Energy-saving temperature control method for drying drum of asphalt mixing station and related equipment

Abstract

The invention discloses an asphalt mixing station drying roller energy-saving temperature control method and related equipment, and the method comprises the steps: firstly building a soft sensing model for predicting the discharge temperature based on the historical operation data of a drying roller, covering the main variable of the discharge temperature, the opening degree of a gas valve and other auxiliary variables; and the model is used as an interaction environment for reinforcement learning agent training, and the agent is trained by taking stable control of the discharging temperature and minimization of the gas usage amount as targets. The trained intelligent agent can output actions according to the current state; and finally, the intelligent agent is deployed to a drying roller control system, the state is collected and input in real time, and the intelligent agent outputs actions to adjust a gas valve and an air blower in real time. The invention aims to realize real-time accurate control of the heating process of the drying roller, stabilize the aggregate discharge temperature, ensure the quality of the finished product material and effectively control the energy consumption.

Description

Technical Field

[0001] This invention belongs to the field of asphalt mixing technology, specifically relating to an energy-saving temperature control method for the drying drum of an asphalt mixing plant and related equipment. Background Technology

[0002] An asphalt mixing plant mixes aggregates and fillers of different particle sizes in a specified ratio, adds asphalt as a binder, and mixes them under set temperature conditions to form a homogeneous finished mixture. The performance of the finished mixture directly affects the quality of the paved asphalt pavement. In a forced intermittent asphalt mixing plant, the drying and heating of cold aggregates are accomplished by a drying drum. The drying drum is a crucial piece of equipment for drying and heating moisture-containing aggregates. It is required to fully dehydrate the high-moisture aggregates and heat them to the target temperature required for mixing within a short time and with the lowest possible energy consumption. This not only helps improve the accuracy of subsequent metering but also ensures that the asphalt can evenly coat the aggregates during the mixing process, thus giving the finished product good paving performance. One end of the drying drum is equipped with a blower and burner to blow in air and burn fuel gas for heating, while the other end is equipped with an induced draft fan and a dust collector. The induced draft fan guides the smoke and dust inside the drum into the dust collector for purification and maintains negative pressure inside the drum to promote stable hot airflow through the drum. The drying drum rotates under the drive of a motor. The material is continuously lifted and scattered by the drum blades, heated by flames and hot air, and moves from the feed end to the discharge end. Therefore, if the aggregate temperature at the discharge end of the drying drum does not meet the standard, it will directly affect the quality of the finished product. At the same time, the drying drum requires a large amount of fuel gas for heating; if not properly controlled, this will not only directly affect overall energy consumption and operating costs but also exacerbate exhaust emissions, harming the environment.

[0003] Currently, research on the heating process of drying drums is mainly based on the fundamental principles of thermodynamics and kinematics. This involves establishing heating process models and performing numerical calculations to analyze the impact of process parameters on heating effects and determine the optimal solution. While the overall energy balance method offers rapid calculations, its model is relatively coarse, making it difficult to capture local details and limiting its real-time adjustment effectiveness. One-dimensional heat-particle models and CFD-DEM coupled simulations can accurately simulate the coupled heat transfer between gas flow and particle motion, revealing the temperature distribution and dynamic characteristics during the heating process; however, these methods require significant computational resources and are time-consuming, typically limiting their application to research. Therefore, none of these methods can meet the real-time operational requirements of production processes. Consequently, the current drying heating process relies heavily on manual experience for real-time adjustments, leading to significant fluctuations in aggregate discharge temperature, affecting the quality of the finished product and consequently impacting paving performance and pavement lifespan. Furthermore, manual adjustments struggle to effectively control energy consumption, failing to meet energy conservation and emission reduction requirements. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides an energy-saving temperature control method and related equipment for the drying drum of an asphalt mixing plant. The purpose is to achieve real-time and precise control of the heating process of the drying drum, stabilize the aggregate discharge temperature, ensure the quality of the finished product, and effectively control energy consumption to meet the requirements of energy conservation and consumption reduction.

[0005] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0006] According to a first aspect of the present invention, an energy-saving temperature control method for the drying drum of an asphalt mixing plant is provided, comprising:

[0007] Based on historical data during the operation of the drying drum, a soft sensing model is constructed to predict the discharge temperature. The historical data includes the discharge temperature as the dominant variable, and the gas valve opening, feed moisture content, feed rate, blower frequency, induced draft fan damper opening, ambient temperature, and ambient humidity as auxiliary variables.

[0008] The soft sensing model is used as the interactive environment for training the reinforcement learning agent. With the optimization objectives of stable control of discharge temperature and minimization of gas consumption, the historical data is used to train the reinforcement learning agent to obtain a trained agent. The agent is used to output actions according to the current state, which includes discharge temperature, feed moisture content, feed rate, induced draft fan damper opening, ambient temperature, and ambient humidity. The actions include gas valve opening and blower frequency.

