AI-based engineering vehicle multi-zone environmental control system

Abstract

The application discloses an AI-based engineering vehicle multi-region environment control system, and particularly relates to the technical field of engineering vehicle environment control. A humidity, electric field, air pressure and dust multi-modal sensing array is arranged in the engineering vehicle multi-region, the timing characteristics of dust particle deposition distribution are extracted, the deposition acceleration coefficient and humidity response coefficient are calculated, the dust deposition risk coefficient is calculated, and the preset value is compared. The region exceeding the threshold value is marked as the region of dust deposition abnormal state. Based on the time series prediction model, the deposition trend is analyzed, the environmental parameters are input, and the dust deposition speed curve of each region of the engineering vehicle in the future set time is predicted. The prospective evaluation of the electrostatic, humidity, air pressure and dust coupling deposition risk is realized, the collaborative intervention strategy is triggered through the dynamic early warning threshold function, the effect of risk rapid suppression is achieved, and the problem of traditional single-parameter monitoring lag and inability to capture the coupling mechanism, which causes the sudden drop of ventilation efficiency, is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of environmental control technology for engineering vehicles, and more specifically, to an AI-based multi-area environmental control system for engineering vehicles. Background Technology

[0002] In dusty environments, such as open-pit mines and arid construction sites, engineering vehicles operate continuously in environments with high dust concentrations, low relative humidity, and large fluctuations in electrostatic potential. Dust particles are easily deposited on the inner walls of air ducts, heat exchange fins, and the surfaces of temperature and humidity / CO2 sensors due to the combined effects of electrostatic adsorption and humidity gradients. Conventional positive pressure air supply and single-stage filters can only trap large-diameter dust particles, while fine dust particles are repeatedly enriched with the circulating airflow, ultimately leading to airflow obstruction, reduced heat exchange efficiency, and decreased cabin air quality.

[0003] The existing system lacks synchronous monitoring and coupled analysis of humidity, potential and dust concentration, and cannot identify abnormal dust deposition triggered by the coupling effect, nor has it established a targeted intervention mechanism. When deposition intensifies, relying solely on passive dust removal or filter replacement is insufficient to restore ventilation efficiency in a timely manner. Although this can partially alleviate the problem, it cannot accurately match the deposition risk under complex working conditions, resulting in wasted resources and poor intervention effects. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, this invention provides an AI-based multi-area environmental control system for engineering vehicles. By predicting deposition risks and implementing coordinated interventions of active air pressure, humidity, and static electricity elimination, it suppresses deposition and sensor contamination caused by the superposition of static electricity, humidity, and dust, thereby solving the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an AI-based multi-area environmental control system for engineering vehicles, comprising:

[0006] The environmental data acquisition module collects environmental parameters, including humidity values, electric field strength, air pressure data, and dust concentration data, by setting humidity sensors, electrostatic field strength meters, air pressure sensors, and dust sensors in each area of ​​the engineering vehicle, and records the dust particle deposition distribution map in each area of ​​the engineering vehicle.

[0007] The dust deposition anomaly identification module extracts the temporal characteristics of the dust particle deposition distribution map, calculates the deposition acceleration coefficient and humidity response coefficient, and calculates the dust deposition risk coefficient based on the deposition acceleration coefficient and humidity response coefficient. It compares the temporal characteristics of the dust particle deposition distribution with preset values ​​and marks areas that exceed the threshold as areas with abnormal dust deposition status.

[0008] The dust deposition risk prediction module analyzes the deposition trend based on a time series prediction model, inputs environmental parameters, and predicts the dust deposition speed curve of each region of the engineering vehicle within a future set time when the region enters a dust deposition abnormal state; the input features of the time series prediction model include: humidity change rate, potential fluctuation trend, deposition speed change rate, deposition amount trend, cross-region airflow feature, and time lag feature;

[0009] The dust deposition intervention module dynamically calculates the collaborative intervention strategy of air pressure adjustment, humidity adjustment, and electrostatic elimination based on the predicted dust deposition speed curve, effectively reducing the speed of dust deposition; wherein the air pressure adjustment preferentially reduces the pressure difference in the cross-region airflow feature to suppress dust migration, the humidity adjustment adaptively controls the dew point difference based on the dewing risk probability to reduce sensor drift, and the electrostatic elimination adjusts the discharge frequency according to the potential fluctuation trend to weaken particle adsorption; the simultaneous monitoring and adjustment module monitors the dewing state of the humidity sensor surface in real time, compensates and corrects the humidity data affected by dewing, and dynamically adjusts the intervention strategy.

[0010] Preferably, the system includes a region division module that divides the engineering vehicle into regions based on the airflow path, equipment layout, and dust deposition characteristics of the engineering vehicle; the dust deposition characteristics are obtained by analyzing the air pipe path, airflow direction, and flow distribution of the airflow path, and the engineering vehicle cabin is divided into zones using K-means clustering.

[0011] Preferably, the deposition acceleration coefficient is obtained by calculating the change rate of dust concentration and air pressure; dust concentration data and air pressure data in each region are collected, the change rate of dust concentration, i.e., the change amount of dust concentration per unit time, is calculated, the change rate of air pressure, i.e., the change amount of air pressure per unit time, is calculated, and the deposition acceleration coefficient is obtained by multiplying the change rate of air pressure and the change rate of dust concentration, thereby representing the acceleration trend of dust deposition; the deposition acceleration coefficient reflects the influence of environmental changes on dust deposition, especially the promoting or slowing effect of air pressure changes on dust deposition speed;

[0012] The humidity response coefficient is obtained by calculating the change rate of humidity and the average humidity level of airflow; the change in humidity in the region is monitored in real time, the rate of change of humidity is calculated, and the average humidity of the region is calculated; after dimensionless processing, the humidity change rate and the average humidity of the region are weighted to obtain the humidity response coefficient, which reflects the influence of humidity change on dust deposition; the greater the humidity change, the more obvious the influence on the deposition process, so this coefficient can quantify the influence of humidity on deposition speed;

[0013] The dust deposition risk coefficient is obtained by multiplying the deposition acceleration coefficient and the humidity response coefficient; the deposition acceleration coefficient reflects the acceleration of dust deposition, and the humidity response coefficient reflects the influence of humidity change on deposition; the dust deposition risk coefficient comprehensively considers the acceleration trend of dust deposition and the influence of humidity on the deposition process, thereby giving a comprehensive risk assessment result; when the dust deposition risk coefficient exceeds the set threshold value, the system will trigger the corresponding intervention measures to reduce the risk of dust deposition and adjust the environment.

