Dynamic control optimization method and system for electric snow-removing roller based on torque feedback

Abstract

This invention discloses a dynamic control optimization method and system for electric snowplows based on torque feedback, relating to the field of industrial control technology. The method includes: S1, periodically collecting snowplow operation monitoring data and performing preprocessing on the data; S2, generating a double-end aligned torque sequence, and then purifying and filtering the sequence using a freeze window, a potential pressure anti-phase blocking segment, and attitude transmission disturbance values, outputting an effective lifting response sequence and a net operating torque sequence; S3, calculating target torque value one, target torque value two, and target torque value three, and outputting torque feedback judgment data; S4, generating a basic lifting adjustment amount and a basic speed compensation amount, and implementing dynamic control of the snowplow after directional projection based on the corresponding encoding results. This solves the problems in existing electric snowplow control, such as difficulty in accurately determining double-end torque, lag in lifting adjustment leading to uneven force distribution, and the difficulty in balancing snow removal effect and component lifespan.

Description

Technical Field

[0001] This invention relates to the field of industrial control technology, specifically to a dynamic control optimization method and system for electric snowplows based on torque feedback. Background Technology

[0002] With the continuous development of new energy special-purpose vehicles, electric actuators, data-driven control methods, and multi-constraint optimization control technologies, motion control, attitude control, and load control technologies for complex working conditions are gradually evolving from single-actuator control to multi-variable coupled control. Especially in the fields of engineering equipment, electric vehicles, mobile robots, and special-purpose work equipment, the continuous development of technologies such as dynamic modeling, predictive control, hierarchical optimization, data-driven identification, and data acquisition and control systems has made building control models based on multi-source state variables and outputting execution control quantities an important development direction for current intelligent equipment control technology.

[0003] For example, application CN121050320A relates to a robot motion control method and system based on center-of-mass dynamics and hierarchical optimization, including: establishing a center-of-mass dynamics model of a four-wheeled robot; constructing a nonlinear model predictive control optimization problem based on the center-of-mass dynamics model, combined with center-of-mass-joint coupling dynamic constraints, wheel-ground rolling constraints, and auxiliary constraints; solving the problem using a real-time solution strategy based on discretization and iterative optimization to obtain the optimal state variables and optimal input variables; and solving the optimal torque and wheel angular velocity of the robot using a hierarchical optimization whole-body control strategy based on the optimal state variables and optimal input variables.

[0004] For example, application CN120972642A relates to a data-driven multi-objective optimization control method for intelligent connected electric vehicles and its application. The method extracts a vehicle model for a single intelligent connected electric vehicle and obtains its state representation in a vehicle queue; it constructs a data-driven prediction model based on subspace identification methods and historical data, establishing a linear input-output mapping relationship; it constructs a multi-objective optimization function and executes optimization control based on the multi-objective optimization results; the method is applied to the controller of the intelligent connected electric vehicles in the vehicle queue, where the controller collects acceleration and braking operation data to track torque, outputs the control quantity for executing optimization control, and realizes queue control.

[0005] However, the aforementioned existing technologies mainly address motion planning, trajectory tracking, queue control, or multi-objective optimization of the entire vehicle in scenarios such as four-wheeled legged robots and multi-vehicle cooperative electric vehicles. Their control objects, operating constraints, and load formation mechanisms differ significantly from the operation process of electric snowplows. They do not yet address a technical solution for dynamically and collaboratively controlling the height and driving state of the snowplow during continuous snowplow operation, based on the torque feedback at both ends of the brush, the displacement response of the lifting cylinder, changes in hydraulic load, and the transmission relationship of vehicle posture disturbances. For electric snowplows, the combined effects of changes in snow thickness and density, road surface undulations, snowplow deflection, and changes in vehicle driving state cause continuous fluctuations in the dual-end operating resistance corresponding to the same lifting position. If fixed-position control or simple threshold adjustment methods are still used, problems such as torque response distortion, lifting adjustment lag, uneven brush load, and difficulty in balancing snow removal effect and component lifespan are likely to occur.

[0006] Therefore, in order to address the above problems, there is an urgent need for a dynamic control optimization method and system for electric snowplows based on torque feedback. Summary of the Invention

[0007] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a dynamic control optimization method and system for electric snowplows based on torque feedback. This solves the problems in existing electric snowplow control, such as the difficulty in accurately determining the torque at both ends, the lag in lifting adjustment, resulting in uneven force distribution, and the difficulty in balancing snow removal effect and component lifespan.

[0008] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a dynamic control optimization method for electric snowplows based on torque feedback, comprising: S1, periodically collecting snowplow operation monitoring data, performing preprocessing on the snowplow operation monitoring data, and outputting preprocessed snowplow operation monitoring data; S2, generating left and right end torque misalignment combinations based on the preprocessed snowplow operation monitoring data and calculating the torque alignment step size, generating a double-end aligned torque sequence, and then performing purification and screening on the double-end aligned torque sequence by combining a freeze window, a potential pressure anti-phase blocking segment, and attitude transmission disturbance values, and outputting an effective lifting response sequence and a net S3: Based on the effective lifting response sequence and net working torque sequence, locate the starting anchor point and torque convergence anchor point of the lifting action, extract the regression event segment and generate the leading torque sample value and convergence torque sample value, calculate the target torque value one, target torque value two and target torque value three, and output torque feedback judgment data; S4: Based on the torque feedback judgment data and the corresponding snowplow operation monitoring data, construct the control input window, generate the basic lifting adjustment amount and basic speed compensation amount, and send them to the hydraulic pump station and snowplow motor controller after the execution direction projection according to the corresponding coding result to implement the dynamic control of the snowplow.

[0009] Furthermore, the specific steps for periodically collecting snowplow operation monitoring data and preprocessing the data to output the preprocessed snowplow operation monitoring data are as follows: A fixed-width sliding time window is set as one sampling period. Snowplow operation monitoring data is periodically collected, including the sampling period identifier, left-end torque value, right-end torque value, snowplow motor speed value, snowplow motor output current value, lifting cylinder extension length value, lifting cylinder rodless chamber pressure value, lifting cylinder rod chamber pressure value, and hydraulic pump station output pressure. The data includes force value, snowplow deflection angle value, vehicle speed value, vehicle pitch angle value, and vehicle roll angle value. For the collected snowplow operation monitoring data, a linear time alignment algorithm is used to perform multi-source time-series synchronization processing; a piecewise cubic spline interpolation algorithm is used to perform missing data completion processing; a local outlier anomaly detection algorithm is used to perform outlier data identification and removal processing; a sliding median filtering algorithm is used to perform mechanical impact noise suppression processing; and a Z-score standardization algorithm is used to perform numerical scale unification processing, outputting the preprocessed snowplow operation monitoring data.

[0010] Furthermore, based on the preprocessed snowplow operation monitoring data, the specific steps for generating left and right end torque misalignment combinations and calculating the torque alignment step size are as follows: Read the preprocessed snowplow operation monitoring data according to the sampling period identifier order; construct a local judgment window in units of N consecutive sampling periods; within each local judgment window, maintain the original order of the left-end torque values ​​and sequentially shift the right-end torque values ​​forward according to the preset misalignment step size, generating multiple sets of left and right end torque misalignment combinations; then calculate the change in left-end torque value and the change in right-end torque value between adjacent sampling periods for each set of left and right end torque misalignment combinations. When the change in the left-end torque value and the change in the right-end torque value are both less than the micro-change threshold for M consecutive sampling periods, the corresponding sampling period segment is marked as a frozen window. For each group of left and right-end torque misalignment combinations, the absolute difference between the left-end torque value and the misaligned right-end torque value is calculated for each sampling period, and all absolute differences are accumulated. After deleting the sampling period segment corresponding to the frozen window, the misalignment step size with the smallest cumulative result is taken as the torque alignment step size corresponding to the current local judgment window. Then, the right-end torque value is rearranged according to the torque alignment step size, and the rearranged right-end torque value is re-paired with the left-end torque value according to the sampling period to obtain the double-end aligned torque sequence.

[0011] Furthermore, the specific steps for purifying and filtering the double-ended aligned torque sequence by combining the freeze window, the phase reversal blocking segment, and the attitude transmission disturbance value, and outputting the effective lifting response sequence and net working torque sequence are as follows: Read the lifting cylinder extension length value, the lifting cylinder rodless chamber pressure value, and the lifting cylinder rod chamber pressure value according to the sampling period corresponding to the double-ended aligned torque sequence. Subtract the lifting cylinder rod chamber pressure value from the lifting cylinder rodless chamber pressure value to obtain the lifting net pressure difference value. Calculate the change in lifting cylinder extension length value and the change in lifting net pressure difference value between adjacent sampling periods. When the direction of change in lifting cylinder extension length value is opposite to the direction of change in lifting net pressure difference value, backtrack K sampling periods from the current sampling period and extend K sampling periods backward, marking the corresponding sampling period segment as the phase reversal blocking segment, and deleting the sampling period corresponding to the phase reversal blocking segment from the double-ended aligned torque sequence. Output the remaining sampling periods as the effective lifting response sequence. Based on the sampling period corresponding to the double-end aligned torque sequence, the vehicle pitch angle, vehicle roll angle, snowplow yaw angle, and vehicle speed are read. The square root of the square of the change in vehicle pitch angle plus the square of the change in vehicle roll angle is used to obtain the attitude synthesis term. The absolute value of the snowplow yaw angle (cosine) is added to the absolute value of the snowplow yaw angle (sine) to obtain the yaw projection term. The vehicle speed is incremented by one, the natural logarithm is taken, and then one is added again to obtain the speed amplification. The attitude synthesis term, yaw projection term, and speed amplification are multiplied sequentially to obtain the attitude transmission disturbance value. The absolute value of the change in torque on the left is added to the absolute value of the change in torque on the right to obtain the torque fluctuation. The attitude transmission disturbance value and the torque fluctuation value are matched item by item. When the attitude transmission disturbance value and the torque fluctuation value increase in the same direction, the corresponding sampling period is deleted from the double-end aligned torque sequence, and the remaining sampling period is output as the net working torque sequence.

[0012] Furthermore, based on the effective lifting response sequence and net working torque sequence, the specific steps for locating the lifting action start point and torque convergence anchor point, extracting regression event segments, and generating leading torque sample values ​​and convergence torque sample values ​​are as follows: Following the sampling period identification order, locate the sampling period in the effective lifting response sequence where the change in the lifting cylinder extension length value changes from zero to non-zero, and mark the corresponding sampling period as the lifting action start point; starting from each lifting action start point, calculate the change in left-end torque value and right-end torque value between adjacent sampling periods in the net working torque sequence. When the left-end... When the change in torque value has the same sign as the change in torque value on the right, and the absolute values ​​of the change in torque value on the left and right are both less than the convergence threshold for W consecutive sampling periods, the corresponding sampling period is marked as the torque convergence anchor point. The sampling period segment between the starting anchor point of the lifting action and the torque convergence anchor point is taken as the regression event segment. The arithmetic mean of the left and right torque values ​​corresponding to the starting position of each regression event segment is recorded as the leading torque sample value, and the arithmetic mean of the left and right torque values ​​corresponding to the ending position of each regression event segment is recorded as the convergence torque sample value.

[0013] Further, the specific steps for calculating target torque value one, target torque value two, and target torque value three are as follows: Density peak clustering algorithm is used to extract cluster centers for all converged torque sample values, and the cluster center value corresponding to the position with the highest sample density is taken as target torque value three; using target torque value three as the boundary, all leading torque sample values ​​are divided into a first leading sample set and a second leading sample set located to the left and right of target torque value three; then, semi-supervised Gaussian mixture clustering algorithm is used to extract boundary clusters for the first and second leading sample sets respectively, generating a first torque regression boundary band and a second torque regression boundary band; the absolute difference between each sample value in the first and second torque regression boundary bands and target torque value three is calculated respectively, and the sample value with the smallest absolute difference in the first torque regression boundary band is taken as target torque value one; the sample value with the smallest absolute difference in the second torque regression boundary band is taken as target torque value two; target torque value one, target torque value two, and target torque value three are output.

