Multi-sensor fused system and method for regulating and controlling moisture content of cold recycled mixture

Abstract

The invention relates to the technical field of automatic control of road engineering, and discloses a multi-sensor fused cold recycled mixture moisture content regulation and control system and a multi-sensor fused cold recycled mixture moisture content regulation and control method. A data preprocessing and tensor construction module; a tensor analysis and state identification module; a moisture content prediction module; an intelligent decision module; a cooperative regulation and control execution module; the method comprises the following steps: fusing and collecting multi-modal parameters of a mixture, and preprocessing the multi-modal parameters to construct a high-dimensional time sequence tensor; identifying an internal state through tensor analysis, decoupling interference, extracting a pure signal and predicting the future moisture content; based on the predicted value and the internal state, the water adding amount is decided through reinforcement learning, and closed-loop water content regulation and control are accurately executed. According to the method, the problems of inaccurate measurement and control lag of a traditional method are solved, prospective and accurate dynamic regulation and control of the moisture content are realized through prediction and state recognition, and the stability and the engineering quality of a cold regeneration process are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of road engineering automation control, and in particular to a multi-sensor fusion cold recycled mixture moisture content control system and method. Background Art

[0002] Cold recycling technology, an economical and environmentally friendly road maintenance method, is widely used in pavement repair projects. In this technology, the moisture content of the cold-recycled mixture is a key process parameter that determines the final project quality. It directly affects the mixing effect, compaction performance, and structural strength of the mixture. Accurate and stable control of moisture content is key to ensuring the performance of cold-recycled pavements.

[0003] Currently, construction sites rely heavily on traditional methods to control moisture content. The mainstream approach relies on operator experience, combined with feedback control from single-point sensors like microwave or infrared sensors deployed on the production line. This involves manually or semi-automatically adjusting the amount of water added based on a single sensor reading. This approach, based on real-time measurement and empirical judgment, is currently the most widely adopted technical solution.

[0004] However, these existing technologies have inherent limitations in their application. The aggregate, temperature, and other components of cold recycled mixtures are complex and variable, which can easily interfere with the measurement of a single sensor, resulting in inaccurate readings and poor stability. At the same time, there is a significant delay in the control process from measurement to adjustment, making it difficult to cope with real-time fluctuations in raw materials, resulting in low moisture content control accuracy. More importantly, these methods have difficulty in gaining insight into key physical and chemical states within the mixture, such as emulsified asphalt demulsification. Their control strategies remain superficial, limiting the improvement of the level of process refinement.

[0005] Therefore, the present invention proposes a multi-sensor fusion cold recycled mixture moisture content control system and method to solve the shortcomings of the existing technology. Summary of the Invention

[0006] In response to the shortcomings of the existing technology, the present invention provides a multi-sensor fusion cold recycled mixture moisture content control system and method, which solves the problems of low moisture content control accuracy, poor robustness and untimely response caused by traditional control methods due to the susceptibility of single-point measurement to interference, lag in control feedback and difficulty in understanding the internal physical and chemical state of the material.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a multi-sensor fusion cold recycled mixture moisture content control system, comprising: Multimodal sensing module: used to collect parameter information of at least two different modes of cold recycled mixture in real time through multiple sensors, the parameter information including dielectric constant data, spectral reflectance data, surface temperature distribution data, volume profile and flow rate data, as well as on-site ambient temperature and humidity; Data preprocessing and tensor construction module: used to clean, normalize and interpolate missing values ​​of the parameter information collected in real time, and construct a fourth-order time series tensor including the parameter information; A tensor analysis and state recognition module is configured to perform tensor representation learning on the fourth-order time series tensor to extract potential features, identify the internal component states of the cold recycled mixture based on the potential features, and decouple interference caused by non-moisture content factors in the parameter information to obtain a pure moisture content-related signal; Moisture content prediction module: used for adaptively predicting the future moisture content of the cold recycled mixture based on the pure moisture content related signal; Intelligent decision-making module: for dynamically deciding the water addition adjustment instruction required for the cold recycled mixture based on the future moisture content prediction value and the internal component state through causal inference and reinforcement learning; Collaborative control execution module: used to adjust the water addition amount according to the water addition amount adjustment instruction to dynamically control the moisture content of the cold recycled mixture.

[0008] Preferably, the multimodal perception module collects the parameter information in real time through the following sensors: A broadband microwave sensor array is deployed at the cold recycled mixture conveyor belt or the mixer outlet to collect dielectric constant data of the mixture; a multispectral near-infrared sensor deployed above the surface of the cold recycled mixture, for collecting spectral reflectance data of the mixture; A high-precision thermal imaging sensor for monitoring the surface temperature distribution of the cold recycled mixture; a high-resolution laser displacement sensor for measuring the volume profile and flow rate of the cold recycled mixture; Ambient temperature and humidity sensors are used to monitor the on-site ambient temperature and humidity of cold recycled mixture construction.

[0009] Preferably, the data preprocessing and tensor construction module includes: Clean the parameter information collected in real time, including outlier removal and noise filtering; Performing Min-Max normalization processing on the parameter information; The parameter information is interpolated in time series to fill in missing data points and constructed as a fourth-order time series tensor ; Where, Indicates the number of sensor modalities; Indicates the number of sampling points in the sensor space; Indicates the length of the time series; Indicates the feature dimension corresponding to each sampling point.

