of the laminator heating plate is constructed. A convolutional neural network is used to extract and analyze the features of the temperature field distribution map, and to identify the temperature deviation value of each independent temperature control zone. The temperature deviation value is calculated as the difference between the real-time temperature of the corresponding zone and the standard temperature of the zone. The standard temperature is preset based on the specifications of the photovoltaic module (including module size and module type) and the type of laminating film. For example, when the laminating film is POE film, the standard temperature is preset to 150℃±2℃. Based on the absolute value of the temperature deviation, the temperature deviations in each zone are divided into three severity levels: Slight deviation: | Deviation value| ≤ 1℃; Moderate deviation: 1℃ < |deviation value| ≤ 3℃; Severe deviation: | Deviation value| > 3℃; The final output includes the temperature deviation value, deviation level, and deviation location information for each independent temperature control zone, which is the location identifier of the corresponding independent temperature control zone.

[0050] In this embodiment, the adjustment parameter correction model is configured to perform the following operations: Based on the deviation identification results, and combined with the current component specifications, environmental conditions, and equipment operating status, the control parameters (core parameters include: heating power adjustment range, control response speed, and power compensation coefficient) are adaptively corrected. The correction rules are shown in Table 1. Table 1: Correction rules for control parameters

[0051] In this embodiment, as Figure 3 As shown, the model self-iterative optimization module is configured to perform the following operations: Obtain defect data (including defect type, defect location, and defect severity) and control effect data (including temperature deviation change after control, lamination component pass rate, and control response time) from subsequent processes (such as the component inspection process after lamination). Using the current temperature field deviation characteristics (preprocessed temperature distribution data), component specification parameters (size, type), and environmental condition data (workshop temperature, humidity, dust concentration) as the state space, the adjustment amount of the convolution kernel parameters of the CNN model and the adjustment amount of the hidden layer weights of the LSTM model as the action space, and the reduction in defect rate and the degree of temperature deviation reduction after adjustment as the reward function, a Q-table or deep Q-network (DQN) for reinforcement learning Q-learning algorithm is constructed. By iteratively updating the Q value, the convolution kernel parameters of the temperature field deviation recognition model and the adjustment parameters are used to correct the hidden layer weights of the model, so as to achieve adaptive optimization of the model for the lamination scenario. Dynamic updates of model parameters are achieved using a dual-mode approach: real-time iteration and batch iteration. After a single control operation is completed, based on the real-time effect data of the control (such as temperature deviation changes within 1-5 minutes after control, and immediate defect feedback in subsequent processes), the local convolution kernel parameters of the temperature field deviation identification model or the short-term memory weights of the control parameter correction model are fine-tuned to achieve a rapid response of the model to the effect of a single control. The system automatically aggregates control effect data, defect data, environmental condition data, and component specification data for all lamination processes during a preset time period each day (e.g., 0:00-2:00 AM). It then uses the Q-learning algorithm to batch optimize the global convolution kernel parameters of the temperature field deviation identification model and the long-term memory weights of the control parameter correction model, thereby improving the model's generalization ability and prediction accuracy under long-term data accumulation. Moreover, the batch iteration process does not occupy the system's computing resources during daytime production hours.

[0052] In the model self-iterative optimization module, the combined index of temperature field deviation correction rate (target value ≥ 98%) and lamination defect rate (target value ≤ 0.1%) is used as the reward function of the reinforcement learning Q-learning algorithm. The temperature field deviation correction rate is the proportion of independent temperature control zones whose temperature deviation value is reduced to the target range (slight deviation ≤ 1℃) after adjustment; the lamination defect rate is the ratio of the number of defects in the laminated components to the total number of components. When the temperature field deviation correction rate is less than 95% or the lamination defect rate is greater than 0.3% under a certain operating condition, the following operations will be automatically triggered: Extract the full-dimensional dataset under this working condition as a new training sample. The full-dimensional dataset includes: preprocessed temperature field distribution data, component specification parameters (size / type), environmental working condition data (workshop temperature / humidity / dust concentration), control parameter instructions (heating power / pressure set value) and corresponding subsequent process defect data (defect type / location / severity). The newly added training samples were added to the training sets of the CNN bias identification model and the LSTM parameter adjustment model. The iterative optimization process of the Q-learning algorithm was re-executed to update the convolution kernel parameters of the CNN bias identification model and the hidden layer weights of the LSTM parameter adjustment model, so as to improve the model's adaptability and prediction accuracy for this abnormal working condition.