[0014] Further, the specific steps for outputting torque feedback judgment data are as follows: In the net operating torque sequence, calculate the first absolute difference between the left-end torque value and the target torque value 3, and the second absolute difference between the right-end torque value and the target torque value 3, period by period; locate the continuous sampling period interval where the left-end torque value and the right-end torque value are both lower than the target torque value 1, and the first absolute difference and the second absolute difference decrease simultaneously in the comparison of adjacent sampling periods, and encode it as the descending entry interval; locate the continuous sampling period interval where the left-end torque value and the right-end torque value are both higher than the target torque value 2, and the first absolute difference and the second absolute difference decrease simultaneously in the comparison of adjacent sampling periods, and encode it as the ascending entry interval; locate the sampling period interval where the left-end torque value and the right-end torque value are located on both sides of the target torque value 3 and are not equal to the target torque value 3, and the change in the lifting cylinder extension length value in the effective lifting response sequence is less than the displacement threshold for Z consecutive times, and encode it as the holding interval; output each sampling period interval and the corresponding encoding result as torque feedback judgment data.

[0015] Furthermore, the specific steps for constructing a control input window based on torque feedback judgment data and corresponding snowplow operation monitoring data to generate basic lifting adjustment and basic speed compensation are as follows: Read the torque feedback judgment data and corresponding snowplow operation monitoring data in the order of sampling cycle identifiers. Construct a control input window for Q consecutive sampling cycles. Use the first absolute difference between the left-end torque value and the target torque value, the second absolute difference between the right-end torque value and the target torque value, and the corresponding encoding results to form a torque regression input sequence. Use the changes in the lifting cylinder extension length, the hydraulic pump station output pressure, the snowplow motor output current, the snowplow motor speed, and the vehicle speed to form an execution load input sequence. Input the torque regression input sequence into the first... One branch uses a dilated causal convolutional network algorithm to convolve layer by layer with different dilation coefficients to obtain the torque regression response sequence corresponding to each sampling period. The execution load input sequence is input into the second branch, and a gated recurrent unit network algorithm is used to recursively update it according to the sampling period to obtain the execution load response sequence corresponding to each sampling period. Then, the torque regression response sequence is used as the query sequence, and the execution load response sequence is used as the key-value sequence to input the cross-gated attention fusion algorithm. The gating weight of the torque regression response sequence to the execution load response sequence is calculated according to the sampling period, and the execution load response sequence is weighted and filtered according to the gating weight. Finally, the weighted execution load response sequence is fed back to the corresponding position of the torque regression response sequence, and the basic lifting adjustment amount and basic speed compensation amount corresponding to each sampling period are output.

[0016] Further, the specific steps for implementing dynamic control of the snowplow are as follows: The basic lifting adjustment amount and basic speed compensation amount are input into the monotonic constraint projection algorithm. When the encoding result corresponds to a descent entering the interval, the basic lifting adjustment amount is projected as the extension direction of the lifting cylinder; when the encoding result corresponds to an ascent entering the interval, the basic lifting adjustment amount is projected as the retraction direction of the lifting cylinder; when the encoding result corresponds to a hold interval, the basic lifting adjustment amount and basic speed compensation amount are simultaneously projected to zero; the projected lifting adjustment amount and speed compensation amount are sent to the hydraulic pump station and snowplow motor controller according to the sampling period. The hydraulic pump station controls the extension or retraction of the lifting cylinder based on the lifting adjustment amount, and the snowplow motor controller corrects the speed of the snowplow motor based on the speed compensation amount.

[0017] The second aspect of this invention provides a dynamic control optimization system for an electric snowplow based on torque feedback, comprising: a data acquisition and processing module, a dual-end alignment purification module, a regression judgment generation module, and a linkage adaptive control module, wherein: the data acquisition and processing module is used to periodically acquire snowplow operation monitoring data, perform preprocessing on the snowplow operation monitoring data, and output preprocessed snowplow operation monitoring data; the dual-end alignment purification module is used to generate left and right end torque misalignment combinations based on the preprocessed snowplow operation monitoring data and calculate the torque alignment step size to generate a dual-end aligned torque sequence, and then perform purification and screening on the dual-end aligned torque sequence by combining a freeze window, a potential pressure anti-phase blocking segment, and an attitude transmission disturbance value. The system outputs an effective lifting response sequence and a net operating torque sequence. A regression judgment generation module is used to locate the starting anchor point and torque convergence anchor point of the lifting action based on the effective lifting response sequence and the net operating torque sequence. It extracts regression event segments and generates leading torque sample values ​​and convergence torque sample values, calculates target torque value one, target torque value two, and target torque value three, and outputs torque feedback judgment data. A linkage adaptive control module is used to construct a control input window based on the torque feedback judgment data and the corresponding snowplow operation monitoring data. It generates basic lifting adjustment and basic speed compensation amounts, and sends the results to the hydraulic pump station and snowplow motor controller after direction projection according to the corresponding coding results, implementing dynamic control of the snowplow.

[0018] The present invention has the following beneficial effects: (1) The method and system for dynamic control optimization of electric snowplow based on torque feedback, by generating the torque misalignment combination of the left and right ends and calculating the torque alignment step, can correct the problem of asynchronous torque response at both ends. This can improve the accuracy of torque correspondence when there is a temporal offset in the force on the left and right ends of the snowplow, and provide a consistent data basis for subsequent purification screening and regression judgment.

[0019] (2) The dynamic control optimization method and system of electric snowplow based on torque feedback introduces a freezing window, a pressure-phase blocking section and attitude transmission disturbance value to perform layered elimination of the static micro-change section, displacement-pressure inverse correlation section and attitude disturbance transmission section in the double-end aligned torque sequence. This can distinguish the torque change caused by the actual lifting action from the pseudo disturbance caused by the vehicle body attitude and abnormal hydraulic transmission, thereby improving the effectiveness of the net working torque sequence.

[0020] (3) The method and system for dynamic control optimization of electric snowplow based on torque feedback constructs a regression event segment by locating the starting anchor point of the lifting action and the torque convergence anchor point, and generates target torque value one, target torque value two and target torque value three based on the leading torque sample value and the convergence torque sample value. This can adaptively extract the torque regression boundary and stable torque center during the operation of the electric snowplow, avoiding the problem that a single fixed threshold is difficult to match different working conditions.

[0021] (4) The dynamic control optimization method and system of electric snowplow based on torque feedback, by constructing the torque regression input sequence and the execution load input sequence, generates the basic lifting adjustment amount and the basic speed compensation amount, and combines the corresponding coding results to perform the direction projection and linkage control of the hydraulic pump station and the snowplow motor controller, which can establish a coordinated control relationship between lifting adjustment and drive speed, so that the snowplow height adjustment and drive state correction are carried out synchronously, thereby improving the snow removal stability under dynamic working conditions. Attached Figure Description

[0022] Figure 1 Flowchart of the dynamic control optimization method for electric snowplows based on torque feedback; Figure 2 The structural diagram of the electric snowplow dynamic control optimization system based on torque feedback; Figure 3 Perturbation corridor diagram for attitude transmission; Figure 4 A schematic diagram of a simulated electric snowplow; In the diagram, ① is the snow removal rolling power assembly; ② is the hydraulic pump station; ③ is the transfer case; ④ is the lifting cylinder; ⑤ is the drive shaft; ⑥ is the snow removal roller brush and protective cover; ⑦ is the drive sprocket; ⑧ is the driven sprocket; and ⑨ is the left and right swing cylinders. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figures 1-4 This invention provides a technical solution: a dynamic control optimization method for electric snowplows based on torque feedback, comprising: S1, periodically collecting snowplow operation monitoring data, performing preprocessing on the snowplow operation monitoring data, and outputting preprocessed snowplow operation monitoring data; S2, generating left and right end torque misalignment combinations based on the preprocessed snowplow operation monitoring data and calculating the torque alignment step size to generate a double-end aligned torque sequence, and then performing purification and screening on the double-end aligned torque sequence by combining a freeze window, a potential pressure anti-phase blocking segment, and attitude transmission disturbance values, and outputting an effective lifting response sequence and net operating torque. Sequence; S3, based on the effective lifting response sequence and net working torque sequence, locate the starting anchor point and torque convergence anchor point of the lifting action, extract the regression event segment and generate the leading torque sample value and convergence torque sample value, calculate the target torque value one, target torque value two and target torque value three, and output torque feedback judgment data; S4, based on the torque feedback judgment data and the corresponding snowplow operation monitoring data, construct the control input window, generate the basic lifting adjustment amount and basic speed compensation amount, and send them to the hydraulic pump station and snowplow motor controller after the execution direction projection according to the corresponding coding result to implement the dynamic control of the snowplow.

[0025] Specifically, the steps for periodically collecting snowplow operation monitoring data and preprocessing the data to output the preprocessed snowplow operation monitoring data are as follows: A fixed-width sliding time window is set as one sampling period, with the window length ranging from 200 milliseconds to 800 milliseconds. The sampling period is continuously divided according to the controller's internal clock, and a unique sampling period identifier is written for each period. A full data acquisition is triggered at the beginning of each sampling period. The snowplow operation monitoring data includes the sampling period identifier, left-end torque value, right-end torque value, snowplow motor speed value, snowplow motor output current value, and lifting cylinder... The parameters include: extension length value, rodless chamber pressure value of the lifting cylinder, rod chamber pressure value of the lifting cylinder, hydraulic pump station output pressure value, snowplow deflection angle value, vehicle speed value, vehicle pitch angle value, and vehicle roll angle value. Specifically, the left-end torque value is output by a torque sensor located on the left-end drive link of the snowplow, and the right-end torque value is output by a torque sensor located on the right-end drive link of the snowplow. The snowplow motor speed value is output by the snowplow motor encoder, and the snowplow motor output current value is sampled and output by the motor controller. The lifting cylinder extension length value is output by a linear displacement sensor installed on the lifting cylinder. The rodless chamber pressure value and the rod chamber pressure value of the lifting cylinder are respectively determined by parameters installed on the hydraulic lines of the two chambers of the lifting cylinder. The pressure sensor outputs pressure values ​​from the hydraulic pump station, the hydraulic pump station outlet pressure sensor outputs pressure values, the snowplow deflection angle is output from the angle encoder at the snowplow's pivot point, the vehicle speed is output from the vehicle speed sensor, and the vehicle pitch and roll angles are output from the vehicle attitude sensor. During data acquisition, the current output of each sensor is bound to a sampling period identifier and written into the buffer, ensuring that all data within the same sampling period forms a complete sampling record. This unified sampling period binding method avoids cross-period mismatches in subsequent processing of left and right end torque values, lifting cylinder extension length values, hydraulic pressure data, and attitude data. Linear time is used for the collected snowplow operation monitoring data. The alignment algorithm performs multi-source timing synchronization processing. Specifically, it reads the original timestamp of each data item according to the sampling period identifier, calculates the offset of each data item's timestamp relative to the start time of the sampling period, and then uses the start time of the sampling period as a unified time reference. For data items with non-zero offsets, it linearly maps them to the unified sampling time according to adjacent times. This ensures that the left-end torque value, right-end torque value, lifting cylinder extension length value, pressure data, and attitude data correspond to the same time reference in the same sampling period. The technical principle of this processing method is to use the characteristic that the changes between adjacent sampling points are approximately continuous to compress the original acquisition time difference of the multi-source sensors to a unified time, ensuring that the subsequent torque alignment step size calculation is based on the data at the same moment.A piecewise cubic spline interpolation algorithm is used to perform missing data completion. Specifically, it checks whether there are missing records in each data item according to the sampling period. For data segments with valid values ​​before and after the missing position, a piecewise cubic spline function is constructed based on adjacent valid sampling points, and the interpolation result corresponding to the missing position is calculated. For sampling segments with continuous missing lengths exceeding a preset length, the missing mark is retained and no interpolation is performed. The technical principle of this processing method is that the cubic spline function can maintain first-order and second-order continuity between segments, which is suitable for completing data such as torque, displacement, pressure, and attitude that change smoothly over time, and can avoid sharp jumps caused by direct linear completion. A local outlier anomaly detection algorithm is used to identify abnormal data. The removal process involves constructing local density distributions of left-end torque value, right-end torque value, lifting cylinder extension length value, lifting cylinder rodless chamber pressure value, lifting cylinder rod chamber pressure value, hydraulic pump station output pressure value, snowplow deflection angle value, vehicle speed value, vehicle pitch angle value, and vehicle roll angle value within the neighborhood of each sampling period. The local outlier factor of the current sampling record relative to its neighborhood samples is calculated. When the local outlier factor exceeds a preset outlier threshold, the corresponding sampling record is marked as an anomaly and deleted from the subsequent calculation sequence. The technical principle behind this processing method is that outliers often exhibit significantly lower local density than adjacent sampling points. Utilizing local density differences, it is possible to identify outliers caused by sensor jitter, Isolated outliers caused by transient impacts and sudden communication distortions are addressed using a sliding median filtering algorithm to suppress mechanical impact noise. Specifically, a sliding window is constructed using multiple consecutive sampling periods. Within each window, median replacement is applied to the left-end torque value, right-end torque value, lifting cylinder extension length value, lifting cylinder rodless chamber pressure value, lifting cylinder rod chamber pressure value, hydraulic pump station output pressure value, vehicle pitch angle value, and vehicle roll angle value. The technical principle behind this method is that median calculation can suppress spike impulse noise caused by road bumps, snow impacts, and hydraulic pulsations, while preventing the edges of torque changes and displacement trends from being significantly flattened. A Z-score normalization algorithm is then used for numerical scaling. The unified processing involves calculating the mean and standard deviation of the following values ​​in the current batch: left-end torque value, right-end torque value, lifting cylinder extension length value, lifting cylinder rodless chamber pressure value, lifting cylinder rod chamber pressure value, hydraulic pump station output pressure value, snowplow deflection angle value, vehicle speed value, vehicle pitch angle value, and vehicle roll angle value. Then, each data item is subtracted from its corresponding mean and divided by its corresponding standard deviation to obtain data results under a unified scale. The technical principle of this processing method is to eliminate the original dimensional differences between torque, displacement, pressure, angle, and speed, so that subsequent functions such as frozen window identification, pressure-phase blocking segment identification, attitude transmission disturbance value calculation, and torque regression determination are based on a unified numerical scale.After completing multi-source time-series synchronization processing, missing data completion processing, outlier data identification and removal processing, mechanical impact noise suppression processing, and numerical scale unification processing, the preprocessed snow removal rolling operation monitoring data is output.

