Intelligent temperature control method for blast furnace pulverizing system

Abstract

The invention discloses an intelligent temperature control method for a blast furnace pulverizing system, and the method comprises the steps: collecting and storing multi-dimensional working data such as the outlet temperature of a coal mill and the raw coal flow in real time, and providing a data basis for subsequent analysis; data are processed by means of mean filtering, noise is removed, and meanwhile a training set, a verification set and a prediction set are scientifically divided. A mill outlet temperature prediction model is constructed by adopting a random forest machine learning algorithm, hyper-parameters are optimized in combination with a Bayesian optimization algorithm, an effect is evaluated by a mean square error, an optimal model is determined, and prediction accuracy is improved. The optimal opening value of the gas regulating valve is determined based on the model, and through comparison with the last period and setting of a threshold value and duration, accurate and stable control is achieved, stable operation of the system is guaranteed, safety is improved, and energy consumption is reduced.

Description

Technical Field

[0001] The present application relates to the technical field of temperature control for blast furnace pulverizing, and in particular to an intelligent temperature control method for a blast furnace pulverizing system. Background Art

[0002] In blast furnace pulverizing systems, the pulverizer outlet temperature is a critical factor affecting system stability and production quality. Traditional temperature control methods for blast furnace pulverizing systems have numerous shortcomings. For one thing, traditional control methods struggle to comprehensively and in real time collect and process the numerous factors that influence pulverizer outlet temperature, such as raw coal flow rate, raw coal moisture, and gas and air parameters. This untimely and incomplete data collection results in a lack of accurate control data, making it difficult to respond promptly to complex and changing operating conditions.

[0003] On the other hand, traditional methods for constructing temperature control models often rely on simple algorithms or empirical formulas, which are unable to effectively handle the nonlinear and time-varying system characteristics. Given the complex physical and chemical reactions and dynamic changes in blast furnace pulverizing systems, these models suffer from low predictive accuracy, making it difficult to accurately predict the changing trends of mill outlet temperature. This leads to control lag and an inability to achieve precise control.

[0004] Furthermore, traditional control strategies for key regulating components, such as gas control valves, lack a scientific and rational adjustment mechanism. Frequent adjustments can cause system fluctuations, increase equipment wear and energy consumption, or untimely adjustments can lead to significant temperature deviations, impacting production safety and product quality. These deficiencies severely restrict the efficient and stable operation of blast furnace pulverizing systems. Summary of the Invention

[0005] In an exemplary embodiment of the present application, a temperature intelligent control method for a blast furnace pulverizing system is provided, which comprehensively and accurately controls the outlet temperature of the pulverizer in real time.

[0006] The present application provides a temperature intelligent control method for a blast furnace pulverizing system, which comprises the following steps: S1: Collecting the working data of the coal mill in real time, including outlet temperature, raw coal flow, raw coal moisture, gas regulating valve opening, gas pressure, gas concentration, air regulating valve opening and air pressure, and storing the collected working data in a database; S2: performing mean filtering on the working data with a time period of 10 seconds, dividing the data of the working data after the mean filtering process that have completed the mill outlet temperature control into a training set and a validation set in a ratio of 8:2, and the data that have not completed the mill outlet temperature control is used as a prediction set; S3: constructing a mill outlet temperature prediction model for the training set using a random forest machine learning algorithm, optimizing the hyperparameters of the mill outlet temperature prediction model using a Bayesian optimization algorithm and the data in the validation set, wherein the hyperparameters to be optimized include the number of decision trees of the mill outlet temperature prediction model, the maximum depth of the tree, the minimum sample ratio required for splitting, and the minimum sample ratio of leaf nodes, evaluating the fitting effect of the mill outlet temperature prediction model on the validation set using the mean square error as the objective function, and selecting the hyperparameters with the minimum mean square error to determine the optimal mill outlet temperature prediction model; S4: Predicting the data in the prediction set according to the determined optimal mill outlet temperature prediction model, determining the optimal opening value of the gas regulating valve and storing it in a database; S5: Compare the optimal opening value of the gas regulating valve determined in step S4 with the optimal opening value of the gas regulating valve in the previous statistical period. If the absolute value of the change in the optimal opening value of the gas regulating valve determined in step S4 compared with the optimal opening value of the gas regulating valve in the previous statistical period exceeds a preset threshold and lasts for a preset time, the optimal opening value of the gas regulating valve determined in step S4 is transmitted to the controller and the opening value of the gas regulating valve is adjusted to the optimal opening value of the gas regulating valve. Otherwise, the opening value of the gas regulating valve is not adjusted.

