Production parameter regulation system for reflective solder strip suitable for n-type heterojunction cell

Abstract

This invention relates to the field of photovoltaic cell module manufacturing technology, specifically to a system for controlling production parameters of reflective solder strips adapted for N-type heterojunction cells. The system includes a data acquisition module, a state modeling module, a target generation module, an optimization control module, and an execution module. The data acquisition module collects real-time process data such as solder strip substrate temperature, coating thickness, reflective layer coating speed, and tension roller pressure. The state modeling module constructs a production state evolution model based on this data, characterizing the dynamic coupling relationships between multiple parameters. The target generation module matches the model output with the cell performance specifications to generate a production target vector. The optimization control module inputs the target vector and real-time data into an improved model predictive control algorithm. This algorithm introduces an interior point penalty function based on the model gradient to handle coupling constraints during rolling optimization, outputting the setpoints for solder strip substrate temperature, coating current, coating pump speed, and tension. The execution module sends the setpoints for execution.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic cell module manufacturing technology, and in particular to a reflective solder ribbon production parameter control system adapted to N-type heterojunction cells. Background Technology

[0002] In photovoltaic module manufacturing, the application of reflective solder ribbons can effectively improve the light utilization efficiency of the cells. N-type heterojunction cells have more stringent requirements for solder ribbon performance; its reflectivity, surface morphology, and coating thickness must be precisely matched to the cell structure to achieve optimal optical gain. Existing reflective solder ribbon production lines are typically equipped with independent controllers for temperature, tension, coating current, and coating speed. Each process parameter is set and adjusted independently by operators based on offline sampling test results. This control method relies on experience, treating the continuous production process as a combination of multiple static stages.

[0003] The existing production technology has significant shortcomings. Due to the strong dynamic coupling between the solder strip substrate temperature, coating deposition rate, reflective layer coating leveling, and production line tension, adjusting any parameter will trigger a chain reaction of changes in other state variables. For example, increasing tension to improve production line speed may alter the substrate's heat dissipation conditions, affecting coating crystallization and causing uneven wet film coating thickness. Existing independent control loops cannot detect and compensate for this coupling interference, leading to increased fluctuations in coating thickness and surface roughness under high-speed production, and large dispersion in reflectivity indicators. This instability in production conditions makes it difficult to meet the high standards of N-type heterojunction solar cells in terms of product performance consistency, resulting in a bottleneck in the yield of high-quality products.

[0004] To stably produce high-performance reflective solder ribbon, two fundamental, interrelated problems need to be addressed. First, how to accurately characterize and predict the real-time interaction and dynamic evolution of multiple key parameters such as substrate temperature, coating thickness, coating speed, and tension / pressure during production—this is a prerequisite for precise control. Second, based on this, how to collaboratively calculate a set of optimal setpoints within each control cycle, ensuring that multiple conflicting process objectives (such as desired coating thickness, required surface roughness, and reflectivity) are simultaneously met under complex process constraints, and guaranteeing stable and smooth execution of the control system. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a reflective ribbon production parameter control system adapted to N-type heterojunction solar cells.

[0006] To achieve the above objectives, the present invention employs the following technical solution: a reflective solder ribbon production parameter control system adapted to N-type heterojunction solar cells, comprising: The data acquisition module collects real-time process data during the production of reflective welding strips. The real-time process data includes online measurements of the welding strip substrate temperature, coating thickness, reflective layer coating speed, and tension roller pressure. The state modeling module, based on the real-time process data, constructs a production state evolution model that reflects the dynamic coupling relationship between the substrate temperature of the welding strip, the coating thickness, the coating speed of the reflective layer, and the pressure of the tension roller. The target generation module matches the real-time output of the production state evolution model with the preset performance specifications of the N-type heterojunction battery for the reflective welding ribbon, and generates a production target vector that includes the target coating thickness range, target surface roughness and target reflectivity. The optimized control module inputs the production target vector and the real-time process data into the improved model predictive control algorithm. The improved model predictive control algorithm introduces an interior point penalty function based on the gradient information of the production state evolution model during the rolling optimization process to handle the coupling constraint between the coating thickness and the reflective layer coating speed, and outputs the set values ​​of the solder strip substrate temperature, coating current, coating pump speed, and tension. The execution module sends the set values ​​of the welding strip substrate temperature, coating current, coating pump speed, and tension to the corresponding production equipment actuators.

[0007] As a further aspect of the present invention, the acquisition of real-time process data during the production of reflective welding strips includes: The surface temperature of the welding strip at the annealing furnace outlet is collected by an infrared thermometer and used as the welding strip substrate temperature. The online X-ray fluorescence thickness gauge was used to measure the coating thickness on the surface of the solder strip after the coating process. The roller rotation speed of the reflective coating machine is measured by an encoder, and the coating speed of the reflective layer is calculated by combining the known roller diameter. The positive pressure applied to the welding strip by the tension regulating roller is measured by a pressure sensor and is taken as the tension roller pressure.

[0008] As a further aspect of the present invention, the construction of a production state evolution model reflecting the dynamic coupling relationship between the substrate temperature of the welding strip, the coating thickness, the reflective layer coating speed, and the tension roller pressure includes: The online measurements of the welding strip substrate temperature, coating thickness, reflective layer coating speed, and tension roller pressure were used as state variables. The control inputs of the production equipment are used as input variables, including heater power, plating tank current, coating pump speed and tension motor torque; Based on historical production data, the recursive least squares method with a forgetting factor is used to identify the coefficients of the multi-input multi-output state space equation between state variables and input variables online. The output of the state space equation is the predicted value of the state variable at the next time step, and the state space equation is the production state evolution model.

[0009] As a further aspect of the present invention, the online identification of the coefficients of the multi-input multi-output state-space equation between state variables and input variables using the recursive least squares method with a forgetting factor includes: Initialize the parameter matrix and covariance matrix of the state-space equation; In each sampling period, the control input from the previous moment is combined with the observed state variables at the current moment to form an augmented observation vector; Based on the augmented observation vector and the forgetting factor, the parameter matrix and covariance matrix are recursively updated. The updated parameter matrix is ​​used to construct the production state evolution model at the current moment.

