AI-powered intelligent temperature control system and method for dried bean curd production line integrating environmental perception

Abstract

This invention discloses an AI-powered intelligent temperature control system and method for a tofu skin production line that incorporates environmental perception. Relating to the field of AI-powered intelligent temperature control, the system includes: establishing a comparison table of process objectives and staged parameter boundaries covering all technological stages of tofu skin production; real-time data collection, processing, and storage in a time-series database to construct an environment-process correlation database; setting an attention mechanism for sudden changes in environmental parameters, dynamically adjusting the attention weights of each parameter to adapt to production needs; constructing a neural network model mapping humidity, airflow velocity, and temperature, outputting the temperature adjustment step size and its effective probability; and constructing a comprehensive evaluation system with dual-model coupling, prioritizing temperature compensation for humidity and flow velocity effects, and adjusting multiple parameters in conjunction with probability-based judgments when failure occurs. The advantages of this invention are: effectively combining environmental perception data from the tofu skin production line to achieve intelligent and refined temperature control, effectively balancing the control needs of normal operating conditions and abnormal scenarios.

Description

Technical Field

[0001] This invention relates to the field of AI intelligent temperature control, specifically to an AI intelligent temperature control system and method for a dried bean curd production line that incorporates environmental perception. Background Technology

[0002] As a crucial step in traditional soy product processing, the quality of dried bean curd sticks is significantly affected by environmental parameters, especially the synergistic effect of humidity, air velocity, and temperature. In actual production, humidity is easily affected by workshop ventilation, raw material moisture content, and external weather, while air velocity fluctuates due to equipment layout and personnel operation. Both are difficult to control precisely and stably, becoming key bottlenecks restricting the uniformity of dried bean curd stick quality. For example, a sudden increase in humidity during the drying process can lead to uneven dehydration and mold growth, while unstable air velocity can cause differences in surface hardening speed, resulting in "mottled" defects. In contrast, temperature parameters can be controlled in a closed loop through heating devices, offering fast response and high precision. It is currently the most easily stable environmental variable to control in dried bean curd stick production lines. Therefore, the industry generally adopts the control approach of "compensating for the effects of humidity and wind speed fluctuations with temperature." By dynamically adjusting temperature parameters, the interference of abnormal humidity and air velocity on the quality of dried bean curd sticks (such as moisture content, elasticity, and color) is offset. This strategy has become the core means of balancing environmental uncertainty and production stability.

[0003] With the development of intelligent manufacturing technology, artificial intelligence algorithms are gradually being applied to the environmental control of tofu skin production lines. However, there is still room for optimization in complex environment correlation modeling and dynamic compensation strategy generation. It is urgent to build a more adaptable intelligent control system for the scenario characteristics of "uncontrollable humidity and wind speed - precise temperature compensation".

[0004] Existing technologies are insufficient to address the uncontrollability of humidity and airflow velocity. Most solutions employ a multi-parameter synchronous forced control mode, stabilizing humidity and flow velocity by adding dehumidification equipment and closed-loop control using wind speed sensors. However, such equipment is costly and energy-intensive, and is prone to coupling interference in multi-stage workshop scenarios (such as a sudden drop in local temperature caused by the activation of the dehumidification device). Furthermore, existing technologies do not fully utilize the easily controllable characteristics of temperature and lack temperature-based compensation logic, resulting in low control efficiency, which in turn leads to increased energy consumption and decreased resource utilization efficiency. Summary of the Invention

[0005] To address the aforementioned technical issues, this paper provides an AI-powered intelligent temperature control system and method for a tofu skin production line that incorporates environmental perception. This technical solution solves the problems mentioned in the background technology, such as high equipment cost, high energy consumption, and easy coupling interference in multi-stage linkage scenarios in the workshop. Furthermore, the existing technology does not fully utilize the easily controllable characteristics of temperature and lacks temperature-based compensation logic, resulting in low control efficiency, which in turn leads to increased energy consumption and decreased resource utilization efficiency.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] An AI-powered intelligent temperature control method for a dried bean curd production line that incorporates environmental sensing includes:

[0008] Based on the quality requirements of dried bean curd production, a comparison table of process objectives and phased parameter boundaries covering all process links in dried bean curd production was formulated.

[0009] High-precision environmental sensors and key quality feature detection devices are deployed in each process step of the dried bean curd production line. The data is collected and processed in real time and stored in a time-series database to build an environment-process correlation database.

[0010] Based on the comparison table, an attention mechanism is set up for sudden changes in environmental parameters, and the attention weight of each parameter is dynamically adjusted to adapt to production needs.

[0011] A neural network model mapping humidity, air velocity, and temperature is constructed with the goal of minimizing the deviation between each key quality characteristic and the process target value, and the temperature adjustment step size and its effective probability are output.

[0012] A comprehensive evaluation system based on dynamic attention weights is constructed using a dual-model coupling mechanism. It prioritizes compensating for the effects of humidity and flow rate through temperature, and adjusts multiple parameters in a coordinated manner when failure is detected by probability.

[0013] Preferably, the step of setting an attention mechanism for sudden changes in environmental parameters based on a lookup table, and dynamically adjusting the attention weights of each parameter to adapt to production needs, specifically includes:

[0014] Based on the comparison table and the data in the environment-process correlation database, and based on the expert scoring method, the basic weight values ​​of each environmental parameter corresponding to different process links and stages under normal operating conditions of dried bean curd production are set.

[0015] Based on the boundary thresholds of environmental parameters and combined with data from the environmental-process correlation database, a trigger threshold for sudden changes in environmental parameters is set.

[0016] Based on the trigger threshold of sudden changes in environmental parameters, calculate the weight increment of each environmental parameter relative to the basic weight after the sudden change.

[0017] Based on the weight increment of each environmental parameter relative to the basic weight after a sudden change, an attention mechanism for sudden changes in environmental parameters is constructed, and the attention weight of each parameter is dynamically adjusted to adapt to production needs.

[0018] Preferably, the weight increment of each environmental parameter relative to the basic weight after the sudden change specifically includes:

[0019] Based on the environment-process correlation database, obtain the processed environmental sensing data;

[0020] Based on the processed environmental perception data, the intensity of sudden changes in each environmental parameter is calculated;

[0021] Based on the processed environmental perception data, the initial influence weight of each environmental parameter in each process of dried bean curd production is calculated by multinomial correlation coefficient, and then the initial weight is corrected based on expert experience.