[0007] Furthermore, the preset threshold is 0.2, and the preset time is 5s.

[0008] Furthermore, the mean filtering process in S2 adopts a sliding window algorithm, and the window length is 30 sampling points.

[0009] Furthermore, the mean square error in S3 is calculated using a sliding window averaging method, where the window covers the validation set data within the last hour.

[0010] Furthermore, the optimization range of the number of decision trees in S3 is 50-200, and the optimization range of the maximum depth of the tree is 10-30 layers.

[0011] Furthermore, the construction of the decision tree in the random forest machine learning algorithm in S3 is based on the Gini index to select features for splitting.

[0012] Furthermore, the Bayesian optimization algorithm in S3 adopts a Gaussian process proxy model.

[0013] The embodiments of the present application have the following beneficial effects: The present invention provides a temperature intelligent control method for a blast furnace pulverizing system, which focuses on the temperature intelligent control of the blast furnace pulverizing system, realizes efficient and precise control through multi-step collaboration, collects and stores multi-dimensional working data such as the mill outlet temperature and raw coal flow in real time, and provides a comprehensive and accurate data basis for subsequent analysis, ensuring the timeliness and integrity of the data. The working data is processed using mean filtering to effectively remove noise interference and improve data quality; the data is scientifically divided into training sets, validation sets and prediction sets to provide reasonable data support for model construction and optimization. The mill outlet temperature prediction model is constructed using the random forest machine learning algorithm, and the hyperparameters are optimized in combination with the Bayesian optimization algorithm. The fitting effect is evaluated using the mean square error as the objective function to determine the optimal model, which greatly improves the accuracy and reliability of the temperature prediction and enables the model to more accurately reflect the actual working conditions. The prediction set data is predicted based on the optimal model to determine the optimal opening value of the gas regulating valve, providing a scientific basis for system adjustment. Finally, by comparing the currently determined optimal opening value with the previous cycle, setting a reasonable change threshold and duration, avoiding frequent and ineffective adjustments, and achieving accurate and stable control of the gas regulating valve opening, the blast furnace pulverizing system can be ensured to operate under stable temperature conditions, thereby improving the safety and stability of the system operation and reducing energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0015] Figure 1 A flow chart of an intelligent temperature control method for a blast furnace pulverizing system provided in an embodiment of the present application is exemplified. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0017] To further illustrate the technical solutions provided by the embodiments of the present application, the following is a detailed description of the technical solutions in conjunction with the accompanying drawings and specific implementation methods. Although the embodiments of the present application provide the method operation steps as shown in the following embodiments or drawings, more or fewer operation steps may be included in the method based on routine or no creative work. In steps where there is no necessary causal relationship logically, the execution order of these steps is not limited to the execution order provided in the embodiments of the present application.

[0018] refer to Figure 1 As shown, the present application provides a temperature intelligent control method for a blast furnace pulverizing system, which includes the following steps: S1: Collect the working data of the coal mill in real time, including outlet temperature, raw coal flow, raw coal moisture, gas regulating valve opening, gas pressure, gas concentration, air regulating valve opening and air pressure value, and store the collected working data in the database.

[0019] S1 is designed to provide comprehensive and accurate basic data support for subsequent temperature prediction and control. In blast furnace pulverizing systems, the operating status of the coal mill is influenced by a variety of interrelated factors. Real-time collection of operating data such as the mill outlet temperature, raw coal flow rate, raw coal moisture, gas control valve opening, gas pressure, gas concentration, air control valve opening, and air pressure accurately reflects the mill's operating status at any given moment.

