Multi-process mode intelligent switching method and system for vehicle-mounted sludge treatment

Abstract

The invention provides a multi-process-mode intelligent switching method and system for vehicle-mounted sludge treatment, and the method comprises the steps: carrying out the time-space alignment and standardization of multi-dimensional sensing data such as sludge characteristics, environmental parameters and equipment states, carrying out the real-time reasoning of a process switching critical value through combining with a lightweight incremental regression model, generating a self-adaptive threshold value band with a confidence interval, and carrying out the real-time reasoning of the process switching critical value. Dynamic monitoring and mode switching pre-judgment of actual operation parameters are realized; in the switching process, a weighted fusion progressive transition control strategy is adopted, the transfer rate is adaptively adjusted according to real-time feedback, the operation stability is guaranteed, and the intelligence, reliability and switching smoothness of process switching of the vehicle-mounted sludge treatment system are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for vehicle-mounted sludge treatment systems, and in particular to a method and system for intelligent switching of multiple process modes in vehicle-mounted sludge treatment. Background Technology

[0002] Existing vehicle-mounted sludge treatment systems generally employ intelligent switching technology for multiple process modes to improve sludge treatment efficiency and environmental adaptability. Their control decision-making modules primarily rely on static threshold methods for mode determination and switching. Traditional technologies often set switching conditions using fixed parameter boundaries (such as key process indicators like moisture content, energy consumption, and load) to achieve switching between process modes, supplemented by certain predictive strategies and actuator calibration to ensure equipment operational safety. However, current mainstream control schemes are prone to insufficient decision-making adaptability when dealing with complex and variable on-site sludge characteristics and environmental conditions (such as the impact of different seasons, humidity, and temperature changes on equipment operating conditions). This leads to engineering problems such as inaccurate system mode switching timing, transient shocks during transitions, uneven discharge, or drastic energy consumption fluctuations. Existing technologies typically focus on optimizing a single process or improving control accuracy under certain operating conditions. For example, they utilize multi-sensor fusion to achieve precise sensing of sludge moisture and particle distribution, or employ PID closed-loop control to enhance the response speed and stability of processing modules. However, in the switching between multiple process modes, constant thresholds set manually based on experience are still used as the criterion, lacking the ability to adapt to the dynamic evolution of operating conditions. Furthermore, they cannot achieve intelligent sensing and flexible threshold adjustment for newly emerging abnormal operating conditions (such as highly viscous sludge or extreme temperature and humidity environments). In recent years, some technical solutions in the industry have attempted to introduce multi-dimensional data fusion and simple statistical discrimination; however, these technologies still lag behind in responding to high-frequency changes in operating conditions in mobile vehicle scenarios, failing to effectively mitigate the risks of decision-making misjudgments and equipment shocks caused by sudden changes in operating conditions. Existing intelligent control solutions for vehicle-mounted sludge treatment are mainly limited to engineering scenarios where the physical properties and environmental indicators of sludge fluctuate relatively little. For nonlinear changes and complex coupling phenomena that occur during long-term operation or mobile applications, their fixed threshold modes are prone to switching delays, misjudgments, or system instability. Furthermore, these solutions generally lack proactive transition mechanisms for smooth mode switching, leading to abnormal phenomena such as mechanical shocks, sudden increases in energy consumption, and equipment backflow in actuators during process switching due to sudden command changes, affecting the overall system efficiency and safety. Summary of the Invention

[0003] In order to solve the above-mentioned technical problems, the present invention provides a method for intelligent switching of multiple process modes for vehicle-mounted sludge treatment.

[0004] The technical solution of this invention is implemented as follows: a method for intelligent switching of multiple process modes in vehicle-mounted sludge treatment, comprising: S1: Acquire multi-dimensional sensing data under the current operating conditions of the vehicle-mounted sludge treatment system. The multi-dimensional sensing data includes sludge moisture content, organic matter concentration, particle size distribution, ambient temperature, ambient humidity, as well as equipment load, energy consumption rate and processing efficiency fed back by the processing module. Align the above data with timestamps to form a comprehensive operating condition feature vector with spatiotemporal consistency. S2: Normalize the comprehensive working condition feature vector to eliminate the influence of different physical dimensions on the model input, generate a standardized working condition feature vector, and input it into a lightweight incremental support vector regression model deployed in the control decision-making process. This model is constructed based on a historical effective switching event sample set within a sliding time window and is used to dynamically model the nonlinear mapping relationship between mode switching timing and working condition parameters. S3: The incremental support vector regression model is used to infer and calculate the feature vector of the current standardized working condition, and outputs the critical value of the control parameter associated with the predicted optimal switching point corresponding to the switching operation between each process mode. Combined with the confidence index output by the model, an adaptive threshold band with upper and lower floating boundaries is generated as the dynamic decision basis for subsequent mode switching judgment. S4: Real-time monitoring of actual operating parameters in the execution processing module, including torque change rate, dehydration rate change slope and vibration amplitude, and determining whether the actual operating parameters enter the adaptive threshold range generated by S3. If the entry condition is met, the pre-switching coordination mechanism is started to prepare for execution mode transition control. S5: After confirming that the actual operating parameters have entered the adaptive threshold band, activate the parallel coordination control logic between the two process modes, and use a weighted fusion strategy to synchronously output the control commands of the source mode and the target mode. The initial weight value is biased towards the source mode, and the control weight is gradually transferred from the source mode to the target mode over time according to the exponential decay function, generating a continuous and progressive transition control command sequence. S6: Send the transition control command sequence to the execution processing module to drive the actuator to complete the smooth switching of the control mode. During the switching process, continuously collect status feedback signals, including material discharge uniformity, instantaneous energy consumption fluctuation and mechanical vibration intensity, and construct a stability evaluation index for the switching process based on these signals. S7: Determine whether the stability assessment index generated by S6 exceeds the preset safety tolerance range. If it does not exceed the limit, confirm that the mode switch is successful and maintain the target mode operation. If it exceeds the limit, trigger the rollback mechanism, immediately restore the control command output of the original process mode, and mark the switch as a failure case. S8: Each completed switching event, whether successful or not, is stored in the historical event database in the form of a data structure containing the comprehensive operating condition feature vector before switching, the actual switching time, the adaptive threshold parameters, and the stability evaluation results. This data is then used to periodically update the training sample set of the incremental support vector regression model described in S2, enabling online learning and adaptive optimization of the model parameters.

[0005] The present invention also provides an intelligent switching system for multiple process modes of vehicle-mounted sludge treatment, which uses the above-mentioned intelligent switching method for multiple process modes of vehicle-mounted sludge treatment to switch between multiple process modes of vehicle-mounted sludge treatment.

[0006] The present invention provides a method and system for intelligent switching of multiple process modes for vehicle-mounted sludge treatment, which has the following beneficial effects: (1) This invention constructs a control decision system with adaptive evolution capabilities by introducing a dynamic working condition sensing unit and a lightweight online learning mechanism. This system can integrate sludge physical property parameters (such as moisture content, organic matter concentration, and particle size distribution) with environmental conditions (temperature and humidity) and equipment operating status (load, energy consumption rate, and processing efficiency) in real time to form a high-dimensional comprehensive working condition feature vector. It also uses an incremental support vector regression model based on a sliding time window to continuously model and update parameters for historical successful switching events, realizing dynamic prediction of the optimal switching point and online generation of threshold intervals. This significantly improves the system's decision sensitivity and robustness in the face of raw material fluctuations and external interference in mobile operation scenarios, effectively avoiding the risk of decreased processing efficiency or equipment overload caused by misjudgment or delayed triggering.

[0007] (2) This invention designs a predictive triggering mechanism and weighted transition control logic based on dynamic threshold bands, realizing smooth connection between different processing modes. The model not only outputs the central critical value of the optimal switching point, but also generates upper and lower floating boundaries to form a threshold band in combination with confidence evaluation. When the parameter enters this area, pre-action preparation is started to realize early response. In the formal switching stage, an exponential decay weight function is introduced to gradually fuse the control commands of the source mode and the target mode in the time domain, avoiding mechanical stress concentration and process interruption caused by sudden changes in control quantity. At the same time, the system is equipped with a closed-loop self-verification module to continuously monitor key indicators such as vibration amplitude, energy consumption fluctuation and material output uniformity. Once an abnormality is detected, a safety rollback is executed and failure cases are recorded for subsequent learning and optimization, thereby constructing a complete closed-loop control link of "perception-decision-execution-feedback-evolution", which greatly improves the continuity and reliability of system operation, especially suitable for vehicle-mounted mobile application scenarios with limited space and frequent changes in working conditions. (3) This invention abandons the complex architecture of traditional multi-model parallel or offline parameter tuning, and adopts a lightweight incremental learning strategy, which significantly reduces the consumption of computing resources and the difficulty of engineering deployment while ensuring model accuracy. Since there is no need to repeatedly load the full data for retraining, knowledge accumulation can be completed only through local parameter updates, which greatly reduces the algorithm running latency and storage overhead, and meets the real-time requirements of vehicle embedded systems. In addition, this method has good interpretability and scalability. The dynamic threshold band it generates can intuitively reflect the trend of switching sensitivity changes under the current working conditions, which is convenient for operation and maintenance personnel to understand and intervene. Attached Figure Description

