Multi-temperature zone welding temperature intelligent control method and system based on deep learning

Abstract

This invention provides a deep learning-based intelligent control method and system for multi-temperature zone welding temperature, relating to the field of welding temperature control technology. The method includes: acquiring real-time temperature data of multiple temperature zones for feature extraction and time-series analysis, predicting temperature changes and dividing control levels, calculating temperature deviation values ​​to generate initial control parameters, iteratively optimizing the parameter set through an adaptive weighting algorithm, performing sliding window analysis on historical temperature data, and performing temperature compensation based on proportional-integral control and fuzzy rules.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of welding temperature control, in particular to a multi-temperature zone welding temperature intelligent control method and system based on deep learning. BACKGROUND

[0002] In the electronic manufacturing industry, the welding process is a key link in the production of circuit boards, and the temperature control of multi-temperature zone welding equipment directly affects the welding quality and product yield. The traditional welding temperature control mainly relies on preset temperature curves and simple feedback adjustment mechanisms. With the application of the concept of intelligent manufacturing in the circuit board welding production line, the welding process of setting temperature target values, real-time acquisition of actual temperature in the temperature zone, and proportional, integral, and differential operation control of the output power of the heating element according to the error are widely used.

[0003] However, the existing multi-temperature zone welding temperature control technology still has the problems of relying only on real-time feedback adjustment, being unable to effectively cope with temperature lag, resulting in large temperature fluctuations, being prone to temperature overshoot or undershoot when switching temperature zones, being unable to optimize parameters for different temperature zone thermal characteristics, being difficult to meet the fine requirements of temperature uniformity and stability in complex circuit board welding, having limited adaptability to interference factors, and requiring manual intervention to adjust parameters. SUMMARY

[0004] The embodiments of the present application provide a multi-temperature zone welding temperature intelligent control method and system based on deep learning, which can at least solve some of the problems in the prior art.

[0005] In a first aspect, the embodiments of the present application provide a multi-temperature zone welding temperature intelligent control method based on deep learning, comprising:

[0006] Obtaining real-time temperature data of each temperature zone of a multi-temperature zone welding equipment in a circuit board welding production line and performing feature extraction and time series analysis to obtain temperature change prediction results;

[0007] Based on the temperature change prediction results, dividing each temperature zone into multiple control levels, calculating the temperature deviation value of each control level and generating initial weight coefficients, and fitting the temperature gradient change data by the least squares method to iteratively calculate the temperature compensation coefficient and the initial control parameters of each temperature zone;

[0008] Determining a quantitative index according to the initial control parameters, calculating the influence factor of the quantitative index and the comprehensive evaluation score of the initial control parameters by combining an adaptive weight algorithm, and iteratively optimizing the initial control parameters based on the comprehensive evaluation score to obtain an optimized temperature zone control parameter group;

[0009] The historical temperature data is analyzed based on the optimized temperature zone control parameter set in a sliding window, the temperature change rate and the temperature fluctuation standard deviation are calculated, when the temperature change rate exceeds a preset threshold or the temperature fluctuation standard deviation is greater than an allowed range, a temperature compensation value is calculated based on a proportional integral control algorithm and a heating power adjustment amount is determined based on a fuzzy control rule to generate a temperature compensation control parameter;

[0010] The temperature control instruction is generated based on the optimized temperature zone control parameter set and the temperature compensation control parameter and is executed.

[0011] In an optional implementation,

[0012] Real-time temperature data of each temperature zone of a multi-temperature zone soldering device in a circuit board soldering production line is acquired and feature extraction and time series analysis are performed to obtain a temperature change prediction result including:

[0013] Real-time temperature data of each temperature zone of a multi-temperature zone soldering device in a circuit board soldering production line is acquired and feature extraction is performed on the real-time temperature data to obtain a temperature change rule;

[0014] A temperature change trend matrix is established based on the temperature change rule and time series analysis is performed on the temperature change trend matrix;

[0015] A temperature change prediction value of each temperature zone is calculated based on a result of the time series analysis to obtain a temperature change prediction result.

[0016] In an optional implementation,

[0017] Based on the temperature change prediction result, each temperature zone is divided into multiple control levels, a temperature deviation value of each control level is calculated and an initial weight coefficient is generated, and a temperature compensation coefficient and an initial control parameter of each temperature zone are iteratively calculated by fitting temperature gradient change data through a least square method including:

[0018] Based on the temperature change prediction result, a temperature coupling degree between different temperature zones of a circuit board and a difference in thermal capacity of components are calculated and each temperature zone is divided into multiple control levels, for each control level, a temperature deviation value is calculated by subtracting a real-time temperature value obtained in advance from a target temperature value set in advance, and an initial weight coefficient is calculated based on the temperature deviation value;

[0019] Temperature gradient change data are calculated based on a temperature change trend matrix obtained in advance, the temperature deviation value and the temperature gradient change data of each temperature zone are fitted through a least square method, a fitting result is obtained, and a temperature compensation coefficient corresponding to the current iteration is calculated based on the fitting result and a temperature compensation coefficient obtained in the last iteration;

[0020] Based on the initial weight coefficient, the temperature deviation value and the temperature compensation coefficient, an initial control parameter is calculated by a pre-set temperature control performance evaluation function combined with a gradient descent method.

[0021] In an alternative embodiment,

[0022] Based on the initial weight coefficient, the temperature deviation value and the temperature compensation coefficient, an initial control parameter is calculated by a pre-set temperature control performance evaluation function combined with a gradient descent method.

[0023] The initial weight coefficient, the temperature deviation value and the temperature compensation coefficient are combined as an input vector and input into a pre-set adaptive fuzzy neural network, and a fuzzy processing is performed by a Gaussian membership function to obtain a membership parameter;

[0024] A wavelet transform is performed on the membership parameter to obtain a multi-scale feature coefficient, a hierarchical analysis decision tree is constructed based on the multi-scale feature coefficient, and a local extreme point under different scales is recursively calculated to determine a membership subset of the local extreme point, calculate mutual conditional information of the membership subset, and perform dimensionality reduction reconstruction, and fuse the reconstructed membership subset to obtain an optimized membership parameter;

[0025] A fuzzy rule base is established according to the optimized membership parameter and a pre-set temperature control performance evaluation function, a frequent item set in the fuzzy rule base is mined by an Apriori algorithm, and a support degree between different fuzzy rules is calculated, an association matrix is constructed based on the support degree and singular value decomposition is performed to construct a rule activation sequence, the fuzzy rule base is layered based on the rule activation sequence, and a rule output value is obtained by processing the optimized membership parameter combined with a local linear function, and an optimized intermediate parameter is solved.

[0026] When the temperature deviation value exceeds a pre-set deviation threshold, a compensation control parameter is calculated based on the optimized intermediate parameter combined with an exponential moving average and output as an initial control parameter.

[0027] In an alternative embodiment,

[0028] A quantitative index is determined according to the initial control parameter, an influence factor of the quantitative index and a comprehensive evaluation score of the initial control parameter are calculated by combining an adaptive weight algorithm, and an optimized temperature zone control parameter set is obtained by iteratively optimizing the initial control parameter based on the comprehensive evaluation score.

[0029] A control performance index corresponding to the initial control parameter is obtained, a quantitative index is mapped by combining a pre-set mapping rule, an environmental temperature fluctuation value and a process parameter fluctuation value are obtained, and a state space vector is constructed by combining the quantitative index.

[0030] The proportion corresponding to each quantization index in the state space vector is calculated and information entropy is determined, an influence factor corresponding to each quantization index is calculated based on the information entropy, a temperature change sequence is collected within a preset time window and time sequence features and nonlinear features are extracted to calculate a parameter sensitivity index;

[0031] The initial control parameters are encoded as position information of firefly individuals based on an adaptive firefly algorithm, a fluorescence intensity evaluation function is constructed based on the control performance index and an attraction coefficient is initialized based on the environmental temperature fluctuation value, an optimal individual is determined and the position information of the optimal individual is decoded as a parameter estimate value, a Lyapunov function value corresponding to the parameter estimate value is calculated, convergence is judged and system stability is determined, the quantization index is subjected to fuzzy evaluation to obtain a comprehensive evaluation score based on the parameter sensitivity index and the system stability, a gradient of the initial control parameters with respect to the comprehensive evaluation score is calculated and the initial control parameters are updated according to the gradient direction, the updating is repeated until the comprehensive evaluation score converges, and an optimized temperature zone control parameter group is obtained.

[0032] In an alternative embodiment,

[0033] The historical temperature data are subjected to sliding window analysis based on the optimized temperature zone control parameter group, a temperature change rate and a temperature fluctuation standard deviation are calculated, a temperature compensation value is calculated based on a proportional integral control algorithm and a heating power adjustment amount is determined in combination with fuzzy control rules when the temperature change rate exceeds a preset threshold or the temperature fluctuation standard deviation is greater than an allowable range, and a temperature compensation control parameter is generated, including:

[0034] A sliding window and a step length are set based on the optimized temperature zone control parameter group, historical temperature data collected and stored during a welding process are sampled to obtain a temperature sequence, and a temperature change rate and a temperature fluctuation standard deviation corresponding to the historical temperature data are calculated based on the temperature sequence;

[0035] When the absolute value of the temperature change rate is greater than a preset temperature change rate threshold or the temperature fluctuation standard deviation is greater than a preset fluctuation standard deviation threshold, a temperature error is calculated by calculating the difference between a preset expected temperature and a current temperature, the temperature error is divided into a preheating section, a reflow section and a cooling section by a piecewise adaptive Smith prediction compensation algorithm, compensation parameters are set for each section, the lag time of each section is calculated in combination with a particle swarm algorithm, and a temperature compensation value is obtained by feedforward compensation of the lag time;

[0036] The temperature compensation value is input into preset fuzzy control rules, a domain division value corresponding to the optimized temperature zone control parameter group is calculated, a temperature field topological feature map is constructed in combination with the thermal stress of the circuit board material and the pre-set deformation control constraint, the weight distribution of different regions of the temperature field is calculated based on the temperature field topological feature map, the heating power adjustment amount is calculated in combination with the temperature compensation value, and the temperature compensation control parameter is solved based on the heating power adjustment amount and the expected temperature.