[0009] The trained agent is deployed to the drying drum control system. The current status is collected in real time and input into the agent. The agent outputs actions and adjusts the gas valve and blower of the drying drum in real time according to the actions.

[0010] In one possible implementation of the first aspect, the construction of a soft-sensing model for predicting the discharge temperature based on historical data during the operation of the drying drum specifically includes:

[0011] S101. Perform moving average decomposition on the historical data of the collected dominant and auxiliary variables, decompose the historical data of each variable into trend items and detail items, and standardize the decomposed trend items and detail items respectively.

[0012] S102. Based on the standardized data, a training dataset for training the soft sensing model is constructed using the sliding window method. The soft sensing model consists of a dominant variable trend term prediction model and a dominant variable detail term prediction model. The dominant variable prediction value of the soft sensing model is the sum of the inverse standardized results of the prediction values ​​of the dominant variable trend term prediction model and the inverse standardized results of the prediction values ​​of the dominant variable detail term prediction model. The training dataset includes a dominant variable trend term prediction training dataset for training the dominant variable trend term prediction model and a dominant variable detail term prediction training dataset for training the dominant variable detail term prediction model. The dominant variable trend term prediction training dataset includes standardized auxiliary variable trend term data and dominant variable trend term data, and the dominant variable detail term prediction training dataset includes standardized auxiliary variable detail term data and dominant variable detail term data.

[0013] S103. Train a first long short-term memory neural network based on the dominant variable trend term prediction training dataset to obtain a dominant variable trend term prediction model. Train a second long short-term memory neural network based on the dominant variable detail term prediction training dataset to obtain a dominant variable detail term prediction model.

[0014] In one possible implementation of the first aspect, the step of performing a moving average decomposition on the collected historical data of the dominant and auxiliary variables, decomposing the historical data of each variable into trend and detail terms, specifically involves:

[0015]

[0016] q k =y k -p k k = 1, 2, ..., n

[0017] In the formula, is the trend term of auxiliary variable i at time k; w is the window length of the moving average; Let i be the observed value of auxiliary variable i at time j; Let be the observed value of auxiliary variable i at time kj; n is the sampling length; For the details of auxiliary variable i at time k; p is the observed value of auxiliary variable i at time k; k The trend term of the dominant variable at time k; y j y represents the observed value of the dominant variable at time j; k-j q represents the observed value of the dominant variable at time kj; k The details of the dominant variable at time k; y k The dominant variable is the observed value at time k.

[0018] In one possible implementation of the first aspect, the standardization of the decomposed trend term and detail term is specifically as follows:

[0019]

[0020] In the formula, The standardized result of the trend term of auxiliary variable i at time k; This represents the trend term of auxiliary variable i at time l; The standardized result of the details of the auxiliary variable i at time k; For the details of auxiliary variable i at time l; p represents the standardized result of the trend term of the dominant variable at time k; l The trend term of the dominant variable at time l; The standardized result of the details of the dominant variable at time k; q l This refers to the details of the dominant variable at time l.

[0021] In one possible implementation of the first aspect, the construction of a training dataset for training the soft sensing model based on the standardized data using a sliding window method specifically involves:

[0022]

[0023] In the formula, The trend term data of the auxiliary variable has been standardized. The standardized result of the trend term of the auxiliary variable m at time t-L+1; The standardized result of the trend term of the auxiliary variable m at time t-L+2; The standardized result of the trend term of the auxiliary variable m at time t; L is the window length; The dominant variable trend item data after standardization; The standardized result of the trend term of the dominant variable at time t+1; This refers to the standardized auxiliary variable details. The standardized result of the details of the auxiliary variable m at time t-L+1; The standardized result of the details of the auxiliary variable m at time t-L+2; The standardized result of the details of the auxiliary variable m at time t; This refers to the standardized details of the dominant variable. The standardized results of the details of the dominant variable at time t+1.

[0024] In one possible implementation of the first aspect, the predicted value of the dominant variable of the soft sensing model is the sum of the inverse standardized result of the predicted value of the dominant variable trend term prediction model and the inverse standardized result of the predicted value of the dominant variable detail term prediction model, specifically expressed as:

[0025]

[0026] In the formula, These are the predicted values ​​of the dominant variables in the soft sensing model; The predicted values ​​of the trend term prediction model for the dominant variable are the results of inverse standardization. The predicted values ​​of the model for the details of the dominant variable are the results of inverse standardization. The predicted value of the trend term of the dominant variable is the predicted value of the model; p j The trend term of the dominant variable at time j; The predicted value of the model for the detail term of the dominant variable; q j For the details of the dominant variable at time j; LSTM trend (·) is the dominant variable trend term prediction model; LSTM detail (·) is the predictive model for the details of the dominant variable.