[0010] As a further aspect of the present invention, the improved model predictive control algorithm introduces an interior point penalty function based on the gradient information of the production state evolution model during the rolling optimization process to handle the coupling constraint between the coating thickness and the reflective layer coating speed, including: In each rolling optimization time domain of the model predictive control algorithm, an objective function is defined that includes the output tracking error and the change in control increment. An inner point penalty function term is added to the objective function. The inner point penalty function is a function of the predicted coating thickness and the predicted reflective layer coating speed. The specific form of the interior point penalty function is as follows: when the combination of the predicted coating thickness and the predicted coating speed of the reflective layer violates the preset coupling constraint, the interior point penalty function value increases sharply according to the gradient information of the corresponding prediction point based on the production state evolution model, forcing the optimization search path to remain within the feasible region defined by the coupling constraint. By combining the objective function with an interior point penalty function term with the production state evolution model, a quadratic programming problem with inequality constraints is constructed and solved to obtain the optimal control input sequence. The first term of the optimal control input sequence is output as the setpoint for the welding strip substrate temperature, the setpoint for the coating current, the setpoint for the coating pump speed, and the setpoint for the tension.

[0011] As a further aspect of the present invention, the preset performance specifications for the reflective solder ribbon of the N-type heterojunction solar cell include the target coating thickness range, target surface roughness, and target reflectivity. The production target vector and the real-time process data are jointly input into the improved model predictive control algorithm, including: The target coating thickness range is transformed into upper and lower bound constraints on the online measurement values ​​of the coating thickness. The target surface roughness and target reflectivity are converted into setting ranges for the coating speed of the reflective layer and setting ranges for the pressure of the tension roller, respectively, through a pre-calibrated process mapping table. The constraint range is compared with the predicted values ​​of state variables obtained from the production state evolution model, and the deviation is used as part of the output tracking error in the model predictive control algorithm.

[0012] As a further aspect of the present invention, the step of combining the objective function with an interior point penalty function term with a production state evolution model to construct and solve a quadratic programming problem with inequality constraints to obtain the optimal control input sequence includes: Based on the production target vector and the predicted values ​​of future time-domain state variables predicted by the production state evolution model, a quadratic objective function is defined that includes the tracking error of the output variable and the increment of the control input. An interior point penalty function term is introduced into the quadratic objective function to construct an augmented objective function. The interior point penalty function term dynamically adjusts the penalty function value according to the distance between the predicted coating thickness and the predicted reflective layer coating speed relative to the preset coupling constraint boundary. The augmented objective function, the production state evolution model, and the inequality constraints for the control input and state variables are collectively expressed as the mathematical form of a standard quadratic programming problem. The standard quadratic programming problem is solved using the effective set method or the interior point method to obtain a set of optimal control input sequences in the future control time domain. The first control quantity corresponding to the control time domain is extracted from the optimal control input sequence and output as the strip substrate temperature setting value, coating current setting value, coating pump speed setting value and tension setting value to the execution and dispatch module.

[0013] As a further aspect of the present invention, the setting values ​​for the solder strip substrate temperature, coating current, coating pump speed, and tension are sent to the corresponding production equipment actuators, including: The temperature setpoint of the welding strip substrate is sent to the heater temperature controller of the annealing furnace through the analog output module; The coating current setting value is sent to the rectified power supply of the electroplating tank through the current regulator. The coating pump speed setting value is sent to the metering pump frequency converter of the reflective layer coating machine through the digital output module; The tension setpoint is sent to the servo motor of the tension control roller via a servo driver.

[0014] As a further aspect of the present invention, the method of sending the welding strip substrate temperature setting value, coating current setting value, coating pump speed setting value, and tension setting value to the corresponding production equipment actuator also includes: Monitor the actual feedback values ​​of each actuator, including the actual temperature of the heater, the actual current of the rectifier power supply, the actual speed of the metering pump, and the actual torque of the servo motor; The actual feedback value is compared with the corresponding set value. If the deviation continues to exceed the execution tolerance and reaches the warning time, an actuator fault alarm is triggered, the output of the model predictive control algorithm is frozen, and the system switches to the preset safe manual control mode.

[0015] As a further aspect of the present invention, the system further includes: The closed-loop update module is used to acquire new real-time process data generated after actuator adjustments and input it into the production state evolution model to update the rolling optimization process of the improved model predictive control algorithm in a closed loop. Specifically, it includes: At the next sampling moment, the new solder strip substrate temperature, the new online measurement value of the coating thickness, the new reflective layer coating speed and the new tension roller pressure are collected again by the sensor as new real-time process data. The new real-time process data is used as the state variable observations of the production state evolution model to update the initial state of the production state evolution model; Based on the updated production state evolution model, the improved model predictive control algorithm is re-executed in the next rolling optimization time domain to generate and issue a new setpoint.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: By constructing a production state evolution model that reflects the dynamic coupling relationship between the substrate temperature of the solder strip, the coating thickness, the coating speed of the reflective layer, and the pressure of the tension roller, the interactive influence of multiple key physical quantities on the production line is mathematically and explicitly represented. This model receives process data in real time and can map the dynamic trajectory of parameter adjustments on intermediate states such as coating growth and coating morphology over a future period. This changes the past decision-making model that relied on static experience or isolated parameter feedback, enabling the control system to predict the potential impact of current control actions on subsequent process steps and the performance of the final product based on model predictions. This provides a dynamic and forward-looking process state basis for multi-variable collaborative decision-making.