[0014] Preferably, the system comprises a warning module connected with the dust deposition risk prediction module, which compares the predicted dust deposition rate curve with the warning threshold value, and when the warning threshold value is exceeded, the system issues a warning and executes the dust deposition intervention module; the response window is defined based on the response time of the intervention strategy and the confidence interval of the historical deposition rate data, the dust deposition amount in the response window is calculated by the time series prediction model; the static basic warning threshold value is associated with the change rate of the predicted deposition rate to generate a dynamic warning threshold function; the warning threshold value is tightened with the increase of the dust deposition rate in the response window; the warning threshold value is scaled and corrected according to the real-time humidity difference and the electric field intensity deviation, to suppress the false high warning threshold value in high humidity or strong electric field environment; when the cumulative deposition amount reaches a preset proportion or the running period exceeds the limit, the dynamic warning threshold function is recalibrated to eliminate the influence of sensor drift and environmental aging.

[0015] Preferably, the time series prediction model is obtained by the following steps:

[0016] Step S1: data acquisition and preprocessing: obtaining the real-time monitoring time series data of humidity value, electric field intensity, dust concentration, current dust deposition rate and dust deposition amount of each area of the engineering vehicle; performing synchronous preprocessing by Z-score normalization processing and sliding window denoising to form a standardized data sequence to be analyzed;

[0017] Step S2: feature extraction: extracting the features of the data sequence to be analyzed, and outputting a feature vector, including: humidity change rate; potential fluctuation trend; deposition rate change rate, deposition amount trend, cross-region airflow characteristics, time lag characteristics;

[0018] Step S3: Model construction: using the feature vector as input, a hybrid network of graph convolution and bidirectional LSTM is adopted, where the graph convolution layer extracts spatial correlation based on the pre-computed airflow adjacency matrix, which is generated by offline CFD simulation covering airflow distribution under typical working conditions; in real-time operation, the adjacency matrix weight is dynamically adjusted through online interpolation method (such as weighted interpolation based on air pressure sensor and wind speed sensor data) to reflect airflow changes, and bidirectional LSTM captures time series dependence, and adjusts the adjacency matrix weight adaptively at each cycle to reflect turbulence evolution, predicts the dust deposition velocity curve of each area of the engineering vehicle within a set time in the future; taking the dust deposition velocity curve prediction error as the loss function, the model parameters are optimized through cross-validation; the output is the dust deposition velocity curve of each area, representing the change trend of deposition velocity per unit time over time;

[0019] Step S4: Model verification and calibration: based on historical operation data and real-time measurement data, the accuracy of the predicted dust deposition velocity curve is verified; when the deviation between the predicted value and the actual measured value exceeds the preset value, the adaptive calibration program is triggered.

[0020] Preferably, the method for obtaining the time lag feature comprises the following steps:

[0021] The nonlinear correlation between humidity, potential, deposition velocity and dust deposition is calculated, and the time lag relationship between each parameter and dust deposition is modeled through mutual information measure and nonlinear cross-correlation analysis to generate a feature vector describing the nonlinear time lag between parameters;

[0022] A neural network regression model is used, which includes an input layer, one or more hidden layers and an output layer; the input layer receives humidity, potential, air pressure difference and deposition velocity parameters, the hidden layer uses a fully connected layer, and the output layer predicts the time lag relationship of humidity, potential, air pressure difference parameters on dust deposition; trained through back propagation algorithm, the loss function is mean square error, and gradient descent method is used for optimization;

[0023] According to the real-time monitoring data, the model parameters of the time lag feature are adjusted through an adaptive algorithm, which compares the actual monitoring results and the model prediction results at each feedback to adjust the network weight to ensure the accuracy of the time lag feature;

[0024] Based on the airflow adjacency matrix generated by CFD simulation, the airflow spatial topological relationship in the adjacency matrix is used as input, and the time lag feature calculation is optimized in combination with the fluid dynamics model; by optimizing the input features, the modeling accuracy of nonlinear time lag is improved;

[0025] The calculated time delay characteristics are fed back to the time series prediction model. When the predicted deposition velocity deviates from the actual measured data by more than a preset error, an adaptive calibration program is triggered to adjust the neural network regression model through a retraining process.

[0026] Preferably, by deploying micro air flow velocity sensors at the corners of the air duct, narrow spaces or air flow intersections, the velocity gradient and turbulence characteristics of the local micro air flow are monitored in real time. After the fusion of the micro air flow disturbance data, the local micro air flow disturbance intensity, including the velocity gradient and turbulence characteristics, is generated. When the monitored value exceeds the threshold value, the adjustable air valve or auxiliary small fan at the corresponding position is controlled to adjust the air flow direction or flow rate to weaken the turbulence or enhance the scouring effect, thereby reducing dust deposition.

[0027] Preferably, to solve the problem of data drift of the humidity sensor caused by local condensation, which leads to misjudgment of the dust deposition abnormal state, the dew point temperature is calculated in real time. When the difference between the ambient temperature and the dew point exceeds a preset value, a dew point compensation correction program is triggered to compensate and correct the humidity measurement value affected by condensation in real time, thereby eliminating the interference of humidity sensor drift on the prediction result and improving the accuracy of dust deposition abnormal state judgment.