[0027] In one possible implementation of the first aspect, the step of using the soft sensing model as an interactive environment for training a reinforcement learning agent, with the optimization objectives of stable control of discharge temperature and minimization of gas consumption, and training the reinforcement learning agent using the historical data, specifically includes:

[0028] The discharge temperature in the state is defined as a performance characteristic, the feed moisture content, feed rate, induced draft fan damper opening, ambient temperature and ambient humidity in the state are defined as environmental characteristics, and the gas valve opening and blower frequency in the action are defined as operable characteristics.

[0029] With the optimization objectives of stable control of discharge temperature and minimization of gas consumption, a reward function is designed, which includes a temperature error reward function and an energy consumption reward function.

[0030] The historical data consists of multiple continuously collected data sequences. During the training of the reinforcement learning agent, each training round is based on only one continuous data sequence, and the next training round is based on the next continuous data sequence. After all continuous data sequences have participated in training once, the training starts again from the first continuous data sequence. Each training round includes multiple iterations, and the number of iterations is the length of the continuous data sequence corresponding to the current training round. When the reward reaches convergence, the current training round ends, and the reinforcement learning agent training is completed.

[0031] Each iteration of each training round is as follows: Extract the environmental features at time t and the performance features at time t output by the soft sensing model from the historical data; input the environmental features and performance features at time t into the reinforcement learning agent to obtain the operable features at time t; replace the gas valve opening and blower frequency at time t in the historical data with the operable features at time t; use the operable features from time t-L+1 to time t as input to the soft sensing model to output the performance features at time t+1; calculate the reward at time t using the reward function based on the gas valve opening and the performance features at time t+1 in the operable features at time t; establish an experience quadruple using the state, action, and reward at time t and the state at time t+1; import the experience quadruple into the experience replay pool for training the reinforcement learning agent; extract the experience quadruple from the experience replay pool to train the reinforcement learning agent.

[0032] In one possible implementation of the first aspect, the reward function is designed with the optimization objectives of stable control of the discharge temperature and minimization of gas consumption as the main objectives, specifically as follows:

[0033] r = r temp +r energy

[0034]

[0035] r energy =-w e ·a

[0036] In the formula, r is the reward function; r temp Here is the temperature error reward function; r energy T is the energy reward function; t T represents the current discharge temperature. c Target discharge temperature; a is the gas valve opening; u is the tolerance boundary; b is the deviation constant; w t Temperature error is assigned a reward weight; w e Energy consumption reward weight.

[0037] According to a second aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the aforementioned energy-saving temperature control method for drying drums in asphalt mixing plants.

[0038] According to a third aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned energy-saving temperature control method for the drying drum of an asphalt mixing plant.

[0039] Compared with the prior art, the present invention has at least the following beneficial effects:

[0040] Traditional research methods based on fundamental principles of thermodynamics and kinematics, such as the overall energy balance method, are coarse and have limited real-time adjustment capabilities. One-dimensional thermo-particle models and CFD-DEM coupled simulations require large computational resources and are time-consuming, making them difficult to meet the real-time operational requirements of production processes. This results in the drying and heating process relying mainly on manual experience for real-time adjustment, leading to large fluctuations in the discharge temperature. This invention, however, constructs a soft-sensor model for predicting the discharge temperature and uses this model as an interactive environment for training a reinforcement learning agent. This trains an agent that can output actions based on the current state. After being deployed to the drying drum control system, the current state can be collected in real-time and input into the agent. The agent then outputs actions in real-time to adjust the gas valve and blower of the drying drum, achieving real-time and precise control of the drying drum heating process. This effectively avoids the lag and inaccuracy problems associated with manual adjustments.

[0041] This invention prioritizes stable discharge temperature control as one of its optimization objectives. It utilizes reinforcement learning algorithms to train an agent that intelligently responds to real-time changes in system status during control, achieving stable discharge temperature control and ensuring the aggregate discharge temperature remains within the target range, thus guaranteeing the quality of the finished product. Furthermore, the drying drum heating requires a large amount of fuel gas. Improper control not only increases overall energy consumption and operating costs but also exacerbates emissions, harming the environment. Traditional methods struggle to effectively control energy consumption. This invention addresses this by minimizing fuel gas consumption as another optimization objective. Reinforcement learning algorithms are used to train the agent's control strategy, balancing temperature control accuracy with energy efficiency. In actual operation, the agent's control actions directly adjust the gas valve opening and blower frequency, ensuring stable discharge temperature while effectively controlling energy consumption, meeting energy conservation and emission reduction requirements, lowering operating costs, and reducing emissions.

[0042] In summary, the method of this invention achieves automated real-time control of the drying drum heating process, eliminating the need for frequent manual intervention and reducing human error. Furthermore, by enabling real-time, precise control and stable discharge temperature, it avoids rework due to substandard discharge temperature, thereby improving production efficiency.

