Asphalt mixing station intelligent monitoring method and system based on Internet of Things data

Abstract

The invention relates to the technical field of road construction quality control, in particular to an asphalt mixing station intelligent monitoring method and system based on Internet of Things data, and aims to solve the problems that in the prior art, technological parameters of an asphalt mixing station cannot be dynamically optimized, state vectors cannot be structured and defined as action spaces, and the working efficiency of the asphalt mixing station cannot be improved. The stability and convergence efficiency of strategy updating cannot be ensured, and long-term optimal control cannot be realized; the state and action space is constructed through the reinforcement learning strategy construction module, the multi-target reward function is combined, the reinforcement learning model is trained through the PPO algorithm, dynamic optimization of the technological parameters of the asphalt mixing station is achieved, the environment perception and regulation and control capacity of the model is enhanced through the structured state and the executable action, and the dynamic optimization of the technological parameters of the asphalt mixing station is achieved. The weighted reward mechanism overall plans quality, energy consumption and stability, and the PPO algorithm ensures efficient and stable training and supports long-term optimal control.

Description

Technical Field

[0001] The present invention relates to the technical field of road construction quality control, and more specifically, to an intelligent monitoring method and system for an asphalt mixing station based on Internet of Things data. Background Art

[0002] Traditional asphalt mixing plants rely on manual recording of material ratios, temperature and other parameters, which are subject to data lag, easy errors, large monitoring blind spots, and complex wiring, making it difficult to achieve real-time monitoring of the entire area. Although existing IoT technology solutions attempt to improve data collection efficiency, they are limited by technical bottlenecks such as short transmission distances and weak anti-interference capabilities, and cannot meet the needs of large-scale construction scenarios, resulting in delayed quality control, slow abnormal response, low management efficiency, and difficulty in ensuring the quality of asphalt mixture production.

[0003] Patent application with reference publication number CN112327781A discloses a remote monitoring and management system for production quality of an asphalt mixing plant based on Internet of Things technology. The data transmission module is connected to the central monitoring computer of the mixing plant to transmit the data collected from the sensor monitoring data acquisition terminal. The data is then uploaded to the data analysis and information management platform through the operator's GPRS public base station. The collected monitoring data is compared and analyzed with the pre-entered standard configuration data, and the analysis results or alarm information is sent to the data analysis and information management platform and the user data receiving terminal to achieve real-time error correction and strict implementation of the mix ratio. The real-time collection, analysis and transmission of production data ensures that supervisors can timely, dynamically and comprehensively control the production and processing status in the mixing plant. The system can send graded alarm information when the production data is abnormal, realize active control of the production process, and provide data statistical analysis for process improvement after production.

[0004] However, the above-mentioned reference patent constructs an intelligent monitoring system for an asphalt mixing station based on the Internet of Things, realizing real-time supervision of data throughout the entire process, abnormal classification warning and process optimization, achieving active production control and quality efficiency improvement, but cannot dynamically optimize the process parameters of the asphalt mixing station, cannot structure the state vector and define it as an action space, cannot ensure the stability and convergence efficiency of the strategy update, and is not conducive to achieving long-term optimal control; at the same time, it cannot build a complete closed-loop control system from perception, decision-making to execution and optimization, and cannot realize dynamic regulation and closed-loop control of the process parameters of the asphalt mixing station, thereby reducing the intelligence level and production control efficiency of the asphalt mixing station.

[0005] To this end, we propose an intelligent monitoring method and system for asphalt mixing stations based on Internet of Things data to address the above problems. Summary of the Invention

[0006] The purpose of the present invention is to provide an intelligent monitoring method and system for an asphalt mixing plant based on Internet of Things data. This method solves the problems that the existing technology cannot dynamically optimize the process parameters of the asphalt mixing plant, cannot structure the state vector and define it as an action space, cannot ensure the stability and convergence efficiency of the strategy update, and is not conducive to achieving long-term optimal control. At the same time, it cannot build a complete closed-loop control system from perception, decision-making to execution and optimization, cannot realize dynamic regulation and closed-loop control of the process parameters of the asphalt mixing plant, and reduces the intelligence level and production control efficiency of the asphalt mixing plant.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] The intelligent monitoring method for asphalt mixing plants based on IoT data includes:

[0009] Step 1: Collect key monitoring data generated during the production process of the asphalt mixing plant and perform pre-processing operations on the collected key monitoring data;

[0010] Step 2: Collect key quality indicators of asphalt mixture and use machine learning models to build a quality indicator prediction model. Based on real-time key monitoring data, the model performs online prediction and anomaly detection.

[0011] Step 3: Based on the preprocessed key monitoring data, construct the state space and action space, design the reward function, and use the PPO algorithm to train the reinforcement learning model for dynamic scheduling and process parameter optimization of the asphalt mixing plant;

[0012] Step 4: Based on the pre-processed key monitoring data and quality prediction results, a reinforcement learning model is used to dynamically adjust key process parameters, update the control strategy based on feedback information, and build a closed-loop production control process.

[0013] As a preferred embodiment of the present invention, the process of processing the key quality indicators of the asphalt mixture in step 2 includes:

[0014] Obtain historical key quality indicators of asphalt mixtures and key monitoring data generated during the production process of asphalt mixing plants. Key quality indicators include Marshall stability, void ratio, dynamic stability, and freeze-thaw splitting strength ratio. Key monitoring data include aggregate silo material weight, asphalt heating temperature, aggregate heating temperature, mixture discharge temperature, mixing pot stirring time, mixing pot speed, and mineral powder addition ratio. Generate a collection cycle and divide the collection cycle into multiple collection periods.