[0010] Preferably, the tensor analysis and state identification module includes: Performing tensor representation learning on the fourth-order time series tensor to obtain a core tensor and a factor matrix by non-negative tensor factorization, wherein the core tensor and the factor matrix represent potential characteristics of free water, bound water, aggregate, asphalt content, and temperature components in the cold recycled mixture; Based on the core tensor and the factor matrix, identifying the internal component state of the cold recycled mixture through a tensor neural network, the internal component state including the real-time demulsification degree, free water content, and bound water content of the emulsified asphalt; According to the internal component state, tensor independent component analysis or orthogonal projection method is used to decouple interference caused by non-water content factors from the fourth-order time series tensor to obtain the pure water content related signal.

[0011] Preferably, the step of obtaining the core tensor and factor matrix by non-negative tensor factor decomposition comprises: For the fourth-order time series tensor Perform non-negative Tucker decomposition, which is expressed as: ; Where, is the core tensor; 、 、 and are the factor matrices corresponding to sensor modality, spatial sampling points, time series, and feature dimensions respectively; Represents a tensor along the Modulo product, modal index , corresponding to sensor modality, spatial sampling points, time series and feature dimensions respectively.

[0012] Preferably, the moisture content prediction module includes: constructing the pure moisture content related signal as an input time series; Processing the input time series through a gated recurrent unit network with an integrated attention mechanism to learn temporal dependencies and generate weighted hidden state representations; The weighted hidden state representation is input into a fully connected output layer to generate and output a predicted value of the future moisture content of the cold recycled mixture at one or more future time points.

[0013] Preferably, the weighted hidden state representation Calculated by the following formula: ; Where, The gated recurrent unit network at time step The hidden state of the output; The hidden state calculated by the attention mechanism The weight of is the length of the input time series.

[0014] Preferably, the intelligent decision-making module includes: Using a causal inference method, constructing and updating a structural causal model, wherein the structural causal model is used to quantify the causal effects of the internal component state, historical water addition, and environmental parameters on the moisture content; The reinforcement learning agent is based on a state composed of the future moisture content prediction value, the internal component state, and the causal effect, and takes minimizing the deviation between the future moisture content prediction value and the preset target moisture content as the optimization goal. Through a deep Q network, the reinforcement learning agent dynamically decides and outputs the water addition adjustment instruction.

[0015] Preferably, the collaborative control execution module includes: receiving the water addition amount adjustment instruction and parsing the water addition amount adjustment instruction into a target flow rate setting value; generating an actuator control signal based on a deviation between the target flow rate set value and the actual flow rate value fed back in real time by the water flow meter through a proportional-integral-differential controller; According to the actuator control signal, the electronic control valve or the variable frequency water pump is driven to accurately adjust the real-time water addition amount entering the cold recycled mixture, so as to dynamically control the moisture content.

[0016] The present invention also provides a method for controlling the moisture content of cold recycled mixture based on multi-sensor fusion, comprising the following steps: Collecting parameter information of at least two different modes of the cold recycled mixture in real time through multiple sensors, the parameter information including dielectric constant data, spectral reflectance data, surface temperature distribution data, volume profile and flow rate data, and on-site ambient temperature and humidity; Cleaning, normalizing, and interpolating missing values ​​of the parameter information collected in real time, and constructing a fourth-order time series tensor including the parameter information; performing tensor representation learning on the fourth-order time series tensor to extract potential features, identifying internal component states of the cold recycled mixture based on the potential features, and decoupling interference caused by non-moisture content factors in the parameter information to obtain a pure moisture content-related signal; Adaptively predicting a future moisture content value of the cold recycled mixture based on the pure moisture content related signal; Based on the future moisture content prediction value and the internal component state, dynamically determining the water addition adjustment instruction required for the cold recycled mixture through causal inference and reinforcement learning; According to the water addition amount adjustment instruction, the water addition amount is adjusted to dynamically control the moisture content of the cold recycled mixture.

[0017] The present invention provides a multi-sensor fusion cold recycled mixture moisture content control system and method. It has the following beneficial effects: 1. This invention utilizes a multimodal sensing module and a tensor analysis and state recognition module to construct a high-dimensional time series tensor using data such as dielectric constant and spectral reflectance collected through multi-sensor fusion. This data is then deeply processed using tensor analysis methods. This approach effectively decouples interference from non-moisture factors in the mixture, such as aggregate and temperature, and extracts a pure moisture-related signal. This fundamentally addresses the vulnerability of traditional single-point or single-modal measurement methods to complex field conditions and their poor robustness, significantly improving the accuracy and reliability of moisture content monitoring in cold recycled mixtures.

[0018] 2. By building a moisture prediction module and an intelligent decision-making module, this invention adaptively predicts the future moisture content of the mixture based on pure moisture-related signals and formulates water addition adjustment instructions by combining causal inference and reinforcement learning. This shifts the control mode from "passive response" to "active prediction," effectively overcoming the inherent delays caused by material transportation and water penetration in the production line. This makes dynamic control more forward-looking and precise, ensuring that the moisture content of the final mixture remains stable within the target range over the long term.