[0017] In the rolling optimization stage of the model predictive control algorithm, an interior point penalty function based on the gradient information of the production state evolution model is introduced. When dealing with constraints with strong coupling relationships, such as "coating thickness" and "reflective layer coating speed," this method can proactively guide the optimization search direction towards the feasible region using model gradient information, effectively avoiding optimization oscillations or infeasibility caused by coupled constraints. This enables the optimization controller to calculate a set of process setpoints more robustly and smoothly in each control cycle. While satisfying multi-objective performance vectors, it automatically coordinates the rate matching between coating deposition and coating processes, ensuring the stability and executability of setpoint adjustments, and suppressing production process fluctuations caused by drastic changes in setpoints from the source of optimization decisions. Attached Figure Description

[0018] Figure 1 This is a timing diagram of the reflective solder ribbon production parameter control system adapted to N-type heterojunction solar cells described in this invention. Figure 2 A workflow diagram for building a production state evolution model for the state modeling module; Figure 3 A flowchart illustrating the workflow for handling coupled constraints in an improved model predictive control algorithm. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 This invention provides a reflective solder ribbon production parameter control system adapted to N-type heterojunction solar cells, specifically comprising: The system comprises a data acquisition module, a state modeling module, a target generation module, an optimization control module, and an execution deployment module. The data acquisition module collects real-time process data during the production of reflective welding strips. This real-time process data specifically includes online measurements of the welding strip substrate temperature, coating thickness, reflective layer coating speed, and tension roller pressure. Based on the collected real-time process data, the state modeling module constructs a production state evolution model that reflects the dynamic coupling relationship between the welding strip substrate temperature, coating thickness, reflective layer coating speed, and tension roller pressure. The target generation module matches the real-time output of the production state evolution model with the preset performance specifications of the reflective welding strip for N-type heterojunction solar cells, generating a quantified production target vector. This vector includes the target coating thickness range, target surface roughness, and target reflectivity. The optimization control module inputs the production target vector and the real-time process data into an improved model predictive control algorithm. This algorithm introduces an interior point penalty function based on the gradient information of the production state evolution model during the rolling optimization process. This function specifically addresses the coupling constraint between the coating thickness and the reflective layer coating speed, and ultimately outputs a set of optimized setpoints, including the solder strip substrate temperature setpoint, coating current setpoint, coating pump speed setpoint, and tension setpoint. The execution and distribution module then distributes these optimized setpoints to the corresponding production equipment actuators, driving their actions to adjust the production process.

[0022] In one embodiment of the present invention, real-time process data acquisition during the production of reflective welding ribbon is accomplished in the following manner, see reference. Figure 2 The surface temperature of the welding strip at the annealing furnace outlet is collected using an infrared thermometer; this surface temperature is used as the welding strip substrate temperature. The online X-ray fluorescence thickness gauge measures the coating thickness on the welding strip surface after the coating process. An encoder measures the roller speed of the reflective coating machine, and the reflective coating speed is calculated based on the known roller diameter. A pressure sensor measures the positive pressure applied to the welding strip by the tension regulating roller; this positive pressure is used as the tension roller pressure. In some embodiments, a production state evolution model reflecting the dynamic coupling relationship between the welding strip substrate temperature, coating thickness, reflective coating speed, and tension roller pressure is constructed. The online measured values ​​of the welding strip substrate temperature, coating thickness, reflective coating speed, and tension roller pressure are used as state variables, and the control inputs of the production equipment are used as input variables. The control inputs include heater power, plating tank current, coating pump speed, and tension motor torque. In practical implementation, based on historical production data, a recursive least squares method with a forgetting factor is used to identify the coefficients of the multi-input multi-output state-space equation between state variables and input variables online. The parameter matrix and covariance matrix of the state-space equation are initialized. In each sampling period, the control input from the previous time step and the observed values ​​of the state variables at the current time step are combined to form an augmented observation vector. Based on the augmented observation vector and the forgetting factor, the parameter matrix and covariance matrix are recursively updated. Optionally, the parameter matrix update formula is expressed as follows:

[0023] in: This represents the updated parameter matrix at the current time. This represents the parameter matrix from the previous time step. This represents the updated gain matrix calculated at the current time. This represents the vector of state variables observed at the current moment. This represents the augmented observation vector composed of historical input and output data. This represents the transpose operation of a vector or matrix. The updated parameter matrix is ​​used to construct the current production state evolution model, and the output of the production state evolution model is the predicted value of the state variables for the next time step. It can be understood that historical production data comes from past production batch records stored in the database, and the forgetting factor is set between 0.95 and 0.99. It can be understood that through online identification using recursive least squares with a forgetting factor, the production state evolution model can continuously reflect the real-time coupling relationships in the reflective welding strip production process.

[0024] In the specific implementation process, online identification using recursive least squares with a forgetting factor is the core step in realizing the dynamic updating of the production state evolution model. This method differs from conventional offline modeling or fixed parameter models. The uniqueness of this project lies in the deep integration of the algorithm's recursive process with the real-time data stream of the production line. The implementation process begins with an initialization phase, requiring the setting of appropriate initial values ​​for the parameter matrix and covariance matrix. These initial values ​​can be initially estimated using batch least squares based on historical stable production data, thus providing a reasonable starting point for online recursion. Once the production line starts, the system enters each fixed sampling cycle. Within each sampling cycle, the system synchronously captures the current state variable observations obtained from sensors, including the solder strip substrate temperature, online measurement of coating thickness, reflective layer coating speed, and tension roller pressure. Simultaneously, it reads the control inputs issued and executed in the previous sampling cycle, including heater power, plating tank current, coating pump speed, and tension motor torque. These latest data are immediately combined into an augmented observation vector, which reflects the immediate information on the impact of inputs on the system state within the most recent control cycle. The recursive update is triggered, and the algorithm uses this new augmented observation vector and a preset forgetting factor to mathematically update the parameter matrix and covariance matrix. The forgetting factor exponentially decays the influence of historical data, giving higher weight to new data. This allows the model parameters to gradually "forget" earlier data that may no longer reflect the current process conditions, ensuring that the production state evolution model always tracks the latest dynamic characteristics of the production line, such as the response speed of coating thickness to current changes or the real-time relationship between temperature and coating effect. After the update is complete, the new parameter matrix defines the production state evolution model available at the current moment. This model will be immediately used for state prediction in the next sampling cycle. The entire identification process is automatically and cyclically executed in each operation cycle of the control system, forming a continuously self-adjusting mathematical model learning process embedded in the real-time control loop.