[0015] Obtaining the weight of the aggregate silo material of the asphalt mixing plant during multiple collection periods, calculating the arithmetic average of the obtained multiple aggregate silo material weights, and recording the arithmetic average of the multiple aggregate silo material weights as the average aggregate silo material weight;

[0016] The average asphalt heating temperature, average aggregate heating temperature, average mixture discharge temperature, average mixing pot stirring time, average mixing pot speed and average mineral powder addition ratio can be obtained by the same method as that used to calculate the average aggregate bin material weight.

[0017] As a preferred embodiment of the present invention, the process of using a machine learning model to construct a quality indicator prediction model in step 2 includes:

[0018] The calculated average aggregate bin material weight, average asphalt heating temperature, average aggregate heating temperature, average mixture discharge temperature, average mixing pot stirring time, average mixing pot speed, and average mineral powder addition ratio are combined to form an input feature matrix X. The feature matrix X is used as the input of the machine learning model, and the quality index prediction matrix for the future period corresponding to each set of feature matrices X is used as the output of the machine learning model. The quality index prediction matrix for the future period is used as the prediction target, and the sum of the prediction errors of all training data is minimized as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the training is stopped to obtain the quality index prediction model. The quality index prediction model is expressed as follows:

[0019]

[0020] in represents the quality index prediction matrix for a period of time in the future, f(·) represents an arbitrary form of mapping function, ε represents the error term, MS represents the predicted value of Marshall stability, AVC represents the predicted value of void fraction, DS represents the predicted value of dynamic stability, and TSR represents the predicted value of freeze-thaw splitting strength ratio.

[0021] As a preferred embodiment of the present invention, the process of performing online prediction and anomaly detection based on real-time key monitoring data and using a model in step 2 includes:

[0022] Obtain real-time key monitoring data, convert it into the corresponding feature matrix X and input it into the quality indicator prediction model. The quality indicator prediction model is used to obtain the real-time quality indicator prediction matrix for a period of time in the future. The real-time prediction values of the key quality indicators in the real-time quality indicator prediction matrix are compared with their corresponding preset thresholds:

[0023] If the real-time predicted values of Marshall stability, void ratio, dynamic stability and freeze-thaw splitting strength ratio are all greater than their corresponding preset thresholds, it indicates that the overall performance of the currently produced asphalt mixture is good;

[0024] If the real-time predicted values of Marshall stability, void ratio, dynamic stability and freeze-thaw splitting strength ratio are not all greater than their corresponding preset thresholds, it indicates that there are quality problems with the currently produced asphalt mixture. An alarm will be triggered immediately and relevant personnel will be notified, and corresponding measures will be taken to deal with it.

[0025] As a preferred embodiment of the present invention, the process of constructing the state space and action space based on the pre-processed key monitoring data in step 3 includes:

[0026] Obtain the key monitoring data after preprocessing. The state space is positioned as the current process state and production operation status of the asphalt mixture. Based on the key monitoring data after preprocessing, the state vector is defined as follows:

[0027]

[0028] where s t is the state vector at time step t, represents the weight of the aggregate bin material at time step t, represents the asphalt heating temperature at time step t, represents the aggregate heating temperature at time step t, represents the discharge temperature of the mixture at time step t, represents the stirring time of the stirring pot at time step t, represents the stirring pot speed at time step t, Indicates the proportion of mineral powder added at time step t;

[0029] The action space is defined as executable control behaviors. Actions can include dynamic adjustments to the following process parameters:

[0030]

[0031] where a t is the action vector at time step t, is the adjustment amount for the i-th parameter at time step t.

[0032] As a preferred embodiment of the present invention, the process of designing a reward function based on the pre-processed key monitoring data in step 3 includes:

[0033] The goal of the reward function is to select a scheduling strategy that not only meets quality requirements but also ensures optimal energy consumption and stable production. The reward function is designed as a weighted combination of multiple sub-goals:

[0034]

[0035] where r t is the reward value at time step t, λ1, λ2, and λ3 represent the weight coefficients of quality reward, energy consumption reward, and stability reward, respectively;

[0036] Calculated by the following formula

[0037] in is the quality reward at time step t, is the predicted value of the i-th quality indicator at time step t, is the target value of the i-th quality indicator;

[0038] Calculated by the following formula

[0039] in is the energy consumption reward at time step t, α, β, and γ represent the influence coefficients of asphalt heating temperature, aggregate heating temperature, and mixing pot speed on energy consumption, respectively;

[0040] Calculated by the following formula

[0041] in is the stability reward at time step t.

[0042] As a preferred embodiment of the present invention, the process of using the PPO algorithm to train the reinforcement learning model for dynamic scheduling and process parameter optimization of the asphalt mixing plant in step 3 includes:

[0043] The PPO algorithm is used to train the model. The core training process is as follows:

[0044] Define the policy network, policy function π θ (a t |s t ) is a deep neural network with input state s t , output action a t The distribution parameters of

[0045] Define the value function network and estimate the state value V φ (s t ), used to calculate the advantage function A t ;

[0046] Strategy optimization objective function, using the clipping probability ratio method to define the optimization objective:

[0047] L CLIP (θ)=E t [min(r t (θ)A t ,clip(r t (θ),1-η,1+η)A t )]

[0048] A t =δ t =rt+γV(s t+1 )-V(s t );

[0049] Among them L CLIP (θ) is the loss function of the PPO algorithm, θ is the parameter of the policy network, E t is the expectation for all time steps t, r t (θ) is the probability ratio, A t is the advantage function, clip(x,a,b) is the clipping function, and η is the clipping threshold;

[0050] The specific steps for training a reinforcement learning model using the PPO algorithm are as follows:

[0051] T1: Initialize the policy network and value network parameters;

[0052] T2: Sampling trajectory in asphalt mixing plant simulation or digital twin environment (s t ,a t ,r t ,s t+1 );

[0053] T3: Calculate the advantage function A t ;

[0054] T4: Update the policy network and value network using gradient descent;

[0055] T5: Repeat training until convergence.