[0008] Figure 1 A flowchart of a multi-process mode intelligent switching method for vehicle-mounted sludge treatment according to the present invention; Figure 2 This is a sub-flowchart of a multi-process mode intelligent switching method for vehicle-mounted sludge treatment according to the present invention. Figure 3 This is another sub-flowchart of the intelligent switching method for multi-process modes in vehicle-mounted sludge treatment according to the present invention. Detailed Implementation

[0009] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0010] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0011] like Figure 1 As shown, this invention provides a method for intelligent switching of multiple process modes in vehicle-mounted sludge treatment, specifically including: S1: Acquire multi-dimensional sensing data under the current operating conditions of the vehicle-mounted sludge treatment system. The multi-dimensional sensing data includes sludge moisture content, organic matter concentration, particle size distribution, ambient temperature, ambient humidity, as well as equipment load, energy consumption rate and processing efficiency fed back by the processing module. Align the above data with timestamps to form a comprehensive operating condition feature vector with spatiotemporal consistency. S2: Normalize the comprehensive working condition feature vector to eliminate the influence of different physical dimensions on the model input, generate a standardized working condition feature vector, and input it into a lightweight incremental support vector regression model deployed in the control decision-making process. This model is constructed based on a historical effective switching event sample set within a sliding time window and is used to dynamically model the nonlinear mapping relationship between mode switching timing and working condition parameters. S3: The incremental support vector regression model is used to infer and calculate the feature vector of the current standardized working condition, and outputs the critical value of the control parameter associated with the predicted optimal switching point corresponding to the switching operation between each process mode. Combined with the confidence index output by the model, an adaptive threshold band with upper and lower floating boundaries is generated as the dynamic decision basis for subsequent mode switching judgment. S4: Real-time monitoring of actual operating parameters in the execution processing module, including torque change rate, dehydration rate change slope and vibration amplitude, and determining whether the actual operating parameters enter the adaptive threshold range generated by S3. If the entry condition is met, the pre-switching coordination mechanism is started to prepare for execution mode transition control. S5: After confirming that the actual operating parameters have entered the adaptive threshold band, activate the parallel coordination control logic between the two process modes, and use a weighted fusion strategy to synchronously output the control commands of the source mode and the target mode. The initial weight value is biased towards the source mode, and the control weight is gradually transferred from the source mode to the target mode over time according to the exponential decay function, generating a continuous and progressive transition control command sequence. S6: Send the transition control command sequence to the execution processing module to drive the actuator to complete the smooth switching of the control mode. During the switching process, continuously collect status feedback signals, including material discharge uniformity, instantaneous energy consumption fluctuation and mechanical vibration intensity, and construct a stability evaluation index for the switching process based on these signals. S7: Determine whether the stability assessment index generated by S6 exceeds the preset safety tolerance range. If it does not exceed the limit, confirm that the mode switch is successful and maintain the target mode operation. If it exceeds the limit, trigger the rollback mechanism, immediately restore the control command output of the original process mode, and mark the switch as a failure case. S8: Each completed switching event, whether successful or not, is stored in the historical event database in the form of a data structure containing the comprehensive operating condition feature vector before switching, the actual switching time, the adaptive threshold parameters, and the stability evaluation results. This data is then used to periodically update the training sample set of the incremental support vector regression model described in S2, enabling online learning and adaptive optimization of the model parameters.

[0012] Step S1: Acquire multi-dimensional sensor data under the current operating conditions of the vehicle-mounted sludge treatment system. This multi-dimensional sensor data includes sludge moisture content, organic matter concentration, particle size distribution, ambient temperature, ambient humidity, and equipment load, energy consumption rate, and processing efficiency fed back by the processing module. The above data is then aligned by timestamps to form a comprehensive operating condition feature vector with spatiotemporal consistency. Specifically, this includes: S1.1: Obtain real-time sludge characteristic parameters output from the process pattern recognition module, including sludge moisture content, organic matter concentration and particle size distribution, as well as ambient temperature and humidity data collected from environmental sensors, as basic input signals to characterize external processing conditions, so as to construct an original sensing dataset reflecting the current working environment. In the control and decision-making module of the vehicle-mounted sludge treatment system, a multi-source sensor acquisition method is adopted (parameter: sampling period). =100ms, resolution ≤0.1%), enabling real-time acquisition of sludge characteristic parameters output by the process mode recognition module, including three key dimensions: sludge moisture content, organic matter concentration, and particle size distribution; By using a high-precision temperature and humidity sensor array acquisition method (parameters: temperature accuracy ±0.2℃, humidity accuracy ±1%RH), the physical measurement of ambient temperature and humidity is realized and converted into standardized voltage signals to obtain a basic data stream characterizing external processing conditions. Furthermore, a data filtering and denoising algorithm (parameters: third-order low-pass filter cutoff frequency of 1-20Hz, preferably 5Hz, noise suppression threshold of 0.05 units) is adopted to perform signal preprocessing on sludge characteristic parameters and environmental state parameters, and obtain the effective measurement value matrix after noise suppression. Furthermore, a parameter calibration and linearization mapping algorithm is adopted (parameters: calibration coefficients are provided by laboratory calibration results, and linearization adopts a piecewise fitting method) to convert the nonlinear response of various sensor outputs into a linear scale of physical quantities and generate an original sensing dataset with a unified unit system. By using a multi-dimensional data structure encapsulation processing method, the results of the previous step are transformed into raw sensing datasets with acquisition timestamps, providing accurate basic input for subsequent synchronous acquisition of device operating parameters and time alignment; For example, in a system configured for vehicle-mounted sludge treatment, the process mode recognition module uses a near-infrared spectral sensor to measure the sludge moisture content as 42.7%, organic matter concentration as 18.5%, and particle size distribution D50 value as 0.58 mm, with a data sampling period of 100 ms. The environmental sensor module measures the current operating environment temperature as 28.4℃ and humidity as 65.3%RH, also with a sampling period of 100 ms. The raw signals are processed by a third-order low-pass filter with a cutoff frequency of 5Hz. After noise reduction, the instantaneous fluctuation of the moisture content signal is reduced to ±0.05%, and the instantaneous fluctuations of the temperature and humidity signals are reduced to ±0.1℃ and ±0.3%RH, respectively. Subsequently, based on laboratory-calibrated coefficients, piecewise linearization mapping is performed on the moisture content, concentration, particle size, and temperature and humidity signals, for example, the output voltage of the moisture content sensor. The relationship with the actual moisture content W is mapped as follows: (W is in percentage) (Unit: Volt) The final result is a raw sensing data vector containing five physical quantities, along with a precise timestamp sequence, providing consistent input conditions for the multi-source heterogeneous data fusion in the next sub-step. S1.2: Synchronously collect and execute the equipment operation status parameters fed back by the processing module, including equipment load, energy consumption rate and processing efficiency. Based on the industrial bus communication protocol, timestamp the above parameters to ensure the time consistency of the data, and incorporate them into the original sensing dataset to form a multi-source heterogeneous sensing data set containing process status and system response. The input conditions are real-time sampling signals of equipment operating status parameters from the execution processing module, including multi-dimensional data streams such as equipment load, energy consumption rate and processing efficiency. These data streams need to be combined with the original sensing dataset of sludge characteristics and environmental status from the preceding S1.1 to form a fusionable multi-source heterogeneous information set. A real-time data acquisition method conforming to the industrial bus communication protocol (parameters: bus type = CANOpen, sampling period = 100ms, data frame length = 64 bytes) is adopted to realize the synchronous data acquisition function of equipment load sensor, energy consumption metering unit and processing efficiency calculation module; Furthermore, through a timestamp labeling algorithm (parameters: unified clock source = GPS timing module, time accuracy = 1ms), the precise time positioning of the device's operating status parameters is achieved, and the time sequence label matrix of each data frame is obtained; Furthermore, through a data format standardization processing method (parameters: numerical encoding format = IEEE754 single-precision floating point, field order = [load, energy consumption, efficiency]), a unified structured encapsulation of multi-dimensional parameters of device operating status is achieved, and a collection of status data objects with a single parsing interface is generated; Furthermore, through a multi-source data merging algorithm (parameters: merging strategy = index matching, main index field = timestamp), the primary key alignment of the equipment operation status data with the preceding sludge characteristics and environmental status datasets is achieved, and a combined data form is generated, the structure of which is a multi-source parameter row vector corresponding to the timestamp index; Through the above-mentioned industrial bus acquisition, timestamp annotation, data formatting and multi-source merging processing methods, the equipment operation status sampling signal of the previous step is transformed into a multi-source heterogeneous sensor data set with time consistency, realizing the synchronous fusion of process status and system response data in the numerical domain and time domain, providing a stable input for the subsequent time series alignment and feature fusion in S1.3; For example, in the vehicle-mounted sludge treatment device, the processing module includes a load sensor (range 0-50kN), a three-phase active power acquisition unit (sampling accuracy 0.1kW), and an online processing efficiency calculation module (based on the discharge quality sensor). The industrial bus uses the CANOpen protocol, with a sampling period of 100ms. A single frame of data consists of three floating-point fields, representing the load (kN), energy consumption (kW), and efficiency (kg / h). The timestamp is synchronized with GPS with an accuracy of 1ms, and the output format is a UNIX timestamp (millisecond level). In one acquisition cycle, the load sensor output value is 12.3kN, the energy consumption unit output value is 15.4kW, and the calculated processing efficiency value is 240kg / h. After being received by the industrial bus, a standardized data object {time:1627543210123, load:12.3, energy:15.4, efficiency:240} is generated. The parameters such as sludge moisture content and ambient temperature at the same time stamp as the preceding S1.1 are indexed, matched, and merged to obtain a combined row vector [1627543210123, moisture content=0.65, organic matter concentration=0.12, particle size=0.8mm, temperature=28.4°C, humidity=58%, load=12.3kN, energy consumption=15.4kW, efficiency=240kg / h]. This vector serves as a record in the multi-source heterogeneous sensor data set, and subsequent time series alignment can completely maintain the spatiotemporal consistency and numerical accuracy of parameters across sources. S1.3: Perform time series alignment processing on each parameter in the multi-source heterogeneous sensing data set, use interpolation algorithm to compensate for the time offset caused by the difference in sensor sampling period, and map all parameters to the same timestamp node based on a unified time reference to generate an aligned sensing dataset with spatiotemporal consistency, so as to ensure the accuracy of subsequent feature fusion. S1.4: The aligned sludge characteristic parameters, environmental state parameters and equipment operating parameters are vectorized according to a preset dimension order to construct a high-dimensional numerical vector. This vector serves as the preliminary form of the comprehensive operating condition feature vector and is used to characterize the complete operating condition of the current system under the coupling effect of multiple physical fields. S1.5: Perform integrity verification and outlier detection on the generated comprehensive operating condition feature vector. Based on statistical methods, identify and remove invalid data items that exceed the reasonable range or have abrupt transitions. If the verification passes, retain the vector as a valid input for subsequent normalization processing and model inference to ensure the reliability and stability of the input data in the control decision-making process.