[0022] Calculate the attenuation coefficient of the sudden change in environmental parameters based on the duration of the sudden change.

[0023] Based on the intensity of the sudden change, the impact weight, and the duration decay coefficient, the weight increment of each environmental parameter relative to the basic weight is calculated after the sudden change.

[0024] Preferably, the construction of the humidity, air velocity, and temperature mapping neural network model, with the objective of minimizing the deviation between each key quality characteristic and the process target value, and outputting the temperature adjustment step size and its effective probability, specifically includes:

[0025] Based on the environment-process correlation database, preprocessed environmental perception data of temperature, humidity and air velocity, aligned with process steps, and corresponding key quality characteristic data are extracted.

[0026] Based on historical data or by designing multiple sets of orthogonal gradient combination test experiments, for different combinations of humidity and air velocity gradients, the temperature is controlled by PID segmented adjustment mode to obtain the correlation data between humidity, air velocity and temperature and key quality characteristics.

[0027] Based on the above-mentioned related data, the data is randomly divided and standardized in a 7:2:1 ratio to construct the training set, validation set, and test set required for training the neural network model.

[0028] Temperature, humidity, and air velocity are used as inputs, key quality feature data are used as supervision signals to calculate model loss, and temperature adjustment step size and its effective probability are used as outputs.

[0029] With the goal of minimizing the deviation between each key quality characteristic and the process target value, a loss function for a neural network model mapping humidity, air velocity, and temperature is constructed based on the mean square error formula.

[0030] A three-channel LSTM neural network layer was set up to extract the temporal features of humidity, air velocity and temperature respectively;

[0031] Based on the Transformer attention mechanism, attention weights for humidity-temperature time series features, air velocity-temperature time series features, and humidity and air velocity-temperature time series features are extracted respectively.

[0032] Attention weights are fused with multivariate features through a weighted summation method, and then concatenated with the temporal features output by the LSTM before being input into the fully connected layer.

[0033] A neural network model mapping humidity, air velocity, and temperature is constructed, denoted as Model 1. Based on the collected environmental perception data of temperature, humidity, and air velocity, as well as the corresponding key quality feature data, the model outputs the temperature adjustment step size and its effective probability.

[0034] Preferably, the comprehensive evaluation system based on dynamic attention weights and dual-model coupling prioritizes compensating for the effects of humidity and flow rate through temperature, and adjusts multiple parameters in a coordinated manner when failure is probabilistically determined. Specifically, this includes:

[0035] Based on the environment-process correlation database, environmental perception parameter features are used as inputs, and key quality feature data are used as supervision signals to calculate model loss.

[0036] The adjustment step size of each environmental perception parameter is used as the output quantity.

[0037] With the goal of minimizing the deviation between each key quality characteristic and the process target value, a loss function for the characteristics of all environmental sensing parameters is constructed based on the mean square error formula.

[0038] Based on the Transformer attention mechanism and the attention mechanism for sudden changes in environmental parameters, a comprehensive attention mechanism is constructed as a custom attention layer after the input layer of the neural network.

[0039] Based on the LSTM neural network, a prediction model for the adjustment step size of environmental perception parameter features is constructed, denoted as Model 2. The adjustment step size of environmental perception parameter features is dynamically predicted by collecting environmental perception data.

[0040] Based on Model 1 and Model 2, a comprehensive evaluation system with dual-model coupling is constructed. The step size is adjusted by comprehensively evaluating and outputting environmental perception parameter features based on the effective probability output by Model 1.

[0041] Determine whether the effective probability in the output of Model 1 is greater than the preset value. If it is, it means that the temperature control is effective and outputs the temperature adjustment step size generated by Model 1. If it is not, it means that the temperature control is ineffective and outputs the adjustment step size of each parameter feature of the environmental perception output by Model 2.

[0042] When the effective probability is less than the preset value, the temperature adjustment step size is a comprehensive temperature adjustment step size obtained by coupling the temperature adjustment step size output by Model 1 and the temperature adjustment step size output by Model 2.

[0043] If the temperature adjustment step size output by Model 1 is in the opposite direction to the temperature adjustment step size output by Model 2, then when the effective probability is less than the preset value, the temperature adjustment step size output by Model 2 shall prevail.

[0044] Furthermore, this solution proposes an AI-powered intelligent temperature control system for a bean curd stick production line that incorporates environmental sensing, to achieve the aforementioned AI-powered intelligent temperature control method for the bean curd stick production line, including:

[0045] The comparison table module is used to formulate a comparison table of process objectives and stage parameter boundaries covering each process link in the production of dried bean curd sticks, based on the quality requirements of dried bean curd stick production.

[0046] The database module is used to deploy high-precision environmental sensors and key quality feature detection devices in each process link of the dried bean curd production line, collect and process data in real time and store it in the time series database to build an environment-process association database.

[0047] The parameter control module is used to set an attention mechanism for sudden changes in environmental parameters according to a lookup table, and dynamically adjust the attention weight of each parameter to adapt to production needs; construct a neural network model mapping humidity, air velocity, and temperature, with the goal of minimizing the deviation between each key quality characteristic and the process target value, and output the temperature adjustment step size and its effective probability; construct a comprehensive evaluation system based on dynamic attention weights and dual-model coupling, which prioritizes temperature compensation for the influence of humidity and flow rate, and adjusts multiple parameters in a coordinated manner when failure is probabilistically determined.

[0048] Preferably, the parameter control module includes:

[0049] An attention mechanism unit is used to set an attention mechanism for sudden changes in environmental parameters according to a lookup table, and to dynamically adjust the attention weight of each parameter to adapt to production needs.

[0050] The temperature control unit is used to construct a neural network model that maps humidity, air velocity and temperature, with the goal of minimizing the deviation between each key quality characteristic and the process target value, and outputs the temperature adjustment step size and its effective probability.

[0051] The parameter control unit is used to construct a comprehensive evaluation system with dual-model coupling based on dynamic attention weights. It prioritizes compensating for the effects of humidity and flow rate through temperature, and adjusts multiple parameters in a coordinated manner when failure is probabilistically determined.