[0053] In this embodiment, the anomaly warning and emergency decision-making submodule is configured to perform the following operations: Identify sudden anomalies in the temperature field (such as a sudden rise / fall in temperature in a certain zone, or simultaneous deviations exceeding the standard in multiple zones), determine the cause of the anomaly (sensor failure, heating module failure, pressure anomaly, etc.), trigger graded early warnings, and execute emergency control, as shown in Table 2.

[0054] Table 2: Tiered Early Warning and Emergency Response

[0055] In this embodiment, as Figure 4 As shown, the control module includes: The execution equipment group includes 12 independent zone heating modules of the laminator (each heating module supports stepless power adjustment from 0-100%), pressure regulating valves, backup sensors, audible and visual early warning devices, and operation and maintenance terminals. The execution control unit includes an edge controller and a PLC controller. The edge controller is configured to receive control commands output by the AI ​​algorithm processing module. The control commands include the power adjustment values ​​and temperature field deviation compensation coefficients of each independent zone heating module. The PLC controller is communicatively connected to the edge controller and is configured to drive the corresponding independent zone heating module to adjust its power according to the control commands, and synchronously adjust the pressure regulating valve of the corresponding area to achieve temperature and pressure coordinated control of heating power and chamber pressure, thereby improving heat transfer efficiency. The data feedback unit is configured to collect temperature data (including the temperature of each independent zone heating plate and the surface temperature of the components) and equipment operation data (including the real-time power, current, voltage of each heating module and the opening degree of the pressure regulating valve) in real time during the execution of the control command. The collected data is then fed back to the AI ​​algorithm processing module for the verification of deviation correction effect and the self-iterative optimization of the temperature field deviation identification model and the control parameter correction model.

[0056] In practical applications, the adaptive control implementation of the heating module of a large-size 210mm module under low temperature and high humidity conditions in winter or the aging conditions of the heating module of a medium-size 182mm module is taken as an example: In winter, under low temperature and high humidity conditions, a photovoltaic module factory produces 210mm large-size photovoltaic modules (compatible with POE film, standard lamination temperature 150℃±2℃) in winter. The workshop environment temperature is 18℃ (<20℃), humidity is 65%RH (>60%RH, high humidity condition), and dust concentration is 0.2mg / m³. 3 All heating modules have a runtime of 3000 hours (<5000 hours for new modules), with no equipment aging issues. During production, it is necessary to address the problems of rapid heat loss at the edges of large-sized components and slow heating due to low temperature and high humidity.

[0057] Step 1: Initial Data Pre-collection 1. Temperature data: Temperature data was collected from 12 PT100 temperature measurement points in the temperature control zones (T...). 1-1 ~T 12-3 Real-time temperature (initial range 142~146℃), temperature measurement point T at the center of the module. C =145℃, edge temperature measuring point T E1 ~T E4=142~143℃, upper chamber T U =144℃, Lower chamber T D =145℃; 2. Equipment operating data: Heating power of each zone P1~P 12 =85%~88%, current I1~I 12 =8.2~8.5A, voltage U1~U 12 =220±5V, chamber pressure P chamber =0.2MPa, heating module running time t run =3000h, cumulative operating mileage of equipment S total =12000km; 3. Module specifications: 210mm large-size module, POE film, lamination standard temperature 150℃±2℃; 4. Environmental data: Workshop temperature (T) workshop =18℃, RH workshop =65%RH, dust concentration C dust =0.2mg / m 3 .

[0058] Step 2: Initial Data Preprocessing 1. Noise reduction processing: using T E1 Taking the data from the edge temperature measurement point as an example, the accurate temperature after noise reduction at the previous moment was 142.2℃, and the current original collected value is 141.8℃ (with slight fluctuations). According to the noise reduction logic, the predicted value is 142.2℃, and the weight is... =0.8, the corrected accurate temperature = 142.2 + 0.8 × (141.8 - 142.2) = 141.88℃, filtering out minor fluctuations; 2. Remove outlier extreme data: The average value of all temperature data is calculated to be μ = 144.2℃, the dispersion σ = 1.1℃, and 3σ = 3.3℃. All data satisfy | x i - μ|≤3.3℃, ​​no abnormal data, all data retained; 3. Standardize data range: For temperature data, xmin = -50℃, xmax = 200℃, and after standardization, T C = (145 + 50) / 250 = 0.78, T E1 = (141.88 + 50) / 250 = 0.767, after power data standardization, P1~P 12 =0.85~0.88.