[0013] Step S2: Normalize the comprehensive operating condition feature vector to eliminate the influence of different physical dimensions on the model input, generate a standardized operating condition feature vector, and input it into a lightweight incremental support vector regression model deployed in the control decision-making process. This model is constructed based on a historical effective switching event sample set within a sliding time window and is used to dynamically model the nonlinear mapping relationship between mode switching timing and operating condition parameters. Specifically, it includes: S2.1: Based on the comprehensive operating condition feature vector output in the previous step S1, obtain its original data for each dimension, including sludge moisture content, organic matter concentration, particle size distribution, ambient temperature, ambient humidity, equipment load, energy consumption rate and treatment efficiency. Using the normalized boundary parameters formed by the maximum and minimum values ​​of each feature in historical operating data, perform range normalization processing on the comprehensive operating condition feature vector to eliminate the dimensional differences and order of magnitude deviations between multi-source sensor data and generate a dimensionless preliminary standardized vector. Based on the comprehensive operating condition feature vector output from the preceding step S1, numerical range analysis is performed on the raw data of each dimension to provide reference boundary parameters for normalization. The raw data includes sludge moisture content, organic matter concentration, and particle size distribution from the sludge characteristic detection unit; ambient temperature and humidity from the environmental status sensing unit; and equipment load, energy consumption rate, and processing efficiency from the equipment operation monitoring unit. Using historical data analysis (parameter: sampled datasets from the most recent M running cycles), the maximum value of each feature dimension under historical operating conditions is extracted. and minimum value This is used as a set of normalized boundary parameters for subsequent range normalization calculations. The above process is implemented using sliding window retrieval and statistical analysis functions to ensure dynamic updating of the boundaries. Furthermore, a range normalization algorithm (parameter: set of boundary parameters) is used to eliminate dimensional differences and order-of-magnitude deviations among multi-source sensor data. This algorithm follows the formula...

[0014] Perform dimensionless mapping on each dimension parameter, where This is the original value for the current dimension. and These are the corresponding historical boundary parameters; Furthermore, the normalized values ​​obtained from the calculation of the parameters in each dimension are combined into a preliminary standardized vector. During the combination process, the order of the dimensions is kept consistent with the comprehensive working condition feature vector to avoid disrupting the feature mapping relationship. This combination adopts a vector concatenation method, storing the dimensionless parameters into a new vector structure according to a set index sequence. Furthermore, utilizing a dynamic update mechanism for boundary parameters, the current normalized scaling factor sequence is recorded simultaneously with the generation of the initial standardized vector, serving as a reference for subsequent zero-mean and standard deviation calculations. This scaling factor is the value for each dimension. This is appended to the vector metadata in numerical form; By using the above range normalization method, the comprehensive working condition feature vector obtained in the previous step S1 is transformed into a dimensionless preliminary standardized vector, realizing the mapping of multi-source heterogeneous parameters in the same numerical space, providing a unified dimensional basis for subsequent zero-mean normalization and model input, and improving the stability of data fusion and the controllability of computation. For example, during 30 consecutive operating cycles (sampling interval of 1 second per cycle) of a vehicle-mounted sludge treatment system, the historical maximum sludge moisture content was 82.5% and the minimum was 68.1%; the historical maximum organic matter concentration was 450 mg / L and the minimum was 310 mg / L; the historical maximum particle size distribution was 3.8 μm and the minimum was 1.2 μm; the historical maximum ambient temperature was 38.4°C and the minimum was 18.9°C; the historical maximum ambient humidity was 88.3°C and the minimum was 45.2°C; the historical maximum equipment load was 15.2 rpm and the minimum was 8.4 rpm; the historical maximum energy consumption rate was 5.8 rpm and the minimum was 3.1 rpm; and the historical maximum treatment efficiency was 94.6 rpm and the minimum was 82.7 rpm. For the current sampled values ​​(moisture content 74.3%, organic matter concentration 376%, particle size distribution 2.5%, temperature 26.7%, humidity 65.4%, equipment load 11.3%, energy consumption rate 4.2%, treatment efficiency 88.1%), the range normalization formula is applied. Calculate the normalized values ​​for each dimension separately. For example, the normalized value for the moisture content dimension is... = This process is repeated to obtain an 8-dimensional dimensionless vector, which is then arranged according to the dimensional order of the original comprehensive working condition feature vector to form a preliminary standardized vector. Through this processing, the data of each dimension is mapped to a unified [0,1] interval, significantly improving the convergence stability of subsequent models for multi-source inputs and the accuracy of nonlinear mapping. S2.2: The dimensionless preliminary standardized vector generated in S2.1 is subjected to zero-mean processing. Based on the mean vector of the N most recent valid working condition samples within the sliding time window as a dynamic benchmark, the offset of the current vector from the benchmark is calculated, and the offset is divided by the standard deviation sequence of the corresponding feature to obtain a standardized working condition feature vector with statistical consistency, thereby improving the convergence stability of the input data for the incremental learning model. S2.3: Based on the standard working condition feature vector output by S2.2, as the core input data, combined with the model initialization configuration parameters (including kernel function type, penalty coefficient, and insensitive loss parameter), the lightweight incremental support vector regression model deployed in the control decision module is activated. The model structure is initialized based on the initial weight vector and bias term obtained by training the previous historical effective switching event sample set, and enters the online inference preparation state. S2.4: Based on the model initialization completed in S2.3, the standardized operating condition feature vector is used as the independent input sample at the current moment and input into the loaded incremental support vector regression model. It is converted into an implicit expression vector in the high-dimensional feature space through the kernel function mapping mechanism. Based on the support vector set and Lagrange multiplier sequence currently maintained by the model, forward inference calculation is performed to obtain the preliminary estimation results of the critical values ​​of the control parameters associated with the predicted optimal switching point corresponding to the switching operation between each process mode. S2.5: Based on the preliminary estimation results output by S2.4, and combined with the historical prediction error sequence within the sliding time window maintained internally by the model, calculate the confidence index of the current prediction result. Use this confidence index to adjust the prediction critical value for uncertainty compensation, and generate an adaptive threshold parameter set containing upper and lower floating boundaries. This serves as the input basis for the next stage of dynamic decision-making, completing the key transformation from raw perceived data to executable decision parameters.