[0037] In an alternative embodiment,

[0038] The temperature control instruction is generated based on the optimized temperature zone control parameter group and the temperature compensation control parameter, and the temperature control instruction is executed, including:

[0039] The control period and the temperature sampling interval are obtained from the optimized temperature zone control parameter group, and periodic temperature sampling is performed on each temperature zone of the circuit board to obtain temperature sampling values and circuit board state parameters.

[0040] The temperature sampling values are compared with the temperature compensation control parameter to obtain sampling temperature deviation values, and the temperature control adjustment amount is obtained by correcting the sampling temperature deviation values based on the optimized temperature zone control parameter group.

[0041] The temperature control instruction containing the heating power and the heating time is generated based on the temperature control adjustment amount and the circuit board state parameters, and the temperature control instruction is executed to adjust the temperature of the temperature zone.

[0042] In a second aspect of the embodiment of the present application, a multi-temperature zone welding temperature intelligent control system based on deep learning is provided, including:

[0043] The first unit is configured to obtain real-time temperature data of each temperature zone of the multi-temperature zone welding equipment in the circuit board welding production line, and perform feature extraction and time series analysis to obtain a temperature change prediction result.

[0044] The second unit is configured to divide each temperature zone into a plurality of control levels based on the temperature change prediction result, calculate a temperature deviation value of each control level and generate an initial weight coefficient, and fit the temperature gradient change data by the least square method to iteratively calculate a temperature compensation coefficient and an initial control parameter of each temperature zone.

[0045] The third unit is configured to determine a quantitative index according to the initial control parameter, calculate an influence factor of the quantitative index and a comprehensive evaluation score of the initial control parameter in combination with an adaptive weight algorithm, and iteratively optimize the initial control parameter based on the comprehensive evaluation score to obtain an optimized temperature zone control parameter group.

[0046] The fourth unit is used to perform sliding window analysis on historical temperature data based on the optimized temperature zone control parameter group, calculate the temperature change rate and temperature fluctuation standard deviation. When the temperature change rate exceeds the preset threshold or the temperature fluctuation standard deviation is greater than the allowable range, the temperature compensation value is calculated based on the proportional-integral control algorithm and the heating power adjustment amount is determined in combination with the fuzzy control rules to generate temperature compensation control parameters.

[0047] The fifth unit is used to generate and execute temperature control commands based on the optimized temperature zone control parameter set and temperature compensation control parameters.

[0048] A third aspect of the present invention provides an electronic device, comprising:

[0049] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0050] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0051] In this invention, deep learning is used to extract features and perform time-series analysis on real-time temperature data of multi-temperature zone welding equipment, enabling accurate prediction of temperature changes. This allows for the identification of temperature fluctuation trends, effectively reducing temperature deviations, improving welding quality and first-pass yield. By dividing control levels and combining them with an adaptive weighting algorithm, intelligent optimization of control parameters for different temperature zones is achieved, significantly improving temperature control accuracy. Dynamic adjustment of temperature compensation control is realized, enabling rapid response to abnormal temperature fluctuations, reducing energy consumption, improving production efficiency, and extending equipment lifespan. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating the intelligent multi-temperature zone welding temperature control method based on deep learning, as described in an embodiment of the present invention.

[0053] Figure 2 This is a flowchart illustrating the temperature compensation parameter generation process of the multi-temperature zone welding temperature intelligent control method based on deep learning, as described in an embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0056] Figure 1 This is a flowchart illustrating the intelligent multi-temperature zone welding temperature control method based on deep learning, as described in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0057] Real-time temperature data of each temperature zone in the multi-temperature zone welding equipment of the circuit board welding production line are obtained, and feature extraction and time series analysis are performed to obtain temperature change prediction results.

[0058] Based on the temperature change prediction results, each temperature zone is divided into multiple control levels. The temperature deviation value of each control level is calculated and an initial weighting coefficient is generated. The temperature gradient change data is fitted by the least squares method, and the temperature compensation coefficient and the initial control parameters of each temperature zone are obtained by iterative calculation.

[0059] The quantitative index is determined based on the initial control parameters. The influence factor of the quantitative index and the comprehensive evaluation score of the initial control parameters are calculated by combining the adaptive weight algorithm. The initial control parameters are iteratively optimized based on the comprehensive evaluation score to obtain the optimized temperature zone control parameter set.

[0060] Based on the optimized temperature zone control parameter group, a sliding window analysis is performed on historical temperature data to calculate the temperature change rate and temperature fluctuation standard deviation. When the temperature change rate exceeds the preset threshold or the temperature fluctuation standard deviation is greater than the allowable range, the temperature compensation value is calculated based on the proportional-integral control algorithm and combined with fuzzy control rules to determine the heating power adjustment amount, thereby generating temperature compensation control parameters.

[0061] Temperature control commands are generated and executed based on the optimized temperature zone control parameter group and temperature compensation control parameters.

[0062] In one alternative implementation,

[0063] Real-time temperature data of each temperature zone in a multi-temperature zone welding equipment in a circuit board welding production line is acquired, and feature extraction and time series analysis are performed to obtain temperature change prediction results, including:

[0064] Real-time temperature data of each temperature zone in the multi-temperature zone welding equipment in the circuit board welding production line are obtained, and feature extraction is performed on the real-time temperature data to obtain the temperature change pattern.

[0065] A temperature change trend matrix is ​​established based on the temperature change pattern, and a time series analysis is performed on the temperature change trend matrix.

[0066] Based on the results of the time series analysis, the predicted temperature changes for each temperature zone are calculated, and the predicted temperature changes are obtained.

[0067] Temperature data is collected by a network of temperature sensors distributed across various temperature zones, with a sampling frequency of once per second. The data collected by the temperature sensors is transmitted to the data processing unit through an industrial IoT module to form a complete temperature data stream. For example, a welding equipment has 5 temperature zones, labeled Z1 to Z5. During continuous production, the temperature data collected from each temperature zone forms a time series {T1, 1, T1, 2, ..., T1, n} to {T5, 1, T5, 2, ..., T5, n}, where Ti,j represents the temperature value of the i-th temperature zone at the j-th time point.

[0068] Feature extraction is performed on the collected real-time temperature data. The feature extraction process includes data preprocessing, trend analysis, and fluctuation feature identification. In the data preprocessing stage, the raw temperature data is filtered to eliminate the influence of random noise. A moving average filtering method is used to average the temperature data within each time window, with the window size set to 10 sampling points. The filtered data is used to calculate the temperature change rate, which is the temperature difference between adjacent time points divided by the time interval. By analyzing the distribution characteristics of the temperature change rate, the acceleration pattern of temperature change is identified. Furthermore, the temperature statistical characteristics of each temperature zone in different time periods are calculated, including the average, maximum, minimum, and standard deviation. For example, in a welding process, the average temperature of temperature zone Z3 under stable operating conditions is 210°C, the standard deviation is 1.2°C, and the average temperature change rate is 0.05°C / second. These data reflect the temperature change characteristics of this temperature zone.

[0069] Based on the extracted temperature change features, a temperature change trend matrix is ​​constructed. This matrix is ​​a three-dimensional data structure containing three dimensions: temperature zone number, time point, and temperature characteristics. For each temperature zone, key parameters characterizing the temperature change trend are extracted based on the historical temperature data change patterns. These parameters include the trend slope, periodic fluctuation amplitude, and phase information. These key parameters are arranged in a time series within the matrix, forming a complete trend description. The construction of the temperature change trend matrix considers the mutual influence between temperature zones. Correlation analysis is used to determine the temperature conduction relationship between different temperature zones, quantifying this relationship as a temperature transfer coefficient. For example, in practical applications, temperature changes in zone Z2 typically affect zone Z3 after 3-5 seconds, with an influence coefficient of approximately 0.7, and this is encoded into the trend matrix.

[0070] A time-series analysis was performed on the temperature change trend matrix using deep learning methods, specifically a Long Short-Term Memory (LSTM) network model. The LSM network model consists of an input layer, multiple LSM layers, and an output layer. The input layer receives the normalized temperature trend matrix data. Each LSM layer contains 64 neurons to capture the time dependence of temperature changes. Historical temperature data was used as the training set, and the model parameters were optimized using stochastic gradient descent. The mean squared error loss function was chosen. To prevent overfitting, dropout was employed with a dropout rate of 0.2. After training, the average prediction error on the test set was less than 1.5°C, meeting the accuracy requirements for industrial welding.

[0071] Based on the results of time-series analysis, predicted temperature changes for each temperature zone are calculated. The prediction process is divided into short-term and long-term predictions. Short-term predictions focus on temperature changes within the next 30 seconds, primarily for real-time control adjustments. Long-term predictions focus on temperature trends within the next 5 minutes, used for process optimization and anomaly warnings. The prediction algorithm comprehensively considers the current temperature status, historical trends, and environmental factors. For each temperature zone, based on the output of the Long Short-Term Memory network and combined with the zone's characteristic parameters, predicted temperature values ​​for each future time point are calculated. The calculation of the predicted values ​​takes into account the heat conduction effect between temperature zones; temperature changes in adjacent temperature zones will proportionally affect the future temperature of the target temperature zone. In a practical application case, the current temperature of zone Z4 of a welding equipment is 250°C. Based on the prediction algorithm, it is predicted that the temperature of this zone will stabilize within the range of 248-252°C within the next 30 seconds, and will slowly decrease to around 245°C after the next 3 minutes.