[0020] S2: Perform mean filtering on the working data with a time period of 10 seconds. The data that have completed the mill outlet temperature control in the working data after mean filtering are divided into a training set and a validation set in a ratio of 8:2, and the data that have not completed the mill outlet temperature control are used as the prediction set.

[0021] The S2 mean filtering process uses a sliding window algorithm with a window length of 30 sampling points.

[0022] The purpose of step S2 is to provide a high-quality input dataset for the subsequent temperature prediction model through data preprocessing and structured partitioning, while ensuring the real-time and stable control decisions of the system. This step plays a key role in improving data quality, enhancing model generalization capabilities, and adapting control timing within the overall solution.

[0023] Step S2 uses a sliding window mean filter with a window length of 30 sampling points, corresponding to continuous data collection over a 10-second period. This process effectively eliminates transient noise interference by averaging the data within the window. This is particularly effective for parameters in the blast furnace pulverizing system that are susceptible to operating fluctuations, such as mill outlet temperature and gas pressure. It can also filter out spikes caused by equipment vibration and sensor anomalies.

[0024] Step S2 strategically partitions the dataset, dividing the historical data from completed temperature control into a training set and a validation set in an 8:2 ratio. This partitioning ratio, designed based on the temporal correlation characteristics of industrial process data, preserves sufficient historical patterns for model learning (80% of the training set) while also reserving sufficient recent data to verify the model's adaptability to new operating conditions (20% of the validation set). Specifically, the latest data from uncompleted control is independently set as a prediction set, establishing a "historical-real-time" dual-channel data processing mechanism. This allows the model to both learn long-term operating patterns and dynamically respond to recent changes in operating conditions.

[0025] The principle of allocating 80% of the training data to the training set ensures that the model can fully learn the steady-state operating patterns and typical disturbance response patterns in historical operating conditions, covering the complete data distribution of common operating conditions such as raw coal moisture fluctuations and gas pressure changes. The remaining 20% of the validation set data serves as the latest samples at the end of the historical data, reflecting the time continuity characteristics of equipment operating status, and is used to simulate the model's predictive capabilities under near-real-time operating conditions.

[0026] The large-scale training set provides a sufficient sample base for the random forest algorithm, supporting the feature space partitioning and ensemble learning process of multiple decision trees. This ensures the integrity of the numerical distribution of continuous variables such as raw coal flow and air pressure. The validation set retains sufficient data diversity to test the model's generalization performance on unknown data, avoiding the risk of overfitting and misjudgment due to insufficient validation data.

[0027] A three-tiered data architecture is constructed with a fixed ratio between training and validation sets and a dynamic update mechanism for the prediction set. 80% of historical data from completed temperature control is used for model training and parameter tuning, 20% of recent data is used to verify the model's adaptability to new operating conditions, and real-time data from uncompleted control is used independently as a prediction set. This structure isolates data from the model training, validation, and application phases, preventing biased model evaluations caused by future information leaks.

[0028] Step S2, through the preset 10-second processing cycle and data window parameters, achieves a precise match between control decision frequency and process response speed. The lag time constant for mill outlet temperature control is approximately 45 seconds. The 30-sampling point window set in this step (corresponding to a 30-second data span) covers approximately 66% of the lag period, preventing excessively short windows from causing frequent model false triggers and excessively long windows from impairing control timeliness.

[0029] Step S2 establishes the basic framework for feature engineering through structured data preprocessing. The smoothed time series features formed by filtering parameters such as raw coal moisture and gas concentration synergize with the feature selection mechanism of the random forest algorithm.

[0030] S3: Use the random forest machine learning algorithm to construct a mill outlet temperature prediction model for the training set, and use the Bayesian optimization algorithm and the data in the validation set to optimize the hyperparameters of the mill outlet temperature prediction model. The optimized hyperparameters include the number of decision trees of the mill outlet temperature prediction model, the maximum depth of the tree, the minimum sample ratio required for splitting, and the minimum sample ratio of leaf nodes. The mean square error is used as the objective function to evaluate the fitting effect of the mill outlet temperature prediction model on the validation set, and the hyperparameters with the smallest mean square error are selected to determine the optimal mill outlet temperature prediction model.