[0052] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0053] This solution provides an AI-powered intelligent temperature control system and method for a tofu skin production line that incorporates environmental perception. By using temperature as the core control variable and constructing a neural network model mapping humidity, airflow velocity, and temperature, this solution prioritizes the controllability of temperature to compensate for the impact of humidity and flow velocity fluctuations on product quality. This avoids the problems of high equipment cost, high energy consumption, and coupling interference caused by traditional multi-parameter forced control, significantly improving economic efficiency while ensuring control effectiveness. Furthermore, by setting an attention mechanism for sudden changes in environmental parameters, and using dynamic calculations of basic weights, sudden change trigger thresholds, and weight increments, the system captures abnormal fluctuations in environmental perception parameters in real time and adjusts the attention weights of each parameter accordingly, thereby proactively strengthening the control of sudden environmental changes. The impact of key parameters is mitigated to avoid the impact of sudden environmental changes on the quality of dried bean curd sticks, significantly improving the production line's adaptability to complex environments. Finally, by constructing a comprehensive evaluation system with dual-model coupling, under normal operating conditions, the mapping model outputs precise temperature adjustment step sizes through LSTM temporal feature extraction and Transformer attention mechanism; when the effective probability of temperature control is insufficient, the parameter adjustment model is immediately linked, and multi-parameter collaborative optimization is achieved based on dynamic attention weights. This mode of single-parameter fine control and multi-parameter emergency linkage not only solves the shortcomings of traditional single models in taking into account both steady-state and abnormal scenarios, but also achieves smooth switching through effective probability judgment, ensuring the stability of dried bean curd stick quality throughout the entire production cycle. Attached Figure Description

[0054] Figure 1 This is a flowchart of the AI-powered intelligent temperature control method for a dried bean curd production line that incorporates environmental sensing, as described in this invention.

[0055] Figure 2 According to the reference table of this invention, an attention mechanism for sudden changes in environmental parameters is set up, and the attention weight of each parameter is dynamically adjusted to adapt to the production demand flowchart.

[0056] Figure 3 This is a flowchart illustrating the weight increment of each environmental parameter relative to the basic weight after the sudden change in this invention.

[0057] Figure 4 To construct a neural network model that maps humidity, air velocity and temperature in this invention, with the goal of minimizing the deviation between each key quality feature and the process target value, the model outputs a flowchart of the temperature adjustment step size and its effective probability.

[0058] Figure 5 The comprehensive evaluation system based on dynamic attention weights for constructing a dual-model coupling prioritizes compensating for the effects of humidity and flow rate through temperature, and adjusts the multi-parameter flowchart in response to probability-based failure. Detailed Implementation

[0059] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0060] Reference Figure 1 As shown, an AI-powered intelligent temperature control method for a dried bean curd production line, incorporating environmental sensing, includes:

[0061] Based on the quality requirements of dried bean curd production, a comparison table of process objectives and phased parameter boundaries covering all process links in dried bean curd production was formulated.

[0062] High-precision environmental sensors and key quality feature detection devices are deployed in each process step of the dried bean curd production line. The data is collected and processed in real time and stored in a time-series database to build an environment-process correlation database.

[0063] Based on the comparison table, an attention mechanism is set up for sudden changes in environmental parameters, and the attention weight of each parameter is dynamically adjusted to adapt to production needs.

[0064] A neural network model mapping humidity, air velocity, and temperature is constructed with the goal of minimizing the deviation between each key quality characteristic and the process target value, and the temperature adjustment step size and its effective probability are output.

[0065] A comprehensive evaluation system based on dynamic attention weights is constructed using a dual-model coupling mechanism. It prioritizes compensating for the effects of humidity and flow rate through temperature, and adjusts multiple parameters in a coordinated manner when failure is detected by probability.

[0066] The environmental sensing parameters mainly include temperature, humidity, air velocity, light intensity, water quality, microbial environment, and air pressure. Due to the fixed production site, indoor production, and production mode, the influence of environmental factors such as light intensity, water quality, microbial environment, and air pressure is relatively constant. Under normal circumstances, without significant environmental changes, the fluctuation range of these environmental sensing parameters within a single batch production cycle is small. Humidity and air velocity, however, are affected by uncontrollable factors such as season, weather, and human factors, resulting in a dynamic effect that requires regulation. Temperature, on the other hand, can be controlled in a closed-loop manner through heating devices, offering fast response and high precision. It is currently the most easily stabilized and controllable environmental variable in the tofu skin production line. Therefore, when temperature can compensate for the dynamic effects of humidity and air velocity, temperature control can be prioritized to compensate for their influence. This solution utilizes an attention mechanism for sudden changes in environmental parameters, a neural network model mapping humidity, air velocity, and temperature, and a predictive model for adjusting the step size of environmental sensing parameter features. It prioritizes temperature compensation for the effects of humidity and air velocity, and adjusts multiple parameters in a coordinated manner when these fail, achieving intelligent temperature control in tofu skin production and improving the stability of tofu skin production quality.

[0067] The aforementioned table, which establishes a comparison of process objectives and phased parameter boundaries for each stage of the tofu skin production process based on quality requirements, specifically includes:

[0068] Based on the quality requirements of dried bean curd production, the production process is divided into several steps, and each step is marked with a number.

[0069] Based on historical testing experience and the quality requirements of dried bean curd production, key quality characteristics affecting the quality of dried bean curd production in each process step are extracted, and the process targets for each key quality characteristic are quantified.

[0070] Based on the production requirements of each process, each process is divided into several stages, which are then marked with textual descriptions.

[0071] Based on the process objectives of each key quality characteristic, and using historical data or test experiments, the distinguishing boundaries of each stage of each process step are selected through normal distribution.

[0072] Based on historical data or test experiments, the maximum allowable range of environmental parameters for each stage is set, and the two extreme values ​​of the range are used as the boundary thresholds of the environmental parameters. If the environmental parameters exceed the maximum allowable range, it indicates that the quality of dried bean curd production is seriously affected and forced adjustment needs to be triggered.

[0073] Based on process steps, key quality characteristics, process objectives, process stages and boundaries, and environmental parameter boundary thresholds, a comparison table of process objectives and stage parameter boundaries covering all process steps in the production of dried bean curd sticks was developed.