[0028] Explanations, the system collects environmental parameters such as humidity, electric field intensity, air pressure and dust concentration in real time through a multi-modal sensor array, combines a graph convolution-bidirectional LSTM hybrid network, accurately predicts the dust deposition velocity curve, captures the complex coupling effect of electrostatic, humidity, air pressure and dust, and avoids the misjudgment of deposition risk caused by the lag of traditional single-parameter monitoring; the graph convolution layer is based on the air flow adjacency matrix generated by offline CFD simulation to extract the spatial correlation between regions, and dynamically adjusts the weight through air pressure and wind speed sensor data interpolation online to reflect real-time turbulence changes; the bidirectional LSTM excavates the long-term dependence in the time series to ensure the adaptability of the prediction model to dynamic working conditions; the dynamic intervention strategy uses a multi-objective optimization algorithm to consider the interaction constraints of air pressure adjustment, humidity control and electrostatic elimination, suppresses dust migration by preferentially reducing the air pressure difference between regions, adaptively adjusts the dew point difference to reduce the drift of the humidity sensor caused by condensation, and optimizes the electrostatic elimination frequency according to the potential fluctuation trend to weaken the particle adsorption effect, thereby effectively reducing the dust deposition on the inner wall of the air duct and the heat exchange fins; the nonlinear coupling calibration mechanism further integrates features such as charge heterogeneity, local micro air flow disturbance intensity, condensation risk and oil mist complex ratio to dynamically correct the deposition velocity curve in the form of exponential decay integral, match the inertial changes of the engineering vehicle running conditions, and ensure the robustness of the prediction accuracy and intervention effect; logically form a complete chain from multi-source perception to accurate prediction to coordinated intervention, significantly improve the ventilation efficiency, sensor reliability and system maintenance economy, and reasonably solve the deposition problem in complex dust environment.

[0029] Preferably, the multi-objective optimization algorithm includes:

[0030] The air pressure adjustment target adjusts the air pressure in real time to optimize the air flow and prevent dust deposition acceleration by minimizing the deposition acceleration coefficient;

[0031] The humidity adjustment target adjusts the humidity range in real time to prevent abnormal dust deposition caused by excessive humidity fluctuations by calculating the humidity response coefficient;

[0032] The static electricity elimination target optimizes the static electricity elimination strategy based on the comprehensive model of electric field intensity and humidity changes to reduce the impact of static electricity accumulation on dust deposition;

[0033] The air pressure adjustment target, humidity adjustment target, and static electricity elimination target are integrated through a weighted target function, wherein the weights of the targets are dynamically adjusted according to real-time environmental data and prediction results, thereby achieving the coordinated optimization of air pressure, humidity, and static electricity elimination.

[0034] Technical effects and advantages of the present application:

[0035] (1) The present application provides a scheme that generates a dust deposition speed curve in real time by deploying humidity, electric field, air pressure, and dust multi-modal sensing array in multiple regions of the engineering vehicle, and combining a time series prediction model, to realize forward-looking assessment of electro-humidity-dust coupling deposition risk, and trigger a coordinated intervention strategy through a dynamic early warning threshold function, to achieve the effect of rapid risk containment, effectively solving the problem of sudden reduction of ventilation efficiency caused by traditional single-parameter monitoring lag and inability to capture coupling mechanisms.

[0036] (2) The present application provides a scheme that calculates the coordinated intervention curve of air pressure, humidity, and static electricity elimination through multi-objective optimization, to achieve adaptive reduction of pressure difference scouring and dew deviation without affecting the comfort of the vehicle cabin, and continuously updates the time series prediction model with drift detection and nonlinear coupling calibration mechanism, effectively solving the problems of short filter core life, large sensor drift, and high maintenance cost under high dust working conditions, while significantly reducing energy consumption and improving system reliability and maintenance efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 The figure is a structural diagram of the AI-based multi-region environmental control system of the engineering vehicle. DETAILED DESCRIPTION

[0038] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0039] It should be understood, of course, that the detailed description and specific examples, while indicating preferred embodiments, are given by way of illustration only and are not by way of limitation.

[0040] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the application or its application and uses.

[0041] Techniques, methods, and apparatus known to those of ordinary skill in the relevant art can not be discussed in detail herein, but should be considered within the scope of the disclosure where appropriate.

[0042] Embodiment 1, see Figure 1 The AI-based engineering vehicle multi-zone environment control system structure diagram provides an AI-based engineering vehicle multi-zone environment control system as shown in Figure 1 The AI-based engineering vehicle multi-zone environment control system comprises:

[0043] An environment data acquisition module acquires environment parameters including humidity values, electric field intensity, air pressure data, and dust concentration data in each region of the engineering vehicle by respectively setting humidity sensors, electrostatic field intensity meters, air pressure sensors, and dust sensors in each region of the engineering vehicle; and records dust particle deposition distribution maps of each region of the engineering vehicle using high-speed photography or electrostatic imaging methods;

[0044] A dust deposition anomaly identification module extracts time sequence characteristics of the dust particle deposition distribution maps, calculates a deposition acceleration coefficient and a humidity response coefficient, and calculates a dust deposition risk coefficient based on the deposition acceleration coefficient and the humidity response coefficient; compares the dust deposition risk coefficient with a preset value, and marks a region that exceeds a threshold value as a region in a dust deposition abnormal state;

[0045] A dust deposition risk prediction module analyzes a deposition trend based on a time sequence prediction model for a region that enters the dust deposition abnormal state, inputs environment parameters, and predicts a dust deposition speed curve of each region of the engineering vehicle within a future set time; input features of the time sequence prediction model include a humidity change rate, a potential fluctuation trend, a deposition speed change rate, a deposition amount trend, a cross-region air flow feature, and a time lag feature;

[0046] A dust deposition intervention module dynamically calculates a collaborative intervention strategy of air pressure adjustment, humidity adjustment, and electrostatic elimination based on the predicted dust deposition speed curve using a multi-objective optimization algorithm to effectively reduce the speed of dust deposition; wherein the air pressure adjustment preferentially reduces a pressure difference in the cross-region air flow feature to suppress dust migration, the humidity adjustment adaptively controls a dew point difference based on a dewing risk probability to reduce sensor drift, and the electrostatic elimination adjusts a discharge frequency based on the potential fluctuation trend to weaken particle adsorption; a monitoring and adjusting module simultaneously monitors a dewing state of a humidity sensor surface, compensates and corrects humidity data affected by dewing, and dynamically adjusts the intervention strategy.

