A seasoning frying reaction kettle temperature adaptive batch control method

Abstract

A seasoning frying reaction kettle temperature adaptive batch control method, steps include: (1) preprocessing the historical data of multiple batches of kettle temperature, removing abnormal points by Savitzky-Golay filtering, aligning multiple batches of trajectories by dynamic time warping algorithm, and then extracting process common characteristics by functional data analysis; (2) build a double-layer architecture of mechanism model iterative correction layer and real-time dynamic control layer; based on real-time kettle temperature and steam valve opening data, online update mechanism model parameters and real-time parameter setting using adaptive batch PID controller, compensate high-frequency disturbance and saturate the output; (3) fuse the outputs of mechanism model iterative correction layer and real-time dynamic control layer to generate the final steam valve control command of the reaction kettle. This method does not require additional sensors, only relies on conventional measurement data, improves temperature control accuracy, working condition adaptability and multi-batch production stability.

Description

Technical Field

[0001] This invention belongs to the fields of adaptive control, PID control and food processing technology, and specifically relates to an adaptive batch control method for the temperature of a condiment stir-frying reactor. Background Technology

[0002] The stir-frying process is a crucial step in condiment production, directly determining the product's color, flavor, and quality stability. In industrial production, the stir-frying process is typically carried out in a reaction vessel equipped with a jacketed heating and stirring device. Because condiments are not made from single chemical agents but rather from a variety of plant and animal biological materials, the stir-frying process involves complex physicochemical changes, including Maillard reactions, caramelization, water evaporation, and the formation of flavor compounds. These reactions are highly nonlinear, time-varying, and strongly coupled, requiring extremely precise control of process parameters such as temperature, pressure, stirring speed, and time.

[0003] Currently, the stir-frying reactors for seasonings generally employ a classic PID control strategy, using methods such as adjusting the opening of steam valves to achieve constant or programmed temperature control within the reactor. However, several challenges exist in actual production: First, the material properties, such as viscosity and specific heat capacity, dynamically change during the stir-frying process, leading to significant uncertainties and time-varying characteristics in the controlled object model. Second, the initial states (such as moisture content and composition) of different batches of raw materials vary, and scaling may occur in the reactor during long-term operation, causing a slow drift in heat transfer characteristics. Traditional fixed-parameter PID controllers struggle to adapt to these complex dynamics, often exhibiting problems such as excessive overshoot, excessively long settling times, and inconsistent product quality across different batches, impacting production efficiency and product yield.

[0004] To improve control quality, existing research has attempted to introduce intelligent methods such as fuzzy PID control and neural network PID to tune parameters online. These methods have improved system adaptability to some extent, but still have limitations: fuzzy rules rely on expert experience, resulting in a large design workload and insufficient generalization ability; neural networks require a large amount of training data and suffer from overfitting and poor real-time performance bottlenecks in engineering applications. Furthermore, these methods mostly focus on setpoint tracking, and have limited ability to suppress periodic disturbances (such as steam network pressure fluctuations and periodic feed interference) and random disturbances (ambient temperature changes) during the roasting process, lacking active learning and compensation mechanisms for disturbance characteristics.

[0005] In recent years, iterative learning control (ILC) has demonstrated advantages in controlling intermittent processes with repetitive operation characteristics, as it can iteratively optimize control inputs through historical batch data. However, although seasoning stir-frying is a batch process, the differences in raw materials between batches, variations in initial conditions, and slow drifts in equipment states all constitute "non-repetitive" disturbances, and pure ILC strategies are insufficiently robust to such disturbances. On the other hand, active disturbance rejection control (ADRC) estimates and compensates for the "total disturbance" in real time through an extended state observer, but when directly applied to long-term, multi-stage stir-frying processes, there is still room for improvement in the estimation accuracy and convergence speed of slowly time-varying parameter drifts.

[0006] Therefore, given the characteristics of the seasoning stir-frying reactor, which features strong nonlinearity, time-varying parameters, batch-to-batch raw material differences, and the coexistence of slow time-varying disturbances and high-frequency disturbances, existing control strategies are unable to simultaneously meet the requirements of control accuracy, operating condition adaptability, and consistency across multiple batches. Summary of the Invention

[0007] The present invention aims to provide a method for adaptive batch control of temperature in a condiment frying reactor. Through data preprocessing and a two-layer adaptive control architecture, it achieves precise temperature tracking and improves the stability and consistency of different batches of production.