[0074] This can be explained by the fact that the various processes in a bean curd stick production line include: raw material screening, soaking, grinding and separating, boiling, peeling (forming), drying, cutting and packaging, etc. Different processes involve different stages; for example, boiling includes a heating stage and a heat-preserving stage, while drying includes pre-drying and final drying. The specific stage divisions can be made according to the actual needs of bean curd stick production. By extracting key quality characteristics of each process (such as the fineness of the soy milk, moisture content, elasticity, color, and film thickness of the bean curd stick) and quantifying its process objectives, the quality requirements of each stage can be transformed from vague descriptions into measurable and verifiable specific indicators (such as a moisture content target of ≤12% in the drying stage). This provides a clear basis for quality assessment during the production process and is based on historical data. Through experiments, the maximum allowable range and boundary thresholds of environmental parameters (temperature, humidity, air velocity, etc.) were set for different stages of each process (such as the heating stage and the heat preservation stage in the boiling process). The boundaries between normal operating conditions and sudden environmental changes were clarified (such as the need for forced cooling when the temperature exceeds 65℃ in the drying stage). This was done to avoid irreversible impacts on quality caused by abnormal environmental parameters. By using a reference table, the process links, stages, quality objectives, and environmental parameter thresholds were deeply correlated. This provided a benchmark for the attention mechanism (determining when to adjust weights), neural network model training (clarifying optimization objectives), and dual-model coupling control (determining the direction of adjustment) for subsequent sudden changes in environmental parameters. This ensured that all control strategies revolved around optimizing the quality of dried bean curd production and provided standard boundaries for subsequent steps.

[0075] The deployment of high-precision environmental sensors and key quality characteristic detection devices in each process stage of the dried bean curd production line, the real-time collection and processing of data, and the storage of data in a time-series database to construct an environment-process correlation database specifically includes:

[0076] High-precision environmental sensors and key quality feature detection devices are deployed in each process stage of the dried bean curd production line to collect environmental perception data and key quality feature data of each process stage in real time.

[0077] The PTP protocol is used to synchronize the hardware clock of environmental sensors and key quality feature detection devices, and all data records are recorded with a unified timestamp.

[0078] Kalman filtering is performed in real time on high-frequency continuous environmental data, and moving average filtering is performed on low-frequency discrete key quality characteristic data after buffering and accumulation, and timestamps are aligned by interpolation.

[0079] Based on min-max normalization, the collected environmental perception data and key quality characteristic data are normalized to eliminate the influence of data units.

[0080] An environment-process correlation database is constructed to store processed environmental perception data and key quality characteristic data, providing data support for subsequent model training and control decisions.

[0081] This can be explained by the fact that the acquisition and preprocessing of data on the weight increments of each environmental parameter relative to the basic weights after a sudden change are crucial for subsequent model training and control decisions. Hardware clock synchronization of all sensors is achieved through the PTP protocol, Unix timestamps and nanosecond offsets are added to the data, and clock drift is calibrated every 10 seconds. Kalman filtering (state noise Q=0.01, observation noise R=0.1) is applied to high-frequency continuous environmental data. For low-frequency discrete quality data, a weighted moving average filter is applied after accumulating five samples, and a time series with the same frequency as the continuous data is generated through cubic spline interpolation. This ensures a precise correlation between changes in environmental parameters and the response of quality characteristics in the time dimension. This time alignment provides a basis for subsequent queries of key quality characteristic feedback results after environmental parameter adjustments. Thus, the changes in quality characteristics after adjusting environmental parameters can be traced using timestamps, providing a temporal basis for verifying control effects and optimizing the model.

[0082] Reference Figure 2 As shown, the step of setting an attention mechanism for sudden changes in environmental parameters based on a lookup table, and dynamically adjusting the attention weights of each parameter to adapt to production needs, specifically includes:

[0083] Based on the comparison table and the data in the environment-process correlation database, and based on the expert scoring method, the basic weight values ​​of each environmental parameter corresponding to different process links and stages under normal operating conditions of dried bean curd production are set.

[0084] Based on the boundary thresholds of environmental parameters and combined with data from the environmental-process correlation database, a trigger threshold for sudden changes in environmental parameters is set.

[0085] Based on the trigger threshold of sudden changes in environmental parameters, calculate the weight increment of each environmental parameter relative to the basic weight after the sudden change.

[0086] Based on the weight increment of each environmental parameter relative to the basic weight after a sudden change, an attention mechanism for sudden changes in environmental parameters is constructed, and the attention weight of each parameter is dynamically adjusted to adapt to production needs.

[0087] This can be explained by the fact that, due to the relatively constant influence of environmental factors such as light, water quality, microbial environment, and air pressure in a fixed production site, indoor production, and production mode, the impact of these environmental perception parameters fluctuates within a small range during a single batch production cycle, provided there are no significant sudden environmental changes. However, sudden changes in environmental parameters due to changes in water sources, production sites, or other influencing factors may lead to inaccurate subsequent model predictions. Therefore, it is necessary to set up an attention mechanism for sudden changes in environmental parameters to adjust model parameters in a timely manner and reduce model output bias when sudden environmental changes occur. Thus, this solution, based on a reference table and combined with data from the environment-process correlation database, sets up a dynamic attention mechanism for sudden changes in environmental parameters. This allows for timely detection and adjustment of model parameters through an attention mechanism, thereby improving the accuracy of model predictions. Specifically, when the deviation between the monitored environmental parameter value and the historical steady-state mean exceeds the normal fluctuation threshold, it is determined to be a sudden change in the parameter, triggering an adjustment of the attention weight. The normal fluctuation threshold is obtained based on the 3σ principle of historical data. Under normal operating conditions, static environmental parameters (such as light, water quality, and air pressure) have a lower base weight, while dynamic parameters (temperature, humidity, and air velocity) have a higher weight. When a sudden change occurs, only the weight of the parameter that has changed suddenly is adjusted. The weight value of the parameter can also provide priority support for the adjustment step size of the environmental parameters output by the subsequent dual-model coupled comprehensive evaluation system. That is, the higher the weight value, the higher the adjustment priority.

[0088] The expression for the attention mechanism of the sudden change in environmental parameters is as follows:

[0089]

[0090] In the formula, For the first time after the sudden change Attention weights for each environmental parameter, Based on the weight value, The attenuation coefficient is... For the first time after the sudden change The weight increment of each environmental parameter relative to the base weight. For the first An environmental parameter at time The monitoring value, For the first Historical steady-state mean of each environmental parameter For the first The normal fluctuation thresholds for each environmental parameter are obtained based on the 3σ principle of historical data.

[0091] Reference Figure 3 As shown, the weight increments of each environmental parameter relative to the base weights after the sudden change specifically include:

[0092] Based on the environment-process correlation database, obtain the processed environmental sensing data;

[0093] Based on the processed environmental perception data, the intensity of sudden changes in each environmental parameter is calculated;

[0094] Based on the processed environmental perception data, the initial influence weight of each environmental parameter in each process of dried bean curd production is calculated by multinomial correlation coefficient, and then the initial weight is corrected based on expert experience.

