Integrated control system for electric modification of electric tractor

Abstract

The invention discloses an integrated control system for electric modification of an electric tractor, and belongs to the technical field of agricultural machinery. The system comprises a signal processing module, a health state evaluation module, a control decision module, an execution module and an optimization module, all the modules realize data interaction through the Ethernet, and the system is adaptive to heterogeneous hardware integration requirements after electric modification of a traditional fuel tractor. According to the invention, the health state evaluation module and the prediction model realize quick response and graded protection of high-voltage insulation, battery overheating and actuator fault scenes through clear quantification, and realize quantitative scoring of the health degree in the limitation of temperature, load and attitude coupling complex scenes; meanwhile, the time domain and frequency domain characteristics of the signals are extracted to provide a quantitative basis for evaluation, different actions of power reduction, speed limiting and emergency shutdown are triggered according to the health degree state, the situation that operation efficiency is affected by excessive protection is avoided, early warning can be conducted in the early stage of a fault, rapid protection can be conducted when the fault occurs, and the running safety and reliability of the tractor are improved.

Description

Technical Field

[0001] This invention relates to the field of agricultural machinery technology, and in particular to an integrated control system for the electrification conversion of electric tractors. Background Technology

[0002] Currently, the electrification of fuel-powered tractors commonly employs a piecemeal approach, using drive motors, energy storage batteries, and electric actuators from various manufacturers. The control modules of each component operate independently, lacking a unified communication protocol and signal standard. This results in heterogeneous signals from multiple sources, encompassing a wide range of signal types including voltage, current, temperature, load, attitude, and operational parameters. Furthermore, these signals suffer from high noise levels and temporal / spatial asynchrony. The lack of unified integration, scheduling, and signal processing capabilities prevents efficient fusion and precise interaction of these heterogeneous signals, leading to high signal latency, delayed data transmission between components, and indirectly causing mismatches between operating conditions and power output. Summary of the Invention

[0003] The purpose of this invention is to provide an integrated control system for the electrification conversion of electric tractors, in order to solve the problems of high signal interaction delay and data transmission lag between components in the prior art, which indirectly leads to the derivative problem of mismatch between working conditions and power output.

[0004] To achieve the above objectives, this application proposes an integrated control system for the electrification conversion of electric tractors, which specifically includes: The signal processing module is used to acquire and process the multi-source heterogeneous signals output by the modified tractor; A health status assessment module is used to assess the system health and generate a health status identifier based on the multi-source heterogeneous signals. The control decision module is used to generate a prediction model based on the health status identifier and the processed multi-source heterogeneous signal, and output cooperative control commands based on the prediction model. The execution module is used to respond to the coordinated control command, execute the corresponding vehicle control actions, and generate execution feedback information; An optimization module is used to optimize the prediction model based on the execution feedback information.

[0005] In some embodiments, the tractor modification includes: High-voltage energy storage system, electric drive system, agricultural implement operation system, steering control system, attitude sensing system; The high-voltage energy storage system is used to provide high-voltage power to the vehicle. The electric drive system is used to drive the tractor and output power. The agricultural implement system is used for tilling, sowing, and harvesting operations. The steering control system is used for vehicle steering adjustment; The attitude perception system is used to collect the vehicle's operating attitude and position information.

[0006] In some embodiments, the multi-source heterogeneous signal includes at least one of the following: Heterogeneous signals of the vehicle's high-voltage battery status, power output shaft speed and torque, implement resistance, steering motor speed and steering angle, and vehicle attitude and position.

[0007] In some embodiments, the processing of the multi-source heterogeneous signals includes: For the aforementioned multi-source heterogeneous signals, a bus protocol is used for protocol conversion and format parsing to extract valid multi-source heterogeneous signal data; For the multi-source heterogeneous signal, the Kalman filter algorithm is used to remove high-frequency noise and abnormal pulses in the multi-source heterogeneous signal, and extreme outliers are removed by the 3σ criterion to ensure the stability of the multi-source heterogeneous signal. For the multi-source heterogeneous signals, the Z-score normalization method is used to convert the parsed and eliminated multi-source heterogeneous signals into signals of a unified dimension, eliminating the difference in dimensions and providing a standardized data foundation for subsequent modules.

[0008] In some embodiments, the health status assessment module includes a security rule unit, which includes at least one of the following rules: If the high-voltage battery pack fusion temperature exceeds the first threshold and the implement is under heavy load, a performance-limited flag will be generated. If a critical actuator response deviation is detected to continuously exceed the tolerance range, a protective intervention flag is generated, and an action is taken to switch the vehicle to a safe state.