[0043] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the specific embodiments of the present invention, the drawings used in the description of the specific embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0045] Figure 1 This is a flowchart of an energy-saving temperature control method for the drying drum of an asphalt mixing plant according to the present invention;

[0046] Figure 2 This is a schematic diagram of the drying drum heating process;

[0047] Figure 3 This is a schematic diagram of interactive training of reinforcement learning agents based on a soft sensing model. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Combination Figure 1 and Figure 2 As shown, this invention provides an energy-saving temperature control method for the drying drum of an asphalt mixing plant, specifically including the following steps:

[0050] S1. Based on historical data during the operation of the drying drum, a soft sensing model is constructed to predict the discharge temperature. The historical data includes the discharge temperature as the dominant variable, and the gas valve opening, feed moisture content, feed rate, blower frequency, induced draft fan damper opening, ambient temperature, and ambient humidity as auxiliary variables.

[0051] It should be noted that during the operation of the asphalt mixing plant, data on dominant and auxiliary variables are collected through relevant sensors; the selection of auxiliary variables is based on the process mechanism of drying and heating of the drying drum.

[0052] In detail, consider the factors affecting the discharge temperature of the drying drum: the opening degree of the gas valve (a larger opening degree results in a larger gas flow rate and more heat generated by combustion, directly increasing the temperature inside the drum); the moisture content of the feed material at the feed end of the drying drum (higher moisture content requires more latent heat of vaporization to evaporate this moisture, leading to a lower discharge temperature); the feed rate at the feed end of the drying drum (increasing the feed rate dilutes the heat distribution and lowers the discharge temperature); the frequency of the blower and the opening degree of the induced draft fan damper (both determine the airflow velocity, flow rate, and negative pressure state inside the drum, affecting the contact time and heat exchange efficiency between the hot airflow and the material); the ambient temperature (affects the heat loss of the drying drum and the inlet air temperature); and the ambient humidity (high humidity carries more moisture, reducing the ability of the hot air to evaporate moisture from the material, affecting drying efficiency).

[0053] In this embodiment, the collected auxiliary variable data is represented as X∈R m*n m represents the number of auxiliary variables, n represents the sampling length, and X = {X (1) ,X (2) ,...X (m)}, The collected dominant variable data are represented as Y∈R 1*n Y = {y1, y2, ..., y} n}

[0054] In one feasible approach, a soft-sensing model for predicting the discharge temperature is constructed based on historical data from the operation of the drying drum, as follows:

[0055] S101. Perform moving average decomposition on the historical data of the collected dominant and auxiliary variables, decompose the historical data of each variable into trend items and detail items, and standardize the decomposed trend items and detail items respectively.

[0056] It should be understood that the trend term refers to the overall trend and long-term pattern of the variable's evolution over time, while the detail term refers to the short-term fluctuations and local characteristics remaining after removing the trend term.

[0057] Specifically, the historical data of the collected dominant and auxiliary variables are decomposed using a moving average, breaking down the historical data of each variable into trend and detail components, as follows:

[0058]

[0059] q k =y k -p k k = 1, 2, ..., n

[0060] In the formula, is the trend term of auxiliary variable i at time k; w is the window length of the moving average; Let i be the observed value of auxiliary variable i at time j; Let be the observed value of auxiliary variable i at time kj; n is the sampling length; For the details of auxiliary variable i at time k; p is the observed value of auxiliary variable i at time k; k The trend term of the dominant variable at time k; y j y represents the observed value of the dominant variable at time j; k-j q represents the observed value of the dominant variable at time kj; k The details of the dominant variable at time k; y k The dominant variable is the observed value at time k.

[0061] It should be understood that the window length w of the moving average needs to be determined based on the actual situation.

[0062] Specifically, the decomposed trend and detail items are standardized as follows:

[0063]

[0064]

[0065] In the formula, The standardized result of the trend term of auxiliary variable i at time k; This represents the trend term of auxiliary variable i at time l; The standardized result of the details of the auxiliary variable i at time k; For the details of auxiliary variable i at time l; p represents the standardized result of the trend term of the dominant variable at time k; l The trend term of the dominant variable at time l; The standardized result of the details of the dominant variable at time k; q l This refers to the details of the dominant variable at time l.

[0066] S102. Based on the standardized data, a training dataset for training the soft sensing model is constructed using the sliding window method. The soft sensing model consists of a dominant variable trend term prediction model and a dominant variable detail term prediction model. The dominant variable prediction value of the soft sensing model is the sum of the inverse standardized results of the prediction values ​​of the dominant variable trend term prediction model and the inverse standardized results of the prediction values ​​of the dominant variable detail term prediction model.

[0067] The training dataset includes a dominant variable trend term prediction training dataset for training the dominant variable trend term prediction model, and a dominant variable detail term prediction training dataset for training the dominant variable detail term prediction model. The dominant variable trend term prediction training dataset includes standardized auxiliary variable trend term data and dominant variable trend term data; the dominant variable detail term prediction training dataset includes standardized auxiliary variable detail term data and dominant variable detail term data.