[0025] In terms of the specific form of the model construction, a multi-input multi-output state-space equation is adopted as the mathematical carrier of the production state evolution model. This choice is to accurately characterize the cross-coupling relationships between multiple variables in the production process. The approach of this project is to incorporate all key measurable states and control inputs into a unified dynamic framework. The implementation process requires defining a state variable vector, which consists of four physical quantities that directly characterize the production state: the online measurement of the welding strip substrate temperature, the online measurement of the coating thickness, the reflective layer coating speed, and the tension roller pressure. Simultaneously, an input variable vector is defined, whose elements are four directly adjustable control quantities: heater power, plating tank current, coating pump speed, and tension motor torque. The structure of the state-space equation describes how the control input and state variables at the current moment jointly affect the state variables at the next moment. Its core consists of the system matrix to be identified, the control input matrix, and the possible direct transmission matrix. During the online identification process, the parameter matrix updated by the recursive least squares method with a forgetting factor contains elements of all matrices in these state-space equations. For example, the updated parameter matrix implicitly defines how minute changes in the plating bath current affect the plating thickness several sampling periods later, and also defines how adjustments to the tension motor torque dynamically influence the reflective layer coating speed and the solder strip substrate temperature. Through this multi-input multi-output model structure, the optimization control algorithm, when calculating the heater power setpoint, can simultaneously predict the indirect impact of this adjustment on plating thickness and coating speed, thus making coordinated decisions rather than viewing individual control loops in isolation. This modeling approach, treating the production line as a multivariable coupled dynamic system, is the foundation for achieving high-precision coordinated control in this project. It enables the model predictive control algorithm to perform optimization calculations on a predictive model that considers all intrinsic correlations, thereby outputting a set of mutually matching control setpoints.

[0026] In one embodiment of the present invention, see [reference] Figure 3The improved model predictive control algorithm has a specific construction in each rolling optimization time domain. It defines an objective function that includes output tracking error and control increment change. The output tracking error refers to the deviation between the future state variables predicted by the production state evolution model and the set trajectory transformed from the production target vector. The control increment change refers to the magnitude of the change in the control input within adjacent control cycles. An interior penalty function term is added to the objective function. The interior penalty function is a function of the predicted coating thickness and the predicted reflective layer coating speed. In some embodiments, the specific form of the interior penalty function is designed such that when the combination of the predicted coating thickness and the predicted reflective layer coating speed violates a preset coupling constraint, the interior penalty function value increases sharply based on the gradient information of the production state evolution model at the corresponding prediction point. The gradient information reflects the sensitive relationship between the coating thickness and the reflective layer coating speed, forcing the optimization search path to remain within the feasible region defined by the coupling constraint. Optionally, a specific mathematical expression for the interior penalty function can be written as:

[0027] in: This represents the sequence of predicted values ​​based on coating thickness across all prediction steps. and the predicted value sequence of reflective layer coating speed Calculate the total penalty function value. It is a penalty parameter weight coefficient that is greater than zero. This represents the total number of prediction steps in the model predictive control algorithm. Indicates the first Predicted coating thickness values ​​for each prediction step size Indicates the first Predicted value of reflective layer coating speed for each predicted step size. It is a description of the first The dimensionless function representing the position of a predicted point relative to the coupled constraint boundary always has an output value greater than zero. The closer the output value is to zero, the closer the predicted point is to the constraint boundary. This represents the natural logarithm operation. It can be understood as a function. The predicted values ​​for coating thickness and reflective layer coating speed are constructed by normalizing their distances to their respective constraint boundaries, thus ensuring that the output is a dimensionless pure numerical value, making the natural logarithm calculation meaningful. When the optimized path drives the predicted point closer to the constraint boundary, The value approaches zero, causing its logarithmic value to approach negative infinity, but due to the preceding negative sign, the entire penalty function value... This will tend towards positive infinity, thus generating a large penalty term in the objective function, effectively preventing the optimization algorithm from choosing a control strategy that violates or closely follows the constraint boundaries. The interior point penalty function term, together with the output tracking error and the control increment change term, constitutes the augmented objective function, which is used for subsequent optimization solutions.

[0028] In practical implementation, the pre-defined performance specifications of the N-type heterojunction solar cell for the reflective ribbon are transformed into constraints that can be handled by the model predictive control algorithm. These performance specifications include the target coating thickness range, target surface roughness, and target reflectivity. The target coating thickness range is directly converted into upper and lower bound constraints on the online measurement value of the coating thickness. For example, when the specification requires a coating thickness of 10 to 12 micrometers, the corresponding hard or soft constraint boundary for the predicted coating thickness is set in the model predictive control algorithm. The target surface roughness and target reflectivity are respectively converted into setting ranges for the reflective layer coating speed and tension roller pressure through a pre-calibrated process mapping table. The process mapping table is a data table establishing the correspondence between the reflective layer coating speed, tension roller pressure, and the final surface roughness and reflectivity of the ribbon product through numerous process experiments. In some embodiments, the converted reflective layer coating speed setting range and tension roller pressure setting range, along with the corresponding predicted values ​​of state variables obtained from the production state evolution model, are sent to the optimization control module. The optimization control module calculates the deviation between the predicted values ​​of the state variables and the boundary of the setting range. This deviation is used as part of the output tracking error in the model predictive control algorithm to participate in the calculation of the objective function. It can be understood that by inversely transforming final product performance indicators such as surface roughness and reflectivity into intermediate control targets for key state variables in the production process through a process mapping table, the improved model predictive control algorithm can perform real-time and forward-looking regulation during the production process. Optionally, the process mapping table is stored in the control system in the form of a two-dimensional lookup table or a fitting function, and is called and interpolated each time a new target surface roughness and target reflectivity command is received.