[0019] 3. This invention uses tensor analysis and state recognition modules to identify the internal component states of the mixture, providing online insights into key process indicators such as the real-time demulsification level of the emulsified asphalt and the content of free and bound water. This enables the system to not only adjust the total moisture content but also adaptively optimize and control the mixture based on its inherent physical and chemical changes. This control strategy, based on deep state understanding, significantly enhances the refined management of the cold regeneration process, ensuring the ultimate performance and engineering quality of the mixture. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a system architecture diagram of the present invention; Figure 2 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0022] Please see the attached Figure 1 The embodiment of the present invention provides a multi-sensor fusion cold recycled mixture moisture content control system, including: Multimodal sensing module: used to collect parameter information of at least two different modes of cold recycled mixture in real time through multiple sensors. The parameter information includes dielectric constant data, spectral reflectance data, surface temperature distribution data, volume profile and flow rate data, as well as on-site ambient temperature and humidity; In this embodiment, the core task of the multimodal perception module is to construct a comprehensive, three-dimensional, high-dimensional real-time data stream, providing rich and reliable input for subsequent data processing, state recognition, and intelligent decision-making. The design concept of the multimodal perception module is that the moisture content of the cold recycled mixture is not an isolated physical quantity, but rather the result of a dynamic coupling of multiple factors such as the material's internal composition, physical form, chemical reaction process, and external environment. Any single sensing method is difficult to fully and accurately characterize its true state. Therefore, this invention adopts a multi-sensor information fusion strategy to achieve deep insight into the mixture's state.

[0023] Specifically, the multimodal sensing module collects parameter information of cold recycled mixture in real time by collaboratively deploying a series of heterogeneous sensors. Preferably, the role of these sensors and their functions in the system are described as follows: First, to obtain information on the overall moisture content within the mixture, this embodiment deploys a broadband microwave sensor array. This array is preferably installed beneath the cold recycled mixture conveyor belt or at the mixer's exit channel to achieve non-contact, penetrating measurement of the material flow. The underlying principle is that the complex dielectric constant of water is much higher than that of other components, such as aggregate and asphalt, in the microwave frequency band. The complex dielectric constant is expressed as: ; Where, is the complex dielectric constant; is the real part of the dielectric constant, reflecting the polarization and energy storage capacity of the material; is the dielectric loss factor, which represents the material's ability to absorb microwave energy; Is an imaginary unit.

[0024] Real part of dielectric constant and dielectric loss factor The dielectric constant of the material is closely related to the moisture content and form (free or bound water). By emitting a broad-spectrum microwave signal and detecting its attenuation and phase shift after passing through the material, the system can calculate the dielectric constant data that characterizes the overall volumetric moisture content, providing a basis for macroscopic control of moisture content.

[0025] Secondly, to compensate for the microwave sensor's limited sensitivity to surface moisture and to provide additional identification of water form, this embodiment employs a multispectral near-infrared sensor mounted above the surface of the cold recycled mixture. This sensor exploits the fact that water molecules exhibit strong characteristic absorption peaks in the near-infrared spectral region (preferably around 1450nm and 1940nm). By emitting near-infrared light of a specific wavelength band onto the mixture surface and measuring the spectral reflectance data from its diffuse reflection, the free water content in the material's surface can be accurately quantified. This data, combined with the overall moisture content measured by the microwave sensor, helps the system distinguish between internal and surface water, providing critical information for subsequently determining the state of internal components, such as the progress of emulsified asphalt demulsification.

[0026] Furthermore, the temperature state of the mixture is an important window to reveal its internal chemical and physical processes. For this reason, a high-precision thermal imaging sensor is integrated in this embodiment. The sensor is used to monitor the continuous surface temperature distribution of the cold recycled mixture in real time during transportation or mixing. The temperature distribution information has a dual meaning: on the one hand, the demulsification process of emulsified asphalt is an exothermic reaction. By monitoring the area of ​​abnormal temperature increase, the location and severity of the demulsification can be indirectly judged; on the other hand, the evaporation of water is an endothermic process, which will cause the surface of the material to cool down. The cooling rate is directly related to the free water content, the temperature of the material itself, the wind speed, and the ambient temperature and humidity. Therefore, the surface temperature distribution data provides an indispensable thermodynamic perspective for systematic understanding of the dynamic change process of moisture content and energy balance.

[0027] In addition, in order to achieve precise proportional control of the amount of water added, the real-time output or flow of the mixture must be known. In this embodiment, a high-resolution laser displacement sensor is used. This sensor is usually installed above the conveyor belt and obtains its dynamic three-dimensional volume profile by scanning the surface of the moving mixture at high speed. By integrating the profile data of each frame, the instantaneous cross-sectional area of ​​the material flow can be calculated. Combined with the real-time conveyor belt speed obtained by the speed measuring device , the system can accurately calculate the instantaneous volume flow of the mixture The volume profile and flow rate data are the core basis for the subsequent coordinated control execution module to calculate the target water addition flow rate, ensuring the accuracy of control.

[0028] Finally, the external environment is a crucial boundary condition that influences moisture content changes. This embodiment also includes ambient temperature and humidity sensors to monitor the temperature and humidity of the construction site. These parameters directly influence the evaporation rate of the mixture and serve as important input variables for environmental compensation and disturbance suppression in moisture content prediction models and intelligent decision-making models.