[0068] In this embodiment, a training dataset for training the soft sensing model is constructed based on the standardized data using a sliding window method, specifically as follows:

[0069]

[0070] In the formula, The trend term data of the auxiliary variable has been standardized. The standardized result of the trend term of the auxiliary variable m at time t-L+1; The standardized result of the trend term of the auxiliary variable m at time t-L+2; The standardized result of the trend term of the auxiliary variable m at time t; L is the window length; The dominant variable trend item data after standardization; The standardized result of the trend term of the dominant variable at time t+1; This refers to the standardized auxiliary variable details. The standardized result of the details of the auxiliary variable m at time t-L+1; The standardized result of the details of the auxiliary variable m at time t-L+2; The standardized result of the details of the auxiliary variable m at time t; This refers to the standardized details of the dominant variable. The standardized results of the details of the dominant variable at time t+1.

[0071] In this embodiment, the predicted value of the dominant variable of the soft sensing model is the sum of the inverse standardized result of the predicted value of the dominant variable trend term prediction model and the inverse standardized result of the predicted value of the dominant variable detail term prediction model, specifically expressed as:

[0072]

[0073] In the formula, These are the predicted values ​​of the dominant variables in the soft sensing model; The predicted values ​​of the trend term prediction model for the dominant variable are the results of inverse standardization. The predicted values ​​of the model for the details of the dominant variable are the results of inverse standardization. The predicted value of the trend term of the dominant variable is the predicted value of the model; p j The trend term of the dominant variable at time j; The predicted values ​​of the model for the details of the dominant variable (predicted values ​​of the details of the dominant variable); q j For the details of the dominant variable at time j; LSTM trend (·) is the dominant variable trend term prediction model; LSTM detail (·) is the predictive model for the details of the dominant variable.

[0074] S103. Train the first long short-term memory neural network based on the dominant variable trend term prediction training dataset to obtain the dominant variable trend term prediction model. Train the second long short-term memory neural network based on the dominant variable detail term prediction training dataset to obtain the dominant variable detail term prediction model.

[0075] It should be noted that both the first and second Long Short-Term Memory (LSTM) neural networks are the same type of LSM neural network. The training process for both networks iteratively performs the following steps until the loss converges:

[0076] Forward propagation:

[0077] Mean square error loss calculate:

[0078] Backpropagation to calculate the gradient of each parameter

[0079] Parameter θ update:

[0080] S2. The soft sensing model is used as the interactive environment for training the reinforcement learning agent. With the optimization objectives of stable control of discharge temperature and minimization of gas consumption, the reinforcement learning agent is trained using historical data to obtain a trained agent. The agent is used to output actions based on the current state, which includes the discharge temperature, feed moisture content, feed rate, induced draft fan damper opening, ambient temperature, and ambient humidity at time t. The actions include the gas valve opening and blower frequency at time t.

[0081] In this embodiment, combined with Figure 3 As shown, the specific implementation process of step S2 is as follows:

[0082] a. Define the discharge temperature in the state as a performance characteristic, define the feed moisture content, feed rate, induced draft fan damper opening, ambient temperature and ambient humidity in the state as environmental characteristics, and define the gas valve opening and blower frequency in the action as operability characteristics.

[0083] b. With the optimization objectives of stable control of discharge temperature and minimization of gas consumption, a reward function is designed, which includes a temperature error reward function and an energy consumption reward function.

[0084] Specifically, the temperature error reward function is used to reward the error between the current temperature and the target temperature, encouraging the temperature to approach the target value; the energy consumption reward function is used to reward the gas valve opening, i.e., to control energy consumption, hoping that the gas valve opening is as small as possible.

[0085] In one feasible approach, the reward function is designed with the optimization objectives of stable control of discharge temperature and minimization of gas consumption as the main objectives, as follows:

[0086] r = r temp +r energy

[0087]

[0088] r energy =-w e ·a

[0089] In the formula, r is the reward function; r temp Here is the temperature error reward function; r energy T is the energy reward function; t T represents the current discharge temperature. c Target discharge temperature; a is the gas valve opening; u is the tolerance boundary; b is the deviation constant; w t Temperature error is assigned a reward weight; w e Energy consumption reward weight.

[0090] c. Historical data consists of multiple continuously collected data sequences. During the training of the reinforcement learning agent, each training round is based on only one continuous data sequence, and the next training round is based on the next continuous data sequence. After all continuous data sequences have participated in training once, the training starts again from the first continuous data sequence. Each training round includes multiple iterations, and the number of iterations is the length of the continuous data sequence corresponding to the current training round. When the reward reaches convergence, the current training round ends, and the reinforcement learning agent training is completed.