[0056] As a preferred embodiment of the present invention, the process of dynamically adjusting key process parameters using a reinforcement learning model and updating the control strategy based on feedback information in step 4 includes:

[0057] Obtain key monitoring data and quality prediction results after pretreatment. The quality prediction results indicate that the overall performance of the asphalt mixture is good or that there are quality problems with the asphalt mixture. Obtain key process parameters during the asphalt mixture production process. Key process parameters include asphalt ratio, mineral powder ratio, and mixing time.

[0058] According to the output of the reinforcement learning model, the three key process parameters are adjusted in real time. If the action output by the model is: a t =[ΔAsphalt Ratio t , Δ mineral powder ratio t , Δ stirring time t ];

[0059] The new process parameter values are:

[0060] New asphalt mix ratio = old asphalt mix ratio + Δ asphalt mix ratio t ;

[0061] New ore powder ratio = old ore powder ratio + Δ ore powder ratio t ;

[0062] New stirring time = old stirring time + Δ stirring time t ;

[0063] The specific process of updating the control strategy based on feedback information is as follows:

[0064] Adjust the process parameters of the asphalt mixing plant according to the actions output by the model;

[0065] When performing action a t After that, the system will generate a new state s t+1 , calculate the reward r according to the reward function defined in step 3 t ;

[0066] The new state-action-reward triple (s t ,a t ,r t ) is input into the reinforcement learning model to update the model parameters.

[0067] As a preferred embodiment of the present invention, the steps of constructing a closed-loop production control process in step 4 are as follows:

[0068] S1: Real-time collection of key monitoring data generated during the production process of the asphalt mixing plant;

[0069] S2: Convert the collected data into the state vector s defined in step 3 t , the state vector s t Input into the trained reinforcement learning model to get action a t ;

[0070] S3: Action a output by the model t , adjust key process parameters in the asphalt mixture production process;

[0071] S4: Predict the quality of the produced asphalt mixture using the quality prediction model in step 3, and use the result as part of the feedback information;

[0072] S5: Calculate the reward r according to the reward function defined in step 3 t , and use (s t ,a t ,r t ) Update the parameters of the reinforcement learning model;

[0073] S6: Repeat S1-S5 to achieve closed-loop control.

[0074] As a preferred embodiment of the present invention, the intelligent monitoring system for asphalt mixing plants based on IoT data includes a data acquisition and processing module, a quality index prediction module, a reinforcement learning strategy construction module, and a process parameter dynamic control module;

[0075] The data acquisition and processing module is used to collect key monitoring data generated during the production process of the asphalt mixing plant and perform pre-processing operations on the collected key monitoring data;

[0076] The quality indicator prediction module is used to collect key quality indicators of asphalt mixtures, build a quality indicator prediction model using machine learning models, and perform online prediction and anomaly detection based on real-time key monitoring data through the model;

[0077] A reinforcement learning strategy building module is used to construct the state space and action space based on preprocessed key monitoring data, design the reward function, and use the PPO algorithm to train the reinforcement learning model for dynamic scheduling and process parameter optimization of asphalt mixing plants;

[0078] The process parameter dynamic control module is used to dynamically adjust key process parameters based on pre-processed key monitoring data and quality prediction results using a reinforcement learning model, update the control strategy according to feedback information, and build a closed-loop production control process.

[0079] Compared with the prior art, the advantages of the present invention are:

[0080] (1) In the present invention, the state and action space are constructed by the reinforcement learning strategy building module and combined with the multi-objective reward function. The PPO algorithm is used to train the reinforcement learning model to achieve dynamic optimization of the process parameters of the asphalt mixing plant. The structured state and executable action enhance the model's environmental perception and control capabilities. The weighted reward mechanism coordinates quality, energy consumption and stability. The PPO algorithm ensures efficient and stable training and supports long-term optimal control.

[0081] (2) In the present invention, the dynamic control module of process parameters integrates reinforcement learning, real-time monitoring and quality prediction to achieve dynamic control and closed-loop control of process parameters of asphalt mixing plants. Through state-action modeling and feedback optimization, the decision-making accuracy and adaptive ability are improved. Combined with boundary constraints to ensure operational safety, a "perception-decision-making-execution-optimization" closed loop is constructed, which significantly enhances the level of intelligence and control efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 This is a flowchart of the steps of the intelligent monitoring method for an asphalt mixing plant according to the present invention;

[0083] Figure 2This is a flowchart of the steps of using the PPO algorithm to train the reinforcement learning model in the present invention;

[0084] Figure 3 A flowchart of the steps for constructing a closed-loop production control process in the present invention;

[0085] Figure 4 This is a system block diagram in embodiment 3 of the present invention. DETAILED DESCRIPTION

[0086] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making creative work shall fall within the scope of protection of the present invention.