[0008] The purpose of this invention is to provide an adaptive batch control method for the temperature of a condiment frying reactor. This method addresses the problems of dynamic changes in physical properties during the frying process and differences in the initial state of raw materials in different batches. It performs in-depth cleaning and feature mining on historical operating data of temperatures in multiple batches of reactors, smooths the time-series signal through Savitzky-Golay filtering, and removes anomalies caused by sensor failures or sudden changes in operating conditions. Then, it uses Dynamic Time Warping (DTW) to align the temperature trajectories of different batches to generate a standardized dataset with the same period. Finally, it uses Functional Data Analysis (FDA) to extract common time-series features and statistical regularities that characterize the essence of the process.

[0009] Based on this, a two-layer collaborative control architecture is constructed, consisting of a mechanistic model iterative correction layer and a real-time dynamic control layer. The mechanistic model iterative correction layer updates model parameters online based on real-time acquired temperature and valve opening data. The real-time dynamic control layer employs an adaptive batch PID controller, which updates proportional, integral, and derivative parameters online based on real-time temperature measurements. This rapidly compensates for high-frequency dynamic disturbances such as valve response lag and raw material fluctuations, and saturates and limits the PID output to ensure that the valve opening remains within a certain range. The safe zone;

[0010] Finally, by smoothing the update rate of the mechanism model parameters, limiting the amplitude of the PID control output, and designing a dynamic scheduling mechanism based on temperature tracking error, the two-layer output is integrated to generate the final steam valve opening command.

[0011] The method of this invention does not require the addition of complex sensors; it can achieve high temperature control accuracy, strong adaptability to operating conditions, and excellent multi-batch stability by relying solely on conventional measurements.

[0012] The specific technical solution of this invention is a method for adaptive batch control of temperature in a condiment frying reactor, comprising the following steps:

[0013] Step (1) preprocesses the historical temperature data of multiple batches of reactors, removes outliers by Savitzky-Golay filtering, aligns the trajectories of multiple batches by Dynamic Time Warping (DTW) algorithm, extracts common process features by functional data analysis (FDA), and finally obtains the optimal curve by moving average filtering.

[0014] Step (II) Construct a two-layer architecture consisting of a mechanism model iterative correction layer and a real-time dynamic control layer;

[0015] The mechanism model iterative correction layer updates the mechanism model parameters online based on real-time reactor temperature and steam valve opening data to compensate for slow time-varying disturbances such as raw material fluctuations and equipment scaling.

[0016] The real-time dynamic control layer adopts an adaptive batch PID controller, which uses the optimal temperature target curve as the tracking reference, tunes the proportional, integral, and derivative parameters in real time, compensates for high-frequency disturbances such as valve hysteresis and steam pressure fluctuations, and saturates and limits the PID output.

[0017] Step (3) Integrate the outputs of the mechanistic model iterative correction layer and the real-time dynamic control layer to generate the final steam valve control command for the reactor. Attached Figure Description

[0018] Figure 1 This is a flowchart of the adaptive temperature control process for the frying pot proposed in this invention;

[0019] Figure 2 This is a flowchart of the data preprocessing and optimal curve extraction process of the present invention;

[0020] Figure 3 This is a flowchart of the adaptive control of the present invention. Detailed Implementation

[0021] The reaction process in the seasoning stir-frying stage involves many complex physicochemical changes, characterized by strong nonlinearity, time-varying parameters, and diverse disturbances. Traditional fixed-parameter PID control and existing intelligent PID methods are difficult to balance control accuracy, operating condition adaptability, and robustness, which can easily lead to fluctuations in product quality.

[0022] To address the aforementioned problems, this invention proposes an adaptive batch control method for the temperature of a basic seasoning stir-frying reactor.