[0029] In one embodiment of the present invention, the improved model predictive control algorithm constructs and solves a quadratic programming problem following explicit steps. Based on the production target vector and the predicted values ​​of state variables in the future time domain predicted by the production state evolution model, a quadratic objective function is defined, including the output variable tracking error and the control input increment. The output variable tracking error is the weighted sum of squares of the deviations between the predicted values ​​of state variables in the future prediction time domain and the setpoint or set trajectory transformed from the production target vector. The control input increment is the weighted sum of squares of the changes in control quantity between adjacent control cycles in the future control time domain. An interior penalty function term is introduced into the quadratic objective function to construct an augmented objective function. The interior penalty function term dynamically adjusts the penalty function value according to the distance between the predicted values ​​of the coating thickness and the predicted values ​​of the reflective layer coating speed and the preset coupling constraint boundary. The closer to the constraint boundary, the larger the penalty value generated by the interior penalty function term. In some embodiments, the augmented objective function, the production state evolution model, and the inequality constraints for the control input and state variables are jointly expressed as the mathematical form of a standard quadratic programming problem. The standard quadratic programming problem can be mathematically formulated as solving for a set of future control input sequences that minimizes a quadratic objective function while satisfying a series of linear equality and inequality constraints. The matrix and vector dimensions used to describe this problem are shown in Table 1.

[0030] Table 1: Relevant Matrix and Vector Dimensions for the Quadratic Programming Problem

[0031] In practice, constructing the quadratic programming problem involves specific matrix filling and controlling the input increment sequence. Includes from the present moment to the future All control input changes within a control step, the number of control inputs is: The objective function is typically in the form of:

[0032] in: It is the scalar value of the augmented objective function to be minimized. The future predicted by the production state evolution model The output variable vector of each prediction step size It is the output reference trajectory vector corresponding to the prediction step size. It is a weighted diagonal matrix of output errors. It is a weighted diagonal matrix that controls the input increment. This refers to the interior point penalty function term defined in the above embodiment. It can be understood that the equality constraint... Derived from the production state evolution model, it controls the input increment sequence. Output prediction They are dynamically linked through the model. Inequality constraints. This includes physical limits on the control input itself, as well as boundary constraints on state variables such as coating thickness and reflective layer coating speed. Optional, a weighting matrix... and The diagonal elements are set based on the relative importance of each output variable and control input component, and the expected adjustment range. The standard quadratic programming problem is solved using the effective set method or interior point method to obtain a set of optimal control input sequences in the future control time domain. The first control quantity corresponding to the control time domain is extracted from the optimal control input sequence; this control quantity is then output to the execution module as the setpoints for the solder strip substrate temperature, coating current, coating pump speed, and tension. It can be understood that the model predictive control algorithm re-solves the quadratic programming problem in each control cycle, achieving rolling optimization based on the latest production status.

[0033] In the specific engineering implementation of model predictive control algorithms, the weighted diagonal matrix and The setting of the diagonal elements is not a simple assignment based on experience, but relies on a structured initialization and automatic tuning process. Engineers, based on their understanding of the priority of each control objective in the production process, assign a weighted diagonal matrix to the output variable tracking error. The diagonal elements are assigned initial values; for example, the weight of the coating thickness tracking error may be higher than the weight of the reflective layer coating speed tracking error, reflecting the higher requirements for coating thickness control accuracy. Control input increment weighted diagonal matrix The initial values ​​are set based on the physical characteristics and allowable range of motion of each actuator. For example, the adjustment inertia of heater power is relatively large, so its corresponding control increment weight is set higher to suppress its drastic fluctuations. More crucial is the subsequent automatic tuning process. The system runs for a period of time under the initial weights, recording the actual output variable tracking error sequence and each control input increment sequence. An offline analysis program is then started. This program calculates the statistical variance of each output error channel and each control increment channel over the most recent running time, and scales the initial weights proportionally based on the variance. Channels with larger output variable tracking error variances have their corresponding weights appropriately reduced to prevent a slightly less performing variable from dominating the entire objective function at the expense of other variables' performance. Channels with excessively large control input increment variances have their corresponding weights increased to further smooth the control signal. Weighted diagonal matrix and The final value is the result of this series of statistical analyses based on process data. These results enable terms of different physical dimensions and orders of magnitude in the objective function to be reasonably integrated into a unified scalar cost for minimization.

[0034] The interior penalty function term is integrated into the solution process of the quadratic programming problem. This involves transforming the non-quadratic interior penalty function into a form that can be handled by the quadratic programming framework. This process is achieved through the idea of ​​sequential quadratic programming. During rolling optimization in each sampling period, the algorithm does not directly solve the original optimization problem containing the logarithmic interior penalty function term, but instead uses an iterative approximation method. Based on the current control input sequence guess and the state prediction sequence obtained from the production state evolution model, the optimization solver calculates the gradient vector and Hessian matrix of the interior penalty function term at the current guess point. Using these two pieces of information, it performs a second Taylor expansion approximation on the original interior penalty function term at the current position, thereby locally approximating the interior penalty function term containing the logarithmic term as a quadratic function. This quadratic approximation function is combined with the original quadratic terms of output tracking error and control increment to form a complete quadratic objective function. For this complete quadratic objective function, under the equality constraints derived from the production state evolution model and the inequality constraints of the control input and state variables, the optimization solver solves a standard quadratic programming subproblem to obtain a new control input sequence guess solution. Starting with this new solution, the gradient of the interior point penalty function and the Hessian matrix are recalculated and approximated twice. A new quadratic programming subproblem is then solved again. This process iterates rapidly several times within a sampling period until the change in the control input sequence obtained from two consecutive iterations is less than a preset tolerance threshold. The first element of the optimal control input sequence obtained from the final iteration of the quadratic programming subproblem is extracted as the setpoint to be issued at the current moment. In this way, the improved model predictive control algorithm effectively handles the coupling constraints represented by the interior point penalty function by solving a sequence quadratic programming problem through several iterations within each control period, achieving rolling optimization.