[0087] Example 1: Figure 1 and Figure 2 As shown, the intelligent monitoring method for asphalt mixing plants based on Internet of Things data proposed by the present invention includes:

[0088] Step 1: Collect key monitoring data generated during the production process of the asphalt mixing plant and perform preprocessing operations on the collected key monitoring data. The preprocessing operations include but are not limited to timestamp alignment, missing value filling, outlier removal, noise filtering, sensor calibration and normalization.

[0089] Data acquisition and preprocessing are the foundation for building an intelligent monitoring system for an asphalt mixing plant. By performing time alignment, gap filling, anomaly removal, noise filtering, sensor calibration, and normalization on multi-source sensor data, data quality, consistency, and accuracy can be effectively improved, environmental interference and equipment errors can be eliminated, and high-quality data support can be provided for subsequent quality prediction and intelligent control, ensuring model performance and system stability.

[0090] Step 2: Collect key quality indicators of asphalt mixture and use machine learning models to build a quality indicator prediction model. Based on real-time key monitoring data, the model performs online prediction and anomaly detection.

[0091] The process of processing the key quality indicators of asphalt mixture in step 2 includes:

[0092] Obtain historical key quality indicators of asphalt mixtures and key monitoring data generated during the production process of asphalt mixing plants. Key quality indicators include Marshall stability, void ratio, dynamic stability, and freeze-thaw splitting strength ratio. Key monitoring data include aggregate silo material weight, asphalt heating temperature, aggregate heating temperature, mixture discharge temperature, mixing pot stirring time, mixing pot speed, and mineral powder addition ratio. Generate a collection cycle and divide the collection cycle into multiple collection periods.

[0093] Obtaining the weight of the aggregate silo material of the asphalt mixing plant during multiple collection periods, calculating the arithmetic average of the obtained multiple aggregate silo material weights, and recording the arithmetic average of the multiple aggregate silo material weights as the average aggregate silo material weight;

[0094] The same method used to calculate the average aggregate bin material weight can be used to obtain the average asphalt heating temperature, average aggregate heating temperature, average mixture discharge temperature, average mixing pot stirring time, average mixing pot speed and average mineral powder addition ratio;

[0095] The process of using a machine learning model to build a quality indicator prediction model in step 2 includes:

[0096] The calculated average aggregate bin material weight, average asphalt heating temperature, average aggregate heating temperature, average mixture discharge temperature, average mixing pot stirring time, average mixing pot speed, and average mineral powder addition ratio are combined to form an input feature matrix X. The feature matrix X is used as the input of the machine learning model, and the quality index prediction matrix for the future period corresponding to each set of feature matrices X is used as the output of the machine learning model. The quality index prediction matrix for the future period is used as the prediction target, and the sum of the prediction errors of all training data is minimized as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the training is stopped to obtain the quality index prediction model. The quality index prediction model is expressed as follows:

[0097]

[0098] in represents the quality index prediction matrix for a period of time in the future, f(·) represents an arbitrary form of mapping function, ε represents the error term, MS represents the predicted value of Marshall stability, AVC represents the predicted value of void fraction, DS represents the predicted value of dynamic stability, and TSR represents the predicted value of freeze-thaw splitting strength ratio;

[0099] In step 2, the process of performing online prediction and anomaly detection based on real-time key monitoring data and using the model includes:

[0100] Obtain real-time key monitoring data, convert it into the corresponding feature matrix X and input it into the quality indicator prediction model. The quality indicator prediction model is used to obtain the real-time quality indicator prediction matrix for a period of time in the future. The real-time prediction values of the key quality indicators in the real-time quality indicator prediction matrix are compared with their corresponding preset thresholds:

[0101] If the real-time predicted values of Marshall stability, void ratio, dynamic stability and freeze-thaw splitting strength ratio are all greater than their corresponding preset thresholds, it indicates that the overall performance of the currently produced asphalt mixture is good;

[0102] If the real-time predicted values of Marshall stability, void ratio, dynamic stability, and freeze-thaw splitting strength ratio are not all greater than their corresponding preset thresholds, it indicates that there is a quality problem with the currently produced asphalt mixture. An alarm will be triggered immediately and relevant personnel will be notified. At the same time, appropriate measures will be taken to address the problem. The specific content of the treatment measures is as follows:

[0103] Suspend the current mixing batch production, mark the abnormal batch, and prevent unqualified materials from flowing into subsequent links;

[0104] Check the operating status of relevant sensors. If there is any deviation, recalibrate them to ensure accurate data collection;

[0105] Adjust corresponding parameters (such as adjusting the mineral powder ratio, increasing the mixing temperature, etc.) for non-compliant indicators, and use intelligent algorithms to recommend optimization solutions to achieve dynamic control;

[0106] Take samples of abnormal batches for testing to verify the model prediction results and further determine whether process adjustments are needed based on experimental data;

[0107] Add new samples to the training set, update the quality prediction model, adjust the closed-loop control system feedback mechanism, and improve control accuracy;

[0108] Check the equipment's operating status, troubleshoot potential faults, and repair or replace parts when necessary;

[0109] Completely record the entire processing process and results, analyze the root causes of the problem, formulate improvement plans, and prevent recurrence;

[0110] Organize operator training, strengthen quality awareness, enhance standardized operation capabilities, and improve overall management level;

[0111] By building a machine learning quality prediction model, online prediction and anomaly detection of key quality indicators of asphalt mixtures are achieved, significantly improving the quality control efficiency and intelligence level of the mixing plant production process; by extracting and modeling features of multi-dimensional monitoring data, the accuracy and robustness of the prediction are enhanced; based on the prediction results of future quality indicators, potential quality problems can be discovered in advance and early warning mechanisms can be triggered, realizing the transition from passive detection to active prevention and control; at the same time, the prediction results provide key feedback basis for the dynamic adjustment of subsequent process parameters, promoting the formation of closed-loop optimization of the entire system, effectively reducing dependence on manual experience, and improving production stability and intelligent decision-making capabilities.