[0047] In one possible embodiment, the multi-objective optimization algorithm includes:

[0048] An air pressure adjustment target, which adjusts the air pressure in real time to optimize air flow and prevent dust deposition acceleration by minimizing the deposition acceleration coefficient;

[0049] A humidity adjustment target, which adjusts the humidity range in real time to prevent abnormal dust deposition caused by excessive humidity fluctuations by calculating the humidity response coefficient;

[0050] An electrostatic elimination target, which optimizes the electrostatic elimination strategy based on a comprehensive model of electric field intensity and humidity changes to reduce the impact of electrostatic accumulation on dust deposition;

[0051] The air pressure adjustment target, humidity adjustment target, and electrostatic elimination target are integrated through a weighted objective function, wherein the weights of the targets are dynamically adjusted according to real-time environmental data and prediction results, thereby achieving the coordinated optimization of air pressure, humidity, and electrostatic elimination.

[0052] In this embodiment, the multi-objective optimization algorithm considers the interaction among air pressure, humidity, and electrostatic elimination when adjusting them; air pressure changes affect air flow speed and dust distribution, humidity changes affect electrostatic elimination efficiency, and electrostatic adjustment also changes environmental humidity; the multi-objective optimization algorithm takes humidity, electric field, and air pressure sensor data as input and simultaneously calculates the optimal values of each parameter; when humidity is too high, it reduces humidity and enhances electrostatic elimination; when the electrostatic field is too high, it enhances electrostatic elimination and adjusts air pressure to promote air flow; through the above parameter correlation settings, this embodiment achieves the coordinated control of the three intervention measures.

[0053] Explanation and explanation, with the output intervention strategy as the control input, the humidity regulator (such as a vehicle-mounted dehumidification device or an ultrasonic humidifier) is dynamically adjusted to accurately control the humidity in the area, and the ion wind generator or ultrasonic atomizing nozzle installed in the air pipe and key positions is activated to actively eliminate static charges in the air and on the surface; at the same time, the electrostatic neutralization rod is started to eliminate the surface charge of the cabin wall and the sensor, avoiding the effect of dust electrostatic adsorption; during the execution process, the humidity, potential, and dust sensor feedback data are continuously monitored and compared with the predicted target value in real time, forming a complete closed-loop control.

[0054] To realize the linkage of the prediction model and feedback control, this embodiment continuously compares the measured deposition speed with the predicted target in real time, and the difference between the two is used as a feedback signal to drive the dynamic adjustment of the execution strategy; when the measured deposition speed is higher than the predicted value, the ion wind intensity can be increased or the electrostatic elimination time can be extended; if it is lower than the predicted value, the control intensity can be correspondingly reduced; the fine-tuning process can use proportional regulation or incremental correction algorithm to dynamically update the parameters according to the deviation size; through this closed-loop feedback, this embodiment can adaptively approach the actual state to the predicted target.

[0055] In a possible embodiment, the system comprises a region division module that divides the engineering vehicle into regions based on airflow passages, equipment layout, and dust deposition characteristics in the engineering vehicle; the dust deposition characteristics are obtained by analyzing the air duct path, airflow direction, and flow distribution of the airflow passages, and the K-means clustering is used to divide the cabin of the engineering vehicle;

[0056] In a possible embodiment, the region is divided based on the physical structure (such as the equipment layout and the ventilation system partition), and the clustering algorithm is used to optimize the boundary to ensure that the division result is consistent with the actual structure and has the data-driven characteristics; the initial region is defined according to the compartment and the ventilation duct partition, and the clustering algorithm is used to refine the fuzzy boundary (such as the equipment gap).

[0057] In a possible embodiment, the deposition acceleration coefficient is obtained by calculating the change rate of dust concentration and air pressure; the dust concentration data and air pressure data in each region are collected, the change rate of dust concentration, that is, the change amount of dust concentration per unit time, is calculated, the change rate of air pressure, that is, the change amount of air pressure per unit time, is calculated, and the deposition acceleration coefficient is obtained by multiplying the change rate of air pressure and the change rate of dust concentration, thereby representing the acceleration trend of dust deposition; the deposition acceleration coefficient reflects the influence of environmental changes on dust deposition, especially the promoting or slowing effect of air pressure changes on the dust deposition speed;

[0058] The humidity response coefficient is obtained by calculating the change rate of humidity and the average humidity level of airflow; the change of humidity in the region is monitored in real time, the change rate of humidity is calculated, and the average value of the humidity of the region is calculated; after dimensionless processing, the humidity change rate and the average value of the humidity of the region are weighted to obtain the humidity response coefficient, which reflects the influence of humidity change on dust deposition; the greater the humidity change, the more obvious the influence on the deposition process, so the coefficient can quantify the influence of humidity on the deposition speed;

[0059] The dust deposition risk coefficient is obtained by multiplying the deposition acceleration coefficient and the humidity response coefficient; the deposition acceleration coefficient reflects the acceleration of dust deposition, and the humidity response coefficient reflects the influence of humidity change on deposition; the dust deposition risk coefficient considers the acceleration trend of dust deposition and the influence of humidity on the deposition process, thereby giving a comprehensive risk assessment result; when the dust deposition risk coefficient exceeds a set threshold, the system triggers corresponding intervention measures to reduce the risk of dust deposition and adjust the environment.

[0060] In a possible embodiment, the system comprises a warning module connected to the dust deposition risk prediction module, compares the predicted dust deposition rate curve with the warning threshold, and when the warning threshold is exceeded, sends a warning to the outside and executes the dust deposition intervention module; define the response window based on the response time of the intervention strategy and the confidence interval of the historical deposition rate data, calculate the dust deposition amount in the response window through the time series prediction model; generate a dynamic warning threshold function by associating the static basic warning threshold with the change rate of the predicted deposition rate; make the warning threshold tighten with the increase of the dust deposition rate in the response window; scale and correct the warning threshold according to the real-time humidity difference and electric field intensity deviation, suppress the false high of the warning threshold in high humidity or strong electric field environment; when the cumulative deposition amount reaches a preset proportion or the running period is over, recalibrate the dynamic warning threshold function to eliminate the influence of sensor drift and environmental aging.

[0061] In a possible embodiment, the cumulative deposition amount is calculated by integrating the deposition rate with respect to time; and an abnormal state is judged based on the cumulative deposition amount and a corresponding threshold.