[0059] Step 3: AI Algorithm Processing 1. CNN Bias Recognition and Classification: After constructing the temperature field feature map, edge partitions (T) are identified. E1 ~TE4 The deviation value for the associated heating zones 4-6 is 141.88-143.2℃ - 150℃ = -6.12--6.72℃ (absolute value 2.8-3.12℃). Zones 4 and 5 have moderate deviation (1℃ < |deviation value| ≤ 3℃), and zone 6 has severe deviation (|deviation value| > 3℃). Output the deviation value, level, and location (edge ​​zones 4-6). 2. LSTM parameter correction: Based on the trigger conditions (ambient temperature < 20℃, high humidity, large-size components, moderate to severe deviation), adjust the parameters according to the rules: increase the control response speed by 20% (low temperature) + 15% (high humidity) + 10% (large size) = 45%, increase the power adjustment range of the edge zone by 15%, take the adjustment range of severe deviation (zone 6) as 12%, take the adjustment range of moderate deviation (zones 4 and 5) as 8%, and increase the power compensation coefficient by 0.1 (low temperature) + 0.06 (high humidity) = 0.16; 3. Q-learning model iteration: Real-time acquisition of control effect data, initial deviation correction rate of 94.5% (<95%), automatic extraction of this working condition data as new training samples, fine-tuning of CNN and LSTM model parameters; 4. Logistic Regression Anomaly Warning: If the deviation value of partition 6 is greater than 3℃ but the duration is 2s (≤5s), a level 2 warning is triggered, the current lamination process is suspended, and the warning information is pushed to the operation and maintenance terminal.

[0060] Step 4: Control the laminator The edge controller receives control commands from the AI ​​algorithm and drives heating modules 4-6 via the PLC: the power of zones 4 and 5 is adjusted from 86% to 86%×(1+8%+15%)=106.18% (taking 100% as the upper limit of the 0-100% stepless adjustment), and the power of zone 6 is adjusted from 85% to 85%×(1+12%+15%)=107.95% (taking 100%), with the compensation coefficient added at 0.16; the chamber pressure P_chamber is adjusted to 0.22MPa (temperature and pressure coordination), and the backup sensor is switched to collect data in real time. Feedback data after the control shows that the temperature of the edge zones gradually rises to 147~149℃, the deviation value decreases to 0.5~2.5℃, the level 2 warning is lifted, and production resumes.

[0061] As can be seen, under this operating condition, the system successfully corrected the temperature deviation at the edge of the large-size component, with a final temperature field deviation correction rate of 98.8% and a lamination defect rate of 0.07% (≤0.1%). No equipment failure occurred, making it suitable for low temperature and high humidity conditions in winter and the production needs of large-size components. It can complete precise control and early warning without manual intervention.

[0062] In the aging process of the heating module section of a medium-sized 182mm photovoltaic module, a photovoltaic module factory produces 182mm medium-sized photovoltaic modules (compatible with EVA film, standard lamination temperature 148℃±2℃) at room temperature. The workshop environment temperature is 25℃ (20~30℃, room temperature), humidity is 50%RH, and dust concentration is 0.35mg / m³. 3 (>0.3mg / m 3 (High dust conditions); Among them, heating module No. 3 has been running for 8500 hours (5000~10000 hours, some aging), and the other heating modules have been running for 4000 hours (new modules). During the production process, the temperature of zone No. 3 is slow and fluctuates slightly, which needs to be compensated for by equipment aging and dust heat dissipation.

[0063] Step 1: Initial Data Pre-collection 1. Temperature data: Temperature data was collected from 12 PT100 temperature measurement points in the temperature control zones (T...). 1-1 ~T 12-3 Real-time temperature (initial range 143~147℃), with temperature measurement point T in zone 3. 3-1 ~T 3-3 =143~144℃, temperature measurement point T at the center of the component C =146℃, edge temperature measuring point T E1 ~T E4 =145~146℃, upper chamber T U =145℃, lower chamber T D =146℃; 2. Equipment operating data: Heating power of each zone P1~P 12 =82%~86% (P3 in zone 3 = 86%, power output is high but temperature is low), current I1~I 12 =8.0~8.3A (Partition 3 I3=7.8A, slightly lower than other partitions), voltage U1~U 12 =220±8V, chamber pressure P chamber =0.19MPa, module 3 t run =8500h, the remaining modules t run =4000h, cumulative operating mileage of equipment S total =18000km; 3. Component specifications: 182mm medium-sized component, EVA film, lamination standard temperature 148℃±2℃; 4. Environmental data: Workshop temperature (T) workshop =25℃, RH workshop =50%RH, dust concentration C dust =0.35mg / m 3 .