[0112] Step 3: Based on the preprocessed key monitoring data, construct the state space and action space, design the reward function, and use the PPO algorithm to train the reinforcement learning model for dynamic scheduling and process parameter optimization of the asphalt mixing plant;

[0113] The process of constructing the state space and action space based on the preprocessed key monitoring data in step 3 includes:

[0114] Obtain the key monitoring data after preprocessing. The state space is positioned as the current process state and production operation status of the asphalt mixture. Based on the key monitoring data after preprocessing, the state vector is defined as follows:

[0115]

[0116] where s t is the state vector at time step t, which is a vector containing 7 elements and represents the current state of the asphalt mixing plant. represents the weight of the aggregate bin material at time step t, represents the asphalt heating temperature at time step t, represents the aggregate heating temperature at time step t, represents the discharge temperature of the mixture at time step t, represents the stirring time of the stirring pot at time step t, represents the stirring pot speed at time step t, Indicates the proportion of mineral powder added at time step t;

[0117] The action space is defined as executable control behaviors. Actions can include dynamic adjustments to the following process parameters:

[0118]

[0119] where a t is the action vector at time step t, is the adjustment amount of the i-th parameter at time step t;

[0120] The process of designing the reward function based on the preprocessed key monitoring data in step 3 includes:

[0121] The goal of the reward function is to select a scheduling strategy that not only meets quality requirements but also ensures optimal energy consumption and stable production. The reward function is designed as a weighted combination of multiple sub-goals:

[0122]

[0123] where r t is the reward value at time step t, which is a scalar indicating how good the model's action is at time step t. λ1, λ2, and λ3 represent the weight coefficients of quality reward, energy consumption reward, and stability reward, respectively. These coefficients need to be adjusted according to actual conditions to balance the importance of different objectives.

[0124] Calculated by the following formula

[0125] in is the quality reward at time step t, is the predicted value of the i-th quality indicator at time step t, is the target value of the i-th quality indicator. The negative sign indicates that the closer the quality indicator is to the target value, the higher the reward.

[0126] Calculated by the following formula

[0127] in is the energy consumption reward at time step t, α, β, and γ represent the influence coefficients of asphalt heating temperature, aggregate heating temperature, and mixing pot speed on energy consumption, respectively. These coefficients need to be calibrated according to actual energy consumption data. The negative sign indicates that the lower the energy consumption, the higher the reward;

[0128] Calculated by the following formula

[0129] in is the stability reward at time step t, where the negative sign indicates that the more stable the parameter adjustment, the higher the reward;

[0130] The process of using the PPO algorithm to train the reinforcement learning model for dynamic scheduling and process parameter optimization of asphalt mixing plants in step 3 includes:

[0131] The PPO algorithm is used to train the model. The PPO algorithm is a strategy optimization algorithm that combines stability and efficiency in reinforcement learning. The core training process is as follows:

[0132] Define the policy network, policy function π θ (a t |st ) is a deep neural network (such as a multilayer perceptron MLP), with the input state s t , output action a t the distribution parameters of (such as mean and variance);

[0133] Define the value function network and estimate the state value V φ (s t ), used to calculate the advantage function A t ;

[0134] Strategy optimization objective function, using the clipping probability ratio method to define the optimization objective:

[0135] L CLIP (θ)=E t [min(r t (θ)A t ,clip(r t (θ),1-η,1+η)A t )]

[0136] A t =δ t =rt+γV(s t+1 )-V(s t );

[0137] Among them L CLIP (θ) is the loss function of the PPO algorithm, θ is the parameter of the policy network, E t is the expectation for all time steps t, r t (θ) is the probability ratio, which represents the ratio of the probability of the current strategy to the old strategy, A t is the advantage function, clip(x,a,b) is the clipping function, which limits the value of x to between [a,b], and η is the clipping threshold, which controls the step size of the strategy update;

[0138] The specific steps for training a reinforcement learning model using the PPO algorithm are as follows:

[0139] T1: Initialize the policy network and value network parameters;

[0140] T2: Sampling trajectory in asphalt mixing plant simulation or digital twin environment (s t ,a t ,r t ,s t+1 );

[0141] T3: Calculate the advantage function A t ;

[0142] T4: Update the policy network and value network using gradient descent;

[0143] T5: Repeat training until convergence;

[0144] Based on the preprocessed monitoring data, the state space and action space are constructed. In combination with a multi-objective reward function, the PPO algorithm is used to train the reinforcement learning model, realizing the intelligent dynamic optimization of the process parameters of the asphalt mixing plant. By abstracting the key operating states into structured state vectors and defining executable parameter adjustment actions as the action space, the model is equipped with the ability to understand and control complex production environments. The multi-dimensional weighted reward mechanism takes into account quality compliance, energy consumption control, and operational stability, thus improving the overall performance of the strategy. The PPO algorithm is used for training to ensure the stability and convergence efficiency of the strategy update, which helps to achieve long-term optimal control. The overall approach provides a solid intelligent decision-making foundation for subsequent closed-loop control and real-time process adjustments.