[0015] like Figure 2 As shown, step S3 involves using the incremental support vector regression model to infer and calculate the feature vector of the current standardized operating condition, outputting the critical value of the control parameter associated with the predicted optimal switching point between each process mode, and combining it with the confidence index output by the model to generate an adaptive threshold band with upper and lower floating boundaries, which serves as the dynamic decision-making basis for subsequent mode switching judgments. Specifically, this includes: S3.1: Based on the standard process condition feature vector output from the previous step S2, it is used as input data and loaded into the lightweight incremental support vector regression model deployed in the control decision module. The model constructs a nonlinear mapping relationship based on the accumulated historical effective switching event sample set within the sliding time window and performs a forward inference process to obtain the initial estimation results of the critical values ​​of the control parameters required for switching of each process mode under the current operating condition. Based on the standard working condition feature vector output by step S2, an incremental support vector regression model loading method (parameters: kernel function type, penalty coefficient C, insensitive loss ε, support vector set) is used to import the high-dimensional dimensionless feature vector into the inference engine within the control decision module. Furthermore, by using a kernel function mapping algorithm (parameters: radial basis kernel width γ, feature space dimension M), the standardized operating condition feature vector is mapped to an M-dimensional high-dimensional feature space, thereby explicitly characterizing the similarity between the current operating condition and historical mode switching events in the nonlinear feature domain, and obtaining a high-dimensional implicit expression vector. Furthermore, by utilizing the support vector set and the corresponding Lagrange multiplier sequence, the predicted output value is constructed using the SVR forward inference calculation formula:

[0016] in, This is the critical value for sudden torque change. For Lagrange multipliers, For kernel function, For bias terms, For the current input feature vector, For support vector features; Furthermore, the initial estimate of the critical value of the control parameter corresponding to the current working condition is calculated using the above formula, and the sliding time window design of the model is combined to ensure that the output value reflects the conditional mode characteristics of the most recent effective switching event. By using SVR inference processing, the above calculation results are transformed into initial estimation data of the critical values ​​of control parameters for each process mode switching operation, thereby realizing the quantitative characterization of the mode switching timing and providing a basis for subsequent confidence assessment and adaptive threshold band generation. For example, the standardized operating condition feature vector contains eight dimensions of data, where the sludge moisture content is 0.62 after normalization, the organic matter concentration is 0.48, the particle size distribution characteristic value is 0.35, the ambient temperature is 0.57, the ambient humidity is 0.63, the equipment load is 0.41, the energy consumption rate is 0.44, and the treatment efficiency is 0.52. The incremental support vector regression model in the control decision module is configured with a Gaussian radial basis function, with the kernel width parameter γ set to 0.2, the penalty coefficient C set to 15, the insensitivity loss ε set to 0.05, and the support vector set size to 24 vectors. The above input vectors are mapped to an M=120 dimensional Gaussian kernel feature space, and the following formula is executed for calculation:

[0017] In this scenario, The value is derived from the incremental learning results of the most recent 50 valid historical switching events. The output y-value is the critical value of torque mutation during the corresponding process mode switching, which is approximately 312 Nm. This prediction result will be used in subsequent steps to construct an adaptive threshold band, and the upper and lower limits will be adjusted through a dynamic boundary expansion algorithm to achieve smoothness and accuracy in switching. In field operation tests, this effectively reduces the amplitude of transient impacts, and the vibration index remains within a low range during the switching process. S3.2: Perform confidence assessment on the initial estimation results of the critical values ​​of the control parameters obtained in S3.1. Utilize the residual distribution and local sample density information output by the incremental support vector regression model to calculate the confidence interval width of the current prediction results and form a quantitative confidence index to reflect the modeling reliability level of the model in the current working condition area. The initial estimation results of the critical values ​​of the control parameters obtained in step S3.1 are used to perform residual analysis (parameters: current prediction result vector, historical true value vector) to achieve quantitative evaluation of the local error of the model. Furthermore, by calculating the variance and mean shift of the predicted residual sequence, residual distribution characteristics reflecting the modeling deviation of the current working condition area are extracted, and residual distribution matrix data are obtained. Furthermore, a local sample density estimation algorithm (parameters: the number of k nearest neighbors k of the current working condition feature vector in the support vector set and the distance metric formula) is adopted to realize the neighborhood density analysis of the prediction results in the feature space and generate the sample density index result; Furthermore, the residual variance characteristics and sample density index are input into the confidence interval width generation function, and the comprehensive confidence interval width is calculated through a weighted synthesis method, where the weighting coefficients are preset based on the sensitivity to operating conditions. The confidence interval width is calculated using the following formula. :

[0018] in The standard deviation of the residuals. These are the residual weighting coefficients. For local sample density index, This refers to the density weighting coefficient; Furthermore, by normalizing and comparing the confidence interval width with the confidence interval distribution of historical stable switching samples, a quantitative confidence index for the current prediction result is generated. This confidence assessment process transforms the initial estimation results from the previous step into quantified confidence data that can be used for dynamic boundary expansion, thereby accurately characterizing the reliability level of the model in the current operating condition region. For example, in the dynamic switching scenario of the vehicle-mounted sludge treatment system, the input condition is the critical value of the predictive control parameter. Nm, corresponding to the historical actual measurement value Nm, residual is Nm. Residual sequence variance Mean shift Residual weighting coefficient The local sample density index is calculated based on the k=5 nearest neighbor to obtain the density value. Density weighting coefficient Substitute into the formula The confidence interval width is calculated. Compared with the historical sample width distribution (mean 0.5, standard deviation 0.1), the confidence index obtained after normalization is: This reflects that the predicted value has high reliability in the current operating condition area modeling, and a narrow boundary offset can be applied in S3.3 to improve the sensitivity of the switching decision; S3.3: Based on the initial estimation results of the control parameter critical value output by S3.1 and the confidence index generated by S3.2, the dynamic boundary expansion algorithm is executed. According to the preset confidence-boundary gain mapping function, a variable offset is applied to the upper and lower critical values ​​to generate an adaptive threshold band containing an upper threshold and a lower threshold. High confidence corresponds to narrow bandwidth and low confidence corresponds to wide bandwidth to balance switching sensitivity and stability. S3.4: Encapsulate the adaptive threshold band generated in S3.3 into a structured decision parameter package, which includes the central critical value, upper boundary value, lower boundary value and corresponding process mode pair identifier, and adds timestamp and operating condition label to form a dynamic decision basis data unit that can be used for subsequent real-time monitoring and comparison, and is called by step S4 to determine whether the actual operating parameters have entered the threshold transition zone. Based on the adaptive threshold band parameter set generated by S3.3, a structured data encapsulation method (parameters: center critical value, upper boundary value, lower boundary value, corresponding process mode pair identifier) ​​is used to realize the formatted description of each boundary parameter; Furthermore, a timestamp appending algorithm (parameter: system local high-precision clock source) is used to mark the timing of the encapsulated data, ensuring that the subsequent comparison process can be performed based on a unified time reference. Furthermore, through the working condition label generation mechanism (parameter: the set of environmental and equipment operating status labels corresponding to the current comprehensive working condition feature vector), the semantic association between structured data and specific working conditions is realized, and decision-making basis items with identifiability are obtained; Furthermore, a dynamic monitoring index construction method (parameters: center critical value, boundary value and process mode mapping table) is adopted to realize the ability to quickly retrieve this structured data packet and generate index key-value pairs to support multi-dimensional query operations; Through the above processing method, the adaptive threshold parameter set of the previous step is transformed into a structured decision parameter package containing center value, upper and lower boundary values, pattern identifier, timestamp and working condition label, so as to achieve the technical effect of being called in step S4 for real-time running parameter and threshold interval dynamic boundary comparison. For example, in a certain vehicle-mounted sludge treatment task, the center threshold value of the adaptive threshold band is set to... The upper boundary value is The lower boundary value is The process mode is identified as "dewatering-drying" mode group, with a timestamp of 2024-07-10T08:15:32.865Z. The operating condition label includes sludge moisture content. Equipment load Key parameters such as kW are included. During the encapsulation process, the index key "ModePair=DH12" corresponds to the mode group, and the index key "TS=20240710081532" corresponds to the timestamp, enabling rapid matching with historical records and real-time monitoring data. When step S4 is invoked, this structured parameter package serves as a dynamic reference input, allowing the system to detect the actual torque change rate... When the sampling period is within the threshold band for two consecutive sampling periods, the mode switching prediction mechanism can be triggered immediately and the transition control phase can be entered, thereby effectively improving the accuracy and response speed of the switching judgment. S3.5: Synchronously write the structured decision parameter package generated in S3.4 into the shared memory buffer of the control decision module and trigger a status notification signal to ensure that the execution processing module and the status feedback module can obtain the latest adaptive threshold band information in a timely manner, and maintain the data consistency and timeliness among multiple modules of the system in the mode switching prediction stage.