[0064] Step 2: Initial Data Preprocessing 1. Noise reduction processing: using partition 3 T 3-1 Taking the data as an example, the accurate temperature after noise reduction at the previous moment was 143.5℃, and the current original acquired value was 142.9℃ (with slight fluctuations). The calculated predicted value was 143.5℃, and the corrected accurate temperature was 143.5 + 0.8 × (142.9 - 143.5) = 143.02℃, thus filtering out fluctuation interference. 2. Remove outlier extreme data: The average value of all temperature data is calculated to be μ = 145.3℃, the dispersion σ = 0.9℃, and 3σ = 2.7℃. The data in partition 3 satisfies | x i - μ|≤2.7℃, no extreme outliers, the data fluctuation frequency of partition 3 is marked as high; 3. Standardize data range: After standardizing temperature data, T 3-1 = (143.02 + 50) / 250 = 0.772, after power data standardization P3 = 0.86, after dust concentration standardization = 0.35 / 0.5 = 0.7.

[0065] Step 3: AI Algorithm Processing 1. CNN Deviation Recognition and Grading: After constructing the temperature field feature map, the deviation value of zone 3 is identified as 143.02~144.1℃-148℃=-3.88~-4.98℃ (absolute value 3.88~4.98℃), which is a severe deviation; the deviation values ​​of the other zones are ≤1.2℃, which are slight deviations. The deviation value, grade (zone 3 is a severe deviation, and the others are slight deviations) and location (zone 3 is a heating zone) are output. 2. LSTM Parameter Correction: Based on the triggering conditions (medium-sized components, partial aging of the heating module, high dust conditions, severe deviation in zone 3, and minor deviations elsewhere), adjust the parameters according to the rules: increase the power compensation coefficient of zone 3 by 0.04 (partial aging), and expand the power adjustment range by 5% (high dust) + 12% (severe deviation) = 17%; for other minor deviation zones, the adjustment range is set to 4%, and the response speed is slowed down by 20%. 3. Q-learning model iteration: Real-time acquisition of control effect data, initial deviation correction rate of 96.2% (≥95%), defect rate of 0.09% (≤0.3%), no need to add new training samples, only real-time fine-tuning; 4. Logistic Regression Anomaly Warning: The frequency of temperature fluctuations in Zone 3 is relatively high, but the conditions for a Level 2 warning have not been met. A Level 1 warning has been triggered, with local audible and visual alerts. No production suspension is required.

[0066] Step 4: Control the laminator The edge controller receives control commands from the AI ​​algorithm and drives heating module 3 via the PLC: the power is adjusted from 86% to 86%×(1+17%)=100.62% (taken as 100%), with a power compensation coefficient of 0.04 added; the power of other slightly deviating zones is adjusted to 82%~86%×(1+4%)=85.28%~89.44%, with a 20% slower response speed; the chamber pressure P_chamber is adjusted to 0.21MPa (temperature and pressure coordination), and data after control is collected in real time. Feedback after control shows that the temperature of zone 3 gradually rises to 147.2~148.5℃, the deviation value is reduced to 0.5~1.8℃, and the first-level warning is automatically lifted.

[0067] As can be seen, under this operating condition, the system successfully compensated for the aging and wear of the heating module and the heat dissipation effect of high dust, accurately corrected the serious deviation of a single zone and the slight deviation of other zones, and finally achieved a temperature field deviation correction rate of 99.1% and a lamination defect rate of 0.05%, realizing stable production under the aging condition of the equipment. Only a first-level warning was triggered throughout the process, and the control could be completed without manual intervention.

[0068] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0069] The above embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of protection of the present invention. Any non-substantial changes and substitutions made by those skilled in the art based on the present invention shall fall within the scope of protection claimed by the present invention.