[0029] The multimodal perception module integrates and synchronizes the clocks of the above-mentioned sensors, and can capture comprehensive information about the state of cold recycled mixture in real time and synchronously from multiple dimensions such as electromagnetic properties, spectral chemistry, thermodynamics, physical form and environmental conditions, forming a high-dimensional time series data set.

[0030] Data preprocessing and tensor construction module: used to clean, normalize and interpolate missing values ​​of the parameter information collected in real time, and construct a fourth-order time series tensor including the parameter information; In this embodiment, the data preprocessing and tensor construction module aims to transform the heterogeneous and noisy raw data stream collected by the upstream multimodal perception module into a structured, high-quality data format suitable for advanced algorithm analysis. In real industrial environments, directly collected sensor data often suffers from signal glitches, random noise, numerical drift, asynchrony, and packet loss. If used directly without processing, it will seriously affect the accuracy and stability of subsequent analytical models.

[0031] Specifically, to achieve the above objectives, the data preprocessing and tensor construction module integrates a series of data processing functions. First, the module performs data cleaning operations on the parameter information collected in real time. This step includes outlier removal and noise filtering. The generation of outliers may be due to transient sensor failures or severe external electromagnetic interference. For this reason, the system can preferably adopt a statistically based identification method, such as a method based on the 3σ criterion or the interquartile range (IQR), to identify and eliminate abnormal data points that significantly deviate from the normal data distribution. After eliminating outliers, in order to further suppress high-frequency random noise (such as interference introduced by mechanical vibration or circuit thermal noise), the system can use a time-domain filtering algorithm, such as a sliding average filter, or more preferably a Kalman filter, to smooth the time series signals of each sensor to extract a more stable potential trend signal.

[0032] After data cleaning, the module performs Min-Max normalization on the parameter information, as the parameter information output by different sensors has very different physical dimensions and numerical ranges (for example, temperature is measured in degrees Celsius, and the dielectric constant is a dimensionless value). To mitigate the potential impact of these scale differences on subsequent model training, which can slow convergence or bias the model towards features with larger values, the module performs Min-Max normalization on the parameter information. This process linearly maps the data for each modality to a uniform numerical range, such as [0, 1] or [-1, 1]. This ensures that features with different physical meanings have equal contribution weight in the model analysis.

[0033] Considering the potential for data loss due to unstable industrial field network communications or intermittent sensor maintenance, this module further performs time series interpolation on parameter information to fill in missing data points. To ensure the continuity and dynamic characteristics of the time series, interpolation methods suitable for time series data can be used. In simple implementations, linear interpolation or forward / backward filling can be used. In preferred embodiments requiring higher data fidelity, methods based on spline interpolation or autoregressive models (such as ARIMA) can be used to estimate missing values ​​based on trend information before and after the data point, thereby maximally restoring the dynamic characteristics of the original signal.

[0034] Finally, after the aforementioned cleaning, normalization, and interpolation, the data preprocessing and tensor construction module fuses and structures these high-quality multimodal data streams into a fourth-order time series tensor. As a high-order data structure, tensors can naturally and losslessly preserve the inherent correlations of multi-source information in various physical dimensions, avoiding the information loss that may occur with traditional matrix or vectorization methods. The fourth-order time series tensor is specifically represented as: ; Where, Indicates the number of sensor modalities; Indicates the number of sampling points in the sensor space, reflecting the distribution of data in the spatial dimension; Indicates the length of the time series and defines the time window size used for an analysis. The system processes the data stream in a sliding manner based on this window. Indicates the feature dimension corresponding to each sampling point. For some sensors, one sampling outputs multiple related feature values, and this dimension is used to characterize these features.

[0035] The data preprocessing and tensor construction module finally outputs a regular, clean, and highly information-concentrated fourth-order time series tensor. .

[0036] Tensor Analysis and State Identification Module: This module is used to perform tensor representation learning on the fourth-order time series tensor to extract potential features, identify the internal component states of the cold recycled mixture based on the potential features, and decouple the interference caused by non-moisture content factors in the parameter information to obtain a pure moisture content-related signal; In this embodiment, the tensor analysis and state recognition module receives the structured fourth-order time series tensor generated by the "data preprocessing and tensor construction module" The tensor analysis and state recognition module uses advanced tensor calculation methods to deeply explore the intrinsic structure of high-dimensional data in order to achieve two closely related goals: first, to extract and identify the internal component states that are difficult to measure directly and characterize the essential properties of cold recycled mixtures from complex data; second, based on a deep understanding of the mixture state, to accurately decouple the pure signals directly related to the moisture content change from the highly coupled multimodal signals, thereby laying a solid foundation for subsequent accurate prediction and decision-making.

[0037] To achieve the above goals, the module first performs a four-order time series tensor Perform tensor representation learning. In a preferred embodiment, this step is accomplished through non-negative tensor factorization, specifically using the Non-negative Tucker Decomposition model. This decomposition aims to approximate a complex high-order tensor as the product of a compact core tensor and a series of factor matrices. Its mathematical expression is as follows: ; Where, It is a core tensor whose element size represents the interaction strength between potential features (or principal components) in different dimensions and can be regarded as the core and hub of the entire data structure; 、 、 and are the factor matrices corresponding to sensor modality, spatial sampling points, time series, and feature dimensions respectively; Represents a tensor along the Modulo product, modal index , corresponding to sensor modality, spatial sampling points, time series and feature dimensions respectively.