[0009] In some embodiments, the health status identifier is based on a security rule unit, which performs a weighted evaluation of the multi-source heterogeneous signal parameters to generate a three-level health status identifier: normal status, warning status, and fault status. Among them, the warning state triggers power reduction control, and the fault state triggers emergency shutdown protection.

[0010] In some embodiments, the prediction model includes a working condition identification unit, a power distribution unit, and an instruction generation unit; The working condition identification unit extracts time-domain and frequency-domain features based on the standardized multi-source heterogeneous signals, and identifies the current tractor working condition by combining historical working condition samples. The power distribution unit dynamically allocates the power ratio of each motor based on the identified operating conditions and health status indicators to ensure that the power output matches the operating conditions. The instruction generation unit converts the power allocation result into specific control signals, generates coordinated control instructions, and sends them to the execution module.

[0011] In some embodiments, the coordinated control commands are used to coordinate a composite command set of the electric drive system, the implement operation system, and the steering control system to achieve smooth switching of power modes, adaptive adjustment of steering assist, and linkage control between implement actions and vehicle movement.

[0012] In some embodiments, the execution module includes a control drive unit, a data acquisition unit, and a data output unit; The control and drive unit is used to execute vehicle power distribution strategy, steering control, and implement start-stop and lifting actions on the high-voltage energy storage system, electric drive system, implement operation system, steering control system, and attitude perception system. The data acquisition unit is used to collect the execution parameters of the high-voltage energy storage system, electric drive system, agricultural implement operation system, steering control system, and attitude perception system in real time. The data output unit is used to send the collected execution parameters as execution feedback information to the optimization module.

[0013] In some embodiments, the adaptive optimization process of the optimization module includes: The execution feedback information is compared with the expected results to generate performance deviation data; If the effect deviation data value exceeds the preset threshold, the feature weights of the prediction model are adjusted based on the correlation between the deviation value and the multi-source heterogeneous signal. If the deviation value is within the preset threshold, the current working condition data, control commands, and execution feedback information will be used as new samples to incrementally train the prediction model.

[0014] Compared with the prior art, the embodiments of this application have at least the following beneficial effects: 1. The health status assessment module and prediction model of this invention achieve rapid response and graded protection in key scenarios such as high-voltage insulation, battery overheating, and actuator failure through explicit quantification. It also achieves quantitative scoring of health status in the face of the limitations of complex scenarios involving temperature, load, and attitude coupling. At the same time, it extracts the time-domain and frequency-domain features of the signal to provide a quantitative basis for assessment. Then, it triggers different actions such as power reduction, speed limiting, and emergency shutdown based on the health status. This avoids over-protection from affecting work efficiency, provides timely warnings in the early stages of a fault, and provides rapid protection when a fault occurs. It adapts to the control requirements after modification, realizes unified integrated scheduling of various components, and improves the safety and reliability of the electric-modified tractor. It solves the safety hazards caused by independent components from the perspective of health assessment.

[0015] 2. This invention achieves accurate identification of four types of operating conditions—plowing, sowing, harvesting, and transportation—through a control decision module. Combined with health status indicators, it realizes dynamic power allocation in two dimensions: operating condition and health status. A baseline power weight is preset for different operating conditions, and the weight allocation is dynamically adjusted based on component health status warnings and fault dynamics. Abnormal components are downgraded or disconnected, while normal components are proportionally weighted, protecting faulty components while maximizing the overall vehicle power output. Furthermore, coordinated control commands achieve composite linkage between electric drive, implement operation, and steering control, such as linear power switching during operating condition changes and adaptive adjustment of implement resistance and driving speed torque during tilling, avoiding mechanical shocks caused by sudden power changes and improving operational smoothness and efficiency.

[0016] 3. The execution module of this invention adopts a modular architecture of control drive, data acquisition, and data output. It clearly defines the hardware interfaces and communication protocols of each unit, adapts to the drive and data acquisition requirements of the modified heterogeneous actuator, the control drive unit can convert control commands into execution signals, the data acquisition unit collects execution parameters in real time, and the data output unit realizes high-speed transmission of feedback information through multi-protocol links; at the same time, it constructs a closed-loop execution link for the entire process of control command issuance, execution action triggering, execution parameter acquisition, and feedback information return, ensuring the accurate execution of control commands and providing real execution data for subsequent prediction model optimization.

[0017] 4. The signal processing module of this invention achieves the cleaning, unification and fusion of multi-source heterogeneous signals by constructing a processing flow; the final output standardized signal has high accuracy and strong consistency, providing reliable data support for health status assessment, operating condition identification and control decision-making. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the drawings used in the embodiments or examples will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained according to these drawings without creative effort.