[0019] like Figure 3 As shown, step S4 involves real-time monitoring of the actual operating parameters in the execution processing module, including torque change rate, dehydration rate change slope, and vibration amplitude. It then determines whether these actual operating parameters fall within the adaptive threshold range generated by S3. If the entry condition is met, a pre-switching coordination mechanism is initiated to prepare for transition control of the execution mode. Specifically, this includes: S4.1: Based on the adaptive threshold band parameters output by S3, a dynamic boundary judgment matrix is ​​constructed. The adaptive threshold band includes the upper and lower floating boundaries and the center line of the control parameter critical value. The interval comparison algorithm in the field of industrial control is used to discretize each boundary and generate a threshold window structure that can be used for real-time comparison as the benchmark input for subsequent parameter discrimination. Based on the adaptive threshold band parameters output from step S3, a boundary analysis method is employed (parameter: central critical value). Upper boundary Lower boundary Operating mode code This allows the upper and lower floating boundaries and centerline of the control parameters to be extracted into a structured numerical set. Furthermore, through an interval comparison algorithm (parameter: sampling boundary resolution) The quantization precision scale is used to discretize the above boundary values, generate a set of discrete thresholds for comparison of running parameters, and obtain the upper and lower limits of each segment. Furthermore, through matrix construction methods (parameter: number of boundary segments) Pattern Dimension This allows the discrete threshold set of each control parameter to be mapped to a two-dimensional matrix in rows and columns, where rows represent control parameter indices and columns represent threshold segment indices, thus obtaining the preliminary form of the dynamic boundary determination matrix. Furthermore, by using the matrix normalization method (parameter: normalization range [min, max]), the boundary values ​​in the decision matrix are converted into a unified dimension space, resulting in normalized dynamic boundary decision matrix data that can be directly compared with standardized operating parameters. The above algorithm transforms the adaptive threshold with structured parameters from the previous step into a dynamic boundary judgment matrix, enabling the benchmark input for real-time comparison of multiple parameters and providing data support for the subsequent S4.2~S4.4 running parameters to enter the region discrimination. For example, in a certain operation of the vehicle-mounted sludge treatment system, the adaptive threshold band center critical value output by S3 is... = upper boundary = lower boundary = The operating mode code is =02. Boundary resolution set to = Quantization precision scale= . The interval Discretize into = Divide into segments, and obtain the segment set { , , ,... Let k=1 represent the torque change rate, k=2 represent the dehydration rate change slope, and k=3 represent the vibration amplitude. Construct a 3×3 dynamic boundary judgment matrix, where the matrix elements are the boundary values ​​of each parameter in different segments. Then, normalize the matrix elements according to the range [min, max] = [ , Perform linear normalization to obtain a set of normalized matrix values, the range of which is located in []. , The interval is used for direct comparison with the current operating state vector collected subsequently. In actual verification, this matrix can significantly improve the interval determination accuracy under different operating conditions, ensuring the robustness of the mode switching prediction process; S4.2: Obtain the real-time actual operating parameter stream collected in the execution processing module. The actual operating parameters include torque change rate, dehydration rate change slope and vibration amplitude. Align them with the local clock of the control decision module through a time synchronization mechanism to form a current operating state vector with time consistency, which serves as the data input object for threshold discrimination. S4.3: Perform normalization mapping on each parameter component in the current running state vector, transform it to the same dimension space as the adaptive threshold band, generate a standardized running parameter sequence, and extract its instantaneous change trend features based on the sliding sampling window to obtain a trend vector for boundary crossing analysis; S4.4: The dynamic boundary judgment matrix is ​​used to perform interval classification judgment on the standardized operating parameter sequence and its trend vector. The multi-dimensional spatial point-domain inclusion detection algorithm is used to calculate whether each parameter enters the adaptive threshold band range. When any key parameter falls into the threshold band for two consecutive sampling periods, it is determined that the entry condition is met and a mode switching pre-trigger signal is generated. Based on the standardized operating parameter sequence and trend vector output from step S4.3 as input objects, a multi-dimensional spatial point-domain inclusion detection algorithm (parameters: dynamic boundary judgment matrix, three-dimensional operating parameter spatial coordinate system) is used to determine the spatial positional relationship between the current operating parameters and the adaptive threshold band. Furthermore, by using the interval assignment operation method (parameters: standardized values ​​of each parameter, upper and lower boundary values ​​of the threshold band), the discretized state identifier of each running parameter within the threshold band is generated, and the interval assignment Boolean matrix of each parameter is obtained as an intermediate judgment result. Furthermore, by using the trend vector direction consistency test algorithm (parameters: parameter increase / decrease direction within the sampling period, threshold band center line position), the trend of the change trend of the running parameters gradually approaching the threshold band center line is confirmed, and trend consistency indicator data is generated; Furthermore, by using the continuous sampling period determination method (the number of periods in the formula is set to 2), a logical AND operation is performed on the Boolean matrix and the trend consistency indicator in the time dimension to obtain the determination event set in which the key parameters are all in the threshold band within two consecutive sampling periods. The mode switching pre-trigger signal generation logic (parameters: decision event set, key parameter identifier set) is adopted to convert any key parameter that meets the above continuous sampling conditions into a pre-trigger signal and output it to the event buffer unit of the control decision module. By combining multi-dimensional spatial point-domain inclusion detection with continuous period determination, the standardized operating parameters and trend vector data from the previous step are transformed into pre-trigger signals, thereby improving the real-time performance and reliability of mode switching judgment. For example, in a vehicle-mounted sludge treatment system, the standardized operating parameter sequence collected is: torque change rate. , slope of dehydration rate change Vibration amplitude The adaptive threshold band's upper and lower boundary parameter set is: upper limit of torque change rate. Lower limit Upper limit of the slope of the dehydration rate change Lower limit upper limit of vibration amplitude Lower limit Using the interval affixation method, the following calculations are performed on each parameter: Includes torque change rate , Includes the slope of the dehydration rate change , Includes vibration amplitude The generated Boolean matrix is In the trend consistency test, the sign of the change direction of each parameter was calculated and compared with the center line of the threshold band. It was found that the slopes of the torque change rate and the dehydration rate both approached the central critical value, indicating a consistent trend. Within two consecutive sampling periods, the logical AND operation result of the Boolean matrix and the trend indicator remained all true. The system triggers a mode switching pre-signal, which is output to the event buffer unit. Subsequently, the control module starts the pre-switching coordination mechanism. In actual operation, the system response time is significantly shortened and the stability of the switching process is improved. S4.5: Based on the generated mode switching pre-trigger signal, activate the pre-switching coordination mechanism, send the mode preparation instruction to the execution processing module, start the preliminary matching verification of the source mode and target mode control parameters, and initialize the initial weight configuration of the weighted fusion controller to complete the logical connection from state monitoring to transition control. Upon receiving the mode switching pre-trigger signal generated by S4.4, the control logic triggering method (parameters: pre-trigger signal identifier bit, trigger priority rule set) is used to activate the pre-switching coordination mechanism inside the control decision module. Furthermore, through the industrial bus communication control method (parameters: bus type CAN, data rate 1Mbps, frame format and verification protocol), the mode preparation command is sent from the control decision module to the execution processing module at high speed and reliably, and the mode switching preparation signal code stream is obtained.