[0023] First, the historical temperature data of multiple batches of reactors are preprocessed. Outliers are removed by Savitzky-Golay filtering and the Dynamic Time Warping (DTW) algorithm is used to align the trajectories of multiple batches. Then, common process features are extracted by functional data analysis (FDA). Finally, the optimal curve is obtained by moving average filtering.

[0024] Subsequently, a two-layer architecture was constructed, consisting of a mechanism model iterative correction layer and a real-time dynamic control layer. The former updates the model parameters online based on real-time temperature and valve opening data, while the latter uses an adaptive batch PID controller to tune the parameters in real time, quickly compensate for high-frequency disturbances, and saturate and limit the output.

[0025] Finally, the outputs of the two layers are merged through a dynamic scheduling mechanism to generate the final valve control command.

[0026] This method requires no additional sensors and relies solely on conventional measurement data. It can effectively improve temperature control accuracy, adaptability to operating conditions, and stability in multi-batch production, which is of great significance for promoting the automation upgrade of condiment production and ensuring product quality consistency.

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

[0028] Step 1: Based on the historical operating data of multiple batches of reactor temperature required for the construction of the adaptive PID temperature control strategy of the seasoning stir-frying reactor of this invention, in order to ensure the accuracy of subsequent data processing and system equation construction, the process data is first thoroughly cleaned and the core features are mined.

[0029] Specifically, the collected historical temperature data inside the reactor is continuous time-series data. Affected by factors such as changes in cooking conditions (e.g., raw material input, heat adjustment) and temperature sensor drift, the data is prone to fluctuations and outliers. If such anomalies are not processed promptly, they will severely impact the reliability of subsequent data trajectory alignment and correlation analysis. Therefore, an outlier detection process is first performed on the collected temperature time-series data. The Savitzky-Golay filtering algorithm is used for smoothing preprocessing of the raw temperature data. This algorithm, through polynomial fitting within a sliding window, effectively suppresses random interference noise while preserving the core trend of temperature data changes.

[0030] S1-1. Define a local window, selecting a portion of the original data with a length of... The filtering window is as follows:

[0031] Define a local window, the existing length of which is Historical data, the value of each data point is set to ,in Select a length of The filtering window, where This indicates the number of data items extending to both sides from the center of the window;

[0032] S1-2, Regarding the current data Starting from the initial end, perform a polynomial least squares fit on the data within the window. The fitting equation is as follows:

[0033]

[0034] in Represents the polynomial coefficients. It is the position within the window relative to the center point, using Describe the trend and shape of local data;

[0035] S1-3, Using fitted values Replace the center point and slide the window to obtain the value of the fitted polynomial at the center point. That is the target point. The smoothed new value (i.e., using the obtained value) Replace the original target data points (To generate smooth data), then slide the window to the right and repeat the above steps until all target points have been processed and smoothed data is obtained.

[0036] Step 2: Given that the basic data is still multi-batch sequence data in a continuous time dimension, the Dynamic Time Warping (DTW) algorithm is used to perform trajectory synchronization and alignment processing on the multi-batch temperature sequence data in different time periods.

[0037] Specifically, the time axis and core feature dimensions of each batch of data are first defined. A set of typical batch data that conforms to the standard roasting process, has stable operating conditions, and is complete is selected as the reference trajectory, and the cleaned data of the remaining batches are used as the trajectories to be aligned. Then, a time warping path optimization model is constructed using the DTW algorithm. With the goal of minimizing the cumulative Euclidean distance between the trajectories of the two batches of data, the time axis of the trajectories to be aligned is allowed to be adaptively scaled and translated. Under the premise of completely preserving the core feature information of the temperature data of each batch, the accurate matching of different batches of data at the core process nodes is achieved.

[0038] S2-1. Consider two one-dimensional time series: a reference sequence and a time series sequence. , length is The sequence to be matched is the data smoothed in step 1, i.e. , length is , .

[0039] In the formula, It is a known time series sequence used as a matching benchmark. It is the target sequence that needs to be time-normalized and aligned with the reference sequence.