[0072] In this embodiment, by extracting features from the real-time temperature data of each temperature zone, representative temperature change features can be extracted from the raw data, improving the effectiveness and usability of temperature information expression. By establishing a temperature change trend matrix based on the temperature change pattern and performing time series analysis, the dynamic characteristics of temperature evolution over time can be reflected more accurately, improving the accuracy of temperature change pattern modeling and prediction. By calculating the predicted temperature change values ​​of each temperature zone based on the time series analysis results, the temperature change trend can be obtained, improving the foresight of prediction and the stability of control during the temperature control process.

[0073] In one alternative implementation,

[0074] Based on the temperature change prediction results, each temperature zone is divided into multiple control levels. The temperature deviation value of each control level is calculated and an initial weighting coefficient is generated. The temperature gradient change data is then fitted using the least squares method, and the temperature compensation coefficient and the initial control parameters for each temperature zone are obtained through iterative calculation.

[0075] Based on the temperature change prediction results, the temperature coupling degree between different temperature zones of the circuit board and the difference in heat capacity of components are calculated, and each temperature zone is divided into multiple control levels. For each control level, the difference between the preset target temperature value and the preset real-time temperature value is calculated to obtain the temperature deviation value. The initial weighting coefficient is calculated based on the temperature deviation value.

[0076] Based on the pre-acquired temperature change trend matrix, the temperature gradient change data is calculated. The temperature deviation value and the temperature gradient change data of each temperature zone are fitted by the least squares method to obtain the fitting result. Combined with the pre-acquired temperature compensation coefficient of the previous iteration, the temperature compensation coefficient corresponding to the current iteration is calculated.

[0077] Based on the initial weighting coefficient, the temperature deviation value, and the temperature compensation coefficient, the initial control parameters are calculated using a pre-set temperature control performance evaluation function combined with the gradient descent method.

[0078] After obtaining the temperature change prediction results, the temperature coupling degree between different temperature zones on the circuit board is calculated. The temperature coupling degree represents the intensity of heat conduction influence between different temperature zones. Correlation analysis is used in the calculation, comparing the time series data of temperature changes in each temperature zone to calculate the correlation coefficient matrix. For example, in the soldering process of a certain circuit board, the temperature coupling matrix of the five temperature zones Z1 to Z5 shows that the coupling coefficient between Z2 and Z3 is 0.85, indicating that the temperature change of Z2 will directly affect the temperature state of Z3, indicating a high degree of coupling; while the coupling coefficient between Z1 and Z5 is only 0.32, indicating a weaker thermal influence between distant temperature zones. Simultaneously, the calculation of the component heat capacity difference is based on the thermal characteristic parameters of different types of components on the circuit board. By analyzing the circuit board design and component distribution, combined with the density distribution and material properties of components within each temperature zone, the equivalent heat capacity of each temperature zone is calculated. For example, the heat capacity of a region containing large capacitors or transformers can be as high as 245 J / K, while the heat capacity of a region containing only small resistors and diodes may only be 85 J / K.

[0079] Based on the degree of temperature coupling and differences in heat capacity, each temperature zone is divided into multiple control levels. The control level division employs cluster analysis, comprehensively considering the coupling relationship, heat capacity characteristics, and temperature control importance of each temperature zone. Typically, the temperature zones are divided into three levels: a critical control layer, a coordinating control layer, and an auxiliary control layer. The critical control layer contains the temperature zones with the highest welding quality requirements, such as the reflow peak temperature zone; the coordinating control layer contains temperature zones with high coupling to the critical temperature zone; and the auxiliary control layer contains the remaining temperature zones. For example, the level division result for a six-temperature-zone welding equipment is as follows: Z3 and Z4 are assigned to the critical control layer with a control accuracy requirement of ±1℃; Z2 and Z5 are assigned to the coordinating control layer with a control accuracy requirement of ±2℃; and Z1 and Z6 are assigned to the auxiliary control layer with a control accuracy requirement of ±3℃. After the control levels are divided, the temperature deviation value is calculated for each level, which is the difference between the preset target temperature value and the real-time temperature value. Taking the critical control layer Z3 as an example, if the target temperature is set to 235℃ and the real-time temperature is 232℃, then the temperature deviation value is 3℃.

[0080] Initial weighting coefficients are calculated based on temperature deviation values. These coefficients reflect the impact of temperature deviations in different temperature zones on the overall control objective. A weighted allocation principle is adopted, with weights related to the absolute value of the temperature deviation and the control level to which the temperature zone belongs. The base weight for the critical control layer is 1.0, for the coordinated control layer it is 0.7, and for the auxiliary control layer it is 0.4. The actual weighting coefficients need to be adjusted according to the magnitude of the temperature deviation; the larger the deviation, the higher the weighting coefficient. The base weight is multiplied by the normalized result of the temperature deviation value to obtain the initial weighting coefficient for each temperature zone. For example, the temperature deviation of the critical control layer Z3 is 3℃, which is 0.6 after normalization, so its initial weighting coefficient is 0.6 × 1.0 = 0.6; the temperature deviation of the coordinated control layer Z2 is 2℃, which is 0.4 after normalization, so its initial weighting coefficient is 0.4 × 0.7 = 0.28.

[0081] Temperature gradient data is calculated based on a pre-acquired temperature change trend matrix. This temperature gradient represents the change in temperature per unit time, obtained by calculating the temperature difference between adjacent time points in the temperature change trend matrix. For a system with a sampling frequency of 1 Hz, the temperature gradient is the temperature change per second. For example, the temperature records for temperature zone Z4 over 10 consecutive seconds are {245.2, 246.1, 246.8, 247.3, 247.7, 248.0, 248.2, 248.3, 248.4, 248.5} ℃. The calculated temperature gradient sequence is {0.9, 0.7, 0.5, 0.4, 0.3, 0.2, 0.1, 0.1, 0.1} ℃ / second, showing a gradually decreasing trend, indicating that the rate of temperature increase is gradually slowing down and approaching a stable state.

[0082] The least squares method was used to fit the temperature deviation and temperature gradient change data to obtain a model of the relationship between temperature change trends and control requirements. During the fitting process, the temperature deviation was used as the dependent variable, and the temperature gradient change data as the independent variable, constructing linear or nonlinear fitting equations. The optimal fitting parameters were determined by minimizing the sum of squared residuals. In a welding equipment example, the fitting results for the Z3 temperature zone showed a clear quadratic relationship between the temperature deviation and the temperature gradient. The fitting parameters were: constant term 1.25, first-order coefficient 0.78, quadratic coefficient -0.15, and fitting accuracy R0. 2 A value of 0.92 indicates a good fit. The fitting result is combined with the temperature compensation coefficient from the previous iteration to calculate the temperature compensation coefficient for the current iteration. The iterative calculation of the temperature compensation coefficient uses a weighted average method: the compensation coefficient for the current iteration = the compensation coefficient from the previous iteration × 0.7 + the newly calculated compensation coefficient × 0.3. For example, the compensation coefficient for temperature region Z3 in the previous iteration was 1.15, and the newly calculated compensation coefficient is 1.23. Therefore, the compensation coefficient for the current iteration = 1.15 × 0.7 + 1.23 × 0.3 = 1.174.

[0083] Based on initial weighting coefficients, temperature deviation values, and temperature compensation coefficients, initial control parameters are calculated using a pre-set temperature control performance evaluation function combined with the gradient descent method. The temperature control performance evaluation function comprehensively considers three aspects: control accuracy, response speed, and stability. Control accuracy is represented by the square of the temperature deviation, response speed by the control time length, and stability by the temperature fluctuation amplitude. These three indicators are weighted and summed in a 5:3:2 ratio to form the value of the evaluation function. The gradient descent method is used to find the control parameters that minimize the evaluation function. For example, initial values ​​and learning rates are set for the control parameters, and through multiple iterations, the parameters are updated along the negative gradient direction of the evaluation function until convergence. In a welding temperature zone control case, the initial control parameters were set to P=2.5, I=0.8, and D=0.3. After 50 gradient descent iterations, the parameters were optimized to P=2.78, I=0.95, and D=0.42, resulting in a significant improvement in control performance.

[0084] In this embodiment, the temperature coupling degree between temperature zones and the difference in heat capacity of components are calculated based on the temperature change prediction results, and the temperature zones are divided into multiple control levels. This makes the temperature control more consistent with the physical characteristics of the production process and improves the precision of the layered control. The temperature deviation value is obtained by calculating the difference between the target temperature value and the real-time temperature value. Furthermore, the temperature gradient change data is combined with the least squares method for fitting, which can more accurately depict the relationship between the temperature change trend and the deviation, and improve the accuracy and stability of the compensation calculation. By combining the initial weight coefficient, the temperature deviation value and the temperature compensation coefficient, and using the temperature control performance evaluation function and the gradient descent method to calculate the initial control parameters, the convergence speed can be accelerated while optimizing the control parameters, thereby improving the response efficiency and overall stability of the temperature control.

[0085] In one alternative implementation,

[0086] Based on the initial weighting coefficient, the temperature deviation value, and the temperature compensation coefficient, the initial control parameters are calculated using a pre-set temperature control performance evaluation function and the gradient descent method, including:

[0087] The initial weighting coefficients, temperature deviation values, and temperature compensation coefficients are combined into an input vector and input into a pre-set adaptive fuzzy neural network. The membership parameters are obtained by fuzzification using a Gaussian membership function.