[0095] Calculate the attenuation coefficient of the sudden change in environmental parameters based on the duration of the sudden change.

[0096] Based on the intensity of the sudden change, the impact weight, and the duration decay coefficient, the weight increment of each environmental parameter relative to the basic weight is calculated after the sudden change.

[0097] It can be explained that the weight increment of each environmental parameter relative to the base weight after a sudden change is the core parameter in the attention mechanism of sudden environmental parameter changes. It needs to be comprehensively evaluated by combining the intensity of the sudden change, the influence weight of each environmental parameter in each process of tofu skin production, and the duration of the sudden change. By analyzing data on the intensity of sudden changes of different environmental parameters, the influence weight of each environmental parameter in each process of tofu skin production, and the duration of the sudden change, the impact of the weight increment of each environmental parameter relative to the base weight after a sudden change can be quantified, thereby improving the environmental adaptability, dynamism, and accuracy of the attention mechanism of sudden environmental parameter changes. The intensity of the sudden change of each environmental parameter is expressed as:

[0098]

[0099] In the formula, For the first The intensity of sudden changes in an environmental parameter, For the first An environmental parameter at time The monitoring value, For the first Historical steady-state mean of each environmental parameter For the first The normal fluctuation threshold of an environmental parameter For the first The upper limit value of an environmental parameter. For the first The lower limit value of each environmental parameter;

[0100] The expression for the attenuation coefficient of the sudden change in environmental parameters is:

[0101]

[0102] In the formula, For the first The decay coefficient of the duration of sudden changes in environmental parameters. For the first The attenuation coefficient of each environmental parameter The duration of the sudden change, The preset stabilization threshold time is used; if this time is exceeded, the value is forcibly reset to 0.

[0103] The expression for the weight increment of each environmental parameter relative to the base weight after the sudden change is as follows:

[0104]

[0105] In the formula, For the first time after the sudden change The weight increment of each environmental parameter relative to the base weight. This is the global adjustment coefficient. For the first time after the sudden change The influence weight of each environmental parameter in each process of dried bean curd production;

[0106] The upper and lower limits of environmental parameters are determined based on the characteristics of the dried bean curd production process, using the failure thresholds of key quality indicators (such as breakage rate, mold rate, and surface black spot density) as critical points, through experimental data fitting and production verification. The preset empirical time threshold for the system to recover to steady state after a sudden change in environmental parameters is determined based on the average time taken for parameters to return to the normal fluctuation range in historical sudden change events. For the first time after the sudden change The influence weights of each environmental parameter in each process stage of tofu skin production are the influence weight values ​​after correction based on expert experience, building upon the initial influence weights. Specifically, in the expression for the attenuation coefficient of the duration of abrupt changes in environmental parameters... When a sudden change in environmental parameters occurs, it can be understood as a one-time change in water quality and air pressure parameters caused by a change in water quality or site. Subsequently, due to the determination of the water source and site, there will be no significant changes. Therefore, it is necessary to focus on the parameter at the moment when the sudden change occurs and set the influence intensity to 1 to increase the attention weight of the parameter under the condition of sudden environmental change.

[0107] Reference Figure 4 As shown, the constructed humidity, air velocity, and temperature mapping neural network model aims to minimize the deviation between each key quality characteristic and the process target value, and outputs the temperature adjustment step size and its effective probability, specifically including:

[0108] Based on the environment-process correlation database, preprocessed environmental perception data of temperature, humidity and air velocity, aligned with process steps, and corresponding key quality characteristic data are extracted.

[0109] Based on historical data or by designing multiple sets of orthogonal gradient combination test experiments, for different combinations of humidity and air velocity gradients, the temperature is controlled by PID segmented adjustment mode to obtain the correlation data between humidity, air velocity and temperature and key quality characteristics.

[0110] Based on the above-mentioned related data, the data is randomly divided and standardized in a 7:2:1 ratio to construct the training set, validation set, and test set required for training the neural network model.

[0111] Temperature, humidity, and air velocity are used as inputs, key quality feature data are used as supervision signals to calculate model loss, and temperature adjustment step size and its effective probability are used as outputs.

[0112] With the goal of minimizing the deviation between each key quality characteristic and the process target value, a loss function for a neural network model mapping humidity, air velocity, and temperature is constructed based on the mean square error formula.

[0113] A three-channel LSTM neural network layer was set up to extract the temporal features of humidity, air velocity and temperature respectively;

[0114] Based on the Transformer attention mechanism, attention weights for humidity-temperature time series features, air velocity-temperature time series features, and humidity and air velocity-temperature time series features are extracted respectively.

[0115] Attention weights are fused with multivariate features through a weighted summation method, and then concatenated with the temporal features output by the LSTM before being input into the fully connected layer.

[0116] A neural network model mapping humidity, air velocity, and temperature is constructed, denoted as Model 1. Based on the collected environmental perception data of temperature, humidity, and air velocity, as well as the corresponding key quality feature data, the model outputs the temperature adjustment step size and its effective probability.

[0117] This can be explained by using temperature as the core control variable, prioritizing its controllability to compensate for the impact of humidity and flow rate fluctuations on product quality. This avoids the problems of high equipment cost, high energy consumption, and coupling interference caused by traditional multi-parameter forced control, significantly improving economic efficiency while ensuring control effectiveness. Based on the powerful fitting ability of neural networks, by inputting environmental parameter data, the model aims to minimize the deviation between each key quality feature and the process target value. A loss function for a neural network model mapping humidity, air flow rate, and temperature is constructed based on the mean square error formula. This dynamically adjusts the temperature value by reducing the deviation between each key quality feature and the process target value, achieving dynamic optimization. Key quality feature data serves as a monitoring signal used to calculate the model loss. The detection results of key quality features at the same time are retrieved from the environment-process correlation database and compared with the process target value to obtain the deviation, which is used to adjust the temperature compensation strategy. This post-validation feedback mechanism improves the model's control accuracy by dynamically correcting the temperature adjustment step size.

[0118] The feature extraction of the three-channel LSTM layer specifically includes:

[0119] The three-channel LSTM receives time-series data of humidity, air velocity, and temperature respectively (the time-series window length is set to 10 sampling points, and the input dimension of a single channel is batch, 10, 1).