[0145] Embodiment 2: The technical solution of this embodiment of the present invention differs from that of Embodiment 1 in that:

[0146] like Figure 1 and Figure 3 As shown, step 4: Based on the pre-processed key monitoring data and quality prediction results, the reinforcement learning model is used to dynamically adjust the key process parameters, update the control strategy according to the feedback information, and build a closed-loop production control process;

[0147] The process of using the reinforcement learning model to dynamically adjust key process parameters and update the control strategy based on feedback information in step 4 includes:

[0148] Obtain key monitoring data and quality prediction results after pretreatment. The quality prediction results indicate that the overall performance of the asphalt mixture is good or that there are quality problems with the asphalt mixture. Obtain key process parameters during the asphalt mixture production process. Key process parameters include asphalt ratio, mineral powder ratio, and mixing time.

[0149] According to the output of the reinforcement learning model, the three key process parameters are adjusted in real time. If the action output by the model is: a t =[ΔAsphalt Ratio t , Δ mineral powder ratio t , Δ stirring time t ];

[0150] The new process parameter values are:

[0151] New asphalt mix ratio = old asphalt mix ratio + Δ asphalt mix ratio t ;

[0152] New ore powder ratio = old ore powder ratio + Δ ore powder ratio t ;

[0153] New stirring time = old stirring time + Δ stirring time t;

[0154] To ensure that the adjusted parameters are feasible in actual production, all new parameter values must be limited to a preset reasonable range. For example, if the new asphalt mix ratio exceeds the set maximum value, it will be set to the maximum allowable value. If it is below the minimum value, it will be set to the minimum value. The same tailoring strategy is applied to other parameters to prevent exceeding the limits of equipment capabilities or process requirements.

[0155] According to the feedback information (reward r t and the new state s t+1 )The specific process of updating the control strategy is as follows:

[0156] Adjust the process parameters of the asphalt mixing plant according to the actions output by the model;

[0157] When performing action a t After that, the system will generate a new state s t+1 , calculate the reward r according to the reward function defined in step 3 t ,This reward value reflects the effect of the current action, such as whether the quality of the mixture meets the requirements, whether the energy consumption is reduced, etc. At the same time, more accurate quality indicators can be obtained from online testing or offline testing as feedback;

[0158] The new state-action-reward triple (s t ,a t ,r t ) is input into the reinforcement learning model to update the model parameters. The PPO algorithm will adjust the policy network parameters θ based on this feedback information, so that the model can make better action choices in the future state. This part depends on the specific implementation of the PPO algorithm in step 3;

[0159] The steps to build a closed-loop production control process in step 4 are as follows:

[0160] S1: Real-time collection of key monitoring data generated during the production process of the asphalt mixing plant;

[0161] S2: Convert the collected data into the state vector s defined in step 3 t , the state vector s t Input into the trained reinforcement learning model to get action a t ;

[0162] S3: Action a output by the model t , adjust key process parameters in the asphalt mixture production process;

[0163] S4: Predict the quality of the produced asphalt mixture using the quality prediction model in step 3, and use the result as part of the feedback information;

[0164] S5: Calculate the reward r according to the reward function defined in step 3 t , and use (s t ,a t ,r t ) Update the parameters of the reinforcement learning model;

[0165] S6: Repeat S1-S5 to achieve closed-loop control;

[0166] By introducing a reinforcement learning model and combining real-time monitoring data with quality prediction results, dynamic adjustment and closed-loop control of key process parameters of the asphalt mixing plant are achieved. Based on the predefined state space and action space, the system can make intelligent decisions based on the current production status and continuously optimize the strategy through a reward mechanism to improve control accuracy and adaptability. Boundary clipping is performed on the adjusted parameters to ensure the engineering feasibility of the operation and equipment safety. At the same time, feedback information (new status and reward value) is used to continuously update the model strategy, giving the system self-learning and self-adaptation capabilities, and promoting the evolution of the production process towards high-quality, low-energy consumption, and steady-state operation. The overall process builds a complete closed-loop control system from perception, decision-making to execution and optimization, which significantly improves the intelligence level and production control efficiency of the asphalt mixing plant.

[0167] Embodiment 3: The technical solution of the embodiment of the present invention is different from that of the embodiment 1 and the embodiment 2 in that:

[0168] like Figure 4 As shown in the figure, the intelligent monitoring system of asphalt mixing station based on IoT data includes data acquisition and processing module, quality index prediction module, reinforcement learning strategy construction module, and process parameter dynamic control module;

[0169] The data acquisition and processing module is used to collect key monitoring data generated during the production process of the asphalt mixing plant and perform pre-processing operations on the collected key monitoring data;

[0170] The quality indicator prediction module is used to collect key quality indicators of asphalt mixtures, build a quality indicator prediction model using machine learning models, and perform online prediction and anomaly detection based on real-time key monitoring data through the model;

[0171] A reinforcement learning strategy building module is used to construct the state space and action space based on preprocessed key monitoring data, design the reward function, and use the PPO algorithm to train the reinforcement learning model for dynamic scheduling and process parameter optimization of asphalt mixing plants;

[0172] The process parameter dynamic control module is used to dynamically adjust key process parameters based on pre-processed key monitoring data and quality prediction results using a reinforcement learning model, update the control strategy based on feedback information, and build a closed-loop production control process;

[0173] This intelligent monitoring system for asphalt mixing plants integrates the Internet of Things, machine learning, and reinforcement learning technologies to build an intelligent monitoring system for asphalt mixing plants. The data acquisition and processing module cleans and standardizes multi-source data, providing high-quality input for analysis. The quality prediction module uses machine learning to achieve online prediction and anomaly detection of mixture quality, improving real-time control and accuracy. The reinforcement learning module uses the PPO algorithm to train a control strategy that takes into account quality, energy consumption, and stability, enhancing autonomous decision-making capabilities. The process control module combines feedback information with dynamic optimization strategies to form a closed-loop control process, realizing intelligent management of the entire process from perception to execution.