Claims

1. A photovoltaic module temperature field deviation adaptive system based on AI algorithm, characterized in that, include: The data acquisition module is configured to collect multi-dimensional data during the lamination process; The data preprocessing module is used to perform noise reduction, outlier removal, and data range standardization on the collected raw data to generate preprocessed data. The AI ​​algorithm processing module includes: Temperature field deviation identification model: A convolutional neural network is used to process the preprocessed temperature data, construct a temperature field distribution feature map, identify temperature deviations in each zone, and classify them according to the severity of the deviations; Control parameter correction model: Based on long short-term memory network, the control parameters are adaptively corrected according to the deviation identification results, current component specifications, environmental conditions and equipment operating status. Model self-iterative optimization module: adopts reinforcement learning Q-learning mechanism, and dynamically iteratively optimizes temperature field deviation identification model and control parameter correction model based on defect data and control effect data fed back from subsequent processes; Anomaly warning and emergency decision-making submodule: Based on logistic regression algorithm, it identifies sudden anomalies in the temperature field, determines the cause of the anomaly, triggers graded warnings, and executes emergency control measures; Control module: Based on the control instructions output by the AI ​​algorithm processing module, the heating module and pressure regulating valve of the laminator are driven to make adjustments. At the same time, the execution results are fed back to the AI ​​algorithm processing module in real time to form a closed-loop control, so as to realize the adaptive control of temperature field deviation during the lamination process of photovoltaic modules.

2. The system according to claim 1, characterized in that, Multi-dimensional data includes: Temperature data includes the temperatures of at least 12 independently controlled temperature zones of the laminator heating plate, each zone having at least 3 temperature measuring points; the temperatures of the center and at least 4 edge temperature measuring points on the component surface; and the ambient temperatures of the upper and lower chambers of the laminator. Equipment operating data includes real-time power, current, and voltage of each zone heating module, real-time pressure of the laminator chamber, operating time of the heating modules, and cumulative operating mileage of the equipment; Component specifications; Workshop environmental data, including real-time temperature, humidity, and dust concentration.

3. The system according to claim 1, characterized in that, Noise reduction processing includes the following steps: Data prediction: Based on the accurate data after noise reduction processing at the previous moment. Control quantity at the previous moment Estimate the reasonable data value for the current moment. , represented as: Among them, coefficient A The value is 1, and the coefficient is... B The value is 0. This is the control quantity of the laminator at the previous moment; Predictive uncertainty estimation: based on the predictive uncertainty of the previous moment Calculate the uncertainty of the predicted data at the current moment. , is represented as; in, Q For process noise covariance; Weight calculation: based on the uncertainty of prediction at the current moment Calculate the data weight at the current moment. , represented as: in, H For the observation matrix, R To determine the value of the noise covariance; Data correction: Combine the predicted reasonable data values ​​at the current moment. The raw data directly collected by the sensor at the current moment and data weights Output the accurate data after noise reduction at the current moment. ， Represented as: in, These are the raw data collected by sensors such as temperature, power, and pressure at the current moment. The observed values ​​are for predicting the data.

4. The system according to claim 3, characterized in that, Noise reduction processing also includes an uncertainty update step: Data weights based on the current moment Observation matrix H and the uncertainty of the prediction at the previous moment Calculate the uncertainty of the data after noise reduction at the current moment. , represented as: in, I It is the identity matrix. This uncertainty value is used to quantify the reliability of the data after noise reduction at the current moment, and will be used as the input parameter for the uncertainty estimation step at the next moment, so as to realize the dynamic transmission and iterative update of uncertainty during the noise reduction process.

5. The system according to claim 1, characterized in that, Outlier removal includes the following steps: Data distribution characteristic calculation: For each type of denoised dataset, calculate the average value of the data. μ and degree of dispersion σ ,in: average value μ Calculated using the following formula: in, For a single denoised data point in this type of dataset, n The number of samples in this type of dataset. μ Used to reflect the overall level of data; Dispersion σ Calculated using the following formula: in, σ Used to reflect the range of data fluctuations; Outlier detection and removal: Set the criterion for extreme erroneous data as a single data entry. Compared with the average μ The absolute value of the difference is greater than 3 times the degree of dispersion σ When a piece of data meets the data discrimination criteria, it is directly judged as an extreme error and removed.

6. The system according to claim 1, characterized in that, Unified data scope processing includes the following steps: For each type of valid data after noise reduction and removal of outlier extreme data, determine the minimum value for that type of data. and maximum value ;in, This represents the statistical minimum value of this type of valid data. This represents the maximum statistical value of this type of valid data. For each valid data x The standard data after unification is calculated using the following formula. : Standard data output: As a standardized data output with a unified scope, it is directly used as the input feature for the AI ​​algorithm processing module.