[0120] Each channel uses a single-layer LSTM (hidden layer dimension 64) to extract temporal features independently. After the features are concatenated, a fused feature vector (dimensions batch, 10, 192, i.e. 3×64) is obtained, realizing the hierarchical extraction and preliminary fusion of temporal features of humidity, air velocity, and temperature.

[0121] The Transformer attention mechanism specifically includes:

[0122] The fused features (batch, 10, 192) output by the LSTM are used as input, and a 4-head self-attention mechanism is employed.

[0123] The three types of cross-attention weights are calculated by querying and mapping the humidity-temperature time series features, the air velocity-temperature time series features, and the humidity and air velocity-temperature time series features respectively.

[0124] Attention-weighted features and original fused features are connected via residual connections and normalized through fully connected layers to output fused features that enhance key associations (dimensions batch, 10, 192).

[0125] The fusion features processed by the attention mechanism are reduced in dimensionality (batch, 192) by global average pooling and then connected to a fully connected layer with linear and sigmoid activation layers.

[0126] The linear layer outputs a continuous value for the temperature adjustment step size, and the sigmoid activation layer outputs an effective probability value in the range [0,1].

[0127] The loss function of the neural network model mapping humidity, air velocity, and temperature is:

[0128]

[0129] In the formula, The loss value between key quality characteristics and process target values. The number of key quality characteristics. For the first The weights of the key quality characteristics, For the first The detection values ​​of key quality characteristics, For the first Process target values ​​for key quality characteristics;

[0130] Reference Figure 5 As shown, the comprehensive evaluation system based on dynamic attention weights and dual-model coupling prioritizes compensating for the effects of humidity and flow rate through temperature, and adjusts multiple parameters in a coordinated manner when failure is probabilistically determined. Specifically, this includes:

[0131] Based on the environment-process correlation database, environmental perception parameter features are used as inputs, and key quality feature data are used as supervision signals to calculate model loss.

[0132] The adjustment step size of each environmental perception parameter is used as the output quantity.

[0133] With the goal of minimizing the deviation between each key quality characteristic and the process target value, a loss function for the characteristics of all environmental sensing parameters is constructed based on the mean square error formula.

[0134] Based on the Transformer attention mechanism and the attention mechanism for sudden changes in environmental parameters, a comprehensive attention mechanism is constructed as a custom attention layer after the input layer of the neural network.

[0135] Based on the LSTM neural network, a prediction model for the adjustment step size of environmental perception parameter features is constructed, denoted as Model 2. The adjustment step size of environmental perception parameter features is dynamically predicted by collecting environmental perception data.

[0136] Based on Model 1 and Model 2, a comprehensive evaluation system with dual-model coupling is constructed. The step size is adjusted by comprehensively evaluating and outputting environmental perception parameter features based on the effective probability output by Model 1.

[0137] Determine whether the effective probability in the output of Model 1 is greater than the preset value. If it is, it means that the temperature control is effective and outputs the temperature adjustment step size generated by Model 1. If it is not, it means that the temperature control is ineffective and outputs the adjustment step size of each parameter feature of the environmental perception output by Model 2.

[0138] When the effective probability is less than the preset value, the temperature adjustment step size is a comprehensive temperature adjustment step size obtained by coupling the temperature adjustment step size output by Model 1 and the temperature adjustment step size output by Model 2.

[0139] If the temperature adjustment step size output by Model 1 is in the opposite direction to the temperature adjustment step size output by Model 2, then when the effective probability is less than the preset value, the temperature adjustment step size output by Model 2 shall prevail.

[0140] This can be explained by the fact that the humidity, air velocity, and temperature mapping neural network model, i.e., Model 1, is used to focus on the control of environmental parameters under normal production conditions. That is, in the absence of sudden environmental changes, Model 1 can effectively compensate for the impact of humidity and air velocity on the quality of dried bean curd production through temperature control. However, when the effective probability value output by Model 1 is less than the preset value, the temperature adjustment of Model 1 may fail. Therefore, a more accurate and effective adjustment of environmental parameters is needed to handle situations where the effective probability value is less than the preset value. The preset value is obtained through ROC curve analysis. Therefore, this solution, based on a neural network algorithm, introduces a fitting mapping relationship of all environmental parameter inputs, combined with an attention mechanism for sudden changes in environmental parameters, and aims to minimize the deviation between each key quality characteristic and the process target value. This results in the construction of an environmental perception parameter feature adjustment step size prediction model, i.e., Model 2. The system comprehensively evaluates and outputs the environmental perception parameter features to adjust the step size, and combines this with the output of Model 1 to dynamically fit and output the final environmental parameter adjustment step size. This effectively improves the comprehensive control capability and accuracy of the dual-model coupled comprehensive evaluation system. The specific implementation process based on the Transformer attention mechanism is the same as in Model 1. This layer calculates global cross-attention weights through querying, key-value mapping of multiple parameters including temperature, humidity, air velocity, light intensity, water quality, microbial environment, and air pressure. The feature dimensions remain consistent with the original Transformer output. It is then fused with the attention mechanism for sudden changes in environmental parameters using a linear weighting method to construct a comprehensive attention mechanism. This takes into account local anomaly sensitivity and improves the model's adaptability to complex environments. The expression for the comprehensive attention mechanism is:

[0141]

[0142] In the formula, To incorporate attention weights, These are the weight coefficients for the Transformer attention mechanism. The weight values ​​output by the Transformer attention mechanism;

[0143] Among them, the weight coefficients of the Transformer attention mechanism The weighting coefficients of the temperature adjustment step size output by the neural network model that maps humidity, air velocity, and temperature. The result is obtained by minimizing the loss function of the validation set, specifically by using a grid search method to optimize within the interval [0.1, 0.9] with a step size of 0.1.

[0144] The temperature adjustment step size coupling yields the comprehensive temperature adjustment step size expression as follows:

[0145]

[0146] In the formula, For comprehensive temperature adjustment step size, The weighting coefficients for adjusting the temperature adjustment step size are used to map humidity, air velocity, and temperature to the output of the neural network model. The temperature adjustment step size is mapped to the output of the neural network model that maps humidity, air velocity, and temperature. The temperature adjustment step size output by the prediction model for adjusting the step size of environmental perception parameter features.

[0147] Furthermore, based on the same inventive concept as the aforementioned AI-powered intelligent temperature control method for a bean curd stick production line incorporating environmental perception, this solution proposes an AI-powered intelligent temperature control system for a bean curd stick production line incorporating environmental perception, comprising:

[0148] The comparison table module is used to formulate a comparison table of process objectives and stage parameter boundaries covering each process link in the production of dried bean curd sticks, based on the quality requirements of dried bean curd stick production.