[0020] Step S5: After confirming that the actual operating parameters have entered the adaptive threshold band, the parallel coordination control logic between the two process modes is activated. A weighted fusion strategy is used to synchronously output control commands for the source mode and the target mode. The initial weight value is biased towards the source mode, and the control weight is gradually transferred from the source mode to the target mode over time according to an exponential decay function, generating a continuous and progressive transition control command sequence. Specifically, this includes: S5.1: Based on the adaptive threshold band generated in the preceding step S3 and the trigger signal indicating that the actual operating parameters determined in S4 have entered the threshold band, a parallel coordinated control logic unit is constructed and activated. This unit, as a non-publicly known custom functional module, is used to manage the coordinated control behavior of the source process mode and the target process mode during the switching transition period. Using the mode switching prediction flag bit from the control decision module and the current operating condition feature vector as input conditions, the state machine inside this unit is initialized, causing it to enter the 'transition operation' mode. The output is the activated parallel coordinated control logic environment, providing an operational support framework for subsequent weighted fusion. S5.2: Obtain the current control instruction set of the source process mode and the expected control instruction set of the target process mode. The control instruction set includes key execution parameters such as torque setpoint, speed adjustment parameters, dehydration chamber pressure configuration, and dosing pump frequency. Based on the scheduling mechanism of the parallel coordinated control logic unit, perform spatiotemporal alignment processing on the two sets of control instruction sets to unify the control cycle and communication protocol format, and generate source mode reference instruction vector and target mode reference instruction vector with consistent structure as the input basis for weighted fusion. S5.3: Construct a control weight transfer curve based on an exponential decay function, where the time variable starts at the formal trigger moment of mode switching, and the decay coefficient is dynamically adjusted according to the current sludge characteristics to ensure that the transition time is extended under high viscosity conditions to enhance stability; calculate the initial weight factor to represent the source mode control weight, and obtain the target mode weight at the same time; the output is a set of dynamic weight pairs that evolve over time to achieve a gradual transfer of control responsibility. Based on the parallel coordinated control logic unit, the aligned source mode reference instruction vector and the target mode reference instruction vector are used as input objects for weight calculation. The control weight transfer curve is constructed using an exponential decay function to achieve smooth control weight evolution in the time dimension. The exponential decay weight calculation method is adopted (parameters: time variable t, decay coefficient α). The time of formal triggering mode switching is taken as the time zero point. By analyzing the sludge moisture content, particle size distribution and equipment load in the current comprehensive working condition feature vector, the decay coefficient α is dynamically set to achieve adaptive adjustment of the weight decay rate under different working conditions. Furthermore, the source pattern control weighting factor is calculated using a formula. Complete the exponential decay modeling of source pattern weights over time, and base it on the identity relation. Obtain the target pattern weight A gradual growth curve over time; Furthermore, the influence of different sludge characteristics on the attenuation coefficient α is calculated by the working condition sensitivity analysis method. This value is mapped to the preset α adjustment range, and α is corrected in real time during the generation of the weight transfer curve to ensure that the weight transfer cycle is extended under high viscous sludge or high load conditions to suppress transient shocks. Furthermore, the generated dynamic weight pairs are analyzed using a weight curve smoothness detection algorithm. Second-order difference analysis is performed to remove discontinuities caused by fluctuations in the α value, and a set of smoothed and optimized weight transfer curves is output to ensure the stability of subsequent weighted fusion processing. By modeling with an exponential decay function and adaptively adjusting α, the instruction vector pairs from the previous step are transformed into dynamic weight pairs that evolve over time, thereby achieving a gradual transfer of control responsibility between the source mode and the target mode and smooth control during the transition period. For example, in a vehicle-mounted sludge treatment task, the system detected a sludge moisture content of 0.78%, an organic matter concentration of 0.35%, a particle size distribution concentrated at 75μm, and an equipment load rate of 0.85. The control decision module sets the mode switching trigger time to t=0 and calls the α dynamic adjustment algorithm based on the operating conditions, obtaining an attenuation coefficient α of 0.12. Under these conditions, the source mode control weight is calculated: =1.0, at t=5s, ≈0.548, thus obtaining the target pattern weight. ≈0.452. Before performing weighted fusion, the discontinuous weights affected by load transients at t=3s were eliminated by second-order difference detection. The final output weight curve maintained a smooth change throughout the entire transition period, driving the execution processing module to achieve a smooth mode switching process with a significant reduction in mechanical vibration and a significant suppression of energy consumption peaks. S5.4: Multiply the source mode reference command vector and the target mode reference command vector by the corresponding dynamic weighting factors, perform element-wise weighted operation, and generate the fused transition control command vector; through this weighted fusion process, an intermediate control command sequence with continuous change characteristics is obtained, avoiding sudden changes in the actuator output and suppressing mechanical vibration and energy consumption peaks. The input condition is the dynamic weighting factor calculated by S5.3. And the source mode reference instruction vector and the target mode reference instruction vector after spatiotemporal alignment and structural unification are completed in S5.2; An element-weighted fusion algorithm is used (parameter: source mode reference instruction vector). Target mode reference instruction vector Dynamic weighting factor This allows for the linear combination of two sets of control parameters according to their weights at the current time. Furthermore, through element-wise multiplication (parameter: , ), to obtain the source mode weighted instruction component matrix, and then perform element-wise multiplication (parameter: , ), to obtain the target mode weighted instruction component matrix; Furthermore, by performing element-wise addition, the source mode weighted instruction component matrix and the target mode weighted instruction component matrix are added together to generate the fused transition control instruction vector. The formula is as follows:

[0021] in, To control the weights for the source pattern, Control the weights for the target pattern. For the i-th source mode control parameter, For the i-th target mode control parameters; Furthermore, a sliding time window smoothing algorithm (parameters: fused transition control command vector, window length L) is used to achieve dynamic smoothing of continuous commands, eliminate output jitter caused by sampling interval differences, and obtain a smooth transition control command sequence. Through the above weighted fusion and smoothing process, the result of the combination of source mode and target mode instructions in the previous step is transformed into an intermediate control instruction sequence with continuous gradual change characteristics, so as to achieve the expected technical effect of avoiding sudden response of the actuator during mode switching and suppressing mechanical vibration and instantaneous energy consumption peaks. For example, in a certain vehicle-mounted sludge treatment system, the source mode reference command vector The main parameters include a torque setting of 180 N·m, a speed adjustment parameter of 950 rpm, a dehydration chamber pressure of 50 kPa, and a dosing pump frequency of 8 Hz; the target mode reference command vector. The main parameters include a torque setting of 220 N·m, a speed adjustment parameter of 1100 rpm, a dehydration chamber pressure of 65 kPa, and a dosing pump frequency of 12 Hz. At the moment the mode switch is officially triggered, the dynamic weighting factor calculation result is... =0.88, = 0.12, using an element-weighted formula to combine the four parameters, such as torque control parameters: N·m. As t increases, It gradually decreases according to exponential decay. It gradually increases, and at t = 6s, =0.45, = 0.55, at this time the speed control parameter is calculated by the fusion formula as follows: The command curve obtained after smoothing at rpm remained monotonically changing throughout the transition period, and finally completed the complete migration to the target mode weight in about 12s. System test results showed that the effective value of mechanical vibration was reduced to less than 70% of the original switching method, the instantaneous power peak was significantly reduced, and the smoothness of the switching process was significantly enhanced. S5.5: The generated transition control command sequence is sent frame by frame to the underlying controller of the execution processing module according to the control cycle to drive the coordinated response of each actuator; at the same time, the actual execution status is fed back in each control cycle to form a closed-loop verification link; based on the actual response deviation of the feedback, the decay coefficient α in the exponential decay function is dynamically fine-tuned to optimize the weight transfer rate in subsequent time periods, and the output is a gradual mode switching process with adaptive adjustment capability to ensure the smooth operation and control continuity of the system during mode transition; The received transition control command vector and dynamic weight pairs after fusion from the S5.4 output ( When used as input, the instruction vector is encapsulated frame by frame into a multi-channel control data frame that can be parsed by the underlying controller according to the scheduling plan of the control cycle. The data is then sent down through the industrial bus interface to drive the hydraulic drive unit, the stirring motor frequency converter and the dewatering device actuator to work together within a unified control cycle, thereby achieving a gradual mode switching behavior. A closed-loop state acquisition method is adopted (parameters: underlying controller feedback bus ID, sampling period Δt, state register mapping table). In each control cycle, the real-time response parameters of all actuators are acquired, including actual torque, actual speed, cavity pressure and dosing pump output frequency, etc., which are assembled into feedback state vectors and written into the state buffer of the control decision module to form a timing feedback record that corresponds one-to-one with the control command. Furthermore, the actual response deviation for each cycle is evaluated using a deviation calculation method (parameters: feedback state vector F(t), target fusion command vector U(t)). The calculation formula is as follows:

[0022] in Let be the instantaneous deviation vector of each control channel. For the feedback state vector, This is the command vector to be issued; Adaptive coefficient tuning algorithm is used (parameter: current attenuation coefficient) Instantaneous deviation vector Deviation tolerance threshold This allows for dynamic fine-tuning of the attenuation curve. The adjustment formula is as follows:

[0023] in This is the adjusted attenuation coefficient. To adjust the gain coefficient, This is the tolerance threshold for deviation; Furthermore, through a weighted recalculation method (parameter: adjusted attenuation coefficient) Time variables Update the dynamic weight pairs to satisfy... and The functional relationship ensures that the control weight transfer rate between the source and target modes is dynamically matched according to the actual execution state; Through the above closed-loop adjustment algorithm, the transition control command and feedback state from the previous step are transformed into self-correcting mechanism data, which is converted into a dynamic weight sequence with adaptive decay capability, thereby achieving the expected technical effects of execution stability and control continuity during mode switching. For example, in a certain vehicle-mounted sludge treatment task, the control cycle Δt of the underlying controller is set to 50ms, and the initial decay coefficient is... =0.05, Deviation Tolerance Threshold =2.0, when the deviation between the pressure feedback value of the dehydration chamber monitored during the initial switching phase and the fusion command is 2.0. At that time, according to Calculate and adjust the attenuation coefficient. =0.125, at this time the source mode weight is 0.125 in the 2nd second. The target mode weight is 0.222, with a value of approximately 0.778, which slows down the control weight transfer rate during the initial switching phase and suppresses sudden torque changes and energy consumption spikes. In subsequent control cycles, the weight is continuously fine-tuned based on dynamic changes in the deviation. Ultimately, this allows each actuator to smoothly switch within 4 seconds and significantly improves the uniformity of material output.