[0040] S2-2, Construct a space of size... Local distance matrix , The value is a sequence The Values ​​and sequences The The Euclidean distance between the values ​​is expressed as follows:

[0041]

[0042] S2-3, Construct a space of size... Cumulative distance matrix , The value is from the starting point To the current point The minimum cumulative distance is expressed as follows:

[0043]

[0044] S2-4. Backtrack to find the optimal planned path, according to the following rules:

[0045] In each cell To check which of its three adjacent cells (top left, top right, and left right) is correct. The point with the smallest value is chosen as the first point on the path. This leads to the optimal planned path. That is, the batch of data found, in which By repeating this process multiple times, historical data on multiple batches of reactors performing the same task can be obtained. In the formula, ; and Representing cells respectively The location coordinates.

[0046] Step 3: Based on the standardized process datasets from multiple batches within the same period, feature extraction is performed using the functional data analysis method (FDA).

[0047] This method uses the entire time-series curve as the unit of analysis, rather than isolated data sample points. It effectively integrates the temporal variation characteristics of standardized data from various batches. First, the temperature time-series data of each batch is transformed into a continuously differentiable function curve, achieving a functional representation of discrete data. Then, random noise and minor fluctuations are removed from the data, accurately extracting the common temporal features across all batches. The extracted common features and statistical regularities accurately reflect the essential characteristics of the seasoning cooking process and the core laws governing temperature changes.

[0048] S3-1. Fit multiple batches of data to extract their common features. A linear combination of B-spline basis functions is used to represent the fitting function. The basis function selection and matrix construction are as follows:

[0049] Choose the order and number of nodes of the combined B-spline for each time point. Calculate all B-spline basis functions. The values ​​form the design matrix. ,in It refers to the number of time points. It is the number of basis functions.

[0050] The fitting function takes the following form:

[0051]

[0052] in, For B-spline basis functions, Let be the coefficients of the basis functions to be determined; For time point indexing, For the first Each discrete sampling time point Indicates the first Fitted curves for batch data.

[0053] S3-2. To achieve a balance between data fitting accuracy and curve smoothness, a penalized least squares (P-spline) optimization objective is constructed.

[0054] The fitting term is ,in This is the actual measured value (actual temperature). These are fitted values

[0055] The penalty item is ;

[0056] The expression for the optimization objective can be obtained as follows:

[0057]

[0058] It is a fitted curve Regarding time The second derivative of is used to measure the curvature of the curve; λ is the penalty coefficient used to balance fitting accuracy and smoothness.

[0059] Discretization yields:

[0060]

[0061] in, To design a matrix, its elements are ; It is a smooth matrix, corresponding to the discrete form of the second derivative penalty; For the first A vector of measurement values ​​for a batch.

[0062] S3-3, Transform the function expression Substituting the optimization objective and taking the derivative, we get:

[0063]

[0064] Setting the derivative to zero, we can derive a closed-form solution for the coefficients:

[0065]

[0066] in, To design a matrix, its elements are ; It is a smooth matrix, corresponding to the discrete form of the second derivative penalty; For the first A vector of measurement values ​​for a batch;

[0067] The obtained coefficients Substituting the values ​​into the function representation yields the final smoothed fitting curve. .

[0068] S3-4. Average the fitting coefficients of all batches to obtain the overall average curve. ,in Given the number of batches, the overall average curve can reflect the common trends of data from multiple batches.

[0069] Step 4: Smooth the raw temperature sampling data by applying a moving average filter to the overall average curve. Specific steps include:

[0070] S4-1. Let the length of the sliding window be... The length is defined as One-dimensional cache array Used to store the latest A continuous temperature sample value; the current sampling time is defined as... The corresponding original temperature sampling value is The filtered output value is ;

[0071] S4-2, Constructing sliding window update rules:

[0072] Each time a new temperature sample value is acquired At that time, cache array The earliest stored sample value Remove the elements, move the remaining elements forward in sequence, and then... Store at the end of the cache array Complete the window sliding;

[0073] S4-3. Calculate the moving average filter output, as shown in the following expression:

[0074]

[0075] in, For the first element in cache array B Temperature sample values ​​stored in each location, The length of the sliding window;

[0076] S4-4. Repeat steps S4-2 to S4-3 to achieve real-time smoothing and filtering of the continuous temperature sampling data stream, and obtain the smoothed reactor temperature sequence, which is used for subsequent model identification and as the target curve for reactor temperature tracking, and is used as the setpoint for subsequent adaptive PID control.