[0026] In this implementation plan, by uniformly binding the sampling period, normalizing the time series reference, completing missing data, removing anomalies, suppressing impact noise, and unifying the numerical scale of the snowplow operation monitoring data, it is possible to ensure that the sampling period identifier, left-end torque value, right-end torque value, lifting cylinder extension length value, lifting cylinder rodless chamber pressure value, lifting cylinder rod chamber pressure value, hydraulic pump station output pressure value, snowplow deflection angle value, vehicle speed value, vehicle pitch angle value, and vehicle roll angle value maintain a clear correspondence, continuous trend, and stable numerical distribution within the same sampling period. This improves the reliability of subsequent torque alignment step calculation, freeze window identification, pressure anti-phase blocking segment identification, attitude transmission disturbance value calculation, target torque value generation, torque feedback judgment, basic lifting adjustment amount generation, and basic speed compensation amount generation, so that the dynamic control of the electric snowplow is based on data with consistent time series, controllable quality, and unified scale.

[0027] Specifically, the specific steps for generating left and right end torque misalignment combinations and calculating torque alignment step lengths based on preprocessed snowplow operation monitoring data are as follows: Read the preprocessed snowplow operation monitoring data according to the sampling period identifier. Construct a local judgment window in units of N consecutive sampling periods, where N is 6 to 12 sampling periods. The local judgment window slides recursively along the sampling period identifier by one sampling period, ensuring continuous overlapping sections between adjacent local judgment windows. Within each local judgment window, the left end torque values ​​are arranged in their original order, and the right end torque values ​​are shifted forward sequentially according to a preset misalignment step length, which is 1 to 4 sampling periods, generating multiple sets of left and right end torque misalignment combinations. After the right end torque value is shifted forward, only those values ​​in the same local judgment window as the left end torque value are retained. The effective sampling positions within a fixed window that still have a one-to-one correspondence are used in subsequent calculations. The technical principle behind this approach is that the torque response of the left and right ends of the snowplow is often affected by differences in snow layer entry, elastic lag in the transmission link, and transmission of local attitude disturbances during actual operation, resulting in short-cycle asynchrony. By setting multiple candidate misalignment step sizes and comparing them step by step, the timing difference of the left and right end torques can be compressed to the optimal corresponding state within the same local decision window. Then, for each group of left and right end torque misalignment combinations, the change in the left end torque value and the change in the right end torque value between adjacent sampling periods are calculated. When the change in the left end torque value and the change in the right end torque value are both less than the micro-change threshold for M consecutive sampling periods, the corresponding sampling period segment is marked as a frozen window, where M is 3 to 5 sampling periods and the micro-change threshold is 0.08 to 0 after standardization.15; The determination of the freeze window is implemented using the difference method between adjacent sampling periods. That is, the change in the left-end torque value and the change in the right-end torque value are obtained by subtracting the torque value corresponding to the current sampling period from the torque value corresponding to the previous sampling period. The absolute value is then compared with the micro-change threshold. The technical principle of this method is that the left and right torques remain basically small fluctuations within the freeze window. Such sections do not provide effective discrimination information for the comparison of the misalignment step size. If they are directly included in the accumulation of absolute differences, multiple candidate misalignment step sizes will simultaneously obtain small differences in the static micro-change section, weakening the identification ability of the real time-series offset section. For each combination of left and right torque misalignment, the absolute difference between the left-end torque value and the misaligned right-end torque value is calculated for each sampling period. All absolute differences are accumulated. After deleting the sampling period segment corresponding to the freeze window, the misalignment step size with the smallest accumulated result is taken as the torque alignment step size corresponding to the current local determination window. Among them, when accumulating the absolute difference, only the unaccumulated values ​​are considered. Valid sampling positions marked as frozen windows are accumulated. The accumulated result uses the sum of the absolute differences of all valid sampling positions within the same local judgment window as the discrimination criterion. The technical principle behind this method is that the optimal result corresponding to the torque alignment step size should minimize the overall difference between the left and right torque values ​​within the dynamic change segment. By deleting the frozen window and then comparing the accumulated results, the judgment focus can be concentrated on the actual load change segment. The right torque values ​​are then rearranged according to the torque alignment step size, and the rearranged right torque values ​​are re-paired with the left torque values ​​according to the sampling period, resulting in a double-end aligned torque sequence. Each sampling record in the double-end aligned torque sequence maintains a clear correspondence between the sampling period identifier, the left torque value, and the rearranged right torque value. This ensures that subsequent frozen window identification, potential voltage phase reversal blocking segment identification, attitude transfer disturbance value calculation, and regression event segment truncation are based on the double-end torque data that has undergone time-corrected processing.

[0028] In this implementation scheme, by refining the monitoring data of snowplow operations, including sampling period identification, left and right end torque misalignment combination, small-variation threshold freezing window determination, and torque alignment step size calculation, high accuracy and stability of the data during time alignment can be ensured. By introducing the optimization of the torque alignment step size, errors caused by inconsistent timing or short-term fluctuations are avoided, and static errors in the freezing window are effectively filtered out, making the calculation results of the double-end aligned torque sequence more accurate, thereby improving the response speed and accuracy of subsequent control algorithms. In addition, the rigorous data screening and rearrangement process enhances the reliability of the data, providing more accurate basic data for subsequent attitude disturbance value calculation, regression event segment identification, and control strategy optimization.