[0031] S3 uses a sliding window averaging method to calculate mean squared error, with the window covering the validation set data within the last hour. The optimization range for the number of decision trees in S3 is 50-200, and the maximum tree depth is 10-30 layers. The decision trees in S3's random forest machine learning algorithm are constructed based on the Gini index to select features for splitting. S3's Bayesian optimization algorithm uses a Gaussian process surrogate model.

[0032] Step S3 uses the random forest machine learning algorithm to build a temperature prediction model, leveraging its ensemble learning properties to integrate the nonlinear correlations of multi-source heterogeneous data. The number of decision trees can be set between 50 and 200. By running multiple decision trees in parallel and aggregating their results, the model's learning capacity and computing resource consumption are balanced.

[0033] The maximum tree depth is optimized to 10-30 layers, capturing the deep interactions between raw coal moisture and gas concentration while avoiding overfitting to local noisy data. Each decision tree selects features based on the Gini index, strengthening the identification of causal relationships between key control variables and temperature output.

[0034] Next, in step S3, a Bayesian optimization algorithm is introduced to dynamically optimize the model's hyperparameters. A Gaussian process surrogate model is used to construct a parameter space response surface. A multidimensional optimization framework is established for four core parameters: the number of decision trees, the maximum tree depth, the minimum sample ratio required for splitting, and the minimum sample ratio for leaf nodes.

[0035] This mechanism replaces traditional grid search with a probabilistic proxy model, achieving efficient optimization within a continuous parameter space and ensuring the compatibility of model parameters with real-time operating conditions. The optimization process uses mean squared error as the objective function and evaluates the model's fit on a validation set using a sliding window averaging method. The window covers the last hour of data to reflect process dynamics.

[0036] In terms of model feature processing, step S3 constructs a weighted mapping relationship between process parameters and temperature output based on the Gini index's feature importance assessment mechanism. Input features such as raw coal flow and gas pressure are processed through a multi-tree splitting process in a random forest to form a hierarchical feature selection network.

[0037] This design effectively identifies the sensitivity weights of key control variables, such as gas regulating valve opening, providing a feature-level decision-making basis for subsequent valve regulation. Furthermore, by optimizing parameters for the minimum sample ratio required for splitting and the minimum sample ratio for leaf nodes, the risk of data overfitting under abnormal operating conditions is mitigated.

[0038] Furthermore, the temperature prediction model architecture established in step S3 forms a technical synergy with the dataset preprocessed in step S2. After the feature data, processed by mean filtering, is input into the random forest model, its smoothing properties work in tandem with the splitting rules of the decision tree to enhance the model's ability to capture trend changes.

[0039] The prediction results output by the model and the valve regulation in step S4 form a closed-loop control link, realizing the automation of the entire process from data preprocessing, model prediction to actuator adjustment.

[0040] The mathematical properties of mean squared error (MSE) enable it to comprehensively reflect the overall degree of deviation between predicted and actual values. By amplifying the impact of large deviations through the squaring operation, this metric enhances the model's sensitivity to critical temperature anomalies, prioritizing the suppression of large prediction errors that could potentially pose a safety hazard to the process. Compared to linear metrics like mean absolute error (MAE), the gradient nature of MSE facilitates the Bayesian optimization algorithm's rapid identification of the optimal solution region within parameter space, where model predictions are robust.

[0041] The combination of mean square error and sliding window averaging establishes a model evaluation mechanism for dynamic operating conditions. By limiting the calculation window to the last hour of validation data, the model optimization process is ensured to always focus on the timeliness of the current process state.

[0042] The setting of the mean square error (MSE) forms a technical synergy with the data preprocessing in step S2. After the smoothed data is input into the model through mean filtering, its noise suppression properties work in tandem with the deviation assessment function of the MSE to effectively distinguish the impact of true process fluctuations and data acquisition noise on model training. This collaborative design ensures that the hyperparameter optimization process focuses on improving the ability to model the underlying process laws rather than overfitting random noise.