[0088] Wavelet transform is performed on the membership parameters to obtain multi-scale feature coefficients. A hierarchical analysis decision tree is constructed based on the multi-scale feature coefficients and local extreme points at different scales are recursively calculated. The membership subsets of the local extreme points are determined. The cross-condition information of the membership subsets is calculated and dimensionality reduction and reconstruction are performed. The optimized membership parameters are obtained by fusing the reconstructed membership subsets.

[0089] A fuzzy rule base is established based on the optimized membership parameters and the preset temperature control performance evaluation function. The frequent itemsets in the fuzzy rule base are mined by the Apriori algorithm and the support between different fuzzy rules is calculated. An association matrix is ​​constructed based on the support and singular value decomposition is performed to construct a rule activation sequence. The fuzzy rule base is sealed based on the rule activation sequence. The optimized membership parameters are processed by local linear functions to obtain the rule output values ​​and the optimized intermediate parameters are obtained by solving.

[0090] When the temperature deviation exceeds the preset deviation threshold, the compensation control parameters are calculated based on the optimized intermediate parameters and the exponential moving average, and then output as the initial control parameters.

[0091] The initial weighting coefficients, temperature deviation values, and temperature compensation coefficients are combined into an input vector. Taking a five-temperature-zone welding equipment as an example, the input vector for each temperature zone contains three elements: initial weighting coefficients (range 0-1), temperature deviation values ​​(range ±10℃), and temperature compensation coefficients (range 0.8-1.5). For temperature zone Z3, the actual input vector might be [0.75, 2.3, 1.12], indicating that its initial weighting coefficient is 0.75, the temperature deviation is 2.3℃, and the compensation coefficient is 1.12. The combined input vector is then fed into a pre-set adaptive fuzzy neural network for processing. The adaptive fuzzy neural network adopts a five-layer structure: input layer, fuzzification layer, rule layer, normalization layer, and output layer. The input layer receives the raw data, and the number of nodes equals the dimension of the input vector. The fuzzification layer performs fuzzy transformation on the input using a Gaussian membership function. For each input variable, five fuzzy linguistic variables are set: very low, relatively low, medium, relatively high, and very high. The center point and width parameters of the Gaussian membership function are determined through training with historical data. For example, for the temperature deviation value, the Gaussian function center point for "Medium" is set to 0℃, and the width parameter is 2℃. After fuzzification, the input vector [0.75, 2.3, 1.12] is transformed into a membership parameter matrix, representing the degree of membership of each input on different fuzzy sets. For example, the membership degree of a temperature deviation of 2.3℃ is 0.33 on "Medium" and 0.61 on "High".

[0092] A wavelet transform is performed on the membership parameters using the Haar wavelet basis function. The membership parameters are then decomposed into three levels to obtain low-frequency approximation coefficients and high-frequency detail coefficients. The wavelet transform is implemented using a filter bank: a low-pass filter extracts the smoothing trend of the signal, and a high-pass filter extracts local detail changes. The wavelet transform is applied to each row of the membership parameter matrix to obtain multi-scale feature coefficients. For example, performing a wavelet transform on the membership parameters of temperature region Z3 yields the following three-level decomposition coefficients: first-level approximation coefficients [0.65, 0.72, 0.58], first-level detail coefficients [0.12, -0.08, 0.05], second-level detail coefficients [0.03, -0.02, 0.01], and third-level detail coefficients [0.01, 0.00, -0.01]. A hierarchical analysis decision tree is constructed based on the multi-scale feature coefficients, using the CART algorithm. Each internal node represents a binary test of the feature coefficients, and leaf nodes represent local extrema regions. The decision tree generation process employs a recursive approach, determining the optimal splitting feature and threshold by calculating the Gini coefficient. For example, the first-level split might select the first element of the first-level approximation coefficient with a threshold of 0.6; the second-level split might select the second element of the first-level detail coefficient with a threshold of -0.05. Through recursive calculation of the decision tree, local extrema at different scales are obtained, such as [0.65, 0.12, 0.03] and [0.72, -0.08, -0.02].

[0093] The membership subsets of local extrema are determined. For each local extrema, its corresponding membership subset is determined based on its position in the original membership space. The membership subset is a submatrix of the original membership parameters, reflecting the fuzzy characteristics of the local region. For example, the membership subset corresponding to the extrema [0.65, 0.12, 0.03] might be the top-left 3×3 submatrix of the original membership matrix. The cross-conditional information of the membership subsets is calculated to assess the information redundancy within the subsets. The cross-conditional information is obtained by calculating the information entropy and joint entropy between elements in the subset. Subsets with high information redundancy require dimensionality reduction. Dimensionality reduction and reconstruction uses principal component analysis (PCA), retaining principal components with a cumulative contribution rate of 95%. For example, after PCA, the contribution rates of the first two principal components in a membership subset are 87% and 9%, totaling 96%, so these two principal components are retained for reconstruction. The reconstructed membership subsets are then merged to obtain the optimized membership parameters. The fusion process employs a weighted average method, with the weights proportional to the magnitudes of the extreme points corresponding to the subsets. For temperature region Z3, the dimension of the optimized membership parameter matrix is ​​reduced from 3×15 to 3×8, resulting in a 47% reduction in computational complexity.

[0094] A fuzzy rule base is established based on optimized membership parameters and a preset temperature control performance evaluation function. The temperature control performance evaluation function comprehensively considers three indicators: overshoot, rise time, and steady-state error, and is expressed in a weighted sum form. Fuzzy rules adopt an "if-then" form, such as "if the temperature deviation is large and the rate of change is positive, then the control output is large." The initial rule base, generated through expert experience and historical data analysis, contains approximately 50 rules. The Apriori algorithm is applied to the rule base for frequent itemset mining, setting a minimum support threshold of 0.15 to identify frequently occurring rule combinations. For example, analysis revealed that the items "large temperature deviation" and "large control output" have a support of 0.35 in the rule base, exceeding the threshold, and are therefore identified as frequent itemsets. The support between different fuzzy rules is calculated to obtain a support matrix. An association matrix is ​​constructed based on the support matrix, where the element values ​​represent the association strength between rules.

[0095] Singular value decomposition (SVD) is performed on the correlation matrix to extract key eigenvectors and construct a rule activation sequence. SVD decomposes the correlation matrix into the product of three matrices, and the retained feature dimensions are determined by analyzing the magnitude of the singular values. For example, if the singular value sequence of a certain correlation matrix is ​​[5.2, 2.8, 1.5, 0.9, 0.3], retaining the first three eigenvectors covers 86% of the information. The rule activation sequence is the priority order of rule triggering, obtained by sorting the element values ​​of the eigenvectors. Based on the rule activation sequence, the fuzzy rule base is layered into three levels: high priority, medium priority, and low priority. High-priority rules (e.g., 10 rules) are evaluated first during the control process; if the triggering condition is met, they are directly output; otherwise, medium-priority rules (e.g., 25 rules) are evaluated, and finally, low-priority rules (e.g., 15 rules) are evaluated.

[0096] By combining local linear function processing to optimize membership parameters, rule output values ​​are obtained, and optimization intermediate parameters are solved. The local linear function divides the fuzzy input space into multiple local regions, each using a linear function to describe the relationship between input and output. For temperature region Z4, a linear function with a slope of 0.8 is used when the temperature deviation is within the range of [-2, 2]℃; a linear function with a slope of 1.2 is used when the temperature deviation is within the range of [2, 5]℃. The output values ​​of each rule are fused using a weighted average method, with the weight equal to the trigger strength of the rule, resulting in the final control decision value. For example, within a certain control cycle, the trigger strengths of three high-priority rules are 0.7, 0.5, and 0.3, with corresponding output values ​​of 0.8, 0.6, and 0.4. The fused output value is 0.68, which is then converted into optimization intermediate parameters.

[0097] When the temperature deviation exceeds a preset deviation threshold, a compensation control mechanism is triggered. The preset deviation threshold is typically set to ±2% of the target temperature. For example, if the target temperature is 240℃, the deviation threshold is ±4.8℃. After exceeding the threshold, compensation control parameters are calculated based on optimized intermediate parameters combined with exponential moving averages. The exponential moving average method assigns higher weight to recent data, with a smoothing coefficient set to 0.2. The calculation formula is: Current compensation value = Previous cycle compensation value × 0.8 + Optimized intermediate parameter × 0.2. For example, if the previous cycle compensation value is 1.25 and the current optimized intermediate parameter is 1.35, the calculated compensation control parameter is 1.27. This parameter serves as the initial control parameter output of the PID controller, used to adjust the heating power or cooling intensity to achieve precise temperature control.

[0098] In this embodiment, by inputting the initial weight coefficients, temperature deviation values, and temperature compensation coefficients into an adaptive fuzzy neural network and performing fuzzification processing using a Gaussian membership function, temperature control characteristics can be expressed more flexibly under uncertain environments, improving adaptability to complex nonlinear relationships. By performing wavelet transform on the membership parameters and extracting multi-scale features, and recursively calculating local extrema and membership subsets using a hierarchical analysis decision tree, detailed features of temperature changes can be captured at different scales, improving the accuracy and robustness of membership parameter optimization. By using the Apriori algorithm to mine frequent itemsets in the fuzzy rule base and combining it with singular value decomposition to construct rule activation sequences, redundant rules can be filtered and effective rules strengthened.

[0099] In one alternative implementation,

[0100] Based on the initial control parameters, a quantitative index is determined. The influence factor of the quantitative index and the comprehensive evaluation score of the initial control parameters are calculated using an adaptive weighting algorithm. Based on the comprehensive evaluation score, the initial control parameters are iteratively optimized to obtain an optimized temperature zone control parameter set, including:

[0101] The control performance index corresponding to the initial control parameters is obtained, and the quantitative index is obtained by combining the pre-set mapping rules. The ambient temperature fluctuation value and the process parameter fluctuation value are obtained, and the state space vector is constructed by combining the quantitative index.