[0029] Specifically, the steps for purifying and filtering the double-ended aligned torque sequence by combining the freeze window, the phase-blocking segment of potential pressure, and the attitude transmission disturbance value, and outputting the effective lifting response sequence and the net working torque sequence are as follows: Read the pre-processed lifting cylinder extension length value, lifting cylinder rodless chamber pressure value, and lifting cylinder rod chamber pressure value according to the sampling period corresponding to the double-ended aligned torque sequence. The pre-processed lifting cylinder extension length value, lifting cylinder rodless chamber pressure value, and lifting cylinder rod chamber pressure value are dimensionless data after prior time alignment, anomaly removal, noise suppression, and normalization. The pre-processed lifting cylinder rodless chamber pressure value... The lifting net pressure difference is obtained by subtracting the pre-processed rod chamber pressure of the lifting cylinder from the chamber pressure value. This net pressure difference characterizes the force difference between the two chambers of the lifting cylinder in the current sampling period. An increase in the net pressure difference indicates increased lifting driving force, while a decrease indicates decreased lifting driving force. The actual force change of the lifting cylinder can be extracted from the raw hydraulic pressure data using the net pressure difference. Furthermore, the changes in the pre-processed lifting cylinder extension length and the net pressure difference are calculated between adjacent sampling periods. The change in the pre-processed lifting cylinder extension length characterizes the difference in force between adjacent sampling periods. The direction of displacement change and the change in net lifting pressure difference are used to characterize the direction of hydraulic driving force change between adjacent sampling periods. When the direction of change of the lifting cylinder extension length after preprocessing is opposite to the direction of change of the net lifting pressure difference, it indicates that there is a state in the current sampling period where the direction of displacement response is inconsistent with the direction of hydraulic driving force. This state usually corresponds to abnormal transmission sections caused by hydraulic rebound, mechanical free stroke, local jamming, and short-term inertial delay. If it is directly retained, it will distort the correspondence between subsequent lifting action response and torque change. Therefore, the current sampling period is used as the center to trace back K sampling periods and extend forward K sampling periods. The sampling period is marked as the potential pressure reversal blocking segment, and the sampling period corresponding to the potential pressure reversal blocking segment is deleted from the double-ended aligned torque sequence. K is 2 to 5 sampling periods, preferably 3 sampling periods. The forward and backward extension is used instead of deleting a single sampling period. The reason is that the state where the displacement direction is opposite to the net pressure difference direction generally does not only occur at an isolated moment, but will continue to propagate in several adjacent sampling periods. Forward backtracking and backward extension can completely remove the same abnormal transmission segment, avoiding the abnormal influence remaining in the boundary sampling period. The remaining sampling periods are output as the effective lifting response sequence.Based on the sampling period corresponding to the double-ended aligned torque sequence, the preprocessed vehicle pitch angle, vehicle roll angle, snowplow yaw angle, vehicle speed, left-end torque value, and right-end torque value are read. The preprocessed vehicle pitch angle, vehicle roll angle, snowplow yaw angle, vehicle speed, left-end torque value, and right-end torque value are all dimensionless data after prior dimension unification. The preprocessed vehicle pitch angle is used to characterize the vehicle's front-to-back attitude change; the preprocessed vehicle roll angle is used to characterize the vehicle's left-to-right roll change; the preprocessed snowplow yaw angle is used to characterize the current yaw attitude of the snowplow; and the preprocessed vehicle speed value is used to characterize the intensity of the vehicle's current operating conditions. The changes in the preprocessed vehicle pitch angle and the preprocessed vehicle roll angle are calculated first. The change in pitch angle is taken from the difference between adjacent sampling periods, and the change in roll angle after preprocessing is taken from the difference in roll angle between adjacent sampling periods. The square of the change in pitch angle after preprocessing is added to the square of the change in roll angle after preprocessing, and then the square root is taken to obtain the attitude synthesis term. The purpose of squaring the changes in pitch angle and roll angle after preprocessing is to unify positive and negative changes into non-negative amplitudes, avoiding mutual cancellation when the two types of changes are directly superimposed, so that the pitch disturbance contribution and the roll disturbance contribution can be fully preserved. Adding the squared results can reflect the change intensity of two attitude directions in the same calculation structure. Taking the square root of the sum of squares can preserve the comprehensive amplitude characteristics of the attitude synthesis term on a dimensionless basis, while avoiding excessive numerical amplification caused by directly entering the squared results into subsequent calculations. The absolute values ​​of the cosine and sine of the preprocessed snowplow deflection angle are added together to obtain the deflection projection term. Specifically, taking the cosine of the preprocessed snowplow deflection angle yields the projection component of the snowplow deflection attitude in the longitudinal attitude disturbance transmission direction, while taking the sine of the preprocessed snowplow deflection angle yields the projection component of the snowplow deflection attitude in the lateral attitude disturbance transmission direction. Since the snowplow deflection angle affects the attitude disturbance transmission through geometric projection, using trigonometric functions directly reflects the decomposition effect of deflection angle changes on the disturbance transmission direction. Taking the absolute values ​​of the cosine and sine values ​​aims to preserve the projection intensity without distinguishing between positive and negative deflection directions, ensuring that the deflection projection term always represents the magnitude of the transmission capability. Adding the two together unifies the longitudinal and lateral projection contributions into a single amplified term, allowing the deflection projection term to represent the comprehensive geometric effect of disturbance transmission to the snowplow end under the current deflection attitude.The preprocessed vehicle speed value is incremented by one, and then its natural logarithm is taken, followed by another increment to obtain the speed amplification factor. The purpose of incrementing the preprocessed vehicle speed value by one is to ensure that the input value is always greater than zero, avoiding instability caused by directly taking the natural logarithm when the value is close to zero. Taking the natural logarithm of the preprocessed vehicle speed value after incrementing by one is because the amplification effect of vehicle operating conditions on disturbance transmission increases rapidly in the low-value range and slows down in the higher-value range. Using the natural logarithm allows the differences under lower operating conditions to remain discriminative, and the amplification effect under higher operating conditions to remain gradually changing, preventing the speed value from linearly entering the calculation and excessively dominating the attitude transmission disturbance value. Adding one ensures that the speed amplification factor is always greater than one, guaranteeing that the preprocessed vehicle speed value participates in subsequent product calculations as an amplification factor in all operating conditions. The attitude synthesis term, deflection projection term, and velocity amplification term are multiplied sequentially to obtain the attitude-transmitted disturbance value. The attitude synthesis term represents the basic amplitude of the vehicle's attitude change; the deflection projection term represents the geometric mapping capability of the snowplow deflection attitude to the disturbance transmission direction; and the velocity amplification term represents the amplification effect of vehicle operating conditions on the disturbance propagation intensity. These three terms are multiplied rather than added because attitude changes only significantly affect the transmission of torque fluctuations at both ends when the deflection projection effect exists and is amplified by the operating conditions. The multiplicative form allows the overall transmission value to decrease synchronously when any one factor is weak, which better reflects the actual disturbance transmission mechanism. The absolute value of the pre-processed left-end torque value change is added to the absolute value of the pre-processed right-end torque value change to obtain the torque fluctuation amount. Taking the absolute values ​​of both the pre-processed left-end and right-end torque value changes allows the torque fluctuations to be uniformly converted into amplitude contributions, making the torque fluctuation amount specifically represent the overall intensity of the torque fluctuations at both ends, unaffected by the direction of torque change.The changes in attitude transmission disturbance value and torque fluctuation are calculated based on adjacent sampling periods. The change in attitude transmission disturbance value equals the attitude transmission disturbance value of the current sampling period minus the attitude transmission disturbance value of the previous sampling period, and the change in torque fluctuation equals the torque fluctuation of the current sampling period minus the torque fluctuation of the previous sampling period. When both the change in attitude transmission disturbance value and the change in torque fluctuation are greater than zero, it is determined that the attitude transmission disturbance value and torque fluctuation increase in the same direction in the current sampling period. Specifically, "increasing in the same direction" means that both corresponding changes are positive simultaneously in the same sampling period, rather than solely based on phase... The correlation coefficient or overall trend direction is used to determine the relationship, thereby ensuring that each deleted sampling period has a clear mathematical criterion for each period. In the sampling period that is determined to increase in the same direction, it indicates that the attitude transmission disturbance is enhanced while the torque fluctuation at both ends is enhanced simultaneously. The current torque fluctuation has a direct correspondence with the attitude disturbance transmission. At this time, the change in torque at both ends mainly reflects the transmission effect of attitude disturbance to the snowplow end rather than the actual load change caused by the lifting action. Therefore, the corresponding sampling period is deleted from the double-end aligned torque sequence, and the remaining sampling period is output as the net working torque sequence so that the net working torque sequence more accurately reflects the load change of the snowplow operation.

[0030] The specific formula for calculating the attitude-transferred perturbation value is as follows: ; In the formula, This represents the attitude propagation perturbation value corresponding to the t-th sampling period. This represents the change in the vehicle pitch angle value corresponding to the t-th sampling period. This represents the change in the vehicle roll angle value corresponding to the t-th sampling period. This represents the deflection angle of the snowplow corresponding to the t-th sampling period. This represents the vehicle speed value corresponding to the t-th sampling period.

[0031] In this implementation scheme, by selectively eliminating abnormal displacement response segments and attitude disturbance transmission segments in the double-end aligned torque sequence, the torque changes caused by the actual movement of the lifting cylinder can be separated from the additional fluctuations caused by changes in vehicle body attitude, deflection attitude, and amplified operating conditions. This ensures that both the effective lifting response sequence and the net working torque sequence have clear response relationships, clear sources of disturbance, and stable trends. This improves the accuracy of subsequent lifting action start point positioning, torque convergence anchor point positioning, regression event segment extraction, leading torque sample value generation, convergence torque sample value generation, target torque value one generation, target torque value two generation, target torque value three generation, and torque feedback judgment data output. This allows the dynamic control of the electric snowplow to be based on torque data with more realistic force transmission relationships and more reliable lifting response relationships.

[0032] In this embodiment, Table 1 is a data table of attitude transmission disturbance values, listing the calculated data of attitude transmission disturbance values ​​under five sampling periods. Specifically: In sampling period T1, the change in vehicle pitch angle is 0.6, the change in vehicle roll angle is 0.4, the snowplow yaw angle is 5, the vehicle speed is 3.2, and the attitude transmission disturbance value is 1.90. In sampling period T2, the change in vehicle pitch angle is 1.1, the change in vehicle roll angle is 0.7, the snowplow yaw angle is 8, the vehicle speed is 4.0, and the attitude transmission disturbance value is 3.84. In sampling period T3, the change in vehicle pitch angle is 1.4, the change in vehicle roll angle is 1.0, the snowplow yaw angle is 12, the vehicle speed is 4.8, and the attitude transmission disturbance value is 5.63. In sampling period T4, the change in vehicle pitch angle was 0.9, the change in vehicle roll angle was 0.8, the snowplow yaw angle was 10, the vehicle speed was 3.6, and the attitude transmission disturbance was 3.52. In sampling period T5, the change in vehicle pitch angle was 0.5, the change in vehicle roll angle was 0.3, the snowplow yaw angle was 6, the vehicle speed was 2.8, and the attitude transmission disturbance was 1.50. This data table reflects the differences in the transmission of attitude disturbances to the snowplow end under different sampling periods, and provides a basis for subsequently mapping attitude transmission disturbance values ​​to torque fluctuations, deleting sampling periods with significant attitude disturbance transmission, and outputting the net operating torque sequence.

[0033] Table 1. Attitude Transmitted Perturbation Values

[0034] like Figure 3As shown in the figure, the distribution of attitude-transmitted disturbance values ​​under five sampling periods is displayed in an expanded manner. The horizontal axis represents the sampling period identifiers T1 to T5, and the vertical axis represents the attitude-transmitted disturbance value. The diamond-shaped corridor area corresponding to each sampling period represents the attitude disturbance transmission range under the current sampling period, used to characterize the disturbance intensity range formed by the transmission of vehicle attitude changes to the snowplow end after being amplified by the snowplow deflection projection and vehicle speed. The A value marked at the center of each diamond-shaped corridor area represents the attitude-transmitted disturbance value corresponding to the current sampling period, used to characterize the comprehensive disturbance level after the combined effect of the changes in vehicle pitch angle and vehicle roll angle under this sampling period, after being amplified by deflection projection and speed. The θ value marked above each diamond-shaped corridor area represents the snowplow deflection angle value under the corresponding sampling period, used to characterize the deflection projection state when the attitude disturbance is transmitted to the snowplow end. The v value marked below each diamond-shaped corridor area represents the vehicle speed value under the corresponding sampling period, used to characterize the amplification degree of attitude disturbance transmission under the current operating condition. The attitude transmission disturbance value corresponding to sampling period T1 in the figure is relatively small, indicating that the disturbance caused by the vehicle attitude change to the snowplow end is relatively weak under the current sampling period. The attitude transmission disturbance value corresponding to sampling period T2 increases, indicating that the influence of the vehicle attitude change on the torque fluctuation at both ends is enhanced after deflection projection. The attitude transmission disturbance value corresponding to sampling period T3 is the largest, reflecting that the disturbance transmission from the vehicle attitude change to the snowplow end is most obvious under the combined effect of a large snowplow deflection angle and a high vehicle speed. The attitude transmission disturbance value corresponding to sampling period T4 decreases somewhat, but still remains at a high level. The attitude transmission disturbance value corresponding to sampling period T5 decreases again, indicating that the influence of attitude disturbance on the transmission of torque fluctuation at both ends is weakened under the current sampling period. Figure 3 This invention visually demonstrates the distribution characteristics of attitude transmission disturbance values ​​as the sampling period changes. It can reflect the differences in disturbance transmission under the combined effects of changes in vehicle pitch angle, vehicle roll angle, snowplow deflection angle, and vehicle speed. It also provides a basis for subsequent attitude disturbance removal and output of net working torque sequence for double-end aligned torque sequence.