[0174] The above are only preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes based on the technical solutions and improved concepts of the present invention within the technical scope disclosed by the present invention, and they should be covered by the scope of protection of the present invention.

Claims

1. The intelligent monitoring method of asphalt mixing station based on Internet of Things data is characterized by: include: Step 1: Collect key monitoring data generated during the production process of the asphalt mixing plant and perform pre-processing operations on the collected key monitoring data; Step 2: Collect key quality indicators of asphalt mixture and use machine learning models to build a quality indicator prediction model. Based on real-time key monitoring data, the model performs online prediction and anomaly detection. Step 3: Based on the preprocessed key monitoring data, construct the state space and action space, design the reward function, and use the PPO algorithm to train the reinforcement learning model for dynamic scheduling and process parameter optimization of the asphalt mixing plant; Step 4: Based on the pre-processed key monitoring data and quality prediction results, a reinforcement learning model is used to dynamically adjust key process parameters, update the control strategy based on feedback information, and build a closed-loop production control process.

2. The intelligent monitoring method for asphalt mixing station based on Internet of Things data according to claim 1 is characterized in that: The process of processing the key quality indicators of asphalt mixture in step 2 includes: Obtain historical key quality indicators of asphalt mixtures and key monitoring data generated during the production process of asphalt mixing plants. Key quality indicators include Marshall stability, void ratio, dynamic stability, and freeze-thaw splitting strength ratio. Key monitoring data include aggregate silo material weight, asphalt heating temperature, aggregate heating temperature, mixture discharge temperature, mixing pot stirring time, mixing pot speed, and mineral powder addition ratio. Generate a collection cycle and divide the collection cycle into multiple collection periods. Obtaining the weight of the aggregate silo material of the asphalt mixing plant during multiple collection periods, calculating the arithmetic average of the obtained multiple aggregate silo material weights, and recording the arithmetic average of the multiple aggregate silo material weights as the average aggregate silo material weight; The average asphalt heating temperature, average aggregate heating temperature, average mixture discharge temperature, average mixing pot stirring time, average mixing pot speed and average mineral powder addition ratio can be obtained by the same method as that used to calculate the average aggregate bin material weight.

3. The method for intelligent monitoring of an asphalt mixing station based on Internet of Things data according to claim 2 is characterized in that: The process of using the machine learning model to construct a quality indicator prediction model in step 2 includes: The calculated average aggregate bin material weight, average asphalt heating temperature, average aggregate heating temperature, average mixture discharge temperature, average mixing pot stirring time, average mixing pot speed, and average mineral powder addition ratio are combined to form an input feature matrix X. The feature matrix X is used as the input of the machine learning model, and the quality index prediction matrix for the future period corresponding to each set of feature matrices X is used as the output of the machine learning model. The quality index prediction matrix for the future period is used as the prediction target, and the sum of the prediction errors of all training data is minimized as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the training is stopped to obtain the quality index prediction model. The quality index prediction model is expressed as follows: in represents the quality index prediction matrix for a period of time in the future, f(·) represents an arbitrary form of mapping function, ε represents the error term, MS represents the predicted value of Marshall stability, AVC represents the predicted value of void fraction, DS represents the predicted value of dynamic stability, and TSR represents the predicted value of freeze-thaw splitting strength ratio.

4. The intelligent monitoring method for asphalt mixing station based on Internet of Things data according to claim 3 is characterized in that: The process of performing online prediction and anomaly detection based on real-time key monitoring data and the model in step 2 includes: Obtain real-time key monitoring data, convert it into the corresponding feature matrix X and input it into the quality indicator prediction model. The quality indicator prediction model is used to obtain the real-time quality indicator prediction matrix for a period of time in the future. The real-time prediction values of the key quality indicators in the real-time quality indicator prediction matrix are compared with their corresponding preset thresholds: If the real-time predicted values of Marshall stability, void ratio, dynamic stability and freeze-thaw splitting strength ratio are all greater than their corresponding preset thresholds, it indicates that the overall performance of the currently produced asphalt mixture is good; If the real-time predicted values of Marshall stability, void ratio, dynamic stability and freeze-thaw splitting strength ratio are not all greater than their corresponding preset thresholds, it indicates that there are quality problems with the currently produced asphalt mixture. An alarm will be triggered immediately and relevant personnel will be notified, and corresponding measures will be taken to deal with it.

5. The intelligent monitoring method for asphalt mixing station based on Internet of Things data according to claim 1 is characterized in that: The process of constructing the state space and action space based on the pre-processed key monitoring data in step 3 includes: Obtain the key monitoring data after preprocessing. The state space is positioned as the current process state and production operation status of the asphalt mixture. Based on the key monitoring data after preprocessing, the state vector is defined as follows: where s t is the state vector at time step t, represents the weight of the aggregate bin material at time step t, represents the asphalt heating temperature at time step t, represents the aggregate heating temperature at time step t, represents the discharge temperature of the mixture at time step t, represents the stirring time of the stirring pot at time step t, represents the stirring pot speed at time step t, Indicates the proportion of mineral powder added at time step t; The action space is defined as executable control behaviors. Actions can include dynamic adjustments to the following process parameters: where a t is the action vector at time step t, is the adjustment amount for the i-th parameter at time step t.