[0062] In a possible embodiment, the dynamic warning threshold function represents a dust deposition rate warning threshold that changes with time, and the formula of the dynamic warning threshold function T dynamic (t) is:

[0063]

[0064] Wherein, T base is a static basic warning threshold, determined based on the upper limit of the 95% confidence interval of the historical deposition rate data; is a predicted deposition rate change rate, calculated through a time series prediction model; ΔH(t) is a real-time humidity difference; △E(t) is an electric field intensity deviation; k1 is a change rate sensitive coefficient, reflecting the amplification effect of deposition rate change on the threshold; k2 and k3 are linear scaling coefficients of the humidity difference and the electric field intensity deviation respectively, adjusting the inhibitory effect of environmental conditions on the value.

[0065] In the embodiments of the present application, the acquisition method of the time series prediction model includes the following steps:

[0066] Step S1: data acquisition and preprocessing: obtain the real-time monitoring time series data of humidity value, electric field intensity, dust concentration, current dust deposition speed and dust deposition amount of each area of the engineering vehicle, and output standardized data for analysis after preprocessing;

[0067] The specific implementation is: collecting heterogeneous data on experimental platforms and engineering vehicle sites, specifically including humidity sensor data at different positions, surface electrostatic field strength instrument data, air pressure sensor data, dust concentration sensor data, and recording local particle deposition position distribution; performing denoising and interpolation processing on the heterogeneous data through a sliding window method (synchronously preprocessed through Z-score normalization processing and sliding window denoising);

[0068] Step S2: feature extraction: extracting features of the data sequence to be analyzed, and outputting a feature vector, including: humidity change rate (difference value of relative humidity per unit time); potential fluctuation trend (variance of time series based on electrostatic potential); deposition velocity change rate, deposition amount trend, cross-region air flow characteristics (air flow exchange rate calculated through pressure difference sensor and wind speed sensor), time lag feature (correlation coefficients of humidity, potential, and deposition velocity with respect to dust deposition are calculated through cross-correlation analysis);

[0069] Step S3: model construction: taking the feature vector as input, and using a graph convolution-bidirectional LSTM hybrid network, wherein the graph convolution layer extracts spatial correlation based on a precomputed air flow adjacency matrix, the precomputed air flow adjacency matrix is generated through offline CFD simulation, and covers air flow distribution under typical working conditions; in real-time operation, the adjacency matrix weight is dynamically adjusted through an online interpolation method (such as weighted interpolation based on air pressure sensor and wind speed sensor data) to reflect air flow changes, and the bidirectional LSTM captures time series dependence and adjusts the adjacency matrix weight adaptively at each cycle to reflect turbulent evolution, and predicts the dust deposition velocity curve of each region of the engineering vehicle in a future set time (for example, 1 hour); taking the dust deposition velocity curve prediction error as a loss function, the model parameters are optimized through cross-validation (the number of hidden layer units is 64, the learning rate is 0.001, and the number of training rounds is 100); the output is the dust deposition velocity curve of each region, representing the change trend of deposition velocity per unit time over time;

[0070] Step S4: model verification and calibration: based on historical operation data and real-time measurement data (verified through sediment sensors or optical dust sensors), the accuracy of the predicted dust deposition velocity curve is verified; when the deviation between the predicted value and the actual measured value exceeds a preset value, such as 5%, an adaptive calibration program is triggered.

[0071] In this embodiment, the spatial correlation features are constructed by the airflow adjacency matrix (using CFD simulation or sensor data) and extracted by the graph convolution layer; the time-dependent features are obtained by cross-correlation analysis to get the influence weights of humidity, potential and deposition velocity and form the input vector; the spatial features output by the graph convolution are input into the LSTM together with the time series to capture the time dependence; the graph convolution layer and the LSTM layer interact through feature transmission: the spatial features guide the time series prediction, and the time series output can be used to adjust the graph convolution weights, thereby completing the joint modeling of spatial and temporal features.

[0072] 10、In the embodiments of the present application, the method for obtaining the time delay features includes the following steps:

[0073] Calculate the nonlinear correlation between humidity, potential, deposition velocity and dust deposition, model the time delay relationship between each parameter and dust deposition through mutual information measure and nonlinear cross-correlation analysis, and generate a feature vector to describe the nonlinear time delay between parameters;

[0074] In the implementation process, a suitable mutual information measure method is selected, for example, a mutual information measure based on Shannon entropy, which can effectively evaluate the nonlinear relationship between humidity, potential, deposition velocity and dust deposition; a nonlinear regression model (such as support vector regression) is used to model the time delay relationship between each parameter and dust deposition, an adaptive window length and step are used to capture the nonlinear time delay effect at different time scales, the sensitivity of different time windows and parameters in the experiment is analyzed, the parameters in the correlation calculation are adjusted, and the matching with the actual dust deposition scene is optimized, so that the nonlinear coupling characteristics are more accurately described.

[0075] A neural network regression model is used, which includes an input layer, one or more hidden layers and an output layer; the input layer receives humidity, potential, pressure difference and deposition velocity parameters, the hidden layer uses a fully connected layer, and the output layer predicts the time delay relationship of humidity, potential, pressure difference parameters on dust deposition; training is performed through a back propagation algorithm, the loss function is mean square error, and a gradient descent method is used for optimization;

[0076] In the design of the neural network architecture, each input feature is standardized according to its dimension and unit; the hidden layer is selected to be 2 to 3 layers, the number of neurons in each layer is adjusted according to the size of the actual data set, usually 128 to 512 neurons, the depth of the network and the number of neurons in each layer are determined through cross-validation to prevent overfitting; in the training process, the network weights are optimized through back propagation and mean square error loss function; the training data set and the validation set are divided by time series to ensure the representativeness and diversity of the data, and to avoid the dependence of the model on a certain environmental condition.

[0077] According to real-time monitoring data, the model parameters of the time lag feature are adjusted through an adaptive algorithm. The adaptive algorithm compares the actual monitoring results with the model prediction results at each feedback and adjusts the network weights to ensure the accuracy of the time lag feature.