[0035] Specifically, the steps for locating the lifting action start point and torque convergence anchor point based on the effective lifting response sequence and net operating torque sequence, and for extracting regression event segments and generating leading-edge torque sample values ​​and convergence torque sample values ​​are as follows: Following the sampling period identification order, locate the sampling period in the effective lifting response sequence where the change in the lifting cylinder extension length value changes from zero to non-zero, and mark the corresponding sampling period as the lifting action start point; The change in the lifting cylinder extension length value is obtained by subtracting the lifting cylinder extension length value corresponding to the current sampling period from the lifting cylinder extension length value corresponding to the previous sampling period. When the change in the lifting cylinder extension length value is zero in the previous sampling period and becomes non-zero in the current sampling period, it indicates that the lifting cylinder has entered the action state from a static state. At this time, the corresponding sampling period can characterize the starting position of a real lifting adjustment entering the torque regression process. The technical principle of this determination method is that the lifting cylinder extension length value is a direct representation of the change in snowplow height. When the change in the extension length of the lifting cylinder changes from zero to non-zero, it means that the lifting mechanism has begun to apply actual displacement to the snowplow's posture, and therefore can be used as the starting reference for the regression event segment. Starting from the initial anchor point of each lifting action, the changes in the left and right torque values ​​between adjacent sampling periods are calculated in the net working torque sequence. The change in the left torque value is calculated by subtracting the left torque value corresponding to the current sampling period from the left torque value corresponding to the previous sampling period, and the change in the right torque value is calculated by subtracting the right torque value corresponding to the current sampling period from the right torque value corresponding to the previous sampling period. When the changes in the left and right torque values ​​have the same sign, and the absolute values ​​of the changes in the left and right torque values ​​are simultaneously less than the convergence threshold for W consecutive sampling periods, the corresponding sampling period is marked as the torque convergence anchor point. W is 3 to 6 sampling periods, preferably 4 sampling periods, and the convergence threshold is 0.04 to 0.12 after standardization, preferably 0.08. Using the same sign for the change in torque value at both ends as a prerequisite criterion ensures that the torque at both ends changes in a consistent manner within the current segment, preventing unstable segments where one end is still increasing while the other has decreased from being mistakenly identified as convergence segments. Using W consecutive sampling periods simultaneously being less than the convergence threshold as a torque convergence criterion eliminates false convergence caused by short-term fluctuations in a single sampling period, ensuring that the segment corresponding to the torque convergence anchor point exhibits continuous stability. The technical principle is that after the snowplow height adjustment is truly completed, the torque values ​​at both ends within adjacent sampling periods... The amplitude of the change will decrease synchronously and remain within a small range. Therefore, when multiple consecutive sampling periods simultaneously meet the condition of small change, it can be characterized that the torque has entered a relatively stable state. The sampling period segment between the starting anchor point of the lifting action and the torque convergence anchor point is taken as the regression event segment. Within the regression event segment, the sampling period that exists simultaneously in the effective lifting response sequence and the net working torque sequence is selected as the sample extraction period. This ensures that the lifting action response information and the torque regression information maintain a one-to-one correspondence on the time axis, avoiding sequence misalignment caused by the deletion of the phase blocking segment of the pressure reversal and the elimination of attitude disturbances, which would affect subsequent sample extraction. The arithmetic mean of the left and right torque values ​​corresponding to the starting position of each regression event segment is recorded as the leading torque sample value, and the arithmetic mean of the left and right torque values ​​corresponding to the ending position of each regression event segment is recorded as the convergent torque sample value. The purpose of using the arithmetic mean of the left and right torque values ​​as sample values ​​is to unify the dual-end force state into a single torque representation, avoiding the influence of local force bias on the sample values ​​caused by using only single-end torque values. The leading torque sample value is used to represent the current load level of the snowplow at the beginning of the lifting action, and the convergent torque sample value is used to represent the load level when the dual-end torque returns to a stable state after the lifting action is completed. Through the leading and convergent torque sample values, the load start and load end points corresponding to a complete lifting adjustment process can be extracted, providing a time-sequential sample basis for the generation of subsequent target torque values ​​one, two, and three.

[0036] In this implementation scheme, by establishing a continuous correspondence between the lifting action start point, torque convergence point, regression event segment, leading torque sample value, and convergence torque sample value in the effective lifting response sequence and net working torque sequence, the load start state and load stable state in a real lifting adjustment process can be completely separated from the continuous sampling data. This makes the regression event segment have clear boundaries, clear time sequence, and unified force representation at both ends, thereby improving the stability of the target torque value generation, target torque value generation, target torque value generation, and torque feedback judgment data output. This ensures that the subsequent dynamic control of the electric snowplow is based on a traceable lifting action response process and a quantifiable torque regression process.

[0037] Specifically, the steps for calculating target torque value one, target torque value two, and target torque value three are as follows: Density peak clustering algorithm is used to extract cluster centers for all converged torque sample values. Specifically, a sample distance matrix is ​​first constructed according to the numerical value of the converged torque sample values. Then, all distances between samples in the distance matrix are sorted from smallest to largest. The average distance within the 10th to 20th percentile range of the sorted results is taken as the cutoff distance, preferably the distance value corresponding to the 15th percentile. Subsequently, the local density and relative distance of each converged torque sample value are calculated sequentially. The value corresponding to the sample center with the largest product of local density and relative distance is taken as the cluster center value. The cluster center value corresponding to the position with the highest sample density is taken as target torque value three. The technical principle of using density peak clustering algorithm is that the converged torque sample values ​​originate from the end of the regression event segment, and the numerical distribution usually forms a high-density cluster around the stable load range. By using the joint determination of local density and relative distance, the stable load center can be directly separated from all converged torque sample values, avoiding the bias of a small number of abnormal converged samples on target torque value three. Using the target torque value of 3 as a boundary, all leading torque sample values ​​are divided into a first leading sample set and a second leading sample set located to the left and right of the target torque value of 3. Among them, leading torque sample values ​​less than the target torque value of 3 are written into the first leading sample set, and leading torque sample values ​​greater than the target torque value of 3 are written into the second leading sample set. The purpose of using the segmentation method based on both sides of the target torque value of 3 is to separate the load initial state before regression according to the low load side and the high load side, so as to avoid weakening the boundary distribution characteristics after samples from both sides are mixed into the same clustering process. Then, a semi-supervised Gaussian mixture clustering algorithm is used to extract boundary clusters for the first and second frontier sample sets, respectively, to generate the first and second torque regression boundary bands. Specifically, 2 to 4 Gaussian components are constructed in the first frontier sample set, preferably 3 Gaussian components, and 2 to 4 Gaussian components are constructed in the second frontier sample set, preferably 3 Gaussian components. The sample distribution intervals on both sides of the target torque value are used as the initial constraint boundaries. Then, the mean, variance, and mixture weight of each Gaussian component are updated using the expectation-maximization iterative method. The number of iterations is 50 to 200, preferably 100. Iteration stops when the log-likelihood difference between the first and last iterations is less than 0.001. The technical principle of using the semi-supervised Gaussian mixture clustering algorithm is that the frontier torque sample values ​​located on both sides of the target torque value usually no longer exhibit a single dense center, but are more likely to show a multi-peak structure that is dispersed and clustered along the regression boundary. With the boundary prior provided by the target torque value, and by fitting the sample distribution through multiple Gaussian components, the boundary sample intervals closest to the stable region in the regression frontier state can be extracted.The absolute difference between each sample value in the first torque regression boundary band and the second torque regression boundary band and the target torque value three is calculated respectively. The sample value with the smallest absolute difference in the first torque regression boundary band is taken as the target torque value one; the sample value with the smallest absolute difference in the second torque regression boundary band is taken as the target torque value two. Among them, the target torque value one represents the boundary value of the low load side entering the torque regression zone, the target torque value two represents the boundary value of the high load side entering the torque regression zone, and the target torque value three represents the stable center value after torque regression. The three are determined in the order of target torque value one < target torque value three < target torque value two. The purpose of using the boundary sample with the smallest absolute difference with the target torque value three as the target torque value one and target torque value two is to select the starting boundary closest to the stable center from the first torque regression boundary band and the second torque regression boundary band, so that the subsequent determination of entering the descent interval, determining the entering the ascending interval, and determining the holding interval are based on the boundary value closest to the actual torque regression process. The target torque value one, target torque value two, and target torque value three are output.

[0038] In this implementation scheme, by simultaneously incorporating the stable clustering center in the convergent torque sample values ​​and the boundary distribution of the leading torque sample values ​​on both sides of the target torque value three into the unified generation process, the target torque value one, the target torque value two, and the target torque value three can correspond to the low load side regression boundary, the high load side regression boundary, and the stable load center, respectively. This ensures that the three maintain clear positional relationships, consistent sample sources, and clear regression levels, thereby improving the fit of the torque feedback judgment data to actual load changes. It also ensures that the descent entry interval, the rise entry interval, and the holding interval are established on the torque boundary basis that better conforms to the regression law of real operation.

[0039] Specifically, the steps for outputting torque feedback judgment data are as follows: In the net operating torque sequence, calculate the first absolute difference between the left-end torque value and the target torque value three, and the second absolute difference between the right-end torque value and the target torque value three, periodically. The first absolute difference is the absolute value of the left-end torque value minus the target torque value three, and the second absolute difference is the absolute value of the right-end torque value minus the target torque value three. The first absolute difference characterizes the degree of deviation of the left-end load state from the stable center, and the second absolute difference characterizes the degree of deviation of the right-end load state from the stable center. Using absolute values ​​eliminates the directional cancellation effect of high or low torque in numerical comparisons, ensuring that the amplitudes of the torque deviations from the stable center at both ends are uniformly entered into the subsequent calculations. Continued interval determination; locate the continuous sampling period interval where the left-end torque value and the right-end torque value are simultaneously lower than the target torque value, and the first absolute difference and the second absolute difference decrease simultaneously in adjacent sampling periods, and encode this as the descent entry interval; where synchronous decrease is determined by adjacent sampling periods, i.e., the first absolute difference corresponding to the current sampling period is less than the first absolute difference corresponding to the previous sampling period, and the second absolute difference corresponding to the current sampling period is less than the second absolute difference corresponding to the previous sampling period. When the synchronous decrease condition is met for 3 to 6 consecutive sampling periods, preferably 4 consecutive sampling periods, the corresponding sampling period segment is written into the descent entry interval. The technical principle of this determination method is that when the left-end torque value and the right-end torque value are simultaneously lower than the target torque value, the first absolute difference and the second absolute difference decrease simultaneously in adjacent sampling periods. When the deviation from the target torque value 1 is below the target torque value 3 and continues to converge, it indicates that the load lifting process caused by the extension action of the lifting cylinder is effectively eliminating the low load deviation state. At this time, the corresponding segment can characterize the snowplow's return process after downward adjustment. The continuous sampling period interval where the left-end torque value and the right-end torque value are both higher than the target torque value 2, and the first absolute difference and the second absolute difference decrease simultaneously in the comparison of adjacent sampling periods, is coded as the rising entry interval. Among them, the rising entry interval adopts the continuous convergence judgment method of adjacent sampling periods, and the number of continuous sampling periods is 3 to 6, preferably 4. When the left-end torque value and the right-end torque value are above the target torque value 2 and continue to return to the target torque value 3, it indicates that the lifting cylinder retracts. The load reduction process caused by the operation is effectively eliminating the high load deviation state. At this time, the corresponding section can characterize the snowplow's return process after upward adjustment. The sampling period interval that satisfies the condition that the left and right torque values ​​are located on both sides of the target torque value 3 and are not equal to the target torque value 3, and the change in the extension length of the lifting cylinder in the effective lifting response sequence is less than the displacement threshold for Z consecutive times is encoded as the holding interval. Among them, "on both sides of the target torque value 3" means that the left torque value is less than the target torque value 3 and the right torque value is greater than the target torque value 3, and "another sampling period" means that the left torque value is greater than the target torque value 3 and the right torque value is less than the target torque value 3. Z takes 3 to 8 sampling periods, preferably 5 sampling periods, and the displacement threshold is 0 after standardization.The optimal value is 0.05, ranging from 0.3 to 0.10. The technical principle behind this determination method is that when the left and right torque values ​​are located on either side of the target torque value, and the change in the extension length of the lifting cylinder remains minimal, it indicates that the current snowplow height has entered a fine-tuning termination state. The force at both ends is within a small oscillation range around the stable center. At this point, the corresponding segment can represent the completion of the torque regression process. Each sampling period interval and its corresponding encoding result are output as torque feedback determination data. This torque feedback determination data includes at least the sampling period identifier, the descent entry interval encoding result, the ascent entry interval encoding result, and the hold interval encoding result. This ensures that subsequent generation of basic lifting adjustment and basic speed compensation is based on determination results with clear interval boundaries, clear regression direction, and clear stable state.

[0040] In this implementation plan, by unifying the deviation relationship of the two ends of the net working torque sequence, the boundary position relationship of target torque value one, target torque value two, and target torque value three, and the change state of the lifting cylinder extension length value into the same judgment process, the descent entry interval, the ascent entry interval, and the holding interval can correspond to three different working states: load lifting regression, high load unloading regression, and stable holding regression, respectively. This allows the torque feedback judgment data to simultaneously possess the characteristics of clear interval boundaries, clear regression direction, and distinguishable stable states. This improves the accuracy of the subsequent generation of basic lifting adjustment amount and basic speed compensation amount in response to changes in actual working load, and enables the dynamic control of the electric snowplow to be based on feedback basis with clearer judgment levels and more reliable state switching.