[0077] Step 5: To address the dynamic relationship between the steam valve opening and the internal temperature of the reactor, a system equation is constructed based on historical operating data to establish the relationship between the input (steam valve opening) and the output (internal temperature), thus clarifying the quantitative mapping relationship between the two.

[0078] S5-1. First, based on the historical operating data of the reactor, a discrete transfer function is pre-established. The resulting discrete function is as follows:

[0079]

[0080] In the formula, This represents a transformation operator used to convert a continuous-time system into a discrete-time system, characterizing the delay properties of a discrete-time series;

[0081] S5-2. Transform the mechanistic model established in S5-1 into a linear regression form, as follows:

[0082]

[0083] in, This indicates the dynamic changes in the temperature system of the reactor; Indicates the factors that influence the change. Indicates the weight of each factor. It will adapt to the input of different batches of data.

[0084] This represents the system state variable, i.e., the temperature at the previous moment; This represents the system output variable, namely the current temperature. This represents the control input, i.e., the control quantity output by the PID controller.

[0085] Step 6: Adopt an adaptive batch PID control strategy, use real-time acquired batch data to update the proportional, integral, and derivative parameters of the PID controller online, and build the relationship between the PID output and the actual control quantity to achieve batch adaptive adjustment of controller parameters, thereby improving the temperature control accuracy and stability in multi-batch production processes.

[0086] The required change amount of the controlled object ,in It is the current value. It is the target value (i.e., the specific value corresponding to the optimal temperature curve).

[0087] The adjustment amount is calculated using a PID controller. Then after saturation limiting ( This yields the steam valve opening that meets hardware constraints. (Ensure that the valve is only opened and closed within the specified range.)

[0088]

[0089] In the formula, This represents the unit time calculated by the PID controller. Adjustment amount; The proportional gain coefficient, which is adaptively adjusted over time, is used to adjust for the current deviation. Adjust the ratio to quickly reduce errors; The integral gain coefficient, which is adaptively adjusted over time, is used to compensate for deviations. The integral term is adjusted to eliminate the steady-state error of the system; The differential gain coefficient, which is adaptively adjusted over time, is used to compensate for deviations. Adjusting the rate of change of [something] can suppress system overshoot and improve response speed; This indicates that the PID output is saturated and limited to ensure that the valve opening is always within a certain range. The safe zone; Represents unit of time The valve opening degree;

[0090] The output valve opening degree u controls the degree of opening and closing of the steam valve, changes the heating power of the roasting pot, and thus affects the temperature inside the pot to reach the target value.

[0091] Step 7: The specific control process of adaptive PID is as follows.

[0092] S7-1. Initialize the adaptive PID controller, configuring basic parameters and initial state. The following initialization steps are required:

[0093] (1) System inherent parameters ;

[0094] The inherent parameters a1, a2, b1, b2 are obtained by fitting from the measured data. The inherent parameters of the model are obtained by fitting the measured data using the least squares method. Based on the model structure and fitting method of this invention, the corresponding parameter values ​​can be determined by using measured data from actual application scenarios.

[0095] (2) Define the adaptive gain range ,initial ;

[0096] (3) Initial values ​​for parameter estimation ;

[0097] (4) Covariance matrix ( (It is a 4th-order identity matrix).

[0098] (5) Initialize historical data cache ;

[0099] S7-2. Construct a regression vector based on historical input and output data to extract system state features. The formula for constructing the regression vector is as follows:

[0100]

[0101] in, This is the output from the previous moment. This is the output from the previous time step. The control quantity from the previous moment. This is the control quantity at the previous time step.

[0102] S7-3. Based on the recursive least squares (RLS) method, adaptively update the parameter estimates and dynamically adjust the learning rate in combination with the prediction error.

[0103] Predicted output calculation: ;

[0104] Prediction error calculation: ;

[0105] Learning rate adjustment: ;

[0106] RLS algorithm core update: ;

[0107] S7-4. Calculate the current control quantity based on the updated parameters, and add robust processing and amplitude constraints to ensure control safety. The control quantity is expressed as follows:

[0108]

[0109] in, This will serve as the reference output value for the next time step. For parameter estimates Unpacking parameters.