[0024] Step S6: The transition control command sequence is sent to the execution processing module to drive the actuator to smoothly switch the control mode. During the switching process, status feedback signals are continuously collected, including material discharge uniformity, instantaneous energy consumption fluctuations, and mechanical vibration intensity. Based on these signals, a stability evaluation index for the switching process is constructed. Specifically, this includes: S6.1: The continuous and progressive transition control command sequence output by S5 is parsed, the source mode and target mode control command weight allocation parameters corresponding to each time point are extracted, and the weighted fusion control quantity at the current time is calculated based on the exponential decay function to generate a multi-channel coordinated control command package with time dimension consistency to ensure the continuous output of control signals during mode switching. S6.2: The multi-channel coordinated control instruction package generated by S6.1 is sent to the programmable logic controller (PLC) of the execution processing module through the industrial bus interface. This drives the hydraulic drive unit, the frequency converter of the stirring motor and the actuator of the dewatering device to adjust their operating status synchronously, so as to realize the gradual migration of dynamic parameters from the source process mode to the target process mode and avoid mechanical shock or sudden increase in energy consumption caused by sudden instructions. S6.3: During the mode switching execution, multi-dimensional status signals fed back by the execution processing module are collected in real time, including material uniformity detection sensor data at the discharge port, energy consumption fluctuation value output by the instantaneous power acquisition module, and signals from the triaxial vibration sensor installed on the main frame. The above raw signals are then subjected to bandpass filtering and noise reduction processing to obtain a standardized status monitoring data stream with engineering usability. Throughout the entire process of mode switching, the real-time feedback from the execution processing module is used as the input condition, covering three data sources: the material uniformity detection sensor at the discharge port, the instantaneous power acquisition module, and the triaxial vibration sensor installed on the main frame. A high-speed data acquisition method (sampling frequency ≥ 1kHz, resolution ≥ 16bit) is adopted to achieve parallel acquisition of particle size distribution signal output by material uniformity sensor, instantaneous power signal output by power acquisition module and triaxial acceleration signal output by vibration sensor, and automatically add system timestamp to form multi-channel raw state signal packet. Furthermore, through a multi-channel synchronous calibration algorithm (parameters: the reference clock source is the local clock of the control decision module, and the maximum allowed time deviation is ≤10μs), the signals of each channel are strictly aligned on the sampling time reference, and a multi-dimensional original state signal matrix after time-domain synchronization is output; Furthermore, a bandpass filtering method based on sliding window configuration (filtering parameters: lower cutoff frequency of 0.5Hz, upper cutoff frequency of 200Hz, and filtering order of 4th order Butterworth type) is adopted to extract the frequency domain of the material uniformity signal, instantaneous power signal, and triaxial acceleration signal respectively, so as to remove low-frequency drift and high-frequency noise and extract the effective frequency band of the signal. Furthermore, a wavelet denoising algorithm (parameters: wavelet basis is Daubechies 6, decomposition level is 5, and the soft threshold rule adopts the stationary noise estimation method) is applied to achieve multi-scale noise reduction processing of the filtered signal, eliminating transient impulse interference and retaining key operating condition feature components. By using a normalization method (parameter: range normalization based on the maximum and minimum values ​​within the sliding time window), the three types of denoised signals are mapped to a unified dimension space, generating a standardized state monitoring data stream with a numerical range of [0,1], thereby achieving data consistency and comparability in subsequent evaluation steps. For example, in a vehicle-mounted sludge treatment system with a rated processing capacity of 5 m³ / h, the sampling frequency of the raw signal output by the material uniformity detection sensor at the discharge port is set to 2000 Hz, the sampling frequency of the instantaneous power acquisition module is set to 1000 Hz, and the sampling frequency of the triaxial vibration sensor is set to 5000 Hz. The quantization resolution of all channels is 16 bits. During data synchronization calibration, the reference clock source error is controlled within 5 μs to ensure strict consistency of the data from the three channels on the time axis. In the bandpass filtering stage, a 4th-order Butterworth filter is used, with a lower cutoff frequency of 1 Hz and an upper cutoff frequency of 150 Hz. After filtering, the power signal is free from long-term trend drift and high-frequency spike interference. The vibration signal denoising process uses the Daubechies 6 wavelet basis with a decomposition level of 5. The soft threshold is automatically generated by steady-state noise estimation. After denoising, the energy components of the main mechanical structure resonance frequency band are preserved. After normalization, the material uniformity index signal, energy consumption fluctuation signal and vibration intensity signal are all mapped to the [0,1] interval, which can be simultaneously input into the stability assessment algorithm to generate a state monitoring data stream that is usable in engineering and has consistent dimensions, thereby maintaining the stability and assessment accuracy of the data processing link under complex working conditions. S6.4: Based on the standardized condition monitoring data stream obtained in S6.3, calculate the coefficient of variation of material uniformity, the extreme value of energy consumption change rate per unit time, and the effective value of vibration intensity (RMS), and then linearly weight and fuse the above parameters according to preset weights to generate a comprehensive switching process stability evaluation index, which serves as a quantitative basis for judging the smoothness of this mode switching. S6.5: Output the stability evaluation index of the switching process generated in S6.4 to the state discrimination unit of the control decision module for S7 step to determine whether it exceeds the preset safety tolerance range. At the same time, package the index with the corresponding timestamp and operating condition feature vector into a structured log and temporarily store it in the local cache to provide data support for subsequent model optimization and fault tracing.

[0025] Step S7: Determine whether the stability evaluation index generated by S6 exceeds the preset safety tolerance range. If it does not exceed the tolerance range, confirm that the mode switch is successful and maintain the target mode operation; if it exceeds the tolerance range, trigger the rollback mechanism, immediately restore the control command output to the original process mode, and mark the switch as a failure case. Specifically, this includes: S7.1: Obtain the stability evaluation index of the switching process generated by S6. The stability evaluation index includes the output uniformity index, instantaneous energy consumption fluctuation rate and effective value of mechanical vibration intensity. Use these as input conditions and perform over-limit discrimination processing based on the preset multi-dimensional safety tolerance model to identify whether there are abnormal states affecting the stability of the system during the current switching process. S7.2: Dynamically calibrate the threshold parameters of each dimension in the multidimensional safety tolerance model, and use the statistical distribution characteristics (mean and three times the standard deviation) of the stability evaluation index collected under historical normal operating conditions to generate an adaptive safety boundary to eliminate the risk of misjudgment caused by fixed thresholds, and output a dynamic safety tolerance vector as the benchmark for subsequent limit exceedance judgment. S7.3: Based on the stability evaluation index obtained from S7.1 and the dynamic safety tolerance vector generated from S7.2, perform a dimension-by-dimensional comparison operation. If all indicators do not exceed the safety boundary of the corresponding dimension, generate a 'switching successful' status flag and output an instruction signal to the control decision module to maintain the target mode operation. S7.4: If any stability assessment index in any dimension exceeds its corresponding dynamic safety tolerance boundary, it is determined to be a switching instability event, triggering emergency rollback logic. Based on the real-time status feedback channel of the current control architecture, the target mode control command output is immediately cut off, the PID adjustment parameter configuration of the source process mode is restored, and the execution processing module is ensured to return to the stable operating point. S7.5: At the same time as triggering the rollback operation, generate an abnormal event record containing the type of this switchover failure. The record encapsulates the comprehensive working condition feature vector before the switchover, the actual time of entering the threshold band, the adaptive threshold band parameter configuration, and the type of over-limit index in the form of structured data, and marks it as a negative sample label for subsequent abnormal pattern recognition optimization in model training.

[0026] Step S8: Each completed switching event, whether successful or unsuccessful, is stored in the historical event database in a data structure containing the comprehensive operating condition feature vector before switching, the actual switching time, adaptive threshold parameters, and stability evaluation results. This data structure is used to periodically update the training sample set of the incremental support vector regression model described in S2, enabling online learning and adaptive optimization of the model parameters. Specifically, this includes: S8.1: Based on the handover process stability evaluation index constructed in S6, the adaptive threshold band parameter generated in S3, the actual handover time, and the standard working condition feature vector before handover, a raw event record data structure containing multi-dimensional attributes is constructed. The input conditions are the key state parameters of the entire handover event. The structured data encapsulation technology is used for packaging and processing to form basic event samples with spatiotemporal labels and causal correlation. S8.2: Perform validity verification on the generated basic event samples, filter them based on preset data integrity rules and logical consistency criteria (such as timestamp continuity, parameter range compliance, and stability assessment result Boolean identifier), remove abnormal or incomplete records to ensure the quality and reliability of the data entering the database, and output the verified valid event sample set as the input basis for subsequent storage and training. S8.3: The output valid event sample set is written in batches to the historical event database deployed locally on the control decision module according to a fixed time window or cumulative quantity threshold triggering mechanism. The lightweight embedded database management system is used to perform persistent storage operations to build a traceable and backtrackable mode switching experience knowledge base to support subsequent model iteration and fault diagnosis analysis. S8.4: Based on the sliding time window strategy, extract the most recent N valid switching event samples from the maintained historical event database to form a dynamically updated training data subset. The value of N is set according to the computing resource load capacity of the vehicle system. The sample set is divided by data slicing technology to ensure that the data used for model training is timely and representative of the working conditions. S8.5: The lightweight incremental support vector regression model described in S2 is periodically updated using the generated dynamic training data subset. The incremental learning algorithm is used to fine-tune and optimize the model weight matrix and bias terms, avoiding the high computational overhead caused by full retraining, so as to achieve rapid adaptation of the model to new working conditions and continuous evolution of prediction accuracy.