[0041] Specifically, the steps for constructing a control input window based on torque feedback judgment data and corresponding snowplow operation monitoring data, and generating basic lifting adjustment and basic speed compensation amounts are as follows: Read the torque feedback judgment data and corresponding snowplow operation monitoring data in the order of the sampling period identifier, constructing a control input window for Q consecutive sampling periods, where Q is 4 to 10 sampling periods, preferably 6 sampling periods. The control input window slides along the sampling period identifier in increments of 1 sampling period, ensuring continuous overlap between adjacent control input windows to maintain the continuous temporal correlation between the torque regression process and the load change process; The first absolute difference between the left-end torque value and the target torque value, and the right-end torque value and the target torque value are then compared. The second absolute difference between the three values ​​and their corresponding encoding results form the torque regression input sequence. The first absolute difference characterizes the deviation of the left-end torque value from the stable torque center, the second absolute difference characterizes the deviation of the right-end torque value from the stable torque center, and the corresponding encoding result characterizes which type of regression state the current sampling period belongs to: descending into the interval, ascending into the interval, or maintaining the interval. Writing these three types of data into the same sequence according to the sampling period order allows for the unified compression of the double-end torque deviation intensity, double-end regression direction, and current operating state into a single time-series input chain. This includes the changes in the lifting cylinder extension length, the hydraulic pump station output pressure, the snowplow motor output current, and the snowplow motor speed. The vehicle speed value constitutes the execution load input sequence. The change in the extension length of the lifting cylinder characterizes the current lifting action amplitude; the change in the output pressure of the hydraulic pump station characterizes the change in hydraulic execution resistance; the change in the output current of the snowplow motor characterizes the change in electric drive load; the change in the speed of the snowplow motor characterizes the change in rotational execution state; and the vehicle speed value characterizes the intensity of the operating condition. Arranging these five types of data in the order of sampling period allows the displacement load characteristics, hydraulic load characteristics, electric drive load characteristics, and operating condition characteristics of the execution end to be uniformly written into the same time-series input chain. The torque regression input sequence is input into the first branch, and a dilated causal convolutional network algorithm is used to convolve layer by layer with different dilation coefficients to obtain the results for each sampling period. The corresponding torque regression response sequence is obtained; the number of convolutional layers in the dilated causal convolutional network algorithm is 3 to 5, preferably 4, the dilation coefficient is set to increase in increments of 1, 2, 4, and 8, the kernel length is 2 to 5, preferably 3, and the number of convolutional channels is 16 to 64, preferably 32. The technical principle of using causal convolution is that the output of each sampling period depends only on the current sampling period and the input of the historical sampling period, which can prevent information leakage in future moments; the technical principle of using dilated convolution is that by expanding the receptive field to cover multiple sampling periods, the first branch can sense both short-term torque regression fluctuations and regression convergence trends across multiple sampling periods, thereby converting the first absolute difference, the second absolute difference, and the corresponding encoding results into a torque regression response sequence;The execution load input sequence is input into the second branch, and a gated cyclic unit network algorithm is used to recursively update it according to the sampling period to obtain the execution load response sequence corresponding to each sampling period. The number of hidden units in the gated cyclic unit network algorithm is 16 to 64, preferably 32, and the number of recursion layers is 1 to 3, preferably 2. The technical principle of the gated cyclic unit network algorithm is to use update gates and reset gates to retain and forget historical execution states, enabling the second branch to extract displacement inertia and hydraulic delay from changes in the lifting cylinder extension length, hydraulic pump station output pressure, snowplow motor output current, snowplow motor speed, and vehicle speed. The timing memory characteristics corresponding to hysteresis, electric drive lag, and continuous operating conditions are used to transform the execution load input sequence into an execution load response sequence. Then, the torque regression response sequence is used as the query sequence, and the execution load response sequence is used as the key-value sequence, input into a cross-gated attention fusion algorithm. The gating weight of the torque regression response sequence on the execution load response sequence is calculated according to the sampling period, and a weighted filtering is performed on the execution load response sequence based on the gating weight. Specifically, the number of attention heads in the cross-gated attention fusion algorithm is 2 to 6, preferably 4; the fusion dimension is 16 to 64, preferably 32; and the gating weight ranges from 0 to 1. A larger gating weight indicates a higher execution load state in the current sampling period. The stronger the influence of torque regression on the control output, the smaller the gating weight, indicating a stronger dominance of the torque regression state in the current sampling period on the control output. The technical principle of the cross-gating attention fusion algorithm lies in the active query of the key states in the execution load response sequence from the torque regression response sequence. Based on the query results, the parts of the execution load response sequence with high correlation to the current torque regression demand are strengthened and retained, while the parts with low correlation are suppressed and weakened. This avoids redundant fluctuations in the execution load input sequence from directly interfering with the control output. Subsequently, the weighted execution load response sequence is fed back to the corresponding position in the torque regression response sequence, outputting the basic lift adjustment amount and the base load adjustment amount for each sampling period. The basic speed compensation quantity; where the basic lifting adjustment quantity is used to characterize the displacement adjustment range that should be applied to the lifting cylinder in the current sampling period, and the basic speed compensation quantity is used to characterize the speed correction range that should be applied to the snow removal roller motor in the current sampling period. The technical principle of the feedback fusion output method is to remap the constraint effect of the execution load state on the control action back to the torque regression main chain, so that the final control quantity not only follows the demand change of the two-end torque value to the target torque value three-fold regression, but is also subject to the real-time constraints of the lifting execution state, hydraulic pressure state, motor load state, and operating condition state, thereby ensuring that the generated basic lifting adjustment quantity and basic speed compensation quantity have both regression driving force and executable feasibility.

[0042] In this implementation plan, by organizing the torque feedback judgment data and snowplow operation monitoring data in the same control input window, and by incorporating the first absolute difference, the second absolute difference, the corresponding coding result, the change in the extension length of the lifting cylinder, the change in the output pressure of the hydraulic pump station, the change in the output current of the snowplow motor, the change in the speed of the snowplow motor, and the vehicle speed into the generation process, the basic lifting adjustment amount and the basic speed compensation amount can simultaneously possess the characteristics of clear torque regression direction, sufficient execution load constraint, and continuous temporal correlation. This improves the matching degree between the subsequent directional projection results and the actual operation state, and enables the dynamic control of the electric snowplow to be based on the control basis of the coordinated consistency between torque regression demand and execution load state.

[0043] Specifically, after performing directional projection based on the corresponding encoding results, the data is sent to the hydraulic pump station and snowplow motor controller. The specific steps for implementing dynamic control of the snowplow are as follows: The basic lifting adjustment amount and the basic speed compensation amount are input into the monotonic constraint projection algorithm. The basic lifting adjustment amount characterizes the displacement adjustment amplitude that the lifting cylinder should perform in the current sampling period, and the basic speed compensation amount characterizes the speed correction amplitude that the snowplow motor should perform in the current sampling period. The monotonic constraint projection algorithm performs directional constraints, zero-value freezing, and amplitude limiting processing on the basic lifting adjustment amount and the basic speed compensation amount under the interval encoding constraints corresponding to the current sampling period, and converts the basic lifting adjustment amount and the basic speed compensation amount from continuous values. The space is mapped to a bounded control set consistent with the current torque feedback judgment state to avoid the basic lifting adjustment amount from outputting in the retraction direction within the descent entry range, to avoid the basic lifting adjustment amount from outputting in the extension direction within the rise entry range, and to avoid the basic speed compensation amount from continuing to output disturbance correction amount within the holding range. The bounded control set is jointly determined by the allowable control direction, zero output state, and upper limit of amplitude corresponding to the current encoding result. The monotonic constraint means that the direction of change of the projected lifting adjustment amount is consistent with the control direction corresponding to the current encoding result, and the output of control direction amount opposite to the current encoding result is not allowed within the same encoding range, and the output of non-zero correction amount is not allowed within the holding range.The monotonic constraint projection algorithm is specifically implemented using a bounded mapping method driven by interval coding. Let the basic lifting adjustment amount be U, the basic speed compensation amount be V, the projected lifting adjustment amount be U', the projected speed compensation amount be V', the upper bound of the lifting adjustment amplitude be Um, and the upper bound of the speed compensation amplitude be Vm. The upper bounds of both the lifting adjustment amplitude Um and the speed compensation amplitude Vm are determined based on the equipment's executable capability and operational safety boundaries corresponding to the current sampling period. The upper bound of the lifting adjustment amplitude Um is obtained by reading the current extension length value of the lifting cylinder and the lifting speed compensation amplitude. The maximum permissible extension length of the lifting cylinder, the minimum permissible extension length of the lifting cylinder, and the maximum permissible displacement adjustment of the hydraulic pump station in the current sampling period are determined. The remaining extendable stroke in the current sampling period is obtained by subtracting the current extension length from the maximum permissible extension length. The remaining retractable stroke in the current sampling period is obtained by subtracting the minimum permissible extension length from the current extension length. The remaining extendable stroke and remaining retractable stroke are then compared with the maximum permissible displacement adjustment in the current sampling period. The upper limit of the lifting adjustment amplitude Um is determined by comparing the values ​​and taking the smaller value as the upper limit of the lifting adjustment amplitude Um in the current sampling period. This ensures that the projected lifting adjustment amount does not exceed the mechanical stroke boundary corresponding to the current position of the lifting cylinder, nor does it exceed the maximum displacement adjustment amplitude allowed by the current output capacity of the hydraulic pump station in a single period. The upper limit of the speed compensation amplitude Vm is determined by reading the current speed value, the maximum allowable speed value, the minimum allowable speed value of the snow removal roller motor, and the maximum allowable speed change in a single period corresponding to the current sampling period. The positive remaining speed margin in the current sampling period is obtained by subtracting the current speed value of the snow removal roller motor from the maximum allowable speed value, and the negative remaining speed margin in the current sampling period is obtained by subtracting the minimum allowable speed value of the snow removal roller motor from the current speed value. The positive and negative remaining speed margins are then compared with the maximum allowable speed change in a single period, and the smaller value is taken as the upper limit of the speed compensation amplitude Vm in the current sampling period. This ensures that the projected speed compensation amount does not exceed the speed range and drive safety boundary that the snow removal roller motor can execute.When the encoding result corresponds to the descent into the interval, the basic lifting adjustment amount is projected as the extension direction of the lifting cylinder; where the descent into the interval corresponds to the state where the left-end torque value and the right-end torque value are simultaneously lower than the target torque value, and the first absolute difference and the second absolute difference decrease simultaneously in the comparison of adjacent sampling periods. At this time, the current cutting depth of the snowplow is too small, and the lifting cylinder needs to move along the extension direction to reduce the height position of the snowplow relative to the working surface, so that the double-end torque continues to return to the target torque value; when projecting, the allowable control set corresponding to the basic lifting adjustment amount is limited to the interval [0, Um], and U'=min(ma Projecting x(U, 0), Um) retains the positive component of the basic lifting adjustment that is in the same direction as the extension of the lifting cylinder, while truncating the negative component that is opposite to the retraction of the lifting cylinder to zero. At the same time, the portion exceeding the upper limit of the lifting adjustment amplitude Um is saturated and limited to Um, so that the lifting adjustment output in the descent range always maintains the unidirectional extension characteristic. The basic speed compensation is projected into the range [-Vm, Vm] and limited and retained according to V'=max(min(V, Vm), -Vm), so that the snow removal roller motor can perform torque regression correction in coordination with the lifting action within the allowable speed compensation range. When the encoding result corresponds to the rising into the interval, the basic lifting adjustment amount is projected as the retraction direction of the lifting cylinder. The rising into the interval corresponds to a state where the left-end torque value and the right-end torque value are simultaneously higher than the target torque value, and the first absolute difference and the second absolute difference decrease simultaneously in adjacent sampling periods. At this time, the current cutting depth of the snowplow is too large, and the lifting cylinder needs to move along the retraction direction to raise the height of the snowplow relative to the working surface, so that the double-end torque continues to return to the target torque value. During projection, the allowable control set corresponding to the basic lifting adjustment amount is limited to the interval [-Um, 0], and U'=max(min) is applied. Projecting (U, 0), -Um) retains the negative component of the basic lifting adjustment that is in the same direction as the retraction of the lifting cylinder, and cuts off the positive component that is opposite to the extension of the lifting cylinder to zero. At the same time, the portion exceeding the upper limit of the lifting adjustment amplitude Um is saturated and limited to -Um, so that the lifting adjustment output entering the range always maintains the unidirectional retraction characteristic. The basic speed compensation is projected into the range [-Vm, Vm] and limited and retained according to V'=max(min(V, Vm), -Vm), so that the snow removal roller motor can perform torque regression correction in coordination with the lifting action within the allowable speed compensation range.When the encoding result corresponds to the holding interval, the basic lifting adjustment amount and the basic speed compensation amount are simultaneously projected to zero. The holding interval corresponds to the state where the left-end torque value and the right-end torque value are located on both sides of the target torque value and are not equal to the target torque value. The change in the lifting cylinder extension length value in the effective lifting response sequence is less than the displacement threshold for Z consecutive sampling periods. At this time, the forces at both ends have formed a stable distribution around the target torque value. Continuing to apply the lifting adjustment amount will disrupt the currently established stable load balance, and continuing to apply the speed compensation amount will introduce additional drive disturbances. Therefore, within the holding interval, the allowable control set corresponding to the basic lifting adjustment amount is defined as {0}, and zero-value projection is performed according to U'=0. The allowable control set corresponding to the basic speed compensation amount is defined as {0}, and zero-value projection is performed according to V'=0, so that the control output remains stationary. The aforementioned monotonic constraint projection algorithm constitutes a coding-bound safety constraint layer set between the basic lifting adjustment amount, the basic speed compensation amount, and the actuator. This coding-bound safety constraint layer uses the interval coding result as the basis for determining the control direction, and performs sign constraints, zero-value freezing, and boundary clipping on the basic lifting adjustment amount and the basic speed compensation amount for each sampling cycle. This ensures that the control output in any sampling cycle cannot deviate from the action direction and output range allowed by the current torque feedback determination state, thereby preventing a retraction command when descending into the interval, preventing an extension command when ascending into the interval, and preventing the continued output of non-zero disturbance correction amounts within the interval. The projected lifting adjustment and speed compensation values ​​are sent to the hydraulic pump station and snowplow motor controller according to the sampling period. The sending period corresponds one-to-one with the sampling period identifier. Only one set of projected lifting adjustment and speed compensation values ​​is sent in each sampling period. When sending, the sampling period identifier, lifting adjustment value, and speed compensation value are packaged into a control command record and written into the controller output buffer according to the sampling period. Then, the controller triggers the sending according to the sampling period, so that the control command received by the execution side and the torque feedback judgment data maintain the same period correspondence. This avoids the mis-sending of control values ​​across periods, which would cause a mismatch between lifting action, drive action and load state. The hydraulic pump station controls the extension or retraction of the lifting cylinder based on the projected lifting adjustment amount. Specifically, when the projected lifting adjustment amount is positive, the hydraulic pump station supplies oil to the rodless chamber of the lifting cylinder and releases oil from the rod chamber, causing the lifting cylinder to move in the extension direction. When the projected lifting adjustment amount is negative, the hydraulic pump station supplies oil to the rod chamber of the lifting cylinder and releases oil from the rodless chamber, causing the lifting cylinder to move in the retraction direction. When the projected lifting adjustment amount is zero, the hydraulic pump station maintains the current oil circuit state, keeping the lifting cylinder in its current position.The snowplow motor controller corrects the snowplow motor speed based on the projected speed compensation. Specifically, when the projected speed compensation is positive, the controller increases the output speed based on the current snowplow motor speed; when the projected speed compensation is negative, the controller decreases the output speed based on the current snowplow motor speed; and when the projected speed compensation is zero, the controller maintains the current snowplow motor speed. The technical principle behind this control method is that changes in the snowplow motor speed directly affect the frequency of snow penetration per unit time, the speed of torque build-up at both ends, and the dynamic response speed during torque return. Synchronizing the speed compensation with the lifting adjustment allows height adjustment and drive adjustment to participate in the load return process, thereby improving the continuity of snowplow motor control under dynamic operating conditions. After the control command for the current sampling period is sent, the controller continues to read the left-end torque value, right-end torque value, lifting cylinder extension length value, snowplow motor speed value, and hydraulic pump station output pressure value in the next sampling period. It then regenerates the torque feedback judgment data, basic lifting adjustment amount, and basic speed compensation amount for the corresponding sampling period. This ensures that the actual response after the projection is executed is continuously fed back to the control input of subsequent sampling periods, thereby guaranteeing that the dynamic control process of the snowplow forms a continuous closed loop in the sampling period dimension.