[0110] S7-5. Update the input and output historical data cache, retaining only the data from the most recent 3 steps, in preparation for the construction of the regression vector for the next cycle;

[0111] S7-6. Output the current control result and determine whether to continue control. If to continue, return to step S5-2 to enter the next control cycle.

[0112] 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 above embodiments do not limit the scope of protection of the present invention in any way, and all technical solutions obtained by equivalent substitution or other means fall within the scope of protection of the present invention. Parts not covered in this invention are the same as or can be implemented using existing technology.

Claims

1. A method for adaptive batch control of temperature in a condiment stir-frying reactor, characterized in that the steps include: include: Step (1) preprocesses the historical temperature data of multiple batches of reactors, removes outliers by Savitzky-Golay filtering, aligns the trajectories of multiple batches by Dynamic Time Warping (DTW) algorithm, extracts common process features by functional data analysis (FDA), and finally obtains the optimal temperature curve by moving average filtering, which is used as the control target. Step (II) Construct a two-layer control architecture consisting of a mechanistic model iterative correction layer and a real-time dynamic control layer; The mechanism model iterative correction layer identifies and updates the dynamic mechanism model parameters of the reactor temperature online based on real-time data of reactor temperature and steam valve opening, and compensates for slow time-varying disturbances. The real-time dynamic control layer adopts an adaptive batch PID controller, which uses the optimal temperature target curve as the tracking reference, adjusts parameters in real time, compensates for high-frequency disturbances, and saturates and limits the PID output. Step (3) Integrate the outputs of the mechanistic model iterative correction layer and the real-time dynamic control layer to generate the final steam valve control command for the reactor.

2. The adaptive batch control method for the temperature of the seasoning stir-frying reactor according to claim 1, characterized in that, In step (1), the historical temperature data inside the pot for multiple batches refers to the temperature operation data during the seasoning stir-frying process; For multiple batches of historical in-vessel temperature data, in-depth cleaning and feature mining of process data were carried out. The steps included: 1.1) Perform outlier detection on the collected temperature time series data, use Savitzky-Golay filtering for data smoothing, and identify and remove fluctuation singularities caused by sensor failure or sudden changes in operating conditions. 1.2) The Dynamic Time Warping (DTW) algorithm is used to synchronize and align the trajectories of batch data from different time periods, generating a standardized process dataset with multiple batches and the same period. The method is as follows: First, clarify the timeline and core feature dimensions of each batch of data, select a set of typical batch data that conforms to the standard frying process, has stable operating conditions, and has complete data as a reference trajectory, and use the cleaned data of the remaining batches as the trajectory to be aligned. Subsequently, a time warping path optimization model was constructed using the DTW algorithm, with the goal of minimizing the cumulative Euclidean distance between the two batches of data trajectories, allowing the time axis of the trajectory to be aligned to undergo adaptive nonlinear scaling and translation adjustment; 1.3) Based on multiple batches of data from the same period, the FDA used a functional data analysis method to extract the common time-series characteristics and statistical regularities that characterize the essence of the process. The method is as follows: First, the temperature time series data of each batch is transformed into a continuously differentiable function curve to realize the functional representation of discrete data. Subsequently, random noise and minor fluctuations in the data were removed, and the common temporal features across all batches were extracted to obtain the average curve. 1.4) The average curve is filtered by a simple moving average to retain the main characteristics of the temperature change trend in the reactor and obtain the optimal curve, which is then used as the target curve for subsequent temperature control.

3. The adaptive batch control method for the temperature of the seasoning stir-frying reactor according to claim 2, characterized in that, The specific steps of step 1.1) include: 1.1.1) Define a local window, with an existing length of... Historical data, each data point is set as ,in Select a length of The filtering window, where This indicates the number of data items extending to both sides from the center of the window; 1.1.2) For the current data point Perform in-window polynomial least squares fitting; the fitting equation is as follows: ， in, Represents the coefficients of the polynomial. It is the position within the window relative to the center point, using Describe the trend and shape of local data; 1.1.3) Using fitted values Replace the center point and slide the window to obtain the value of the fitted polynomial at the center point. , as target data points The smoothed new value is specifically obtained using... Replace the original target data points Forming smooth data; Then slide the window to the right and repeat step 1.1.3 until all target points have been processed.