[0102] Calculate the weight of each quantitative index in the state space vector and determine the information entropy. Based on the information entropy, calculate the influence factor corresponding to each quantitative index. Collect temperature change sequence within a preset time window and extract time-series and nonlinear features to calculate the parameter sensitivity index.

[0103] Based on the adaptive firefly algorithm, the initial control parameters are encoded as the location information of individual fireflies. A fluorescence intensity evaluation function is constructed based on the control performance index, and the attraction coefficient is initialized based on the ambient temperature fluctuation value. The optimal individual is determined, and the location information of the optimal individual is decoded into parameter estimates. The Lyapunov function value corresponding to the parameter estimates is calculated, convergence is judged, and system stability is determined. Based on the parameter sensitivity index and the system stability, the quantitative index is fuzzy evaluated to obtain a comprehensive evaluation score. The gradient of the comprehensive evaluation score with respect to the initial control parameters is calculated, and the initial control parameters are updated according to the gradient direction. This update is repeated until the comprehensive evaluation score converges, resulting in an optimized temperature zone control parameter set.

[0104] The control performance indicators corresponding to the initial control parameters are obtained. These indicators include overshoot, rise time, settling time, and steady-state error. Taking a six-temperature zone welding equipment as an example, after applying the initial control parameters P=2.8, I=0.4, and D=0.2 to temperature zone Z2, the measured overshoot is 1.8%, the rise time is 15 seconds, the settling time is 35 seconds, and the steady-state error is 0.5℃. These performance indicators are then mapped to quantized indicators using pre-set mapping rules. The mapping rules employ a piecewise linear mapping method, normalizing each performance indicator to the [0, 1] interval. For example, the overshoot mapping rule is as follows: when the overshoot is less than 1%, the quantized value is 0.9-1.0; when the overshoot is between 1% and 3%, the quantized value is 0.7-0.9; when the overshoot is between 3% and 5%, the quantized value is 0.5-0.7; and so on. According to this rule, an overshoot of 1.8% is mapped to a quantized value of 0.84. Similarly, a rise time of 15 seconds is mapped to 0.76, a settling time of 35 seconds is mapped to 0.68, and a steady-state error of 0.5℃ is mapped to 0.82. Ambient temperature fluctuations and process parameter fluctuations are obtained. The ambient temperature fluctuation range over the past 4 hours, recorded by the plant temperature sensor, is 23.5℃ to 25.8℃, with a standard deviation of 0.7℃. Process parameter fluctuations are obtained by monitoring equipment operating parameters, such as conveyor belt speed fluctuations of ±3% of the standard value and solder supply fluctuations of ±2% of the standard value. A state-space vector is constructed by combining the quantitative indicators, ambient temperature fluctuations, and process parameter fluctuations. For temperature zone Z2, the state-space vector is [0.84, 0.76, 0.68, 0.82, 0.7, 0.03, 0.02], where the first four elements are quantitative indicators of control performance, the fifth element is the quantitative value of ambient temperature fluctuation, and the last two elements are the quantitative values ​​of process parameter fluctuation.

[0105] The importance of each quantitative indicator is determined by calculating its weight in the state space vector. The weight calculation employs the analytic hierarchy process (AHP), constructing a judgment matrix to determine the relative importance of each indicator. This judgment matrix is ​​established based on expert experience and historical data analysis, and its rationality is ensured through consistency checks. For welding temperature control, steady-state error and overshoot are typically more important than rise time and settling time. Calculations show that the weights of each indicator in temperature zone Z2 are as follows: overshoot 0.28, rise time 0.15, settling time 0.12, steady-state error 0.30, ambient temperature fluctuation 0.10, and process parameter fluctuation 0.05. Information entropy is calculated based on these weights, reflecting the uncertainty of the indicator distribution. When calculating information entropy, the weights of each indicator are normalized to a probability distribution, and the entropy value of the probability distribution is calculated. For example, the information entropy of each indicator in temperature zone Z2 is 1.62. Influence factors for each quantitative indicator are then calculated based on the information entropy, representing the degree of influence of the indicator on overall performance. The calculation method involves multiplying the index weight by the normalized value of the information entropy. For example, the influence factor of overshoot is 0.28 × 1.62 ÷ 4.5 = 0.10, where 4.5 is the normalization coefficient. Temperature change sequences are collected and features are extracted within a preset time window. The time window is set to 120 seconds, the sampling frequency is 1Hz, and a total of 120 temperature data points are collected. Temporal features are extracted from the collected sequence data, including statistical features such as mean, standard deviation, skewness, and kurtosis, as well as frequency domain features extracted through Fourier transform. For example, the temporal features of temperature zone Z4 include: mean 245.8℃, standard deviation 1.2℃, skewness 0.15, kurtosis 2.8, and the dominant frequency component around 0.05Hz. Nonlinear feature extraction employs phase space reconstruction and complexity analysis methods to calculate features such as the Lyapunov exponent, entropy value, and fractal dimension. For example, the Lyapunov exponent for temperature zone Z4 is 0.024, indicating certain chaotic characteristics. The parameter sensitivity index is calculated by combining time-series and nonlinear characteristics. Parameter sensitivity represents the degree to which changes in control parameters affect system performance. The calculation method involves applying small disturbances to the control parameters and observing the rate of change of the performance index. For example, when the P parameter increases by 5%, the overshoot increases by 12%, indicating that the sensitivity of the overshoot to the P parameter is 2.4.

[0106] The control parameters are optimized based on the adaptive firefly algorithm, a swarm intelligence-based optimization algorithm that finds the optimal solution by simulating the aggregation behavior of fireflies. Initial control parameters are encoded as the position information of individual fireflies. For a PID controller, each firefly is represented as a three-dimensional vector [P, I, D]. For example, the initial population contains 30 fireflies, and the position of each individual is randomly initialized within an allowable range. For temperature range Z3, the position of an individual in the initial population might be [2.5, 0.4, 0.2]. A fluorescence intensity evaluation function is constructed based on control performance indicators. The evaluation function uses a weighted sum form, with the weight of each performance indicator proportional to its influencing factor. For example, the evaluation function is: 0.35 × overshoot score + 0.15 × rise time score + 0.15 × settling time score + 0.35 × steady-state error score. Fluorescence intensity is proportional to the evaluation function value; the higher the evaluation function value, the stronger the fluorescence intensity. The attraction coefficient is initialized based on the ambient temperature fluctuation value, representing the intensity of mutual influence among firefly individuals. The attraction coefficient is inversely proportional to the fluctuation of ambient temperature; the greater the fluctuation, the smaller the attraction coefficient, indicating that individuals need to explore more autonomously in complex environments. For example, when the standard deviation of ambient temperature fluctuation is 0.7℃, the attraction coefficient is initialized to 0.3.

[0107] In the iterative optimization process, the optimal individual is determined and its position information is decoded into parameter estimates. In each iteration, all firefly individuals move towards individuals with higher fluorescence intensity; the moving distance is related to the attraction coefficient and the distance between individuals. After 50 iterations, the optimal individual position converges to [2.35, 0.55, 0.25], decoded into parameter estimates P=2.35, I=0.55, D=0.25. The Lyapunov function values ​​corresponding to the parameter estimates are calculated to determine the system's stability. The Lyapunov function is constructed based on the system's energy, and the trend of the function value reflects the system's stability. Numerical simulations show a Lyapunov function value of -0.35 and a derivative of -0.08, indicating that the system is asymptotically stable. Convergence is assessed to determine system stability; the convergence criterion is that the optimal value changes by less than 0.01 over 10 consecutive iterations. System stability is graded according to the Lyapunov function value: less than -0.3 indicates high stability, -0.3 to -0.1 indicates medium stability, and greater than -0.1 indicates low stability. For example, a Lyapunov function value of -0.35 in the temperature range Z3 indicates that the system has high stability.

[0108] A comprehensive evaluation score is obtained by fuzzy evaluation of quantitative indicators based on parameter sensitivity and system stability. The fuzzy evaluation adopts a two-layer evaluation model: the first layer evaluates the satisfaction level of each performance indicator, and the second layer comprehensively evaluates the overall performance. A five-level fuzzy set is used for evaluation: excellent, good, average, poor, and very poor. For temperature range Z3, the fuzzy evaluation results are: overshoot - good (0.7), rise time - excellent (0.85), settling time - good (0.75), and steady-state error - excellent (0.8). The comprehensive evaluation score is 0.78, belonging to the "good" level. The gradient of the comprehensive evaluation score with respect to the initial control parameters is calculated to reflect the sensitivity of the score to parameter changes. A numerical differentiation method is used to add small perturbations to each parameter, and the rate of change of the evaluation score is calculated. For example, when the P parameter increases by 0.1, the overall evaluation score changes by -0.05, indicating a gradient of -0.5; when the I parameter increases by 0.1, the overall evaluation score changes by 0.03, indicating a gradient of 0.3; and when the D parameter increases by 0.1, the overall evaluation score changes by 0.02, indicating a gradient of 0.2. The initial control parameters are updated based on the gradient direction, using the formula: New parameter = Old parameter + Learning rate × Gradient. With the learning rate set to 0.2, the updated parameters are: P = 2.35 - 0.2 × (-0.5) = 2.45, I = 0.55 + 0.2 × 0.3 = 0.61, and D = 0.25 + 0.2 × 0.2 = 0.29.