[0078] Explanation: The adaptive algorithm uses an incremental learning strategy based on real-time feedback. By calculating the error between the model prediction and the actual monitoring results, the network weights are dynamically adjusted. The specific method includes inputting the prediction error at each model update, adjusting the learning rate, and flexibly adjusting the model parameters according to the error amplitude. Through the feedback mechanism, the model can adapt to different environmental changes, update the time lag feature in real time, and improve the prediction accuracy.

[0079] Based on the airflow adjacency matrix generated by CFD simulation, the spatial topological relationship in the adjacency matrix is used as input, and the fluid dynamics model is used to optimize the time lag feature calculation; by optimizing the input features, the modeling accuracy of nonlinear time lag is improved;

[0080] Explanation: By fusing the detailed airflow features (including velocity distribution, vortex, etc.) provided by CFD simulation results with the time lag features; CFD simulation results and real-time sensor data (such as airflow velocity, pressure, etc.) are combined through a weighted fusion algorithm to optimize the time lag feature calculation; Kalman filtering and other methods are used to reduce the influence of simulation errors, so as to ensure that the airflow model matches the real situation of dust deposition and improve the prediction accuracy.

[0081] The calculated time lag features are fed back to the time series prediction model. When the predicted deposition velocity and the actual measured data exceed the preset error, the adaptive calibration program is triggered, and the neural network regression model is adjusted through the retraining process;

[0082] Explanation: The calculated time lag features are fused with the historical data and prediction results in the time series prediction model (for example, through a weighted fusion method to fuse with other input features in the time series prediction model), as one of the input features; The feedback mechanism quantifies the contribution of the time lag feature to the deposition velocity prediction through time step adjustment, dynamically updates the model input combined with historical data and real-time sensor data; through recurrent neural network (RNN) or long short-term memory network (LSTM), the prediction weight of the model is adjusted according to the time lag feature in each prediction period, to ensure that its influence on the deposition velocity prediction is accurately captured.

[0083] In one possible embodiment, cross-scale time lag features are extracted through multi-scale wavelet coherence-improved gray mutual information joint analysis, and the main time lag feature components sensitive to the deposition rate are automatically screened through Shapley value ranking.

[0084] In a possible embodiment, the physical information regularized network is used to learn the coupling coefficient online, and the coupling weight is updated iteratively after each 1h sampling window is completed, so as to avoid the inaccuracy of the fixed empirical formula.

[0085] The following is a detailed explanation of the local features and the acquisition method:

[0086] The particle charge distribution reflects the difference in the charge distribution of the dust particles in space due to electrostatic action, and directly affects the deposition behavior; the acquisition method is to collect multi-point potential data (range ± 5kV) through a surface electrostatic field strength instrument, calculate the standard deviation and spatial distribution of the charge distribution;

[0087] The local micro-airflow disturbance intensity is used to describe the velocity variation and turbulence characteristics of the local airflow, and affects the suspension and deposition path of the dust; the acquisition method is to collect velocity data through a micro-airflow velocity sensor (deployed at key positions such as the corner of the air pipe), calculate the velocity gradient (velocity change rate per unit time) and turbulence intensity (based on spatial spectrum analysis);

[0088] The dew risk represents the tendency of dew caused by abnormal humidity fluctuation, and affects the sensor data and deposition rate; the acquisition method is to analyze the time series of humidity sensor data, detect the abnormal peak value or fluctuation frequency of the humidity curve, and calculate the dew probability in combination with the temperature data;

[0089] The oil mist compound ratio reflects the deposition characteristics and adhesion ability of the compound particles formed after the dust and oil mist are mixed; the acquisition method is to calculate the deposition slope (concentration change rate per unit time) and adhesion strength (based on the ratio of the residual amount after airflow scouring to the deposition amount) through the dust concentration sensor and humidity data.

[0090] In a possible embodiment, to solve the problem of abnormal dust deposition caused by local micro-airflow disturbance in the cabin of the engineering vehicle, a micro-airflow velocity sensor is deployed at the corner of the air pipe, narrow space or airflow intersection, to monitor the velocity gradient and turbulence characteristics of the local micro-airflow in real time; after the micro-airflow disturbance data are fused, the local micro-airflow disturbance intensity is generated, including the velocity gradient (reflecting the spatial change rate of the airflow velocity) and the turbulence characteristics (quantifying the turbulence intensity through spatial spectrum analysis); when the monitored value exceeds the threshold value, the adjustable air valve or auxiliary small fan at the corresponding position is controlled to adjust the airflow direction or flow rate, so as to weaken the turbulence or enhance the scouring effect, and reduce the dust deposition.

[0091] Example 2, the difference between the embodiment of the application and example 1 is as follows:

[0092] In one possible embodiment, to address the problem of humidity sensor data drift caused by local condensation leading to misjudgment of dust deposition abnormal state, the dew point temperature is calculated in real time, and when the difference between the ambient temperature and the dew point exceeds a preset value (such as dew point difference ≤ 2℃), the dew point compensation correction program is triggered to start, and the humidity measurement value affected by condensation is compensated and corrected in real time, thereby eliminating the interference of humidity sensor drift on the prediction result and improving the accuracy of dust deposition abnormal state judgment.

[0093] The specific implementation is that the step of compensating and correcting humidity data affected by condensation includes the following steps:

[0094] Step S21: Place a condensation monitoring electrode on the surface of the humidity sensor to monitor the change in resistance or capacitance in real time to detect the condensation state of the sensor surface.

[0095] Step S22: When the change in resistance or capacitance exceeds a preset threshold, determine that condensation has occurred on the surface of the humidity sensor, and start the compensation correction program.

[0096] Step S23: Based on a pre-calibrated condensation deviation curve or an adaptive machine learning model, real-time correction is performed on the humidity measurement affected by condensation.

[0097] Wherein, the pre-calibrated condensation deviation curve is obtained by mapping the condensation state to the humidity deviation through experimental calibration, and the adaptive machine learning model is trained on historical condensation data and corresponding real humidity data to predict humidity deviation based on condensation monitoring signals.

[0098] Step S24: Update the input of the time series prediction model using the corrected humidity data, and dynamically adjust the intervention strategies of air pressure adjustment, humidity adjustment, and static electricity elimination to improve the accuracy of dust deposition anomaly detection.