[0149] The database module is used to deploy high-precision environmental sensors and key quality feature detection devices in each process link of the dried bean curd production line, collect and process data in real time and store it in the time series database to build an environment-process association database.

[0150] The parameter control module is used to set an attention mechanism for sudden changes in environmental parameters according to a lookup table, and dynamically adjust the attention weight of each parameter to adapt to production needs; construct a neural network model mapping humidity, air velocity, and temperature, with the goal of minimizing the deviation between each key quality characteristic and the process target value, and output the temperature adjustment step size and its effective probability; construct a comprehensive evaluation system with dual-model coupling based on dynamic attention weights, prioritizes temperature compensation for the influence of humidity and flow rate, and adjusts multiple parameters in a coordinated manner when failure is probabilistically determined;

[0151] The parameter control module includes:

[0152] An attention mechanism unit is used to set an attention mechanism for sudden changes in environmental parameters according to a lookup table, and to dynamically adjust the attention weight of each parameter to adapt to production needs.

[0153] The temperature control unit is used to construct a neural network model that maps humidity, air velocity and temperature, with the goal of minimizing the deviation between each key quality characteristic and the process target value, and outputs the temperature adjustment step size and its effective probability.

[0154] The parameter control unit is used to construct a comprehensive evaluation system with dual-model coupling based on dynamic attention weights. It prioritizes compensating for the effects of humidity and flow rate through temperature, and adjusts multiple parameters in a coordinated manner when failure is probabilistically determined.

[0155] In summary, the advantages of this invention are: it effectively combines environmental sensing data from the tofu skin production line to achieve intelligent and precise temperature control of the tofu skin production line, effectively balancing the control needs of normal operating conditions and abnormal scenarios.

[0156] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. An AI-powered intelligent temperature control method for a dried bean curd production line that incorporates environmental sensing, characterized in that: include: Based on the quality requirements of dried bean curd production, a comparison table of process objectives and phased parameter boundaries covering all process links in dried bean curd production was formulated. High-precision environmental sensors and key quality characteristic detection devices are deployed in each process link of the tofu skin production line. The data is collected in real time, processed and stored in a time-series database to build an environment-process correlation database. The key quality characteristics include at least the soy milk fineness, moisture content, elasticity, color and film thickness of the tofu skin. Based on the comparison table, an attention mechanism is set up for sudden changes in environmental parameters, and the attention weight of each parameter is dynamically adjusted to adapt to production needs. A neural network model mapping humidity, air velocity, and temperature is constructed with the goal of minimizing the deviation between each key quality characteristic and the process target value, and the temperature adjustment step size and its effective probability are output. A comprehensive evaluation system based on dynamic attention weights and dual-model coupling is constructed. It prioritizes compensating for the effects of humidity and flow rate by temperature, and adjusts multiple parameters in a coordinated manner when failure is probabilistically determined. The aforementioned table, which establishes a comparison of process objectives and phased parameter boundaries for each stage of the tofu skin production process based on quality requirements, specifically includes: Based on the quality requirements of dried bean curd production, the production process is divided into several steps, and each step is marked with a number. Based on historical testing experience and the quality requirements of dried bean curd production, key quality characteristics affecting the quality of dried bean curd production in each process step are extracted, and the process targets for each key quality characteristic are quantified. Based on the production requirements of each process, each process is divided into several stages, which are then marked with textual descriptions. Based on the process objectives of each key quality characteristic, and using historical data or test experiments, the distinguishing boundaries of each stage of each process step are selected through normal distribution. Based on historical data or test experiments, the maximum allowable range of environmental parameters for each stage is set, and the two extreme values ​​of the range are used as the boundary thresholds of the environmental parameters. If the environmental parameters exceed the maximum allowable range, it indicates that the quality of dried bean curd production is seriously affected and forced adjustment needs to be triggered. Based on process steps, key quality characteristics, process objectives, process stages and boundaries, and environmental parameter boundary thresholds, a comparison table of process objectives and stage parameter boundaries covering all process steps in the production of dried bean curd sticks was developed.

2. The AI-powered intelligent temperature control method for a dried bean curd production line based on environmental perception, as described in claim 1, is characterized in that... The deployment of high-precision environmental sensors and key quality characteristic detection devices in each process stage of the dried bean curd production line, the real-time collection and processing of data, and the storage of data in a time-series database to construct an environment-process correlation database specifically includes: High-precision environmental sensors and key quality feature detection devices are deployed in each process stage of the dried bean curd production line to collect environmental perception data and key quality feature data of each process stage in real time. The PTP protocol is used to synchronize the hardware clock of environmental sensors and key quality feature detection devices, and all data records are recorded with a unified timestamp. Kalman filtering is performed in real time on high-frequency continuous environmental data, and moving average filtering is performed on low-frequency discrete key quality characteristic data after buffering and accumulation, and timestamps are aligned by interpolation. Based on min-max normalization, the collected environmental perception data and key quality characteristic data are normalized to eliminate the influence of data units. An environment-process correlation database is constructed to store processed environmental perception data and key quality characteristic data, providing data support for subsequent model training and control decisions.

3. The AI-powered intelligent temperature control method for a dried bean curd production line based on environmental perception, as described in claim 2, is characterized in that... The step of setting up an attention mechanism for sudden changes in environmental parameters based on a lookup table, and dynamically adjusting the attention weights of each parameter to adapt to production needs, specifically includes: Based on the comparison table and the data in the environment-process correlation database, and based on the expert scoring method, the basic weight values ​​of each environmental parameter corresponding to different process links and stages under normal operating conditions of dried bean curd production are set. Based on the boundary thresholds of environmental parameters and combined with data from the environmental-process correlation database, a trigger threshold for sudden changes in environmental parameters is set. Based on the trigger threshold of sudden changes in environmental parameters, calculate the weight increment of each environmental parameter relative to the basic weight after the sudden change. Based on the weight increment of each environmental parameter relative to the basic weight after a sudden change, an attention mechanism for sudden changes in environmental parameters is constructed, and the attention weight of each parameter is dynamically adjusted to adapt to production needs.