[0038] In the present invention, a non-negative constraint (i.e., requiring and all The decomposed latent features are all non-negative) are crucial because they make the decomposed latent features have clear physical interpretability and can directly correspond to the contribution of actual physical quantities such as free water, bound water, aggregate, asphalt content and temperature components.

[0039] In order to obtain the core tensor that can characterize the intrinsic nature of the mixture and the factor matrix Finally, based on these extracted latent features, the module uses a pre-trained tensor neural network (TNN) to identify the internal component states of the cold recycled mixture. Unlike traditional neural networks, TNNs can directly process tensor inputs, fully preserving the structural information of latent features in multidimensional space. This allows for more effective learning of the complex nonlinear mapping relationship between these features and the macroscopic state of the mixture. The internal component states output by this network serve as a key basis for subsequent intelligent decision-making. Preferably, this state includes key performance indicators such as the real-time demulsification level of emulsified asphalt, free water content, and bound water content, which are difficult to directly obtain through a single sensor.

[0040] At the same time, the tensor analysis and state identification module performs another key task: obtaining pure moisture-related signals. Since the signal changes caused by non-moisture content factors (such as fluctuations in aggregate grading, changes in recycled material sources, or fine-tuning of asphalt addition) in the original multimodal parameter information are overlapped with the signals caused by moisture content changes, direct use will lead to huge errors in moisture content prediction. Based on the internal component states identified in the previous step, this module can clarify which components the current main interference sources come from. Based on this, the system preferably adopts signal processing methods such as Tensor Independent Component Analysis (TICA) or orthogonal projection. The principle is to regard the total signal as a linear mixture of multiple statistically independent source signals (moisture content signals, asphalt interference signals, aggregate interference signals, etc.), and to maximize the non-Gaussianity of the output signal through an algorithm or use the orthogonality principle to remove the components that characterize the interference of non-moisture content factors from the original fourth-order time series tensor. Precise separation or projection culling in .

[0041] After this decoupling process, the tensor analysis and state recognition module ultimately outputs a one-dimensional time series—a pure moisture content-related signal. This signal minimizes interference from other physical and chemical factors, and its fluctuations more accurately and sensitively reflect the actual changes in the moisture content of the cold recycled mix. Thus, the tensor analysis and state recognition module successfully transforms high-dimensional, complex, and noisy input data into two clear, high-value outputs: internal state recognition for macroeconomic decision-making, and a pure core signal for precise prediction.

[0042] Moisture content prediction module: used to adaptively predict the future moisture content of cold recycled mixture based on pure moisture content related signals; In this embodiment, the core task of the moisture content prediction module is to accurately and adaptively predict the moisture content of the cold recycled mixture in one or more future time steps based on the pure, interference-free moisture content-related signal provided by the upstream "Tensor Analysis and State Identification Module." In actual production, there are inevitable physical delays (such as water flow and material mixing) and chemical delays (such as emulsified asphalt demulsification and water absorption) between water addition and the corresponding change in the mixture's moisture content.

[0043] The input to the moisture prediction module is the clean moisture-related signal obtained after processing by the previous module. This signal is first constructed as an input time series, which serves as the historical basis for the prediction model. In this embodiment, an attention-based gated recurrent unit (GRU) network with an integrated attention mechanism is preferably used as the core prediction model.

[0044] The Gated Recurrent Unit (GRU) network was chosen because, as an advanced recurrent neural network (RNN) variant, its unique internal "update gate" and "reset gate" structures effectively capture long-term dependencies in time series data, while also significantly alleviating the vanishing or exploding gradient issues that can occur with traditional RNNs when processing long sequences. This enables the model to deeply understand the complex dynamic patterns of moisture content signals evolving over time.

[0045] However, in a sequence of moisture content changes, not all historical moments contribute equally to future predictions. Data from key events (such as a major material change or a sudden environmental change) are far more valuable than data from periods of stable operation. To address this, the present invention integrates an attention mechanism within the GRU network. This mechanism empowers the model with an adaptive "attention" capability, enabling it to dynamically assign different importance weights to each time step in the input time series when making predictions.

[0046] Specifically, the prediction process of this module is as follows: the input time series is sequentially fed into the GRU network. At each time step , the GRU unit combines the current input and the hidden state of the previous moment to generate the hidden state of the current moment After processing the entire length After the input sequence, the attention mechanism will be all generated hidden states Evaluate and calculate a set of corresponding attention weights .

[0047] Subsequently, the moisture content prediction module generates a single, highly information-concentrated context vector by weighted summing of all hidden states, namely, the weighted hidden state representation of The calculation process can be expressed as follows: ; Where, For the gated recurrent unit network at time step The hidden state of the output; The hidden state calculated by the attention mechanism The weight of , whose value reflects the importance of the information at that moment to the final prediction, and satisfies ; is the length of the input time series.

[0048] Finally, this contains a weighted hidden state representation that dynamically focuses on the key information of the entire historical sequence , is input to the fully connected output layer. This output layer acts as a decoder, mapping and decoding this high-dimensional, abstract feature representation into specific physical quantities, thereby generating and outputting the predicted future moisture content of the cold recycled mixture at one or more future time points.