[0091] Each iteration of each training round is as follows: Extract the environmental features at time t and the performance features at time t output by the soft sensing model from historical data; input the environmental and performance features at time t into the reinforcement learning agent to obtain the operable features at time t; replace the gas valve opening and blower frequency at time t in the historical data with the operable features at time t; use the operable features from time t-L+1 to time t as input to the soft sensing model to output the performance features at time t+1; calculate the reward at time t using the reward function based on the gas valve opening and the performance features at time t+1 in the operable features at time t; establish an experience quadruple using the state, action, and reward at time t and the state at time t+1; import the experience quadruple into the experience replay pool for training the reinforcement learning agent; extract the experience quadruple from the experience replay pool to train the reinforcement learning agent.

[0092] As a more preferred implementation, in each iteration, before replacing the gas valve opening and blower frequency data at time t in the historical data with the operable features at time t, in order to ensure that the operable features are smooth in the time dimension, the following is also included: applying an exponentially weighted moving average (EWMA) to the operable features at time t, as follows:

[0093] A t =αa t +(1-α)A t-1

[0094] Where α = 0.96 is the smoothing factor; A t-1 and A t The operable features are smoothed from the previous time step and the current time step; a t The operable features output by the intelligent agent.

[0095] As a more preferred implementation, in order to reduce ineffective or even dangerous actions caused by exploration in the early stage of training, before the training of the reinforcement learning agent begins, the method further includes: establishing a prior experience quadruple based on historical data and importing the prior experience quadruple into the experience replay pool used for training the reinforcement learning agent.

[0096] In this embodiment, the series of empirical quadruples constructed are as follows:

[0097]

[0098] in, Let t represent the state at time t, including discharge temperature, feed moisture content, feed rate, induced draft fan damper opening, ambient temperature, and ambient humidity; b represents the number of variables that make up the state. Let t represent the action at time t, including the gas valve opening and the blower frequency, and c represent the number of variables that make up the action. The reward value at time t is calculated using the reward function; This represents the state at time t+1.

[0099] In one alternative implementation, during each iteration, experience quadruples are extracted from the experience replay pool and trained using the SAC algorithm to train the reinforcement learning agent.

[0100] S3. Deploy the trained agent to the drying drum control system, collect the current status in real time and input it into the agent, the agent outputs actions, and adjusts the gas valve and blower of the drying drum in real time according to the actions.

[0101] In other words, during the actual operation of the drying drum, relevant variables of the heating process (discharge temperature, feed moisture content, feed rate, induced draft fan damper opening, ambient temperature, and ambient humidity) are continuously collected to constitute the current state. This state is input in real time to the trained agent, which outputs the current optimal control action based on its policy network, including the gas valve opening value and the blower frequency value.

[0102] As a more preferred embodiment, an energy-saving temperature control method for the drying drum of an asphalt mixing plant further includes:

[0103] S4. Periodically repeat steps S1 to S3, using newly acquired operational data to update the soft sensing model and retrain the agent to achieve iterative optimization of the model. In other words, this involves retraining the soft sensing model and re-deploying the agent to ensure the long-term control performance of the drying drum system. For example, step S4 is activated when transitioning from summer to winter.

[0104] Specifically, by periodically collecting the latest operational data, updating the soft sensing model, and conducting offline training of the reinforcement learning agent, the trained agent is redeployed to perform online control of the drying drum heating process. By periodically executing this process, the system can adapt to changes in operating conditions and equipment status, maintaining the timeliness of the model and the stability of the control.

[0105] The following is an analysis and explanation of an energy-saving temperature control method for the drying drum of an asphalt mixing plant, based on a specific implementation case.

[0106] Data from the actual drying and heating process of the drying drum was collected, consisting of 15 continuously collected data sequences, including discharge temperature, gas valve opening, feed moisture content, feed rate, blower frequency, induced draft fan damper opening, ambient temperature, and ambient humidity. The first 10 segments were used as the training dataset, and the last 5 segments were used as the test dataset.

[0107] First, a soft sensing model was constructed, with the discharge temperature as the dominant variable and the gas valve opening, feed moisture content, feed rate, blower frequency, induced draft fan damper opening, ambient temperature, and ambient humidity as auxiliary variables. Then, the soft sensing model was used as the interactive environment for training the reinforcement learning agent. The optimization objectives were to stably control the discharge temperature at 200℃ and minimize gas consumption. The reinforcement learning agent was trained using the training dataset, and then tested on the test dataset. The test results are shown in Tables 1 and 2.