[0044] In this implementation scheme, by establishing a one-to-one directional mapping relationship between the basic lifting adjustment amount and the basic speed compensation amount and the descent entry interval, the ascent entry interval, and the holding interval, and by synchronously sending the projected lifting adjustment amount and speed compensation amount according to the sampling period under the data acquisition and control constraints, it is possible to ensure that the lifting cylinder action direction and the snow removal roller motor speed correction direction are always consistent with the current torque return state, avoiding deviation between the control output direction and the load return direction. This allows the lifting adjustment process, speed correction process, and torque return process to be continuously connected under the same time reference, thereby improving the execution consistency, response stability, and state maintenance reliability of the electric snow removal roller dynamic control.

[0045] like Figure 2As shown, the second aspect of the present invention provides a dynamic control optimization system for an electric snowplow based on torque feedback, comprising: a data acquisition and processing module, a dual-end alignment purification module, a regression judgment generation module, and a linkage adaptive control module, wherein: the data acquisition and processing module is used to periodically acquire snowplow operation monitoring data, perform preprocessing on the snowplow operation monitoring data, and output preprocessed snowplow operation monitoring data; the dual-end alignment purification module is used to generate left and right end torque misalignment combinations based on the preprocessed snowplow operation monitoring data and calculate the torque alignment step size, generate a dual-end aligned torque sequence, and then perform purification screening on the dual-end aligned torque sequence in combination with a freeze window, a potential pressure anti-phase blocking segment, and an attitude transmission disturbance value. The system selects and outputs the effective lifting response sequence and net operating torque sequence; the regression judgment generation module is used to locate the starting anchor point and torque convergence anchor point of the lifting action based on the effective lifting response sequence and net operating torque sequence, extract the regression event segment and generate the leading torque sample value and convergence torque sample value, calculate the target torque value one, target torque value two and target torque value three, and output torque feedback judgment data; the linkage adaptive control module is used to construct the control input window based on the torque feedback judgment data and the corresponding snowplow operation monitoring data, generate the basic lifting adjustment amount and basic speed compensation amount, and send the execution direction projection according to the corresponding coding result to the hydraulic pump station and snowplow motor controller to implement dynamic control of the snowplow.

[0046] like Figure 4As shown in the figure, ① represents the snow removal rolling force assembly, which provides the driving force for the rotation of the snow removal roller. The snow removal rolling force assembly includes a permanent magnet synchronous motor for the snow removal roller, a reducer assembly, a motor controller, heat dissipation components, and a power supply control harness; ② represents the hydraulic pump station, which provides hydraulic power to the lifting cylinder and the left and right swing cylinders to realize the lifting and deflection adjustment of the snow removal roller device; ③ represents the transfer case, which is connected to the output end of the snow removal rolling force assembly and distributes the input power to the two transmission links; ④ represents the lifting cylinder, which drives the snow removal roller device to rise or fall vertically; ⑤ represents the drive shaft, which... The drive shaft transmits the rotational power output from the snow removal rolling force assembly to the transfer case, and then from the transfer case to the drive sprockets on both sides; Figure ⑥ shows the snow removal roller brush and protective cover, where the snow removal roller brush is used to actually perform snow removal operations, and the protective cover is used to suppress snow scattering and reduce pollution in the windshield area; Figure ⑦ shows the drive sprocket, which receives the power output from the transfer case and transmits it to the driven sprocket; Figure ⑧ shows the driven sprocket, which receives the power transmitted by the drive sprocket and drives the snow removal roller to rotate, so as to realize the snow removal roller brush operation; Figure ⑨ shows the left and right swing cylinders, which drive the snow removal roller device to deflect to the left or right to adapt to different operating directions. The controller receives control commands from the operator, including at least commands for raising, lowering, turning left, turning right, and a target speed. Based on the target speed command, the controller sends a speed control signal to the motor controller, which drives the snowplow assembly to drive the drive shaft, transfer case, drive sprocket, and driven sprocket to achieve the target speed. Simultaneously, the controller controls the corresponding hydraulic circuit of the hydraulic pump station based on the lifting and deflection commands to drive the lifting cylinder to adjust the snowplow's height and to drive the left and right swing cylinders to adjust its deflection. A hydraulic lock is installed in the hydraulic circuit to limit the passive retraction of the lifting cylinder caused by vehicle vibration, maintaining the snowplow's current height position stably. Limit switches are also installed to limit the maximum extension and retraction positions of the lifting cylinder, preventing excessive lowering or raising of the snowplow. Torque sensors are installed at both ends of the snowplow to collect the torque values ​​at the left and right ends in real time. The controller generates target torque values ​​one, two, and three based on the left and right torque values. When the corresponding state of the left and right torque values ​​is lower than target torque value one, the controller drives the lifting cylinder to move in the extension direction, causing the snowplow to descend until the torque at both ends returns to the range corresponding to target torque value three. When the corresponding state of the left and right torque values ​​is higher than target torque value two, the controller drives the lifting cylinder to move in the retraction direction, causing the snowplow to rise until the torque at both ends returns to the range corresponding to target torque value three. This achieves adaptive adjustment of the snowplow height based on dual-end torque feedback, improving load matching and dynamic control accuracy during snow removal operations.

[0047] In this implementation plan, by continuously organizing the preprocessing of snowplow operation monitoring data, double-end alignment and purification, generation of target torque value one, target torque value two, target torque value three, torque feedback judgment data output, basic lifting adjustment amount generation, and basic speed compensation amount generation according to the same technical chain, the left-end torque value, right-end torque value, lifting cylinder extension length value, hydraulic pump station output pressure value, snowplow motor output current value, snowplow motor speed value, snowplow deflection angle value, vehicle travel speed value, vehicle body pitch angle value, and vehicle body roll angle value are mapped from monitoring status to control commands on a unified data basis. This ensures that the torque regression basis, lifting adjustment basis, and speed compensation basis maintain consistency in source, judgment, and execution, thereby improving the overall coordination, load matching, and operational stability of the electric snowplow dynamic control.

[0048] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0049] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A dynamic control optimization method for electric snowplows based on torque feedback, characterized in that, Includes the following steps: S1 periodically collects snowplow operation monitoring data, performs preprocessing on the snowplow operation monitoring data, and outputs the preprocessed snowplow operation monitoring data. S2, based on the preprocessed snowplow operation monitoring data, generate the left and right end torque misalignment combination and calculate the torque alignment step size, generate the double-end aligned torque sequence, and then combine the freezing window, the pressure anti-phase blocking section and the attitude transmission disturbance value to perform purification and screening on the double-end aligned torque sequence, and output the effective lifting response sequence and net operation torque sequence. S3, based on the effective lifting response sequence and net working torque sequence, locates the lifting action start anchor point and torque convergence anchor point, extracts the regression event segment and generates the leading torque sample value and convergence torque sample value, calculates the target torque value one, target torque value two and target torque value three, and outputs torque feedback judgment data; S4 constructs a control input window based on torque feedback judgment data and corresponding snowplow operation monitoring data, generates basic lifting adjustment amount and basic speed compensation amount, and sends the result of direction projection to the hydraulic pump station and snowplow motor controller according to the corresponding coding result to implement dynamic control of the snowplow.