7. The system according to any one of claims 1 to 6, characterized in that, The temperature field deviation identification model is configured to perform the following operations: Based on the preprocessed temperature data output by the data preprocessing module, a temperature field distribution characteristic map of each independent temperature control zone of the laminator heating plate is constructed. A convolutional neural network is used to extract and analyze the features of the temperature field distribution map, and to identify the temperature deviation value of each independent temperature control zone. The temperature deviation value is calculated as the difference between the real-time temperature of the corresponding zone and the standard temperature of the zone. The standard temperature is preset based on the specifications of the photovoltaic module and the type of lamination film; Based on the absolute value of the temperature deviation, the temperature deviation of each zone is divided into three severity levels: slight deviation, moderate deviation, and severe deviation. The final output includes the temperature deviation value, deviation level, and deviation location information for each independent temperature control zone, which is the location identifier of the corresponding independent temperature control zone.

8. The system according to any one of claims 1 to 6, characterized in that, The model self-iterative optimization module is configured to perform the following operations: Obtain defect data and control effect data from subsequent processes. Use the current temperature field deviation characteristics, component specification parameters, and environmental operating conditions as the state space, the adjustment amount of the convolution kernel parameters of the temperature field deviation identification model and the adjustment amount of the hidden layer weights of the control parameter correction model as the action space, and the reduction of defect rate and the degree of temperature deviation reduction after control as the reward function to construct the Q table or deep Q network of the reinforcement learning Q-learning algorithm. By iteratively updating the Q value, the convolution kernel parameters and control parameters of the temperature field deviation identification model are dynamically adjusted to correct the hidden layer weights of the model, thereby achieving adaptive optimization of the model for the lamination scenario. Dynamic updates of model parameters are achieved using a dual-mode approach: real-time iteration and batch iteration. After a single control operation is completed, based on the real-time effect data of that control, the local convolution kernel parameters of the temperature field deviation identification model or the short-term memory weights of the control parameter correction model are fine-tuned to achieve a rapid response of the model to the effect of a single control. The system automatically summarizes the control effect data, defect data, environmental condition data, and component specification data of all lamination processes during the preset time period each day. It then uses the Q-learning algorithm to batch optimize the global convolution kernel parameters of the temperature field deviation identification model and the long-term memory weights of the control parameter correction model.

9. The system according to claim 8, characterized in that: In the model self-iterative optimization module, the combined index of temperature field deviation correction rate and lamination defect rate is used as the reward function of reinforcement learning Q-learning algorithm. The temperature field deviation correction rate is the proportion of independent temperature control zones whose temperature deviation value is reduced to the target range after regulation; the lamination defect rate is the ratio of the number of defects in the laminated components to the total number of components. When the temperature field deviation correction rate under a certain operating condition is found to be less than the first preset value or the lamination defect rate is found to be greater than the second preset value, the following operation will be automatically triggered: The full-dimensional dataset under this working condition is extracted as a new training sample. The full-dimensional dataset includes: preprocessed temperature field distribution data, component specification parameters, environmental working condition data, control parameter instructions and corresponding subsequent process defect data. Add the new training samples to the training sets of the CNN bias recognition model and the LSTM parameter adjustment model, re-execute the iterative optimization process of the Q-learning algorithm, and update the convolution kernel parameters of the CNN bias recognition model and the hidden layer weights of the LSTM parameter adjustment model.

10. The system according to any one of claims 1 to 6, characterized in that, The control module includes: The execution equipment group includes 12 independent zone heating modules for the laminator, pressure regulating valves, backup sensors, audible and visual early warning devices, and operation and maintenance terminals; The execution control unit includes an edge controller and a PLC controller. The edge controller is configured to receive control commands output by the AI ​​algorithm processing module. The control commands include the power adjustment values ​​and temperature field deviation compensation coefficients of each independent zone heating module. The PLC controller is communicatively connected to the edge controller and is configured to drive the corresponding independent zone heating module to adjust its power according to the control commands, and synchronously adjust the pressure regulating valve of the corresponding area to achieve temperature and pressure coordinated control of heating power and chamber pressure. The data feedback unit is configured to collect temperature data and equipment operation data in real time during the execution of control commands, and feed the collected data back to the AI ​​algorithm processing module for the verification of deviation correction effect and the self-iterative optimization of temperature field deviation identification model and control parameter correction model.

Images