[0043] Optimizing the mean square error directly determines the accuracy of calculating the optimal gas valve opening in step S4. By minimizing temperature prediction deviation, this metric ensures the reliability of the input data to the control command generation module, providing an accurate basis for subsequent threshold determination and valve control. Furthermore, the continuity of the mean square error ensures a stable convergence of the Bayesian optimization algorithm, avoiding optimization path oscillations caused by discretization.

[0044] S4: Predicting the data in the prediction set according to the determined optimal mill outlet temperature prediction model, determining the optimal opening value of the gas regulating valve and storing it in the database.

[0045] Step S4, based on the optimal mill outlet temperature prediction model trained in step S3, performs a reverse calculation of the gas regulating valve opening value for the prediction data set. By inputting real-time process parameters such as raw coal moisture and gas pressure into the prediction model and combining it with the preset temperature control target value, the optimal valve opening parameter that meets the temperature control requirements under the current operating conditions is dynamically calculated.

[0046] This process constructs an inverse mapping relationship between multivariable inputs and control targets, transforming the complex temperature control problem into an optimization calculation problem of valve opening.

[0047] Step S4 establishes a storage mechanism that associates prediction results with historical data, storing the calculated optimal opening value, along with the corresponding timestamp and operating condition feature tag, in a database. This storage structure design includes a time series index and a process parameter hash table, supporting data tracing and parameter comparison within subsequent control cycles.

[0048] By setting the timing alignment rules for data storage, it is ensured that the prediction results and the control execution data are strictly corresponding in the time dimension, providing a complete data chain for the decision analysis in step S5.

[0049] The prediction calculation cycle in step S4 is synchronized with the data preprocessing cycle in step S2, both using 10-second intervals. This design dynamically matches the prediction instruction generation rhythm with process parameter updates, avoiding system disturbances caused by lagging or advancing control instructions.

[0050] The dynamic update mechanism of the prediction set data ensures that the model always performs opening calculations based on the latest operating data, effectively responding to changes in operating conditions such as coal quality fluctuations and load changes.

[0051] Step S4 establishes a secure isolation mechanism between the model prediction layer and the device control layer. By storing the optimal opening value in an independent database partition rather than transmitting it directly to the actuator, a buffered verification link for the control command is formed.

[0052] This design provides data preparation for the threshold determination in step S5, preventing abnormal prediction values from directly triggering device actions. Hierarchical control settings for database access permissions ensure the integrity and tamper-proofing of prediction data during transmission.

[0053] The prediction output of step S4 establishes the basic framework for feedforward control. The calculation of the optimal opening value not only relies on the current process parameters but also incorporates the evolution trend of historical data to form a forward-looking control pre-instruction.

[0054] The pre-instruction is transmitted to the subsequent control module through the database sharing interface, and together with the feedback verification mechanism of step S5, constitutes a composite control system to achieve feedforward-feedback dual-mode collaborative control.

[0055] S5: Compare the optimal opening value of the gas regulating valve determined in step S4 with the optimal opening value of the gas regulating valve in the previous statistical period. If the absolute value of the change in the optimal opening value of the gas regulating valve determined in step S4 compared with the optimal opening value of the gas regulating valve in the previous statistical period exceeds a preset threshold and lasts for a preset time, the optimal opening value of the gas regulating valve determined in step S4 is transmitted to the controller and the opening value of the gas regulating valve is adjusted to the optimal opening value of the gas regulating valve. Otherwise, the opening value of the gas regulating valve is not adjusted. The preset threshold is 0.2 and the preset time is 5s.

[0056] Step S5 establishes a timing association mechanism for control instructions. By aligning and comparing the timestamps of the current opening data with the previous cycle data in the historical database, the temporal consistency of control decisions is ensured. This design avoids control logic conflicts caused by data acquisition delays or misaligned processing cycles, maintaining strict synchronization between the control cycle and the process response cycle.

[0057] Step S5 sets a buffer isolation layer for the control command. The optimal opening value must be double-verified by the threshold and duration before being transmitted to the controller for execution through an independent communication protocol.