[0049] The moisture content prediction module can make full use of the inherent laws and key nodes of historical information to provide reliable judgments on future trends.

[0050] Intelligent decision-making module: used to dynamically determine the water addition adjustment instructions required for cold recycled mixture based on the future moisture content prediction value and internal component status through causal inference and reinforcement learning; In this embodiment, the fundamental task of the intelligent decision-making module is to comprehensively utilize the multi-dimensional information provided by upstream modules, including forward-looking predictions about the future and deep insights into the inherent state of the material. Through an advanced algorithmic framework that transcends traditional control logic, it dynamically determines the optimal water addition adjustment instructions. The core concept of this design is that controlling the moisture content of cold recycled mixtures is a complex, nonlinear, and dynamic process influenced by multiple coupled factors. Simple controllers based on fixed rules or models are difficult to adapt to real-time changes in operating conditions. By integrating causal inference with reinforcement learning, this module aims to enable the system to understand, learn, and autonomously optimize.

[0051] To achieve this advanced decision-making capability, the intelligent decision-making module first employs causal inference to construct and continuously update a structural causal model (SCM). The purpose of introducing causal inference is to enable the system to move beyond "correlation" to "causality." This means understanding not only which variables change with moisture content, but also which variables "cause" the change in moisture content and to what extent. The SCM represents the causal relationships between variables using a directed acyclic graph (DAG). Its nodes include internal component states (such as demulsification) identified by upstream modules, historical water additions, and environmental parameters, while the target node is moisture content. By learning from historical data, the model can quantitatively calculate the causal effect of each upstream variable on moisture content changes. This step is crucial because it helps the system eliminate spurious correlations. For example, if a new batch of recycled aggregate has a high moisture content, the system can identify the increase in moisture content as a result of aggregate changes, rather than water addition, thus avoiding erroneous adjustments.

[0052] After gaining a deep understanding of the causal relationships between system variables, the core of this module, a reinforcement learning agent, begins making decisions. This invention preferably uses a Deep Q-Network (DQN) as the algorithm for implementing this agent. The essence of reinforcement learning is to allow the agent to learn optimal action strategies through trial and error interactions with the environment (or its precise digital twin).

[0053] Specifically, the decision-making process of a reinforcement learning agent follows the following framework: State Construction: The agent's decision-making is based on its observation of the current environment, known as its "state." This invention constructs a rich and comprehensive state representation for the agent. This state is a high-dimensional vector that is not simply the instantaneous value of the current moisture content but rather consists of three key pieces of information: The future moisture content prediction value provided by the "Moisture Content Prediction Module"; Internal component states provided by the "Tensor Analysis and State Identification Module"; The causal effects quantified by the causal inference portion of this module. This state definition gives the agent unprecedented "insight": it can "see" the future (predicted value), "see" through appearances (internal state), and "understand" the true impact of its actions (causal effects).

[0054] Setting an optimization goal: The agent's learning is guided by its optimization goal. In this example, the goal is to minimize the deviation between the predicted future moisture content and a preset target moisture content. The system defines a reward function based on this deviation. The closer the predicted value is to the target value, the higher the reward the agent receives.

[0055] Decision-making and learning: Based on the above high-dimensional state, a deep Q network (a deep neural network) is used to approximate an optimal action-value function This function is used to evaluate the current state Under this condition, take a certain water addition adjustment action The expected value of the long-term cumulative reward that can be brought. When making a decision, the agent inputs the current state into the network, and the network outputs the corresponding values ​​of all optional water-adding actions. value, the agent chooses the one that makes The action that maximizes the value is selected. This selected optimal action is ultimately parsed into a specific water adjustment instruction and output. Through mechanisms such as experience replay and target networks, DQN continuously optimizes its network parameters through continuous interaction with the environment, making its assessment of action value increasingly accurate, thereby learning the optimal control strategy.

[0056] The intelligent decision-making module deeply integrates causal reasoning based on understanding physical laws with reinforcement learning based on learning optimal behaviors, creating a decision-making core capable of autonomous thinking and evolution. It no longer relies on static rules or models, but instead dynamically and intelligently generates each control instruction based on comprehensive and profound state awareness, aiming to achieve long-term optimal control. Collaborative control execution module: used to adjust the water addition amount according to the water addition amount adjustment instruction to dynamically control the moisture content of the cold recycled mixture; In this embodiment, the collaborative control execution module is the final physical execution terminal of the entire dynamic control system. Its core responsibility is to accurately and accurately translate the abstract, high-level water adjustment instructions output by the upstream "intelligent decision-making module" into physical adjustments to the real-time water addition to the cold recycled mix. This module bridges the gap between intelligent decision-making and the physical world, ensuring that the entire closed-loop control system "implements its promise" and achieves precise control.

[0057] Specifically, to achieve the above functionality, the module's workflow adheres to a rigorous closed-loop feedback control logic. First, the module receives a water adjustment instruction. This instruction is a logical command output by the intelligent decision-making module based on its complex algorithmic model. For example, it can be a relative value indicating "increase flow rate by 5%" or an absolute value indicating "adjust to 15 liters / minute." The module's first task is to parse this instruction and convert it into a clear physical quantity usable by the engineering control system: the target flow setpoint.