[0027] The present invention also provides an intelligent switching system for multiple process modes of vehicle-mounted sludge treatment, which uses the above-mentioned intelligent switching method for multiple process modes of vehicle-mounted sludge treatment to switch between multiple process modes of vehicle-mounted sludge treatment.

[0028] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0029] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and rules of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent switching of multiple process modes in vehicle-mounted sludge treatment, characterized in that, Includes the following steps: S1: Acquire multi-dimensional sensing data under the current operating conditions of the vehicle-mounted sludge treatment system, and align the multi-dimensional sensing data with timestamps to form a comprehensive operating condition feature vector. S2: Normalize the comprehensive working condition feature vector to generate a standardized working condition feature vector, and input it into the incremental support vector regression model; S3: Utilize the incremental support vector regression model to perform inference calculations on the feature vector of the current standardized operating condition, output the critical value of the control parameter associated with the predicted optimal switching point corresponding to the switching operation between each process mode, and combine it with the confidence index output by the incremental support vector regression model to generate an adaptive threshold band. S4: Monitor the actual operating parameters in the execution processing module in real time, and determine whether the actual operating parameters fall within the adaptive threshold range. If the entry condition is met, start the pre-switching coordination mechanism to prepare for the execution mode transition control. S5: After confirming that the actual operating parameters have entered the adaptive threshold band, activate the parallel coordination control logic between the two process modes, synchronously output the control instructions of the source mode and the target mode, and transfer the control weight from the source mode to the target mode to generate a transition control instruction sequence. S6: The transition control command sequence is sent to the execution processing module to drive the actuator to complete the smooth switching of the control mode. During the switching process, the status feedback signal is continuously collected, and the stability evaluation index of the switching process is constructed based on the status feedback signal.

2. The intelligent switching method for multi-process modes of vehicle-mounted sludge treatment according to claim 1, characterized in that, Following step S6, the following is also included: S7: Determine whether the stability evaluation index exceeds the preset safety tolerance range. If it does not exceed the limit, confirm that the mode switch is successful and maintain the target mode operation. If it exceeds the limit, trigger the rollback mechanism, immediately restore the control command output of the original process mode, and mark the switch as a failure case. S8: Store each completed switching event in the historical event database in a data structure containing the comprehensive operating condition feature vector before switching, the actual switching time, the adaptive threshold parameters, and the stability evaluation results, and use it to periodically update the training sample set of the incremental support vector regression model.

3. The intelligent switching method for multiple process modes in vehicle-mounted sludge treatment according to claim 1, characterized in that, Step S1 specifically includes: Obtain real-time sludge characteristic parameters from the process pattern recognition module to construct an original environmental perception dataset reflecting the current operating environment. The equipment operating status parameters fed back by the synchronous acquisition and execution processing module are timestamped based on the industrial bus communication protocol and incorporated into the original environmental state sensing dataset to form a multi-source heterogeneous sensing data set. Time series alignment processing is performed on each parameter in the multi-source heterogeneous sensing dataset. Interpolation algorithms are used to compensate for the time offset caused by the difference in sensor sampling periods. Based on a unified time reference, all parameters are mapped to the same timestamp node to generate an aligned sensing dataset. The aligned sensor dataset, the original environmental state perception dataset, and the device operating parameters are vectorized and arranged according to a preset dimension order to construct a high-dimensional numerical vector, which serves as the preliminary form of the comprehensive working condition feature vector. The comprehensive operating condition feature vector is subjected to integrity verification and outlier detection. Invalid data items that exceed the reasonable range or have abrupt transitions are identified and removed based on statistical methods. If the verification passes, the comprehensive operating condition feature vector is retained as a valid input.

4. The intelligent switching method for multiple process modes in vehicle-mounted sludge treatment according to claim 3, characterized in that, Step S1 further includes preprocessing the sludge characteristic parameters and environmental state parameters using a third-order low-pass filter, wherein the filter cutoff frequency is set to 1-20Hz, and linearization is performed in combination with the calibration coefficients obtained from laboratory calibration, forming an original sensing dataset with acquisition timestamps.

5. The intelligent switching method for multiple process modes in vehicle-mounted sludge treatment according to claim 1, characterized in that, Step S2 specifically includes: Based on the comprehensive operating condition feature vector, the original data of each dimension are obtained. Using the normalization boundary parameters formed by the maximum and minimum values ​​of each feature in the historical operating data, the range normalization process is performed on the comprehensive operating condition feature vector to generate a preliminary standardized vector. The initial standardized vector is zero-mean processed. Based on the mean vector of the N most recent valid working condition samples within the sliding time window as a dynamic benchmark, the offset of the current vector from the dynamic benchmark is calculated, and the offset is divided by the standard deviation sequence of the corresponding feature to obtain the standardized working condition feature vector. Based on the standardized operating condition feature vectors, as the core input data, combined with the model initialization configuration parameters, the incremental support vector regression model deployed in the control decision module is activated and enters the online inference preparation state. Based on the completion of model initialization, the standardized working condition feature vector is used as an independent input sample at the current moment and input into the loaded incremental support vector regression model. It is then converted into an implicit expression vector in a high-dimensional feature space. Based on the support vector set and Lagrange multiplier sequence currently maintained by the incremental support vector regression model, forward inference calculation is performed to obtain preliminary estimation results. Based on the preliminary estimation results, and combined with the historical prediction error sequence within the sliding time window maintained internally by the incremental support vector regression model, the confidence index of the current prediction result is calculated. The confidence index is then used to adjust the prediction critical value for uncertainty compensation, thereby generating an adaptive threshold parameter set.

6. The intelligent switching method for multiple process modes in vehicle-mounted sludge treatment according to claim 1, characterized in that, Step S3 specifically includes: Based on the output standard operating condition feature vector, it is used as input data and loaded into the incremental support vector regression model deployed in the control decision module. The incremental support vector regression model constructs a nonlinear mapping relationship based on the accumulated historical effective switching event sample set within the sliding time window, performs a forward inference process, and obtains the initial estimation results of the critical values ​​of control parameters required for switching of each process mode under the current operating condition. The initial estimation results of the critical values ​​of the control parameters are subjected to confidence evaluation. The confidence interval width of the current prediction result is calculated using the residual distribution and local sample density information output by the incremental support vector regression model, thus forming a confidence index. Based on the initial estimation results of the critical values ​​of the controlled parameters and the confidence index, a dynamic boundary expansion algorithm is executed. According to the preset confidence-boundary gain mapping function, a variable offset is applied above and below the critical value to generate an adaptive threshold band. The adaptive threshold band is encapsulated into a structured decision parameter package, and a timestamp and operating condition label are added to form a dynamic decision basis data unit.

7. The intelligent switching method for multiple process modes in vehicle-mounted sludge treatment according to claim 6, characterized in that, The structured decision parameter package includes a central critical value, an upper boundary value, a lower boundary value, and corresponding process mode pair identifiers.

8. The intelligent switching method for multiple process modes in vehicle-mounted sludge treatment according to claim 1, characterized in that, Step S4 specifically includes: Based on the adaptive threshold band parameter, a dynamic boundary judgment matrix is ​​constructed, and the interval comparison algorithm in the field of industrial control is used to discretize each boundary to generate a threshold window structure. The actual running parameter stream collected in real time in the execution processing module is obtained and aligned with the local clock of the control decision module through a time synchronization mechanism to form a current running state vector with time consistency. Normalization mapping is performed on each parameter component in the current running state vector to transform it into the same dimension space as the adaptive threshold band, generating a standardized running parameter sequence, and its instantaneous change trend features are extracted based on the sliding sampling window to obtain the trend vector; The standardized operating parameter sequence and its trend vector are subjected to interval classification judgment using a dynamic boundary judgment matrix. A multi-dimensional spatial point-domain inclusion detection algorithm is used to calculate whether each parameter enters the adaptive threshold band range. When any key parameter falls into the threshold band for two consecutive sampling periods, it is determined that the entry condition is met and a mode switching pre-trigger signal is generated. Based on the pre-trigger signal for mode switching, the pre-switching coordination mechanism is activated, a mode preparation instruction is sent to the execution processing module, the initial matching verification of the source mode and target mode control parameters is initiated, and the initial weight configuration of the weighted fusion controller is initialized, thus completing the logical connection from state monitoring to transition control.

9. The intelligent switching method for multiple process modes in vehicle-mounted sludge treatment according to claim 8, characterized in that, The actual operating parameter stream includes the torque change rate, the slope of the dehydration rate change, and the vibration amplitude.

10. A multi-process mode intelligent switching system for vehicle-mounted sludge treatment, characterized in that: The intelligent switching method for multiple process modes of vehicle-mounted sludge treatment as described in any one of claims 1-9 is used to switch between multiple process modes of vehicle-mounted sludge treatment.

Images