[0108] Table 1 Comparison of Temperature Control Effects

[0109]

[0110] Table 2 Natural Gas Saving Rate Compared to Manual Control

[0111] Test data segment 1 2 3 4 5 Savings rate 5.19％ 5.62％ 11.31％ 33.16％ 17.51％

[0112] It can be seen that manual control deviates significantly from the target value across different test data segments, with high levels of standard deviation, RMSE, and MAE, indicating insufficient control precision and large fluctuations. In contrast, the reinforcement learning agent of this invention achieves a mean value closer to the target value across all test data segments, with significantly reduced standard deviation, RMSE, and MAE, demonstrating higher stability and accuracy. This verifies its superior performance compared to manual control in dynamic environments. Furthermore, the agent exhibits advantages in energy saving across all test segments, with savings ranging from approximately 5% to over 30%. Therefore, the agent not only achieves more precise and stable temperature control in dynamic environments but also demonstrates significant advantages in energy consumption optimization.

[0113] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used in the operation of an energy-saving temperature control method for the drying drum of an asphalt mixing plant.

[0114] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the energy-saving temperature control method for the drying drum of an asphalt mixing plant in the above embodiments.

[0115] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0116] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0117] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0118] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0119] This invention also provides a computer program product, which is used to execute any of the above-described energy-saving temperature control methods for the drying drum of an asphalt mixing plant. Since the computer program product provided by this invention belongs to the same inventive concept as the energy-saving temperature control method for the drying drum of an asphalt mixing plant described above, the computer program product provided by this invention has all the advantages of the energy-saving temperature control method for the drying drum of an asphalt mixing plant described above. Therefore, the beneficial effects of the computer program product provided by this invention will not be elaborated further here.

[0120] In this invention, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0121] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the scope of the technology disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention.

Claims

1. A method for energy-saving temperature control of the drying drum in an asphalt mixing plant, characterized in that, include: Based on historical data during the operation of the drying drum, a soft sensing model is constructed to predict the discharge temperature. The historical data includes the discharge temperature as the dominant variable, and the gas valve opening, feed moisture content, feed rate, blower frequency, induced draft fan damper opening, ambient temperature, and ambient humidity as auxiliary variables. The soft sensing model is used as the interactive environment for training the reinforcement learning agent. With the optimization objectives of stable control of discharge temperature and minimization of gas consumption, the historical data is used to train the reinforcement learning agent to obtain a trained agent. The agent is used to output actions according to the current state, which includes discharge temperature, feed moisture content, feed rate, induced draft fan damper opening, ambient temperature, and ambient humidity. The actions include gas valve opening and blower frequency. The trained agent is deployed to the drying drum control system. The current status is collected in real time and input into the agent. The agent outputs actions and adjusts the gas valve and blower of the drying drum in real time according to the actions.

2. The energy-saving temperature control method for the drying drum of an asphalt mixing plant according to claim 1, characterized in that, The soft sensing model for predicting the discharge temperature, based on historical data from the operation of the drying drum, specifically includes: S101. Perform moving average decomposition on the historical data of the collected dominant and auxiliary variables, decompose the historical data of each variable into trend items and detail items, and standardize the decomposed trend items and detail items respectively. S102. Based on the standardized data, a training dataset for training the soft sensing model is constructed using the sliding window method. The soft sensing model consists of a dominant variable trend term prediction model and a dominant variable detail term prediction model. The dominant variable prediction value of the soft sensing model is the sum of the inverse standardized results of the prediction values ​​of the dominant variable trend term prediction model and the inverse standardized results of the prediction values ​​of the dominant variable detail term prediction model. The training dataset includes a dominant variable trend term prediction training dataset for training the dominant variable trend term prediction model and a dominant variable detail term prediction training dataset for training the dominant variable detail term prediction model. The dominant variable trend term prediction training dataset includes standardized auxiliary variable trend term data and dominant variable trend term data, and the dominant variable detail term prediction training dataset includes standardized auxiliary variable detail term data and dominant variable detail term data. S103. Train a first long short-term memory neural network based on the dominant variable trend term prediction training dataset to obtain a dominant variable trend term prediction model. Train a second long short-term memory neural network based on the dominant variable detail term prediction training dataset to obtain a dominant variable detail term prediction model.

3. The energy-saving temperature control method for the drying drum of an asphalt mixing plant according to claim 2, characterized in that, The process involves performing a moving average decomposition on the historical data of the collected dominant and auxiliary variables, breaking down the historical data of each variable into trend and detail components, specifically: In the formula, is the trend term of auxiliary variable i at time k; w is the window length of the moving average; Let i be the observed value of auxiliary variable i at time j; Let be the observed value of auxiliary variable i at time kj; n is the sampling length; For the details of auxiliary variable i at time k; p is the observed value of auxiliary variable i at time k; k The trend term of the dominant variable at time k; y j y represents the observed value of the dominant variable at time j; k-j q represents the observed value of the dominant variable at time kj; k The details of the dominant variable at time k; y k The dominant variable is the observed value at time k.