[0099] In one possible embodiment, to compensate for the data deviation of the humidity sensor caused by condensation, a condensation monitoring electrode and a micro-heating element are arranged on the surface of the humidity sensor to monitor the change in resistance or capacitance to detect condensation; when the monitoring value exceeds a preset value, the micro-heating element is started to heat for a preset duration to evaporate the condensation.

[0100] In a possible embodiment, the system comprises a mode matching module, based on the dust deposition velocity curve generated by the dust deposition risk prediction module and the real-time environmental parameters provided by the environmental data acquisition module, a high-dimensional feature space containing humidity-potential-dust multi-dimensional time delay characteristics and cross-regional airflow characteristics is constructed through a time series prediction model, and an initial control mode is trained, using historical operation data and actual deposition data fed back by the optical sensor, a matching relationship library of environmental parameters and operation control modes is optimized and generated through supervised learning combined with a clustering algorithm, the matching relationship library records the optimal air pressure adjustment value, humidity adjustment value and electrostatic elimination strategy under different dust concentration, humidity range and electric field intensity conditions; in real-time operation, the mode matching module queries the corresponding control mode from the matching relationship library according to the current environmental parameters and outputs the intervention strategy, and the real-time prediction frequency is reduced; if the environmental parameters are out of the coverage range of the matching relationship library, an adaptive learning mechanism is triggered, based on real-time environmental data and intervention effect feedback, the weights of time delay characteristics in the high-dimensional feature space are dynamically adjusted using a reinforcement learning algorithm, and the matching relationship library is updated to include new working conditions; the mode matching module regularly optimizes the weight parameters of the control mode in the matching relationship library combined with the actual deposition data fed back by the optical sensor, so as to ensure that the intervention strategy is consistent with the deposition trend of the dynamic working condition.

[0101] Summary: In example 2 of the present application, a dew point compensation correction program is introduced to solve the problem of data drift caused by local condensation of the humidity sensor. The condensation state of the humidity sensor surface is detected in real time by a condensation monitoring electrode. The humidity data is corrected in real time based on a pre-calibrated condensation deviation curve or an adaptive machine learning model. The condensation is evaporated by a micro-heating element to ensure the accuracy of the humidity data. At the same time, the mode matching module constructs a high-dimensional feature space containing humidity-potential-dust multi-dimensional time delay characteristics and cross-regional airflow characteristics, trains a matching relationship library of environmental parameters and operation control modes, dynamically adjusts the feature weights using reinforcement learning, and optimizes the intervention strategy, which significantly reduces the real-time prediction frequency, improves the accuracy of dust deposition anomaly detection, effectively solves the misjudgment problem caused by humidity sensor drift, and enhances the stability and reliability of the multi-region environmental control system of the engineering vehicle in complex dust environment.

[0102] Finally: The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An AI-based multi-zone environmental control system for an engineering vehicle, characterized in that, The system comprises: An environmental data acquisition module that acquires environmental parameters of each region, including humidity values, electric field intensity, air pressure data, and dust concentration data, and records the dust particle deposition distribution map of each region of the engineering vehicle; A dust deposition anomaly identification module that extracts the time sequence characteristics of the dust particle deposition distribution map, calculates the deposition acceleration coefficient and the humidity response coefficient, and calculates the dust deposition risk coefficient according to the deposition acceleration coefficient and the humidity response coefficient, and compares it with the preset value, and marks the region that exceeds the threshold value as the region in the dust deposition anomaly state; A dust deposition risk prediction module that analyzes the deposition trend based on the time sequence prediction model for the region entering the dust deposition anomaly state, inputs the environmental parameters, and predicts the dust deposition speed curve of each region of the engineering vehicle within a future set time; The input features of the time sequence prediction model include: humidity change rate, potential fluctuation trend, deposition speed change rate, deposition amount trend, cross-region airflow characteristics, and time lag characteristics; the time sequence prediction model uses a graph convolution-bidirectional LSTM hybrid network, wherein the graph convolution layer extracts spatial correlation based on the airflow adjacency matrix generated by offline CFD simulation, and dynamically adjusts the weight online to reflect the airflow change, and the bidirectional LSTM captures the time dependence to predict the dust deposition speed curve of each region; when the prediction deviation exceeds the preset value, an adaptive calibration program is triggered; A dust deposition intervention module that dynamically calculates the collaborative intervention strategy of air pressure adjustment, humidity adjustment, and static electricity elimination based on the predicted dust deposition speed curve, to effectively reduce the speed of dust deposition; A monitoring and adjustment module that monitors the dewing state of the humidity sensor surface in real time, compensates and corrects the humidity data affected by dewing, and dynamically adjusts the intervention strategy; The deposition acceleration coefficient is obtained by calculating the change rate of dust concentration and air pressure; the dust concentration data and air pressure data in each region are collected, the change rate of dust concentration, i.e., the change amount of dust concentration per unit time, is calculated, the change rate of air pressure, i.e., the change amount of air pressure per unit time, is calculated, and the deposition acceleration coefficient is obtained by multiplying the change rate of air pressure and the change rate of dust concentration, thereby representing the acceleration trend of dust deposition; The humidity response coefficient is obtained by calculating the change rate of humidity and the average humidity level of airflow; the change of humidity in the region is monitored in real time, the rate of humidity change is calculated, and the average value of the humidity of the region is calculated; after dimensionless processing, the humidity change rate and the average value of the humidity of the region are weighted to obtain the humidity response coefficient, which reflects the influence of humidity change on dust deposition; The dust deposition risk coefficient is obtained by multiplying the deposition acceleration coefficient and the humidity response coefficient.

2. The AI-based engineering vehicle multi-zone environmental control system of claim 1, wherein, The system comprises a region division module that divides the engineering vehicle into regions based on the airflow path, equipment layout, and dust deposition characteristics of the engineering vehicle; the dust deposition characteristics are obtained by analyzing the air pipe path, airflow direction, and flow distribution of the airflow path, and the cabin of the engineering vehicle is divided into zones by K-means clustering.