[0035] In one embodiment of the present invention, the setpoints for the solder strip substrate temperature, plating current, coating pump speed, and tension are sent to the corresponding production equipment actuators via a communication link between the control system and the underlying equipment. The setpoint for the solder strip substrate temperature is sent to the heater temperature controller of the annealing furnace via an analog output module. The analog output module converts the digital setpoint for the solder strip substrate temperature obtained from the internal calculations of the control system into a standard analog current or voltage signal. This analog signal is transmitted to the heater temperature controller of the annealing furnace. The temperature controller adjusts the power output of the heating element according to the received signal, thereby stabilizing the actual temperature of the solder strip substrate near the setpoint. The setpoint for the plating current is sent to the rectifier power supply of the electroplating tank via a current regulator. The current regulator receives the digital setpoint for the plating current, and its internal algorithm calculates a control signal based on the setpoint and the actual current value fed back by the rectifier power supply. The control signal drives the rectifier power supply to change its output current, so that the current during the electroplating process accurately follows the setpoint for the plating current. The coating pump speed setpoint is sent to the metering pump inverter of the reflective coating machine via a digital output module. The digital output module sends the coating pump speed setpoint in the form of pulse frequency or bus communication messages. The metering pump inverter parses the instruction and adjusts the output frequency, thereby controlling the speed of the metering pump motor and achieving precise adjustment of the reflective coating speed. The tension setpoint is sent to the servo motor of the tension control roller via a servo driver. The tension setpoint is usually given in the form of target torque or target position. The servo driver receives the setpoint and drives the servo motor to rotate through its built-in closed-loop control algorithm, so that the tension applied by the tension control roller to the welding strip is exactly equal to the tension setpoint. The mapping relationship between the above setpoints and actuators is shown in Table 2.

[0036] Table 2: Mapping Relationship between Control Setpoints and Actuators

[0037] In some embodiments, the actual feedback values ​​of each actuator are monitored synchronously during the setpoint issuance process. These actual feedback values ​​include the actual temperature of the heater, the actual current of the rectifier power supply, the actual speed of the metering pump, and the actual torque of the servo motor. These actual feedback values ​​are collected in real time by corresponding sensors and transmitters and fed back to the control system. The monitoring and judgment process continues, comparing the actual feedback values ​​with the corresponding setpoints and calculating the absolute deviation. If the deviation of a certain actuator channel continuously exceeds the preset execution tolerance of that channel and the duration reaches the warning time, a fault alarm for that actuator is triggered. The execution tolerance is set according to the control accuracy and process requirements of different actuators, and the warning time is used to avoid false alarms caused by momentary interference. Optionally, the deviation judgment logic can be described by the following expression:

[0038] in: Indicates the first Fault alarm trigger flag for each control channel, Indicates triggering, This indicates that it is normal. Indicates the first Control settings for each channel Indicates the first The actual feedback value of each channel Indicates the first The preset execution tolerance for each channel. This indicates the time during which the deviation continues to exceed the performance tolerance. Indicates the first Each channel has a preset warning duration. This means that when a fault alarm is triggered, the control system will freeze the output of the model predictive control algorithm, stop issuing new optimized setpoints, and switch the system control mode from automatic to a preset safe manual control mode. In safe manual control mode, each actuator will maintain the last valid setpoint before the fault occurred, or switch to a set of predefined safe process parameters until manual intervention resolves the fault. Optionally, different actuator faults can correspond to different safe manual control strategies. For example, when the heater temperature controller experiences a communication failure, the safe manual control mode might fix the heater power at a median value to maintain basic production.

[0039] The implementation of the safety manual control mode involves a pre-set, direct command issuance mechanism that operates independently of the optimization controller. When the system triggers an actuator fault alarm and decides to freeze the model predictive control algorithm output, the core processor of the control system immediately interrupts the output channel to the optimization control module and instead reads a set of pre-set safety parameters for the current fault type and process segment from an independent, non-volatile memory. This set of safety parameters is not determined by a single fixed value, but rather by a deep understanding of the production process. For example, when a continuous deviation between the coating pump speed setpoint and the actual feedback value is detected and an alarm is triggered, the system determines that the metering pump or frequency converter may be malfunctioning. In this case, the parameters invoked by the safety manual control mode may include a conservative fixed value for the coating pump speed, a tension setpoint matching the current production line speed, and conservative setpoints for the coating current and temperature required to maintain coating quality. The control system directly sends these safety parameters to each actuator through the corresponding distribution module, bypassing the optimization calculation stage of the model predictive control algorithm. During operation in the safe manual control mode, the control system continuously monitors the feedback signal of the faulty actuator. If the system detects that the actuator's feedback signal has stabilized and the deviation from the manual set value has returned to the normal tolerance range, the system will prompt the operator that the fault may have been eliminated. However, it will not automatically switch back to automatic mode. The operator must confirm and manually execute the switching command. Before switching, the system will perform a state synchronization process, reading the actual state variable values ​​of the current production line into the production state evolution model as the initial state, and restarting the rolling optimization of the optimization control module to ensure a smooth transition from manual to automatic.

[0040] The setting of the two key thresholds, execution tolerance and warning duration, stems from a comprehensive analysis of the static accuracy, dynamic response characteristics, and process quality tolerance of the production equipment. The execution tolerance setting considers the accuracy level of the measurement sensor itself and the noise level during signal transmission. For example, the measurement of the actual heater temperature may include thermocouple errors and conversion errors from the analog input module; the execution tolerance must be greater than this comprehensive measurement uncertainty to prevent false alarms. Secondly, the execution tolerance also needs to consider the sensitivity of process quality to fluctuations. For instance, small fluctuations in the coating current can significantly affect coating uniformity, so its execution tolerance will be set relatively small. Conversely, fluctuations in tension roller pressure within a reasonable range have a smaller impact on the immediate performance of the product, and its execution tolerance can be appropriately relaxed. The warning duration setting is primarily used to distinguish between persistent real faults and transient disturbances. Its determination is mainly based on statistical analysis of the equipment's dynamic response time. The control system records the statistical distribution of the time required for the actual feedback value of the corresponding actuator to stabilize to the new setpoint after each adjustment of the control setpoint in historical data. For example, the current regulation of a rectifier power supply typically stabilizes within hundreds of milliseconds, so its warning duration might be set to 1-2 seconds. For the temperature control of an annealing furnace with greater inertia, the warning duration might be set to tens of seconds or even longer. If the duration of a deviation exceeds the warning duration of that channel, it indicates that it is highly likely not a normal dynamic adjustment process, but a persistent abnormal state, thus triggering an alarm. These tolerance and duration parameters are initialized by engineers based on equipment manuals and process knowledge during the initial commissioning of the production system. After the system has been running for a period of time, fine-tuning and optimization are performed based on accumulated historical performance data and false alarm / missed alarm data, ultimately being fixed in the control system's configuration file.