[0109] The parameter update process was repeated until the comprehensive evaluation score converged. The convergence criterion was that the score change was less than 0.005 for five consecutive iterations. After 15 iterations, the control parameters of temperature zone Z3 were optimized to P=2.42, I=0.58, and D=0.26, with a comprehensive evaluation score of 0.83, which is considered "excellent". The optimization process was repeated for all temperature zones to obtain the optimized temperature zone control parameter set. For example, the optimized parameter set for the six-temperature zone welding equipment is: Z1[1.85, 0.38, 0.15], Z2[2.20, 0.45, 0.22], Z3[2.42, 0.58, 0.26], Z4[2.35, 0.52, 0.24], Z5[2.10, 0.48, 0.20], Z6[1.75, 0.35, 0.15].

[0110] In this embodiment, by combining mapping rules to transform control performance indicators into quantitative indicators, and jointly constructing a state space vector with ambient temperature fluctuation values ​​and process parameter fluctuation values, it is possible to comprehensively characterize the multi-dimensional temperature control state and improve the ability to describe the system's operating characteristics. By calculating the influence factor based on information entropy and extracting parameter sensitivity indicators by combining time series and nonlinear features, the dominant role of key parameters in control performance can be highlighted, thereby improving the pertinence and effectiveness of parameter optimization. By using the adaptive firefly algorithm to encode and iteratively optimize the initial control parameters, and dynamically adjusting the search process by combining the fluorescence intensity evaluation function and the attraction coefficient, the global optimization capability can be enhanced and the trapping in local optima can be avoided, thereby improving optimization efficiency and the stability of parameter convergence.

[0111] In one alternative implementation,

[0112] Based on the optimized temperature zone control parameter set, a sliding window analysis is performed on historical temperature data to calculate the temperature change rate and temperature fluctuation standard deviation. When the temperature change rate exceeds a preset threshold or the temperature fluctuation standard deviation is greater than the allowable range, a temperature compensation value is calculated based on the proportional-integral control algorithm, and the heating power adjustment is determined in conjunction with fuzzy control rules. The temperature compensation control parameters include:

[0113] Based on the optimized temperature zone control parameter group, a sliding window and step size are set, and the historical temperature data collected and stored during the welding process are sampled to obtain a temperature sequence. Based on the temperature sequence, the temperature change rate and temperature fluctuation standard deviation corresponding to the historical temperature data are calculated.

[0114] When the absolute value of the temperature change rate is greater than the preset temperature change rate threshold or the temperature fluctuation standard deviation is greater than the preset fluctuation standard deviation threshold, the temperature error is obtained by calculating the difference between the preset expected temperature and the current temperature. The temperature error is divided into a preheating section, a recirculation section and a cooling section by a segmented adaptive Smith prediction compensation algorithm. Compensation parameters are set for each section. The lag time of each section is calculated by combining the particle swarm algorithm and the lag time is fed forward to obtain the temperature compensation value.

[0115] The temperature compensation value is input into a preset fuzzy control rule, the universe partition value corresponding to the optimized temperature zone control parameter group is calculated, and the temperature field topology feature map is constructed by combining the thermal stress corresponding to the circuit board material and the preset deformation control constraints. The weight distribution of different regions of the temperature field is calculated based on the temperature field topology feature map, and the heating power adjustment amount is calculated by combining the temperature compensation value. The temperature compensation control parameters are obtained by solving based on the heating power adjustment amount and the desired temperature.

[0116] When setting the sliding window and step size based on the optimized temperature zone control parameter group, the sliding window size is determined according to the welding process characteristics. For the standard reflow welding process, the window size is set to 120 seconds, which can cover a complete temperature change cycle during the welding process. The step size is set to 10 seconds to ensure the continuity of data coverage while reducing the amount of computation. Taking an eight-temperature zone welding equipment as an example, the optimized control parameters for each temperature zone have been determined. For example, the parameters for zone Z3 are P=2.42, I=0.58, and D=0.26. Historical temperature data collected and stored during the welding process are sampled at a sampling frequency of 2Hz, that is, temperature data is collected once every 0.5 seconds. For temperature zone Z4, a sliding window contains 240 sampling points, forming a temperature sequence {T1, T2, ..., T240}. For example, a partial temperature sequence in the Z4 temperature zone during a welding process is {210.5, 211.2, 211.8, 212.5, 213.0, 213.6, 214.2, 214.7, 215.2, 215.8}℃. The rate of temperature change is calculated based on this sequence, which is the temperature difference between adjacent time points divided by the time interval. For data with a sampling interval of 0.5 seconds, the rate of change is calculated as the temperature difference between every two adjacent points multiplied by 2, in℃ / second. The corresponding rate of change sequence is {1.4, 1.2, 1.4, 1.0, 1.2, 1.2, 1.0, 1.0, 1.2}℃ / second. The standard deviation of temperature fluctuation is obtained by taking the square root of the average of the sum of squares of the deviations of the temperature sequence from its mean. For the aforementioned temperature sequence, the mean is 213.25℃, and the standard deviation is 1.76℃.

[0117] When the absolute value of the temperature change rate exceeds a preset temperature change rate threshold or the standard deviation of temperature fluctuation exceeds a preset standard deviation threshold, the temperature compensation control mechanism is triggered. The preset temperature change rate threshold is determined according to process requirements; for precision electronic component welding, it is typically set to 2℃ / second. The standard deviation threshold is set to 1% of the process temperature; for example, when the target temperature is 240℃, the standard deviation threshold is 2.4℃. When the temperature change rate in temperature zone Z4 reaches 2.3℃ / second, exceeding the preset threshold of 2℃ / second, the temperature error is calculated by the difference between the expected temperature and the current temperature. Assuming the expected temperature at this moment is 225℃ and the current actual temperature is 219℃, the temperature error is 6℃. The temperature error is processed using a piecewise adaptive Smith prediction compensation algorithm, dividing the welding temperature curve into three stages: preheating, reflow, and cooling. Different compensation parameters are set for the temperature characteristics of different stages. The preheating section (temperature range typically from room temperature to 180°C) is characterized by a rapid temperature rise, with a compensation parameter α set to 0.7; the reflux section (temperature range typically from 180°C to 250°C) is characterized by precise temperature control, with a compensation parameter β set to 1.2; the cooling section (temperature decreases from its peak) is characterized by a controlled temperature decrease, with a compensation parameter γ set to 0.5. Based on the current temperature of 219°C, we are in the reflux section, and a compensation parameter β=1.2 is applicable.

[0118] The lag time for each segment is calculated using the particle swarm optimization (PSO) algorithm. Lag time represents the delay between the control signal and the actual temperature response. The PSO algorithm parameters are set as follows: population size 30, maximum number of iterations 50, inertia weight 0.8, individual learning factor 1.5, and swarm learning factor 2.0. The objective function of the algorithm is to minimize the mean square error between the predicted and actual temperatures. For the recirculation segment in temperature zone Z4, the lag time calculated by the PSO algorithm is 4.5 seconds. Feedforward compensation is applied to the lag time to calculate the temperature compensation value. The compensation calculation formula is: Temperature compensation value = Temperature error × Compensation parameter × Lag time compensation coefficient. The lag time compensation coefficient is inversely proportional to the lag time and is calculated as 1 ÷ (1 + 0.2 × lag time). For temperature zone Z4, the lag time compensation coefficient is 0.53, and the final temperature compensation value = 6 × 1.2 × 0.53 = 3.82℃.

[0119] Temperature compensation values ​​are input into preset fuzzy control rules to further optimize control decisions. These fuzzy control rules are generated from expert experience and historical data analysis, using an "if-then" format. For example, "If the temperature compensation value is large and the temperature change rate is positive, then the heating power will increase moderately." Fuzzy control uses five fuzzy sets: negative large, negative small, zero, positive small, and positive large. The universe of discourse partitioning value corresponding to the optimized temperature zone control parameter group is calculated. The universe of discourse partitioning determines the boundaries and overlap of the fuzzy sets. A non-uniform partitioning method is used, with finer partitioning in the precise control region and coarser partitioning in the transition region. For temperature zone Z4, the universe of discourse partitioning for the temperature compensation value is {-10, -5, -2, -1, 0, 1, 2, 5, 10}℃, and the universe of discourse partitioning for the heating power adjustment is {-20, -10, -5, -2, 0, 2, 5, 10, 20}%. A temperature field topology feature map is constructed by combining the thermal stress corresponding to the circuit board material and the pre-set deformation control constraints. Thermal stress data were obtained through finite element analysis. For a standard PCB board made of FR-4 material, the thermal stress in the edge region was 15 MPa and in the center region was 12 MPa at a temperature of 240℃. The deformation control constraint set the maximum allowable deformation to be 0.2 mm. The temperature field topology feature map was presented in the form of a thermogram, with color depth representing temperature distribution and overlaid contour lines representing thermal stress distribution.

[0120] The weight distribution of different regions of the temperature field is calculated based on the topological feature map of the temperature field. The weight allocation principle is that regions with high thermal stress, large temperature gradients, and high component density have higher weights. A weighted average method is used to calculate the weight of each region. For example, temperature zone Z4 is divided into 9 sub-regions, with the central sub-region having a weight of 0.18, the edge sub-regions having a weight of 0.08, and the remaining sub-regions having weights between 0.10 and 0.15. The heating power adjustment is calculated in conjunction with the temperature compensation value. The adjustment calculation considers the heat capacity characteristics of the temperature zone and the current temperature state. The calculation formula is: Heating power adjustment = Temperature compensation value × Power coefficient × Comprehensive regional weight value. The power coefficient is directly proportional to the heat capacity of the temperature zone and inversely proportional to the current temperature. For temperature zone Z4, the heat capacity characteristic coefficient is 0.85, the current temperature coefficient is 0.95, and the comprehensive regional weight value is 0.14. The calculated heating power adjustment is 3.82 × 0.85 × 0.95 × 0.14 = 0.43 kW.