4. The energy-saving temperature control method for the drying drum of an asphalt mixing plant according to claim 3, characterized in that, The standardization process for the decomposed trend and detail items is as follows: In the formula, The standardized result of the trend term of auxiliary variable i at time k; This represents the trend term of auxiliary variable i at time l; The standardized result of the details of the auxiliary variable i at time k; For the details of auxiliary variable i at time l; The standardized result of the trend term of the dominant variable at time k; p l The trend term of the dominant variable at time l; The standardized result of the details of the dominant variable at time k; q l This refers to the details of the dominant variable at time l.

5. The energy-saving temperature control method for the drying drum of an asphalt mixing plant according to claim 4, characterized in that, The training dataset for training the soft sensing model is constructed using the sliding window method based on the standardized data, specifically: In the formula, The trend term data of the auxiliary variable has been standardized. The standardized result of the trend term of the auxiliary variable m at time t-L+1; The standardized result of the trend term of the auxiliary variable m at time t-L+2; The standardized result of the trend term of the auxiliary variable m at time t; L is the window length; The dominant variable trend item data after standardization; The standardized result of the trend term of the dominant variable at time t+1; This refers to the standardized auxiliary variable details. The standardized result of the details of the auxiliary variable m at time t-L+1; The standardized result of the details of the auxiliary variable m at time t-L+2; The standardized result of the details of the auxiliary variable m at time t; This refers to the standardized details of the dominant variable. The standardized results of the details of the dominant variable at time t+1.

6. The energy-saving temperature control method for the drying drum of an asphalt mixing plant according to claim 5, characterized in that, The predicted value of the dominant variable in the soft sensing model is the sum of the inverse standardized result of the predicted value of the dominant variable trend term prediction model and the inverse standardized result of the predicted value of the dominant variable detail term prediction model, specifically expressed as: In the formula, These are the predicted values ​​of the dominant variables in the soft sensing model; The predicted values ​​of the trend term prediction model for the dominant variable are the results of inverse standardization. The predicted values ​​of the model for the details of the dominant variable are the results of inverse standardization. The predicted value of the trend term of the dominant variable is the predicted value of the model; p j The trend term of the dominant variable at time j; The predicted value of the model for the detail term of the dominant variable; q j For the details of the dominant variable at time j; LSTM trend (·) is the dominant variable trend term prediction model; LSTM detail (·) is the predictive model for the details of the dominant variable.

7. The energy-saving temperature control method for the drying drum of an asphalt mixing plant according to claim 1, characterized in that, The step of using the soft sensing model as the interactive environment for training the reinforcement learning agent, with the optimization objectives of stable control of discharge temperature and minimization of gas consumption, and training the reinforcement learning agent using the historical data, specifically includes: The discharge temperature in the state is defined as a performance characteristic, the feed moisture content, feed rate, induced draft fan damper opening, ambient temperature and ambient humidity in the state are defined as environmental characteristics, and the gas valve opening and blower frequency in the action are defined as operable characteristics. With the optimization objectives of stable control of discharge temperature and minimization of gas consumption, a reward function is designed, which includes a temperature error reward function and an energy consumption reward function. The historical data consists of multiple continuously collected data sequences. During the training of the reinforcement learning agent, each training round is based on only one continuous data sequence, and the next training round is based on the next continuous data sequence. After all continuous data sequences have participated in training once, the training starts again from the first continuous data sequence. Each training round includes multiple iterations, and the number of iterations is the length of the continuous data sequence corresponding to the current training round. When the reward reaches convergence, the current training round ends, and the reinforcement learning agent training is completed. Each iteration of each training round is as follows: Extract the environmental features at time t and the performance features at time t output by the soft sensing model from the historical data; input the environmental features and performance features at time t into the reinforcement learning agent to obtain the operable features at time t; replace the gas valve opening and blower frequency at time t in the historical data with the operable features at time t; use the operable features from time t-L+1 to time t as input to the soft sensing model to output the performance features at time t+1; calculate the reward at time t using the reward function based on the gas valve opening and the performance features at time t+1 in the operable features at time t; establish an experience quadruple using the state, action, and reward at time t and the state at time t+1; import the experience quadruple into the experience replay pool for training the reinforcement learning agent; extract the experience quadruple from the experience replay pool to train the reinforcement learning agent.

8. The energy-saving temperature control method for the drying drum of an asphalt mixing plant according to claim 7, characterized in that, The reward function is designed with the optimization objectives of stable control of discharge temperature and minimization of gas consumption as the following: r＝r temp +r energy r energy =-w e ·a In the formula, r is the reward function; r temp Here is the temperature error reward function; r energy T is the energy reward function; t T represents the current discharge temperature. c Target discharge temperature; a is the gas valve opening; u is the tolerance boundary; b is the deviation constant; w t Temperature error is assigned a reward weight; w e Energy consumption reward weight.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements an energy-saving temperature control method for the drying drum of an asphalt mixing plant as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements an energy-saving temperature control method for the drying drum of an asphalt mixing plant as described in any one of claims 1 to 8.

Images