[0041] In one embodiment of the present invention, the closed-loop update module continuously refreshes the rolling optimization process of the improved model predictive control algorithm through a periodic data sampling and model recalculation cycle. At the next sampling moment, the system re-acquires new online measurements of the solder strip substrate temperature, coating thickness, reflective layer coating speed, and tension roller pressure through the sensor network. These latest measurements constitute a new set of real-time process data. The new real-time process data is immediately transmitted to the state modeling module, which uses this data as the state variable observations of the production state evolution model at the current moment. It uses these observations to update the initial state of the production state evolution model, correcting the starting point for the next prediction step to the latest actual production state.

[0042] In specific implementation, the initial state of the model is updated through direct assignment. New observed strip substrate temperature values ​​are assigned to the corresponding state variables in the production state evolution model, as are new online measured coating thickness values, reflective layer coating speeds, and tension roller pressures. This ensures that the production state evolution model's predictions are based on the latest, real production line conditions, rather than previous predictions. In some embodiments, based on the updated initial state of the production state evolution model, the optimization control module re-executes the improved model predictive control algorithm in the next rolling optimization time domain. The optimization control module reads the production target vector generated or updated by the target generation module according to the latest performance specifications, combines it with the latest real-time process data, constructs and solves a new quadratic programming problem with an interior point penalty function term, and generates new setpoints for strip substrate temperature, coating current, coating pump speed, and tension. The new setpoints are transmitted to the execution module, which then sends these setpoints to the corresponding production equipment actuators, driving the heaters, rectifiers, metering pumps, and servo motors to make corresponding adjustments.

[0043] Optionally, within a complete closed-loop update cycle, the total time from data acquisition to the issuance of the new setpoint is strictly designed to be less than the equivalent time constant of the production process to meet the requirements of real-time control. It can be understood that the production state evolution model upon which the improved model predictive control algorithm relies is a version corrected for the latest data, and therefore its rolling optimization is based on the latest understanding of system dynamics. This process is automatically repeated in each sampling cycle, thus forming a continuous, adaptive closed-loop control loop throughout the entire process of reflective welding strip production. In some embodiments, the sampling cycle of the closed-loop update is consistent with the sampling cycle of the data acquisition module and the control cycle of the improved model predictive control algorithm to ensure temporal synchronization. Optionally, the mathematical relationship between state update and model prediction can be expressed as:

[0044] in: Indicates the first The state vector updated at each sampling time point, used to initiate the production state evolution model for the next prediction time domain. Indicates the first The observation vector is composed of a set of new real-time process data collected at each sampling time. This equation describes the operation of injecting the latest observation data into the model to correct its initial conditions, and is one of the core steps of closed-loop updates.

[0045] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A reflective soldering ribbon production parameter control system adapted for N-type heterojunction solar cells, characterized in that, The system includes: The data acquisition module collects real-time process data during the production of reflective welding strips. The real-time process data includes online measurements of the welding strip substrate temperature, coating thickness, reflective layer coating speed, and tension roller pressure. The state modeling module, based on the real-time process data, constructs a production state evolution model that reflects the dynamic coupling relationship between the substrate temperature of the welding strip, the coating thickness, the coating speed of the reflective layer, and the pressure of the tension roller. The target generation module matches the real-time output of the production state evolution model with the preset performance specifications of the N-type heterojunction battery for the reflective welding ribbon, and generates a production target vector that includes the target coating thickness range, target surface roughness and target reflectivity. The optimized control module inputs the production target vector and the real-time process data into the improved model predictive control algorithm. The improved model predictive control algorithm introduces an interior point penalty function based on the gradient information of the production state evolution model during the rolling optimization process to handle the coupling constraint between the coating thickness and the reflective layer coating speed, and outputs the set values ​​of the solder strip substrate temperature, coating current, coating pump speed, and tension. The execution module sends the set values ​​of the welding strip substrate temperature, coating current, coating pump speed, and tension to the corresponding production equipment actuators.

2. The reflective solder ribbon production parameter control system adapted for N-type heterojunction solar cells according to claim 1, characterized in that, The real-time process data collected during the production of reflective solder strips includes: The surface temperature of the welding strip at the annealing furnace outlet is collected by an infrared thermometer and used as the welding strip substrate temperature. The online X-ray fluorescence thickness gauge was used to measure the coating thickness on the surface of the solder strip after the coating process. The roller rotation speed of the reflective coating machine is measured by an encoder, and the coating speed of the reflective layer is calculated by combining the known roller diameter. The positive pressure applied to the welding strip by the tension regulating roller is measured by a pressure sensor and is taken as the tension roller pressure.

3. The reflective solder ribbon production parameter control system adapted for N-type heterojunction solar cells according to claim 1, characterized in that, The production state evolution model, which reflects the dynamic coupling relationship between the substrate temperature of the welding strip, the coating thickness, the coating speed of the reflective layer, and the pressure of the tension roller, includes: The online measurements of the welding strip substrate temperature, coating thickness, reflective layer coating speed, and tension roller pressure were used as state variables. The control inputs of the production equipment are used as input variables, including heater power, plating tank current, coating pump speed and tension motor torque; Based on historical production data, the recursive least squares method with a forgetting factor is used to identify the coefficients of the multi-input multi-output state space equation between state variables and input variables online. The output of the state space equation is the predicted value of the state variable at the next time step, and the state space equation is the production state evolution model.