[0058] This mechanism blocks the risk of abnormal data directly affecting actuators, preventing equipment malfunctions caused by model prediction failures or data transmission errors. The separation of access rights between the database and the controller further strengthens the security of control command transmission.

[0059] Step S5 implements the coordinated operation of feedforward control and feedback verification. Current opening adjustment decisions are based not only on model predictions but also on historical control performance feedback, forming a composite control strategy with self-correcting capabilities. When the opening demand is detected to change in the same direction over multiple consecutive cycles, this mechanism can predict process trend changes in advance, enhancing the system's ability to suppress long-term disturbances.

[0060] Step S5 balances control sensitivity and system stability through parameterized design of preset thresholds and durations. The threshold parameter defines the triggering boundary of the control action, preventing mechanical wear caused by frequent valve fine-tuning. The duration parameter filters out short-term disturbance signals, ensuring that control commands only respond to persistent deviations, reducing the risk of system oscillation.

[0061] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0062] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0063] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0064] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0065] Obviously, those skilled in the art may make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include these modifications and variations.

Claims

1. A temperature intelligent control method for a blast furnace pulverizing system, characterized in that: It includes the following steps: S1: Collecting the working data of the coal mill in real time, including outlet temperature, raw coal flow, raw coal moisture, gas regulating valve opening, gas pressure, gas concentration, air regulating valve opening and air pressure, and storing the collected working data in a database; S2: performing mean filtering on the working data with a time period of 10 seconds, dividing the data of the working data after the mean filtering process that have completed the mill outlet temperature control into a training set and a validation set in a ratio of 8:2, and the data that have not completed the mill outlet temperature control is used as a prediction set; S3: constructing a mill outlet temperature prediction model for the training set using a random forest machine learning algorithm, optimizing the hyperparameters of the mill outlet temperature prediction model using a Bayesian optimization algorithm and the data in the validation set, wherein the hyperparameters to be optimized include the number of decision trees of the mill outlet temperature prediction model, the maximum depth of the tree, the minimum sample ratio required for splitting, and the minimum sample ratio of leaf nodes, evaluating the fitting effect of the mill outlet temperature prediction model on the validation set using the mean square error as the objective function, and selecting the hyperparameters with the minimum mean square error to determine the optimal mill outlet temperature prediction model; S4: Predicting the data in the prediction set according to the determined optimal mill outlet temperature prediction model, determining the optimal opening value of the gas regulating valve and storing it in a database; S5: Compare the optimal opening value of the gas regulating valve determined in step S4 with the optimal opening value of the gas regulating valve in the previous statistical period. If the absolute value of the change in the optimal opening value of the gas regulating valve determined in step S4 compared with the optimal opening value of the gas regulating valve in the previous statistical period exceeds a preset threshold and lasts for a preset time, the optimal opening value of the gas regulating valve determined in step S4 is transmitted to the controller and the opening value of the gas regulating valve is adjusted to the optimal opening value of the gas regulating valve. Otherwise, the opening value of the gas regulating valve is not adjusted.

2. The temperature intelligent control method according to claim 1, characterized in that: The preset threshold is 0.2, and the preset time is 5s.

3. The temperature intelligent control method according to claim 1, characterized in that: The mean filtering process in S2 adopts a sliding window algorithm with a window length of 30 sampling points.

4. The temperature intelligent control method according to claim 3, characterized in that: The mean square error in S3 is calculated using a sliding window averaging method, where the window covers the validation set data within the last hour.

5. The temperature intelligent control method according to claim 4, characterized in that: The optimization range of the number of decision trees in S3 is 50-200, and the optimization range of the maximum depth of the tree is 10-30 layers.

6. The temperature intelligent control method according to claim 1, characterized in that: The construction of the decision tree in the random forest machine learning algorithm in S3 is based on the Gini index to select features for splitting.

7. The intelligent temperature control method according to claim 6, characterized in that: The Bayesian optimization algorithm described in S3 adopts a Gaussian process proxy model.

Images