[0121] The temperature compensation control parameters are calculated based on the heating power adjustment and the desired temperature. These parameters include the compensated PID parameters and the prediction time constant. The PID parameters are adjusted incrementally, correcting the original PID parameters according to the magnitude and direction of the heating power adjustment. The calculation formula is: Compensated P parameter = Original P parameter × (1 + Heating power adjustment ratio × P correction coefficient), where the heating power adjustment ratio is the percentage of the adjusted power relative to full power, and the P correction coefficient is set to 0.2. The compensated I and D parameters are then calculated. For temperature zone Z4, the original PID parameters are P = 2.35, I = 0.52, D = 0.24, and the heating power adjustment ratio is 0.43 ÷ 2.0 = 0.215. The calculated compensated PID parameters are P = 2.45, I = 0.55, and D = 0.25. The prediction time constant is determined based on the system's dynamic characteristics, and the calculation formula is: Prediction time constant = Lag time × (1 - Heating power adjustment ratio × 0.3). For temperature zone Z4, the calculated prediction time constant is 4.18 seconds.

[0122] In this embodiment, by sampling historical temperature data through a sliding window based on the optimized temperature zone control parameter set and calculating the temperature change rate and temperature fluctuation standard deviation, abnormal temperature changes during welding can be identified in real time, improving the sensitivity and accuracy of temperature anomaly detection. By employing a piecewise adaptive Smith prediction compensation algorithm to process temperature errors in segments and combining it with a particle swarm optimization algorithm to optimize and compensate for the lag time of each segment, the impact of system lag on temperature control accuracy can be effectively reduced, improving the real-time performance and stability of compensation control. By inputting the temperature compensation value into fuzzy control rules and constructing a temperature field topology feature map in conjunction with thermal stress and deformation control constraints, spatial modeling of temperature distribution and optimized regulation under constraints can be achieved, improving the adaptability of the control process to complex working conditions.

[0123] Figure 2 This is a flowchart illustrating the temperature compensation parameter generation process of the multi-temperature zone welding temperature intelligent control method based on deep learning, as described in an embodiment of the present invention.

[0124] In one alternative implementation,

[0125] Based on the optimized temperature zone control parameter set and temperature compensation control parameters, temperature control commands are generated and executed, including:

[0126] The control cycle and temperature sampling interval are obtained from the optimized temperature zone control parameter group. Periodic temperature sampling is performed on each temperature zone of the circuit board to obtain temperature sampling values ​​and circuit board status parameters.

[0127] The sampled temperature value is compared with the temperature compensation control parameters to obtain the sampled temperature deviation value. The sampled temperature deviation value is then corrected according to the optimized temperature zone control parameter group to obtain the temperature control adjustment amount.

[0128] Based on the temperature control adjustment amount and the circuit board status parameters, a temperature control command containing heating power and heating time is generated, and the temperature control command is executed to adjust the temperature of the temperature zone.

[0129] The control cycle and temperature sampling interval are obtained from the optimized temperature zone control parameter set. This set includes various control parameters, where the control cycle determines the update frequency of control commands, and the temperature sampling interval determines the frequency of data acquisition. For precision electronic component soldering processes, the control cycle is typically set to 200 milliseconds, and the temperature sampling interval to 50 milliseconds, meaning each control cycle contains four temperature sampling points. Taking an eight-zone soldering machine as an example, the parameters for zone Z5 in the optimized temperature zone control parameter set include: P=2.15, I=0.48, D=0.22, control cycle=200 milliseconds, and sampling interval=50 milliseconds. Periodic temperature sampling is performed on each zone of the circuit board, using a high-precision thermocouple temperature sensor network distributed across each zone. Multiple sensors are configured for each zone to ensure spatial coverage of temperature sampling. For example, zone Z5 is equipped with five temperature sensors, located at the center and four corners of the zone. The sampled temperature values ​​are processed by a signal conditioning circuit and then transmitted to the control unit. During a certain control cycle, the temperature values ​​collected by the five sensors in temperature zone Z5 were 241.5℃, 240.8℃, 242.3℃, 241.2℃, and 241.0℃, respectively. The weighted average value of the temperature sampled in zone Z5 was calculated to be 241.3℃, with the center point sensor having a weight of 0.4 and the four corner sensors each having a weight of 0.15.

[0130] The circuit board status parameters are acquired, reflecting the physical state and process conditions of the circuit board during the soldering process. These parameters include: conveyor belt speed, circuit board thickness, thermal properties of the circuit board material, component density, and ambient temperature. These parameters are collected in real time through a pre-configured sensor network. For example, the circuit board status parameters acquired during a certain control cycle might be: conveyor belt speed 22 cm / min, circuit board thickness 1.6 mm, thermal properties of the material 1.2 joules / g·°C, component density 85%, and ambient temperature 24.5°C.

[0131] The sampled temperature value is compared with the temperature compensation control parameters, which include the target temperature value and the allowable deviation range. For temperature zone Z5, the target temperature is set at 240℃, and the allowable deviation range is ±2℃. The comparison process calculates the sampled temperature deviation value by measuring the difference between the sampled temperature value and the target temperature. Taking the aforementioned sampled data as an example, the sampled temperature deviation value for temperature zone Z5 is 241.3℃ - 240℃ = 1.3℃, indicating that the current temperature is slightly higher than the target temperature but still within the allowable range. Correcting the sampled temperature deviation value based on the optimized temperature zone control parameter set is a crucial step in improving control accuracy. The correction process is implemented based on the incremental PID control algorithm, which calculates the control increment based on the current deviation value, the rate of change of the historical deviation value, and the integral value of the deviation. Specifically, the P, I, and D parameters in the optimized temperature zone control parameter set are multiplied by the deviation value, the rate of change of the deviation, and the integral value of the deviation, respectively, and the sum is used to obtain the temperature control adjustment amount.

[0132] Taking temperature zone Z5 as an example, the current sampling temperature deviation is 1.3℃, the previous cycle deviation was 1.5℃, the deviation change rate is -0.2℃ / cycle, and the deviation integral value is 3.8℃·cycle. Using optimized PID parameters P=2.15, I=0.48, D=0.22, the calculations are as follows: Proportional term = 2.15 × 1.3 = 2.795, Integral term = 0.48 × 3.8 = 1.824, Derivative term = 0.22 × (-0.2) = -0.044. The sum of these three terms yields a temperature control adjustment of 4.575℃. Considering stability requirements, the maximum adjustment per cycle is set to 5℃, therefore the final temperature control adjustment is 4.575℃. The sign of the temperature control adjustment indicates the direction of adjustment; a positive value indicates a need for cooling, and a negative value indicates a need for heating.

[0133] Temperature control commands are generated based on temperature control adjustments and circuit board status parameters. These commands contain two key parameters: heating power and heating time. The heating power is calculated based on the temperature control adjustment and the circuit board's thermal characteristics. During calculation, the temperature control adjustment is converted into a heat regulation requirement, and the required power adjustment is calculated by considering the circuit board material's heat capacity, area, and heat transfer efficiency. The heating power calculation also needs to consider the influence of the circuit board's status parameters. For example, when the conveyor belt speed is high, the heating power needs to be increased to compensate for heat loss; when the component density is high, the uneven heat distribution needs to be considered, and the heating power distribution needs to be adjusted appropriately.

[0134] In this embodiment, by setting the control cycle and sampling interval based on the optimized temperature zone control parameter group for periodic temperature sampling, continuous monitoring of temperature changes in each temperature zone of the circuit board can be ensured, improving the real-time performance and completeness of temperature acquisition. By comparing the sampled temperature value with the temperature compensation control parameters and correcting the deviation, the impact of measurement errors and environmental disturbances on control accuracy during the sampling process can be effectively reduced, improving the accuracy of temperature regulation. Based on the corrected temperature control adjustment amount and circuit board status parameters, heating power and heating time commands are generated, enabling fine-grained control of the temperature zone, improving the stability and consistency of temperature control, ensuring uniform heating of the circuit board during soldering, improving solder joint quality, and enhancing the reliability of the production process.

[0135] A second aspect of the present invention provides a deep learning-based intelligent control system for multi-temperature zone welding temperature, comprising:

[0136] The first unit is used to acquire real-time temperature data of each temperature zone of the multi-temperature zone welding equipment in the circuit board welding production line, and to perform feature extraction and time series analysis to obtain temperature change prediction results.

[0137] The second unit is used to divide each temperature zone into multiple control levels based on the temperature change prediction results, calculate the temperature deviation value of each control level and generate initial weight coefficients, and fit the temperature gradient change data by the least squares method to iteratively calculate the temperature compensation coefficient and the initial control parameters of each temperature zone.

[0138] The third unit is used to determine the quantitative index based on the initial control parameters, calculate the influence factor of the quantitative index and the comprehensive evaluation score of the initial control parameters by combining the adaptive weight algorithm, and iteratively optimize the initial control parameters based on the comprehensive evaluation score to obtain the optimized temperature zone control parameter set.

[0139] The fourth unit is used to perform sliding window analysis on historical temperature data based on the optimized temperature zone control parameter group, calculate the temperature change rate and temperature fluctuation standard deviation. When the temperature change rate exceeds the preset threshold or the temperature fluctuation standard deviation is greater than the allowable range, the temperature compensation value is calculated based on the proportional-integral control algorithm and the heating power adjustment amount is determined in combination with the fuzzy control rules to generate temperature compensation control parameters.

[0140] The fifth unit is used to generate and execute temperature control commands based on the optimized temperature zone control parameter set and temperature compensation control parameters.