6. The method for intelligent monitoring of an asphalt mixing plant based on Internet of Things data according to claim 5 is characterized in that: The process of designing the reward function based on the pre-processed key monitoring data in step 3 includes: The goal of the reward function is to select a scheduling strategy that not only meets quality requirements but also ensures optimal energy consumption and stable production. The reward function is designed as a weighted combination of multiple sub-goals: where r t is the reward value at time step t, λ1, λ2, and λ3 represent the weight coefficients of quality reward, energy consumption reward, and stability reward, respectively; Calculated by the following formula in is the quality reward at time step t, is the predicted value of the i-th quality indicator at time step t, is the target value of the i-th quality indicator; Calculated by the following formula in is the energy consumption reward at time step t, α, β, and γ represent the influence coefficients of asphalt heating temperature, aggregate heating temperature, and mixing pot speed on energy consumption, respectively; Calculated by the following formula in is the stability reward at time step t.

7. The method for intelligent monitoring of an asphalt mixing station based on Internet of Things data according to claim 6, characterized in that: The process of using the PPO algorithm to train the reinforcement learning model for dynamic scheduling and process parameter optimization of asphalt mixing plants in step 3 includes: The PPO algorithm is used to train the model. The core training process is as follows: Define the policy network, policy function π θ (a t |s t ) is a deep neural network with input state s t , output action a t The distribution parameters of Define the value function network and estimate the state value V φ (s t ), used to calculate the advantage function A t ; Strategy optimization objective function, using the clipping probability ratio method to define the optimization objective: L CLIP (θ)=E t [min(r t (i)A t ,clip(r t (θ),1-η,1+η)A t )] A t =d t =rt+γV(s t+1 )-V(s t ); Among them L CLIP (θ) is the loss function of the PPO algorithm, θ is the parameter of the policy network, E t is the expectation for all time steps t, r t (θ) is the probability ratio, A t is the advantage function, clip(x,a,b) is the clipping function, and η is the clipping threshold; The specific steps for training a reinforcement learning model using the PPO algorithm are as follows: T1: Initialize the policy network and value network parameters; T2: Sampling trajectory in asphalt mixing plant simulation or digital twin environment (s t ,a t ,r t ,s t+1 ); T3: Calculate the advantage function A t ; T4: Update the policy network and value network using gradient descent; T5: Repeat training until convergence.

8. The intelligent monitoring method for asphalt mixing station based on Internet of Things data according to claim 1 is characterized in that: The process of dynamically adjusting key process parameters using the reinforcement learning model and updating the control strategy based on feedback information in step 4 includes: Obtain key monitoring data and quality prediction results after pretreatment. The quality prediction results indicate that the overall performance of the asphalt mixture is good or that there are quality problems with the asphalt mixture. Obtain key process parameters during the asphalt mixture production process. Key process parameters include asphalt ratio, mineral powder ratio, and mixing time. According to the output of the reinforcement learning model, the three key process parameters are adjusted in real time. If the action output by the model is: a t =[ΔAsphalt Ratio t , Δ mineral powder ratio t , Δ stirring time t ]; The new process parameter values are: New asphalt mix ratio = old asphalt mix ratio + Δ asphalt mix ratio t ; New ore powder ratio = old ore powder ratio + Δ ore powder ratio t ; New stirring time = old stirring time + Δ stirring time t ; The specific process of updating the control strategy based on feedback information is as follows: Adjust the process parameters of the asphalt mixing plant according to the actions output by the model; When performing action a t After that, the system will generate a new state s t+1 , calculate the reward r according to the reward function defined in step 3 t ; The new state-action-reward triple (s t ,a t ,r t ) is input into the reinforcement learning model to update the model parameters.

9. The method for intelligent monitoring of an asphalt mixing plant based on Internet of Things data according to claim 8, characterized in that: The steps for constructing a closed-loop production control process in step 4 are as follows: S1: Real-time collection of key monitoring data generated during the production process of the asphalt mixing plant; S2: Convert the collected data into the state vector s defined in step 3 t , the state vector s t Input into the trained reinforcement learning model to get action a t ; S3: Action a output by the model t , adjust key process parameters in the asphalt mixture production process; S4: Predict the quality of the produced asphalt mixture using the quality prediction model in step 3, and use the result as part of the feedback information; S5: Calculate the reward r according to the reward function defined in step 3 t , and use (s t ,a t ,r t ) Update the parameters of the reinforcement learning model; S6: Repeat S1-S5 to achieve closed-loop control.

10. The intelligent monitoring system for asphalt mixing plants based on IoT data is characterized by: Used to execute the asphalt mixing station intelligent monitoring method based on Internet of Things data as claimed in claim 1, the asphalt mixing station intelligent monitoring system based on Internet of Things data includes a data acquisition and processing module, a quality indicator prediction module, a reinforcement learning strategy construction module, and a process parameter dynamic control module; The data acquisition and processing module is used to collect key monitoring data generated during the production process of the asphalt mixing plant and perform pre-processing operations on the collected key monitoring data; The quality indicator prediction module is used to collect key quality indicators of asphalt mixtures, build a quality indicator prediction model using machine learning models, and perform online prediction and anomaly detection based on real-time key monitoring data through the model; A reinforcement learning strategy building module is used to construct the state space and action space based on preprocessed key monitoring data, design the reward function, and use the PPO algorithm to train the reinforcement learning model for dynamic scheduling and process parameter optimization of asphalt mixing plants; The process parameter dynamic control module is used to dynamically adjust key process parameters based on pre-processed key monitoring data and quality prediction results using a reinforcement learning model, update the control strategy according to feedback information, and build a closed-loop production control process.

Images