4. The adaptive batch control method for the temperature of the seasoning stir-frying reactor according to claim 3, characterized in that, The specific steps in step 1.2) include: 1.2.1) Consider two one-dimensional time series sequences, namely: reference sequence , length is ; sequence to be matched , length is ; In the formula, It is a known time series sequence used as a matching benchmark. It is the target sequence that needs to be time-normalized and aligned with the reference sequence; 1.2.2) Construct a space of size The local distance matrix d, The value is a sequence The Values ​​and sequences The The Euclidean distance between the values ​​is expressed as follows: ， Construct a size of Cumulative distance matrix , The value is from the starting position Go to current location The minimum cumulative distance is expressed as follows: ， 1.2.3) Backtracking to find the optimal planning path, according to the following rules: In each cell The algorithm examines the three adjacent cells (top left, top right, and top left) with the smallest distance; this cell is selected as the first point in the desired path. The optimal planned path is then obtained. As batch data, the cells , In the formula, and Representing cells respectively The location coordinates.

5. The adaptive batch temperature control method for the seasoning stir-frying reactor according to claim 4, characterized in that, The specific steps of step 1.3) include: 1.3.1) Fit multiple batches of data to extract their common features; The fitting function is represented by a linear combination of B-spline basis functions, in the form: ， in, For B-spline basis functions, Let be the coefficients of the basis functions to be determined; For time point indexing, For the first Each discrete sampling time point Indicates the first Fitting curves for batch data; The basis function selection and matrix construction are as follows: select the order and number of nodes of the composite B-spline, and for each time point... Calculate all B-spline basis functions The values ​​form the design matrix. ,in It refers to the number of time points. It is the number of basis functions; 1.3.2) To achieve a balance between data fitting accuracy and curve smoothness, a penalized least squares P-spline optimization objective is constructed, expressed as follows: ， in For fitting terms; It is the actual measured value, specifically the measured temperature; These are fitted values; It is a penalty item; It is a fitted curve Regarding time The second derivative; 1.3.3) Representing the function Substituting the optimization objective, we derive the closed-form solution for the basis function coefficients: ， in, To design a matrix, its elements are ; It is a smooth matrix, corresponding to the discrete form of the second derivative penalty; For the first A vector of measurement values ​​for a batch; The obtained coefficients Substituting the values ​​into the function representation, we obtain the final smoothed fitting curve. ; 1.3.4) Average the fitting coefficients of all batches to obtain the overall average curve. ,in The overall average curve reflects the common trend of multiple batches of data, representing the number of batches.

6. The adaptive batch temperature control method for the seasoning stir-frying reactor according to claim 5, characterized in that, Step 1.4) involves smoothing the original temperature sampling data using a moving average filter on the overall average curve. The specific steps include: 1.4.1) Let the length of the sliding window be... The length is defined as One-dimensional cache array Used to store the latest A continuous temperature sample value; the current sampling time is defined as... The corresponding original temperature sampling value is The filtered output value is ; 1.4.2) Constructing sliding window update rules: Each time a new temperature sample value is acquired At that time, cache array The earliest stored sample value Remove the elements, move the remaining elements forward in sequence, and then... Store at the end of the cache array Complete the window sliding; 1.4.3) Calculate the moving average filter output, as shown in the following expression: ， in, For cache array The Middle Temperature sample values ​​stored in each location, The length of the sliding window; 1.4.4) Repeat steps 1.4.2) to 1.4.3) to achieve real-time smoothing and filtering of the continuous temperature sampling data stream, and obtain the smoothed reactor temperature sequence for subsequent model identification and adaptive control calculation.