3. The AI-based engineering vehicle multi-zone environmental control system of claim 1, wherein, The system comprises a pre-warning module connected with the dust deposition risk prediction module, compares the predicted dust deposition rate curve with the pre-warning threshold, and when the pre-warning threshold is exceeded, sends a pre-warning signal to the outside and executes the dust deposition intervention module; defines a response window based on the response time of the intervention strategy and the confidence interval of the historical deposition rate data, and calculates the dust deposition amount in the response window through a time series prediction model; A static basic pre-warning threshold is associated with the change rate of the predicted deposition rate to generate a dynamic pre-warning threshold function; the pre-warning threshold is tightened as the dust deposition rate increases within the response window; the pre-warning threshold is proportionally scaled and corrected according to the real-time humidity difference and electric field intensity deviation to suppress the false high pre-warning threshold in high humidity and strong electric field environment; when the cumulative deposition amount reaches a preset proportion or the operation cycle exceeds the limit, the dynamic pre-warning threshold function is recalibrated to eliminate the influence of sensor drift and environmental aging.

4. The AI-based engineering vehicle multi-zone environmental control system of claim 1, wherein, The acquisition method of the time series prediction model comprises the following steps: Step S1: data acquisition and preprocessing: obtaining the real-time monitoring time series data of humidity value, electric field intensity, dust concentration, current dust deposition rate and dust deposition amount of each area of the engineering vehicle to form a standardized data sequence for analysis; Step S2: feature extraction: extracting the features of the data sequence for analysis and outputting a feature vector, including: humidity change rate; potential fluctuation trend; deposition rate change rate, deposition amount trend, cross-region airflow characteristics, time lag characteristics; Step S3: model construction: taking the feature vector as input, using a graph convolution-bidirectional LSTM hybrid network, the graph convolution layer extracts spatial correlation based on a precomputed airflow adjacency matrix, the precomputed airflow adjacency matrix is generated through offline CFD simulation, covering airflow distribution under typical working conditions; in real-time operation, the adjacency matrix weight is dynamically adjusted through an online interpolation method to reflect airflow changes, and the bidirectional LSTM captures time series dependence and adjusts the adjacency matrix weight at each cycle to reflect turbulent evolution, predicting the dust deposition rate curve of each area of the engineering vehicle within a future set time; taking the dust deposition rate curve prediction error as a loss function, optimizing the model parameters through cross-validation; the output is the dust deposition rate curve of each area, representing the change trend of the deposition rate per unit time over time; Step S4: model verification and calibration: based on historical operation data and real-time measurement data, verify the accuracy of the predicted dust deposition rate curve; when the deviation between the predicted value and the actual measured value exceeds the preset value, trigger the adaptive calibration program.

5. The AI-based engineering vehicle multi-zone environmental control system of claim 3, wherein, The acquisition method of the time lag feature comprises the following steps: Calculate the nonlinear correlation between humidity, potential, deposition rate and dust deposition, model the time lag relationship between each parameter and dust deposition through mutual information measure and nonlinear cross-correlation analysis, and generate a feature vector for describing the nonlinear time lag between parameters; Using a neural network regression model, the neural network regression model includes an input layer, one or more hidden layers, and an output layer; the input layer receives humidity, potential, pressure difference, and deposition velocity parameters, the hidden layer uses a fully connected layer, and the output layer predicts the time lag relationship of humidity, potential, and pressure difference parameters on dust deposition; training is performed through a back propagation algorithm, the loss function is mean square error, and the gradient descent method is used for optimization; According to the real-time monitoring data, the model parameters of the time lag feature are adjusted through an adaptive algorithm, the adaptive algorithm compares the actual monitoring results and the model prediction results at each feedback, and adjusts the network weight to ensure the accuracy of the time lag feature; Based on the airflow adjacency matrix generated by CFD simulation, the topological relationship of airflow space in the adjacency matrix is used as input, and the fluid dynamics model is used to optimize the time lag feature calculation; the modeling accuracy of the nonlinear time lag is improved by optimizing the input features; The calculated time lag feature is fed back to the time series prediction model, and when the predicted deposition velocity and the actual measured data exceed the preset error, the adaptive calibration program is triggered, and the neural network regression model is adjusted through the retraining process.

6. The AI-based engineering vehicle multi-zone environmental control system of claim 1, wherein, By deploying micro air flow velocity sensors at the corners of the air duct, narrow spaces, or air flow intersections, the velocity gradient and turbulence characteristics of local micro air flow are monitored in real time; after the micro air flow disturbance data is fused, the local micro air flow disturbance intensity, including the velocity gradient and turbulence characteristics, is generated, and when the monitoring value exceeds the threshold, the adjustable air valve or auxiliary small fan at the corresponding position is controlled to adjust the air flow direction or flow to weaken the turbulence or enhance the flushing effect, thereby reducing dust deposition.

7. The AI-based engineering vehicle multi-zone environmental control system of claim 1, wherein, In view of the problem that the humidity sensor data drift caused by local condensation leads to misjudgment of the dust deposition abnormal state, the dew point temperature is calculated in real time, and when the difference between the ambient temperature and the dew point exceeds the preset value, the dew point compensation correction program is triggered to compensate and correct the humidity measurement value affected by condensation in real time, thereby eliminating the interference of humidity sensor drift on the prediction result and improving the accuracy of dust deposition abnormal state judgment.

8. The AI-based engineering vehicle multi-zone environmental control system of claim 1, wherein, The multi-objective optimization algorithm includes: The air pressure adjustment target adjusts the air pressure in real time to optimize the air flow and prevent dust deposition acceleration by minimizing the deposition acceleration coefficient; The humidity adjustment target adjusts the humidity range in real time to prevent dust deposition abnormalities caused by excessive humidity fluctuations by calculating the humidity response coefficient; The static electricity elimination target optimizes the static electricity elimination strategy based on the comprehensive model of electric field intensity and humidity change to reduce the influence of static electricity accumulation on dust deposition; The air pressure adjustment target, humidity adjustment target, and static electricity elimination target are integrated through a weighted objective function, wherein the weights of the targets are dynamically adjusted according to real-time environmental data and prediction results, thereby realizing the collaborative optimization of air pressure, humidity, and static electricity elimination.

Images