[0058] After obtaining the target flow setpoint, the core of the module, a closed-loop control unit that preferably uses a proportional-integral-derivative (PID) controller, begins operation. The PID controller is a very classic and robust control algorithm in the field of industrial control. Its purpose is to make a controlled quantity (in this case, the actual water flow) as close to its setpoint as possible through continuous feedback and adjustment.

[0059] To this end, the coordinated control execution module collects the current actual flow value in real time and at high frequency through a water flow meter installed in the water supply pipeline. The controller continuously compares this real-time feedback actual flow value with the aforementioned target flow set value, and the deviation between the two is is used as the basis for the controller calculation. It is the basis for all actions of the controller.

[0060] The proportional-integral-derivative controller is based on this deviation , generates an actuator control signal through its control law Its classic control law can be expressed as follows: ; Where, For in time generated actuator control signals; For in time deviation, that is ; is the proportional gain coefficient, which is used to make rapid adjustments in proportion to the current deviation; The integral gain coefficient is used to accumulate historical deviations in order to eliminate static errors in the system and ensure that the actual flow rate can be accurately stabilized at the target value after long-term operation. is the differential gain coefficient, which is used to adjust according to the rate of change of the deviation, aiming to predict the future trend of the deviation, thereby suppressing the oscillation of the system and improving the stability of the response; is the integration variable.

[0061] Finally, the collaborative control execution module controls the actuator according to the signal , to drive the final physical actuator. In a preferred embodiment, the actuator can be an electronic control valve or a variable frequency water pump. If an electronic control valve is used, the control signal It will be converted into a standard control current or voltage to accurately adjust the valve opening; if a variable frequency water pump is used, the control signal will instruct the frequency converter to adjust the operating frequency of the water pump motor, thereby changing the speed and output flow of the water pump.

[0062] In this way, the coordinated control execution module drives the actuator to accurately adjust the real-time water addition amount entering the cold recycled mixture until the deviation between the actual flow value fed back by the water flow meter and the target flow set value is Approaching zero.

[0063] In summary, the collaborative control execution module ensures that every intention of the upper-level intelligent decision-making can be realized in the physical world with high fidelity, stability and speed through a complete and rigorous closed-loop control process of "instruction analysis-feedback comparison-PID calculation-drive execution", ultimately completing the dynamic control of moisture content.

[0064] Please see the attached Figure 2 The present invention also provides a multi-sensor fusion cold recycled mixture moisture content control method, comprising the following steps: S1. Collecting parameter information of at least two different modes of the cold recycled mixture in real time through multiple sensors, wherein the parameter information includes dielectric constant data, spectral reflectance data, surface temperature distribution data, volume profile and flow velocity data, and on-site ambient temperature and humidity; S2. Cleaning, normalizing, and interpolating missing values ​​of the parameter information collected in real time, and constructing a fourth-order time series tensor including the parameter information; S3. Performing tensor representation learning on the fourth-order time series tensor to extract potential features, identifying the internal component states of the cold recycled mixture based on the potential features, and decoupling interference caused by non-moisture content factors in the parameter information to obtain a pure moisture content-related signal; S4. Adaptively predicting a future moisture content value of the cold recycled mixture based on the pure moisture content related signal; S5. Based on the predicted future moisture content and the internal component states, dynamically determine the water addition adjustment instruction required for the cold recycled mixture through causal inference and reinforcement learning; S6. According to the water addition amount adjustment instruction, adjust the water addition amount to dynamically control the moisture content of the cold recycled mixture.

[0065] The method of this embodiment can be used to execute the above system embodiment. Its principles and technical effects are similar and will not be described in detail here.

[0066] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. Multi-sensor fusion cold recycled mixture moisture content control system, characterized by: include: Multimodal sensing module: used to collect parameter information of at least two different modes of cold recycled mixture in real time through multiple sensors, the parameter information including dielectric constant data, spectral reflectance data, surface temperature distribution data, volume profile and flow rate data, as well as on-site ambient temperature and humidity; Data preprocessing and tensor construction module: used to clean, normalize and interpolate missing values ​​of the parameter information collected in real time, and construct a fourth-order time series tensor including the parameter information; A tensor analysis and state recognition module is configured to perform tensor representation learning on the fourth-order time series tensor to extract potential features, identify the internal component states of the cold recycled mixture based on the potential features, and decouple interference caused by non-moisture content factors in the parameter information to obtain a pure moisture content-related signal; Moisture content prediction module: used for adaptively predicting the future moisture content of the cold recycled mixture based on the pure moisture content related signal; Intelligent decision-making module: for dynamically deciding the water addition adjustment instruction required for the cold recycled mixture based on the future moisture content prediction value and the internal component state through causal inference and reinforcement learning; Collaborative control execution module: used to adjust the water addition amount according to the water addition amount adjustment instruction to dynamically control the moisture content of the cold recycled mixture.

2. The multi-sensor fusion cold recycled mixture moisture content control system according to claim 1 is characterized in that: The multimodal perception module collects the parameter information in real time through the following sensors: A broadband microwave sensor array is deployed at the cold recycled mixture conveyor belt or the mixer outlet to collect dielectric constant data of the mixture; a multispectral near-infrared sensor deployed above the surface of the cold recycled mixture, for collecting spectral reflectance data of the mixture; A high-precision thermal imaging sensor for monitoring the surface temperature distribution of the cold recycled mixture; a high-resolution laser displacement sensor for measuring the volume profile and flow rate of the cold recycled mixture; Ambient temperature and humidity sensors are used to monitor the on-site ambient temperature and humidity of cold recycled mixture construction.