2. The dynamic control optimization method for electric snowplows based on torque feedback according to claim 1, characterized in that: The specific steps for periodically collecting snowplow operation monitoring data, performing preprocessing on the snowplow operation monitoring data, and outputting the preprocessed snowplow operation monitoring data are as follows: A fixed-width sliding time window is set as a sampling period to periodically collect snowplow operation monitoring data. The snowplow operation monitoring data includes sampling period identifier, left end torque value, right end torque value, snowplow motor speed value, snowplow motor output current value, lifting cylinder extension length value, lifting cylinder rodless chamber pressure value, lifting cylinder rod chamber pressure value, hydraulic pump station output pressure value, snowplow deflection angle value, vehicle travel speed value, vehicle pitch angle value, and vehicle roll angle value. For the collected snowplow operation monitoring data, a linear time alignment algorithm is used to perform multi-source time-series synchronization processing; a piecewise cubic spline interpolation algorithm is used to perform missing data completion processing; a local outlier anomaly detection algorithm is used to perform outlier data identification and removal processing; a sliding median filtering algorithm is used to perform mechanical shock noise suppression processing; and a Z-score standardization algorithm is used to perform numerical scale unification processing, outputting the preprocessed snowplow operation monitoring data.

3. The dynamic control optimization method for electric snowplows based on torque feedback according to claim 2, characterized in that: The specific steps for generating a double-end aligned torque sequence based on the preprocessed snowplow operation monitoring data to generate left and right end torque misalignment combinations and calculate the torque alignment step size are as follows: The preprocessed snow removal rolling operation monitoring data is read according to the sampling cycle identifier. A local judgment window is constructed in units of N consecutive sampling cycles. In each local judgment window, the left torque value is kept in its original order, and the right torque value is shifted forward in sequence according to the preset shift step size to generate multiple sets of left and right torque misalignment combinations. Then, the change in left torque value and the change in right torque value between adjacent sampling cycles are calculated for each set of left and right torque misalignment combinations. When the change in left torque value and the change in right torque value are both less than the micro-change threshold for M consecutive sampling cycles, the corresponding sampling cycle segment is marked as a frozen window. For each set of left and right end torque misalignment combinations, the absolute difference between the left end torque value and the misaligned right end torque value is calculated for each sampling period, and all absolute differences are accumulated. After deleting the sampling period segment corresponding to the frozen window, the misalignment step size with the smallest cumulative result is taken as the torque alignment step size corresponding to the current local decision window. Then, the right end torque value is rearranged according to the torque alignment step size, and the rearranged right end torque value is re-paired with the left end torque value according to the sampling period to obtain the double-end aligned torque sequence.

4. The dynamic control optimization method for electric snowplows based on torque feedback according to claim 3, characterized in that: The specific steps for purifying and filtering the double-ended aligned torque sequence by combining the freeze window, the potential pressure phase-blocking segment, and the attitude propagation perturbation value, and outputting the effective lifting response sequence and the net operating torque sequence are as follows: The lifting cylinder extension length, rodless chamber pressure, and rod chamber pressure are read according to the sampling period corresponding to the double-ended aligned torque sequence. The lifting net pressure difference is obtained by subtracting the rod chamber pressure from the rodless chamber pressure. The changes in lifting cylinder extension length and lifting net pressure difference between adjacent sampling periods are also calculated. When the direction of change of lifting cylinder extension length is opposite to the direction of change of lifting net pressure difference, K sampling periods are backtracked from the current sampling period and then extended forward by K sampling periods. The corresponding sampling period segment is marked as the potential pressure anti-phase blocking segment, and the sampling period corresponding to the potential pressure anti-phase blocking segment is deleted from the double-ended aligned torque sequence. The remaining sampling periods are output as the effective lifting response sequence. Based on the sampling period corresponding to the double-ended aligned torque sequence, the vehicle pitch angle, vehicle roll angle, snowplow yaw angle, and vehicle speed are read. The square root of the square of the change in vehicle pitch angle plus the square of the change in vehicle roll angle is used to obtain the attitude synthesis term. The absolute value of the snowplow yaw angle after taking the cosine is added to the absolute value of the snowplow yaw angle after taking the sine to obtain the yaw projection term. The vehicle speed is incremented by one, the natural logarithm is taken, and then one is added again to obtain the speed amplification. The attitude synthesis term, the yaw projection term, and the speed amplification are multiplied sequentially to obtain the attitude transmission disturbance value. Add the absolute value of the change in torque on the left end to the absolute value of the change in torque on the right end to obtain the torque fluctuation. The attitude transmission disturbance value is matched with the torque fluctuation value one by one. When the attitude transmission disturbance value and the torque fluctuation value increase in the same direction, the corresponding sampling period is deleted from the double-ended aligned torque sequence, and the remaining sampling period is output as the net working torque sequence.

5. The dynamic control optimization method for electric snowplows based on torque feedback according to claim 4, characterized in that: The specific steps for locating the lifting action initiation anchor point and torque convergence anchor point based on the effective lifting response sequence and net operating torque sequence, extracting regression event segments, and generating leading torque sample values ​​and convergence torque sample values ​​are as follows: According to the sampling period identification order, locate the sampling period in the effective lifting response sequence where the change in the extension length of the lifting cylinder changes from zero to non-zero, and mark the corresponding sampling period as the starting anchor point of the lifting action. Starting from the initial anchor point of each lifting action, calculate the change in left-end torque value and the change in right-end torque value between adjacent sampling periods in the net working torque sequence. When the change in left-end torque value and the change in right-end torque value have the same sign, and the absolute value of the change in left-end torque value and the absolute value of the change in right-end torque value are both less than the convergence threshold for W consecutive sampling periods, mark the corresponding sampling period as the torque convergence anchor point. The sampling period segment between the starting anchor point of the lifting action and the torque convergence anchor point is taken as the regression event segment. The arithmetic mean of the left-end torque value and the right-end torque value corresponding to the starting position of each regression event segment is recorded as the forward torque sample value. The arithmetic mean of the left-end torque value and the right-end torque value corresponding to the ending position of each regression event segment is recorded as the convergence torque sample value.

6. The dynamic control optimization method for electric snowplows based on torque feedback according to claim 5, characterized in that: The specific steps for calculating target torque value one, target torque value two, and target torque value three are as follows: For all convergent torque sample values, density peak clustering algorithm is used to extract cluster centers. The cluster center value corresponding to the position with the highest sample density is taken as the target torque value three. Using the target torque value three as the boundary, all leading torque sample values ​​are divided into a first leading sample set and a second leading sample set located to the left and right of the target torque value three. Then, a semi-supervised Gaussian mixture clustering algorithm is used to extract boundary clusters for the first and second leading sample sets respectively, generating a first torque regression boundary band and a second torque regression boundary band. The absolute difference between each sample value in the first and second torque regression boundary bands and the target torque value three is calculated respectively. The sample value with the smallest absolute difference in the first torque regression boundary band is taken as the target torque value one. The sample value with the smallest absolute difference in the second torque regression boundary band is taken as the target torque value two. Output target torque value one, target torque value two, and target torque value three.

7. The dynamic control optimization method for electric snowplows based on torque feedback according to claim 6, characterized in that: The specific steps for determining the output torque feedback data are as follows: In the net working torque sequence, the first absolute difference between the left-end torque value and the target torque value three is calculated for each sampling period, and the second absolute difference between the right-end torque value and the target torque value three is calculated. The continuous sampling period interval where the left-end torque value and the right-end torque value are both lower than the target torque value one, and the first absolute difference and the second absolute difference decrease simultaneously in the comparison of adjacent sampling periods, is identified as the falling entry interval. The interval between consecutive sampling periods in which the left-end torque value and the right-end torque value are both higher than the target torque value, and the first absolute difference and the second absolute difference decrease simultaneously in the comparison of adjacent sampling periods, is encoded as the rising entry interval. The left and right torque values ​​are located on either side of the target torque value 3 and are not equal to the target torque value 3. The sampling period interval in which the change in the extension length of the lifting cylinder in the effective lifting response sequence is less than the displacement threshold for Z consecutive times is encoded as the holding interval. Each sampling period interval and the corresponding encoding result are output as torque feedback judgment data.

8. The dynamic control optimization method for electric snowplows based on torque feedback according to claim 7, characterized in that: The specific steps for constructing a control input window based on torque feedback judgment data and corresponding snowplow operation monitoring data to generate basic lifting adjustment and basic speed compensation are as follows: The torque feedback judgment data and the corresponding snowplow operation monitoring data are read in the order of the sampling period identifier. A control input window is constructed with Q consecutive sampling periods. The first absolute difference between the left torque value and the target torque value, the second absolute difference between the right torque value and the target torque value, and the corresponding coding results are combined to form the torque regression input sequence. The change in the extension length of the lifting cylinder, the change in the output pressure of the hydraulic pump station, the change in the output current of the snowplow motor, the change in the speed of the snowplow motor, and the vehicle speed are combined to form the execution load input sequence. The torque regression input sequence is input into the first branch, and the dilated causal convolutional network algorithm is used to convolve layer by layer with different dilation coefficients to obtain the torque regression response sequence corresponding to each sampling period; the execution load input sequence is input into the second branch, and the gated recurrent unit network algorithm is used to recursively update according to the sampling period to obtain the execution load response sequence corresponding to each sampling period. The torque regression response sequence is then used as the query sequence, and the execution load response sequence is used as the key sequence. The cross-gated attention fusion algorithm is then used to calculate the gating weight of the torque regression response sequence on the execution load response sequence according to the sampling period, and the execution load response sequence is weighted and filtered according to the gating weight. The weighted execution load response sequence is then fed back to the corresponding position of the torque regression response sequence, and the basic lifting adjustment amount and basic speed compensation amount corresponding to each sampling period are output.

9. The dynamic control optimization method for electric snowplows based on torque feedback according to claim 8, characterized in that: The specific steps for implementing dynamic control of the snowplow after performing directional projection based on the corresponding encoding result and sending it to the hydraulic pump station and snowplow motor controller are as follows: Input the basic lifting adjustment amount and the basic speed compensation amount into the monotonic constraint projection algorithm. When the encoding result corresponds to the descent entering the interval, the basic lifting adjustment amount is projected as the extension direction of the lifting cylinder; when the encoding result corresponds to the ascent entering the interval, the basic lifting adjustment amount is projected as the retraction direction of the lifting cylinder; when the encoding result corresponds to the holding interval, the basic lifting adjustment amount and the basic speed compensation amount are projected to zero at the same time. The projected lifting adjustment amount and speed compensation amount are sent to the hydraulic pump station and snowplow motor controller according to the sampling period. The hydraulic pump station controls the extension or retraction of the lifting cylinder according to the lifting adjustment amount, and the snowplow motor controller corrects the speed of the snowplow motor according to the speed compensation amount.

10. A dynamic control optimization system for an electric snowplow based on torque feedback, characterized in that, include: The module comprises a data acquisition and processing module, a dual-end alignment and purification module, a regression judgment and generation module, and a linkage adaptive control module, among which: The data acquisition and processing module is used to periodically acquire snowplow operation monitoring data, perform preprocessing on the snowplow operation monitoring data, and output the preprocessed snowplow operation monitoring data. The dual-end alignment purification module is used to generate left and right end torque misalignment combination and calculate torque alignment step size based on preprocessed snow removal rolling operation monitoring data, generate dual-end aligned torque sequence, and then perform purification and screening on the dual-end aligned torque sequence in combination with freezing window, pressure anti-phase blocking segment and attitude transmission disturbance value, and output effective lifting response sequence and net operation torque sequence. The regression judgment generation module is used to locate the starting anchor point and torque convergence anchor point of the lifting action based on the effective lifting response sequence and the net working torque sequence, extract the regression event segment and generate the leading torque sample value and the converged torque sample value, calculate the target torque value one, the target torque value two and the target torque value three, and output the torque feedback judgment data. The linkage adaptive control module is used to construct a control input window based on torque feedback judgment data and corresponding snowplow operation monitoring data, generate basic lifting adjustment amount and basic speed compensation amount, and send the result of direction projection to the hydraulic pump station and snowplow motor controller according to the corresponding coding result to implement dynamic control of the snowplow.

Images