[0141] A third aspect of the present invention provides an electronic device, comprising:

[0142] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0143] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0144] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0145] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep learning-based intelligent temperature control method for multi-temperature zone welding, characterized in that, include: Real-time temperature data of each temperature zone in the multi-temperature zone welding equipment of the circuit board welding production line are obtained, and feature extraction and time series analysis are performed to obtain temperature change prediction results. Based on the temperature change prediction results, each temperature zone is divided into multiple control levels. The temperature deviation value of each control level is calculated and an initial weighting coefficient is generated. The temperature gradient change data is then fitted using the least squares method, and the temperature compensation coefficient and the initial control parameters for each temperature zone are iteratively calculated, including: Based on the temperature change prediction results, the temperature coupling degree between different temperature zones of the circuit board and the difference in heat capacity of components are calculated, and each temperature zone is divided into multiple control levels. For each control level, the difference between the preset target temperature value and the preset real-time temperature value is calculated to obtain the temperature deviation value. The initial weighting coefficient is calculated based on the temperature deviation value. Based on the pre-acquired temperature change trend matrix, the temperature gradient change data is calculated. The temperature deviation value and the temperature gradient change data of each temperature zone are fitted by the least squares method to obtain the fitting result. Combined with the pre-acquired temperature compensation coefficient of the previous iteration, the temperature compensation coefficient corresponding to the current iteration is calculated. The initial weighting coefficients, temperature deviation values, and temperature compensation coefficients are combined into an input vector and input into a pre-set adaptive fuzzy neural network. The membership parameters are obtained by fuzzification using a Gaussian membership function. Wavelet transform is performed on the membership parameters to obtain multi-scale feature coefficients. A hierarchical analysis decision tree is constructed based on the multi-scale feature coefficients and local extreme points at different scales are recursively calculated. The membership subsets of the local extreme points are determined. The cross-condition information of the membership subsets is calculated and dimensionality reduction and reconstruction are performed. The optimized membership parameters are obtained by fusing the reconstructed membership subsets. A fuzzy rule base is established based on the optimized membership parameters and the preset temperature control performance evaluation function. The frequent itemsets in the fuzzy rule base are mined by the Apriori algorithm and the support between different fuzzy rules is calculated. An association matrix is ​​constructed based on the support and singular value decomposition is performed to construct a rule activation sequence. The fuzzy rule base is sealed based on the rule activation sequence. The optimized membership parameters are processed by local linear functions to obtain the rule output values ​​and the optimized intermediate parameters are obtained by solving. When the temperature deviation exceeds the preset deviation threshold, the compensation control parameters are calculated based on the optimized intermediate parameters and the exponential moving average, and then output as the initial control parameters. The quantitative index is determined based on the initial control parameters. The influence factor of the quantitative index and the comprehensive evaluation score of the initial control parameters are calculated by combining the adaptive weight algorithm. The initial control parameters are iteratively optimized based on the comprehensive evaluation score to obtain the optimized temperature zone control parameter set. Based on the optimized temperature zone control parameter group, a sliding window analysis is performed on historical temperature data to calculate the temperature change rate and temperature fluctuation standard deviation. When the temperature change rate exceeds the preset threshold or the temperature fluctuation standard deviation is greater than the allowable range, the temperature compensation value is calculated based on the proportional-integral control algorithm and combined with fuzzy control rules to determine the heating power adjustment amount, thereby generating temperature compensation control parameters. Temperature control commands are generated and executed based on the optimized temperature zone control parameter group and temperature compensation control parameters.

2. The method according to claim 1, characterized in that, Real-time temperature data of each temperature zone in a multi-temperature zone welding equipment in a circuit board welding production line is acquired, and feature extraction and time series analysis are performed to obtain temperature change prediction results, including: Real-time temperature data of each temperature zone in the multi-temperature zone welding equipment in the circuit board welding production line are obtained, and feature extraction is performed on the real-time temperature data to obtain the temperature change pattern. A temperature change trend matrix is ​​established based on the temperature change pattern, and a time series analysis is performed on the temperature change trend matrix. Based on the results of the time series analysis, the predicted temperature changes for each temperature zone are calculated, and the predicted temperature changes are obtained.

3. The method according to claim 1, characterized in that, Based on the initial control parameters, a quantitative index is determined. The influence factor of the quantitative index and the comprehensive evaluation score of the initial control parameters are calculated using an adaptive weighting algorithm. Based on the comprehensive evaluation score, the initial control parameters are iteratively optimized to obtain an optimized temperature zone control parameter set, including: The control performance index corresponding to the initial control parameters is obtained, and the quantitative index is obtained by combining the pre-set mapping rules. The ambient temperature fluctuation value and the process parameter fluctuation value are obtained, and the state space vector is constructed by combining the quantitative index. Calculate the weight of each quantitative index in the state space vector and determine the information entropy. Based on the information entropy, calculate the influence factor corresponding to each quantitative index. Collect temperature change sequence within a preset time window and extract time-series and nonlinear features to calculate the parameter sensitivity index. Based on the adaptive firefly algorithm, the initial control parameters are encoded as the location information of individual fireflies. A fluorescence intensity evaluation function is constructed based on the control performance index, and the attraction coefficient is initialized based on the ambient temperature fluctuation value. The optimal individual is determined, and the location information of the optimal individual is decoded into parameter estimates. The Lyapunov function value corresponding to the parameter estimates is calculated, convergence is judged, and system stability is determined. Based on the parameter sensitivity index and the system stability, the quantitative index is fuzzy evaluated to obtain a comprehensive evaluation score. The gradient of the comprehensive evaluation score with respect to the initial control parameters is calculated, and the initial control parameters are updated according to the gradient direction. This update is repeated until the comprehensive evaluation score converges, resulting in an optimized temperature zone control parameter set.

4. The method according to claim 1, characterized in that, Based on the optimized temperature zone control parameter set, a sliding window analysis is performed on historical temperature data to calculate the temperature change rate and temperature fluctuation standard deviation. When the temperature change rate exceeds a preset threshold or the temperature fluctuation standard deviation is greater than the allowable range, a temperature compensation value is calculated based on the proportional-integral control algorithm, and the heating power adjustment is determined in conjunction with fuzzy control rules. The temperature compensation control parameters include: Based on the optimized temperature zone control parameter group, a sliding window and step size are set, and the historical temperature data collected and stored during the welding process are sampled to obtain a temperature sequence. Based on the temperature sequence, the temperature change rate and temperature fluctuation standard deviation corresponding to the historical temperature data are calculated. When the absolute value of the temperature change rate is greater than the preset temperature change rate threshold or the temperature fluctuation standard deviation is greater than the preset fluctuation standard deviation threshold, the temperature error is obtained by calculating the difference between the preset expected temperature and the current temperature. The temperature error is divided into a preheating section, a recirculation section and a cooling section by a segmented adaptive Smith prediction compensation algorithm. Compensation parameters are set for each section. The lag time of each section is calculated by combining the particle swarm algorithm and the lag time is fed forward to obtain the temperature compensation value. The temperature compensation value is input into a preset fuzzy control rule, the universe partition value corresponding to the optimized temperature zone control parameter group is calculated, and the temperature field topology feature map is constructed by combining the thermal stress corresponding to the circuit board material and the preset deformation control constraints. The weight distribution of different regions of the temperature field is calculated based on the temperature field topology feature map, and the heating power adjustment amount is calculated by combining the temperature compensation value. The temperature compensation control parameters are obtained by solving based on the heating power adjustment amount and the desired temperature.

5. The method according to claim 1, characterized in that, Based on the optimized temperature zone control parameter set and temperature compensation control parameters, temperature control commands are generated and executed, including: The control cycle and temperature sampling interval are obtained from the optimized temperature zone control parameter group. Periodic temperature sampling is performed on each temperature zone of the circuit board to obtain temperature sampling values ​​and circuit board status parameters. The sampled temperature value is compared with the temperature compensation control parameters to obtain the sampled temperature deviation value. The sampled temperature deviation value is then corrected according to the optimized temperature zone control parameter group to obtain the temperature control adjustment amount. Based on the temperature control adjustment amount and the circuit board status parameters, a temperature control command containing heating power and heating time is generated, and the temperature control command is executed to adjust the temperature of the temperature zone.

6. A deep learning-based intelligent control system for multi-temperature zone welding temperature, used to implement the method of any one of claims 1-5, characterized in that, include: The first unit is used to acquire real-time temperature data of each temperature zone of the multi-temperature zone welding equipment in the circuit board welding production line, and to perform feature extraction and time series analysis to obtain temperature change prediction results. The second unit is used to divide each temperature zone into multiple control levels based on the temperature change prediction results, calculate the temperature deviation value of each control level and generate initial weight coefficients, and fit the temperature gradient change data by the least squares method to iteratively calculate the temperature compensation coefficient and the initial control parameters of each temperature zone. The third unit is used to determine the quantitative index based on the initial control parameters, calculate the influence factor of the quantitative index and the comprehensive evaluation score of the initial control parameters by combining the adaptive weight algorithm, and iteratively optimize the initial control parameters based on the comprehensive evaluation score to obtain the optimized temperature zone control parameter set. The fourth unit is used to perform sliding window analysis on historical temperature data based on the optimized temperature zone control parameter group, calculate the temperature change rate and temperature fluctuation standard deviation. When the temperature change rate exceeds the preset threshold or the temperature fluctuation standard deviation is greater than the allowable range, the temperature compensation value is calculated based on the proportional-integral control algorithm and the heating power adjustment amount is determined in combination with the fuzzy control rules to generate temperature compensation control parameters. The fifth unit is used to generate and execute temperature control commands based on the optimized temperature zone control parameter set and temperature compensation control parameters.

7. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 5.

Images