3. The multi-sensor fusion cold recycled mixture moisture content control system according to claim 1 is characterized in that: The data preprocessing and tensor construction module includes: Clean the parameter information collected in real time, including outlier removal and noise filtering; Performing Min-Max normalization processing on the parameter information; The parameter information is interpolated in time series to fill in missing data points and constructed as a fourth-order time series tensor ; Where, is the constructed fourth-order time series tensor; represents the field of real numbers; Indicates the number of sensor modalities; Indicates the number of sampling points in the sensor space; Indicates the length of the time series; Indicates the feature dimension corresponding to each sampling point.

4. The multi-sensor fusion cold recycled mixture moisture content control system according to claim 1 is characterized in that: The tensor analysis and state recognition module includes: Performing tensor representation learning on the fourth-order time series tensor to obtain a core tensor and a factor matrix by non-negative tensor factorization, wherein the core tensor and the factor matrix represent potential characteristics of free water, bound water, aggregate, asphalt content, and temperature components in the cold recycled mixture; Based on the core tensor and the factor matrix, identifying the internal component state of the cold recycled mixture through a tensor neural network, the internal component state including the real-time demulsification degree, free water content, and bound water content of the emulsified asphalt; According to the internal component state, tensor independent component analysis or orthogonal projection method is used to decouple interference caused by non-water content factors from the fourth-order time series tensor to obtain the pure water content related signal.

5. The multi-sensor fusion cold recycled mixture moisture content control system according to claim 4 is characterized in that: The step of obtaining a core tensor and a factor matrix by non-negative tensor factor decomposition comprises: For the fourth-order time series tensor Perform non-negative Tucker decomposition, which is expressed as: ； Where, is the core tensor; 、 、 and are the factor matrices corresponding to sensor modality, spatial sampling points, time series, and feature dimensions respectively; Represents a tensor along the Modulo product, modal index , corresponding to sensor modality, spatial sampling points, time series and feature dimensions respectively.

6. The multi-sensor fusion cold recycled mixture moisture content control system according to claim 1 is characterized in that: The moisture content prediction module includes: constructing the pure moisture content related signal as an input time series; Processing the input time series through a gated recurrent unit network with an integrated attention mechanism to learn temporal dependencies and generate weighted hidden state representations; The weighted hidden state representation is input into a fully connected output layer to generate and output a predicted value of the future moisture content of the cold recycled mixture at one or more future time points.

7. The multi-sensor fusion cold recycled mixture moisture content control system according to claim 6 is characterized in that: The weighted hidden state representation Calculated by the following formula: ； Where, The gated recurrent unit network at time step The hidden state of the output; The hidden state calculated by the attention mechanism The weight of is the length of the input time series.

8. The multi-sensor fusion cold recycled mixture moisture content control system according to claim 1 is characterized in that: The intelligent decision-making module includes: Using a causal inference method, constructing and updating a structural causal model, wherein the structural causal model is used to quantify the causal effects of the internal component state, historical water addition, and environmental parameters on the moisture content; The reinforcement learning agent is based on a state composed of the future moisture content prediction value, the internal component state and the causal effect, and takes minimizing the deviation between the future moisture content prediction value and the preset target moisture content as the optimization goal. Through a deep Q network, the reinforcement learning agent dynamically decides and outputs the water addition adjustment instruction.

9. The multi-sensor fusion cold recycled mixture moisture content control system according to claim 1 is characterized in that: The collaborative control execution module includes: receiving the water addition amount adjustment instruction and parsing the water addition amount adjustment instruction into a target flow rate setting value; generating an actuator control signal based on a deviation between the target flow rate set value and the actual flow rate value fed back in real time by the water flow meter through a proportional-integral-differential controller; According to the actuator control signal, the electronic control valve or the variable frequency water pump is driven to accurately adjust the real-time water addition amount entering the cold recycled mixture, so as to dynamically control the moisture content.

10. A multi-sensor fusion cold recycled mixture moisture content control method, applied to a multi-sensor fusion cold recycled mixture moisture content control system according to any one of claims 1 to 9, characterized in that: The following steps are involved: Collecting parameter information of at least two different modes of the cold recycled mixture in real time through multiple sensors, the parameter information including dielectric constant data, spectral reflectance data, surface temperature distribution data, volume profile and flow rate data, and on-site ambient temperature and humidity; Cleaning, normalizing, and interpolating missing values ​​of the parameter information collected in real time, and constructing a fourth-order time series tensor including the parameter information; performing tensor representation learning on the fourth-order time series tensor to extract potential features, identifying internal component states of the cold recycled mixture based on the potential features, and decoupling interference caused by non-moisture content factors in the parameter information to obtain a pure moisture content-related signal; Adaptively predicting a future moisture content value of the cold recycled mixture based on the pure moisture content related signal; Based on the future moisture content prediction value and the internal component state, dynamically determining the water addition adjustment instruction required for the cold recycled mixture through causal inference and reinforcement learning; According to the water addition amount adjustment instruction, the water addition amount is adjusted to dynamically control the moisture content of the cold recycled mixture.

Images