7. The adaptive batch control method for the temperature of the seasoning stir-frying reactor according to claim 1, characterized in that, In step (ii), the PID controller tuning parameters include the proportional, integral, and derivative parameters of the PID controller; high-frequency disturbances include valve response hysteresis and raw material fluctuations; Step (II) includes the following steps: 2.1) To address the dynamic relationship between the steam valve opening degree and the internal temperature of the reactor, a system equation is constructed based on historical operating data to represent the quantitative mapping relationship between the steam valve opening degree input and the internal temperature output. 2.2) An adaptive batch PID control strategy is adopted, which uses real-time acquired batch data to update the proportional, integral, and derivative parameters of the PID controller online; Furthermore, the relationship between the PID output and the actual control quantity is established to achieve batch adaptive adjustment of the controller parameters.

8. The adaptive batch control method for the temperature of the seasoning stir-frying reactor according to claim 7, characterized in that, The steps for constructing the system equations in step 2.1) include: 2.1.1) Based on the historical operating data of the reactor, a discrete transfer function is pre-established. The resulting discrete function is as follows: ， In the formula, This represents a transformation operator used to convert a continuous-time system into a discrete-time system, characterizing the delay properties of a discrete-time series; 2.1.2) Transform the mechanistic model established in step 2.1.1) into a linear regression form, as follows: ， in This indicates the dynamic changes in the temperature system of the reactor; Indicates the factors that influence the change. Indicates the weight of each factor. Adapts to the input of different batches of data; In the formula, This represents the system state variable, specifically the temperature at the previous moment; This represents the system output variable, specifically the current temperature. This represents the control input, which is the control quantity output by the PID controller.

9. The adaptive batch temperature control method for the seasoning stir-frying reactor according to claim 8, characterized in that, In step 2.2), the relationship between the PID output and the actual control quantity is constructed using the following method: The required change in the controlled object per unit time t ,in It is the current value. It is the target value, specifically the value corresponding to the optimal temperature curve; The adjustment amount is calculated using a PID controller. After saturation limiting To obtain the steam valve opening degree that meets the hardware constraints. To ensure that the valve only opens and closes within a certain range, it is represented as: ， In the formula, This represents the unit time calculated by the PID controller. Adjustment amount; The proportional gain coefficient, which is adaptively adjusted over time, is used to adjust for the current deviation. Adjust the ratio to quickly reduce errors; The integral gain coefficient, which is adaptively adjusted over time, is used to compensate for deviations. The integral term is adjusted to eliminate the steady-state error of the system; The differential gain coefficient, which is adaptively adjusted over time, is used to compensate for deviations. Adjusting the rate of change of [something] can suppress system overshoot and improve response speed; This indicates that the PID output is saturated and limited to ensure that the valve opening is always within a certain range. The safe zone; Represents unit of time The valve opening degree; Output valve opening By controlling the opening and closing of the steam valve, the heating power of the roasting reactor is changed, thereby affecting the temperature inside the reactor to reach the target value.

10. The adaptive batch control method for the temperature of the seasoning stir-frying reactor according to claim 1, characterized in that, In step (iii), the adaptive PID control process includes: 3.1) Initialize the adaptive PID controller, configuring the basic parameters and initial state; The initialization content includes: Inherent parameters ; Define adaptive gain range ,initial ; Initial values ​​of parameter estimation ; covariance matrix , It is a 4th order identity matrix; Historical data cache initialization ; 3.2) Construct a regression vector based on historical input and output data to extract system state features; the formula for constructing the regression vector is as follows: ， in, This is the output from the previous time step. This is the output from the previous time step. The control quantity from the previous moment. The control quantity at the previous time step; 3.3) The parameter estimates are adaptively updated based on the recursive least squares (RLS) method, and the learning rate is dynamically adjusted in combination with the prediction error; Predicted output calculation: ; Prediction error calculation: ; Learning rate adjustment: ; RLS algorithm core update: ; 3.4) Calculate the current control quantity based on the updated parameters, and add robust processing and amplitude constraints to ensure control safety; the control quantity is expressed as follows: ， in, This will serve as the reference output value for the next time step. For parameter estimates Unpacking parameters; 3.5) Update the input and output historical data cache, retaining only the most recent data to prepare for the construction of the regression vector in the next cycle; 3.6) Output the current control result and determine whether to continue control. If to continue, return to step 2.1.2) to enter the next control cycle.

Images