4. The AI-powered intelligent temperature control method for a dried bean curd production line combining environmental perception, as described in claim 3, is characterized in that... The weight increments of each environmental parameter relative to the base weights after the sudden change specifically include: Based on the environment-process correlation database, obtain the processed environmental sensing data; Based on the processed environmental perception data, the intensity of sudden changes in each environmental parameter is calculated; Based on the processed environmental perception data, the initial influence weight of each environmental parameter in each process of dried bean curd production is calculated by multinomial correlation coefficient, and then the initial weight is corrected based on expert experience. Calculate the attenuation coefficient of the sudden change in environmental parameters based on the duration of the sudden change. Based on the intensity of the sudden change, the impact weight, and the duration decay coefficient, the weight increment of each environmental parameter relative to the basic weight is calculated after the sudden change.

5. The AI-powered intelligent temperature control method for a dried bean curd production line incorporating environmental sensing, as described in claim 4, is characterized in that... The constructed neural network model mapping humidity, air velocity, and temperature aims to minimize the deviation between each key quality characteristic and the process target value. The output temperature adjustment step size and its effective probability specifically include: Based on the environment-process correlation database, preprocessed environmental perception data of temperature, humidity and air velocity, aligned with process steps, and corresponding key quality characteristic data are extracted. Based on historical data or by designing multiple sets of orthogonal gradient combination test experiments, for different combinations of humidity and air velocity gradients, the temperature is controlled by PID segmented adjustment mode to obtain the correlation data between humidity, air velocity and temperature and key quality characteristics. Based on the above-mentioned related data, the data is randomly divided and standardized in a 7:2:1 ratio to construct the training set, validation set, and test set required for training the neural network model. Temperature, humidity, and air velocity are used as inputs, key quality feature data are used as supervision signals to calculate model loss, and temperature adjustment step size and its effective probability are used as outputs. With the goal of minimizing the deviation between each key quality characteristic and the process target value, a loss function for a neural network model mapping humidity, air velocity, and temperature is constructed based on the mean square error formula. A three-channel LSTM neural network layer was set up to extract the temporal features of humidity, air velocity and temperature respectively; Based on the Transformer attention mechanism, attention weights for humidity-temperature time series features, air velocity-temperature time series features, and humidity and air velocity-temperature time series features are extracted respectively. Attention weights are fused with multivariate features through a weighted summation method, and then concatenated with the temporal features output by the LSTM before being input into the fully connected layer. A neural network model mapping humidity, air velocity, and temperature is constructed, denoted as Model 1. Based on the collected environmental perception data of temperature, humidity, and air velocity, as well as the corresponding key quality feature data, the model outputs the temperature adjustment step size and its effective probability.

6. The AI-powered intelligent temperature control method for a dried bean curd production line incorporating environmental sensing, as described in claim 5, is characterized in that... The comprehensive evaluation system based on dynamic attention weights and dual-model coupling prioritizes compensating for the effects of humidity and flow rate through temperature, and adjusts multiple parameters in a coordinated manner when failure is probabilistically determined. Specifically, this includes: Based on the environment-process correlation database, environmental perception parameter features are used as inputs, and key quality feature data are used as supervision signals to calculate model loss. The adjustment step size of each environmental perception parameter is used as the output quantity. With the goal of minimizing the deviation between each key quality characteristic and the process target value, a loss function for the characteristics of all environmental sensing parameters is constructed based on the mean square error formula. Based on the Transformer attention mechanism and the attention mechanism for sudden changes in environmental parameters, a comprehensive attention mechanism is constructed as a custom attention layer after the input layer of the neural network. Based on the LSTM neural network, a prediction model for the adjustment step size of environmental perception parameter features is constructed, denoted as Model 2. The adjustment step size of environmental perception parameter features is dynamically predicted by collecting environmental perception data. Based on Model 1 and Model 2, a comprehensive evaluation system with dual-model coupling is constructed. The step size is adjusted by comprehensively evaluating and outputting environmental perception parameter features based on the effective probability output by Model 1. Determine whether the effective probability in the output of Model 1 is greater than the preset value. If it is, it means that the temperature control is effective and outputs the temperature adjustment step size generated by Model 1. If it is not, it means that the temperature control is ineffective and outputs the adjustment step size of each parameter feature of the environmental perception output by Model 2. When the effective probability is less than the preset value, the temperature adjustment step size is a comprehensive temperature adjustment step size obtained by coupling the temperature adjustment step size output by Model 1 and the temperature adjustment step size output by Model 2. If the temperature adjustment step size output by Model 1 is in the opposite direction to the temperature adjustment step size output by Model 2, then when the effective probability is less than the preset value, the temperature adjustment step size output by Model 2 shall prevail.

7. An AI-powered intelligent temperature control system for a dried bean curd production line that incorporates environmental sensing, characterized in that: The method for implementing the AI-powered intelligent temperature control of a dried bean curd production line incorporating environmental sensing as described in any one of claims 1-6 includes: The comparison table module is used to formulate a comparison table of process objectives and stage parameter boundaries covering each process link in the production of dried bean curd sticks, based on the quality requirements of dried bean curd stick production. The database module is used to deploy high-precision environmental sensors and key quality feature detection devices in each process link of the dried bean curd production line, collect and process data in real time and store it in the time series database to build an environment-process association database. The parameter control module is used to set an attention mechanism for sudden changes in environmental parameters according to a lookup table, and dynamically adjust the attention weight of each parameter to adapt to production needs; construct a neural network model mapping humidity, air velocity, and temperature, with the goal of minimizing the deviation between each key quality characteristic and the process target value, and output the temperature adjustment step size and its effective probability; construct a comprehensive evaluation system based on dynamic attention weights and dual-model coupling, which prioritizes temperature compensation for the influence of humidity and flow rate, and adjusts multiple parameters in a coordinated manner when failure is probabilistically determined.

8. The AI ​​intelligent temperature control system for a dried bean curd production line incorporating environmental perception as described in claim 7, characterized in that, The parameter control module includes: An attention mechanism unit is used to set an attention mechanism for sudden changes in environmental parameters according to a lookup table, and to dynamically adjust the attention weight of each parameter to adapt to production needs. The temperature control unit is used to construct a neural network model that maps humidity, air velocity and temperature, with the goal of minimizing the deviation between each key quality characteristic and the process target value, and outputs the temperature adjustment step size and its effective probability. The parameter control unit is used to construct a comprehensive evaluation system with dual-model coupling based on dynamic attention weights. It prioritizes compensating for the effects of humidity and flow rate through temperature, and adjusts multiple parameters in a coordinated manner when failure is probabilistically determined.

Images