4. The reflective solder ribbon production parameter control system adapted for N-type heterojunction solar cells according to claim 3, characterized in that, The method of online identification of the coefficients of the multi-input multi-output state-space equations between state variables and input variables using the recursive least squares method with a forgetting factor includes: Initialize the parameter matrix and covariance matrix of the state-space equation; In each sampling period, the control input from the previous moment is combined with the observed state variables at the current moment to form an augmented observation vector; Based on the augmented observation vector and the forgetting factor, the parameter matrix and covariance matrix are recursively updated. The updated parameter matrix is ​​used to construct the production state evolution model at the current moment.

5. The reflective solder ribbon production parameter control system adapted for N-type heterojunction solar cells according to claim 1, characterized in that, The improved model predictive control algorithm introduces an interior point penalty function based on the gradient information of the production state evolution model during the rolling optimization process to handle the coupling constraint between the coating thickness and the reflective layer coating speed, including: In each rolling optimization time domain of the model predictive control algorithm, an objective function is defined that includes the output tracking error and the change in control increment. An inner point penalty function term is added to the objective function. The inner point penalty function is a function of the predicted coating thickness and the predicted reflective layer coating speed. The specific form of the interior point penalty function is as follows: when the combination of the predicted coating thickness and the predicted coating speed of the reflective layer violates the preset coupling constraint, the interior point penalty function value increases sharply according to the gradient information of the corresponding prediction point based on the production state evolution model, forcing the optimization search path to remain within the feasible region defined by the coupling constraint. By combining the objective function with an interior point penalty function term with the production state evolution model, a quadratic programming problem with inequality constraints is constructed and solved to obtain the optimal control input sequence. The first term of the optimal control input sequence is output as the setpoint for the welding strip substrate temperature, the setpoint for the coating current, the setpoint for the coating pump speed, and the setpoint for the tension.

6. The reflective solder ribbon production parameter control system adapted for N-type heterojunction solar cells according to claim 5, characterized in that, The preset performance specifications for reflective solder ribbons in N-type heterojunction solar cells include target coating thickness range, target surface roughness, and target reflectivity. The production target vector and the real-time process data are input together into an improved model predictive control algorithm, including: The target coating thickness range is transformed into upper and lower bound constraints on the online measurement values ​​of the coating thickness. The target surface roughness and target reflectivity are converted into setting ranges for the coating speed of the reflective layer and setting ranges for the pressure of the tension roller, respectively, through a pre-calibrated process mapping table. The constraint range is compared with the predicted values ​​of state variables obtained from the production state evolution model, and the deviation is used as part of the output tracking error in the model predictive control algorithm.

7. The reflective solder ribbon production parameter control system adapted for N-type heterojunction solar cells according to claim 5, characterized in that, The process of combining the objective function with an interior point penalty function term with the production state evolution model to construct and solve a quadratic programming problem with inequality constraints to obtain the optimal control input sequence includes: Based on the production target vector and the predicted values ​​of future time-domain state variables predicted by the production state evolution model, a quadratic objective function is defined that includes the tracking error of the output variable and the increment of the control input. An interior point penalty function term is introduced into the quadratic objective function to construct an augmented objective function. The interior point penalty function term dynamically adjusts the penalty function value according to the distance between the predicted coating thickness and the predicted reflective layer coating speed relative to the preset coupling constraint boundary. The augmented objective function, the production state evolution model, and the inequality constraints for the control input and state variables are collectively expressed as the mathematical form of a standard quadratic programming problem. The standard quadratic programming problem is solved using the effective set method or the interior point method to obtain a set of optimal control input sequences in the future control time domain. The first control quantity corresponding to the control time domain is extracted from the optimal control input sequence and output as the strip substrate temperature setting value, coating current setting value, coating pump speed setting value and tension setting value to the execution and dispatch module.

8. The reflective solder ribbon production parameter control system adapted for N-type heterojunction solar cells according to claim 1, characterized in that, The setpoints for the solder strip substrate temperature, coating current, coating pump speed, and tension are sent to the corresponding production equipment actuators, including: The temperature setpoint of the welding strip substrate is sent to the heater temperature controller of the annealing furnace through the analog output module; The coating current setting value is sent to the rectified power supply of the electroplating tank through the current regulator. The coating pump speed setting value is sent to the metering pump frequency converter of the reflective layer coating machine through the digital output module; The tension setpoint is sent to the servo motor of the tension control roller via a servo driver.

9. The reflective solder ribbon production parameter control system adapted for N-type heterojunction solar cells according to claim 8, characterized in that, The process of sending the solder strip substrate temperature setting, coating current setting, coating pump speed setting, and tension setting to the corresponding production equipment actuators also includes: Monitor the actual feedback values ​​of each actuator, including the actual temperature of the heater, the actual current of the rectifier power supply, the actual speed of the metering pump, and the actual torque of the servo motor; The actual feedback value is compared with the corresponding set value. If the deviation continues to exceed the execution tolerance and reaches the warning time, an actuator fault alarm is triggered, the output of the model predictive control algorithm is frozen, and the system switches to the preset safe manual control mode.

10. The reflective solder ribbon production parameter control system adapted for N-type heterojunction solar cells according to claim 1, characterized in that, The system also includes: The closed-loop update module is used to acquire new real-time process data generated after actuator adjustments and input it into the production state evolution model to update the rolling optimization process of the improved model predictive control algorithm in a closed loop. Specifically, it includes: At the next sampling moment, the new solder strip substrate temperature, the new online measurement value of the coating thickness, the new reflective layer coating speed and the new tension roller pressure are collected again by the sensor as new real-time process data. The new real-time process data is used as the state variable observations of the production state evolution model to update the initial state of the production state evolution model; Based on the updated production state evolution model, the improved model predictive control algorithm is re-executed in the next rolling optimization time domain to generate and issue a new setpoint.

Images