[0019] Figure 1 This is an architecture diagram of the integrated control system for the electrification retrofit of electric tractors provided in an embodiment of this application; Figure 2 This is an architecture diagram of the execution module of the integrated control system for the electrification retrofit of electric tractors provided in this application embodiment. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0022] The electrification of electric tractors typically involves multiple steps, including the removal of the original fuel-powered system, the installation of electric drive and energy storage components, the electrification of implement actuators, and the adaptation and debugging of the control logic of each component.

[0023] For example, after electric tractors are converted to electric power, problems such as disordered signal interaction and uncoordinated power distribution occur, affecting work efficiency and operational safety.

[0024] To address the technical problem of complex signal types caused by the multi-source heterogeneous characteristics of signals after the electrification of existing electric tractors, this application provides an integrated control system for the electrification of electric tractors. The core technical features are: setting up five collaborative modules: signal processing module, health status assessment module, control decision module, execution module, and optimization module. The system solves the compatibility problem of multi-source heterogeneous signals through a unified signal processing mechanism, achieves accurate fault early warning and hierarchical protection by combining dual health assessment logic, constructs a dynamic prediction model based on the dual dimensions of operating conditions and health status, outputs collaborative control commands and forms a closed-loop optimization, adapts to the control requirements after the modification, and realizes unified integrated scheduling of various components.

[0025] The integrated control system for electric tractor electrification retrofitting provided in this application relates to the field of agricultural machinery technology. This integrated control system for electric tractor electrification retrofitting can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle-mounted terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the integrated control system for electric tractor electrification retrofitting, but is not limited to the above forms.

[0026] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0027] Figure 1This is an architecture diagram of the integrated control system for the electrification retrofit of electric tractors provided in this application embodiment.

[0028] The signal processing module, health status assessment module, control decision module, execution module, and optimization module shown in this application embodiment achieve data interaction and command transmission through a communication link built with the industrial Ethernet protocol. Each module works together to ensure the stable operation of the integrated control for the electrification conversion of electric tractors.

[0029] See appendix Figure 1 This invention discloses an integrated control system for the electrification conversion of electric tractors. The system mainly includes a signal processing module, a health status assessment module, a control decision module, an execution module, and an optimization module. Each module interacts with data through CAN / LIN bus and Ethernet, adapting to the heterogeneous hardware integration requirements after the electrification conversion of traditional fuel tractors.

[0030] The signal processing module is used to acquire and process the multi-source heterogeneous signals output by the modified tractor; A health status assessment module is used to assess the system health and generate a health status identifier based on the multi-source heterogeneous signals. The control decision module is used to generate a prediction model based on the health status identifier and the processed multi-source heterogeneous signal, and output cooperative control commands based on the prediction model. The execution module is used to respond to the coordinated control command, execute the corresponding vehicle control actions, and generate execution feedback information; An optimization module is used to optimize the prediction model based on the execution feedback information.

[0031] In some embodiments, the tractor modification includes: High-voltage energy storage system, electric drive system, agricultural implement operation system, steering control system, attitude sensing system; The high-voltage energy storage system is used to provide high-voltage power to the vehicle. The electric drive system is used to drive the tractor and provide power output; wherein... The agricultural implement system is used for tilling, sowing, and harvesting operations. The steering control system is used for vehicle steering adjustment; The attitude perception system is used to collect the vehicle's operating attitude and position information.

[0032] In some embodiments, the multi-source heterogeneous signal includes at least one of the following: Heterogeneous signals of the vehicle's high-voltage battery status, power output shaft speed and torque, implement resistance, steering motor speed and steering angle, and vehicle attitude and position.

[0033] In some embodiments, the processing of the multi-source heterogeneous signals includes: For multi-source heterogeneous signals, the bus protocol is used for protocol conversion and format parsing to extract valid multi-source heterogeneous signal data; For multi-source heterogeneous signals, the Kalman filter algorithm is used to remove high-frequency noise and abnormal pulses from the multi-source heterogeneous signals, while the 3σ criterion is used to remove extreme outliers to ensure the stability of the multi-source heterogeneous signals. For multi-source heterogeneous signals, the Z-score standardization method is used to convert the analyzed and eliminated multi-source heterogeneous signals into signals of a unified dimension, eliminating the difference in dimensions and providing a standardized data foundation for subsequent modules. The analyzed and eliminated multi-source heterogeneous signals are converted into signals of a unified dimension with a mean of 0 and a standard deviation of 1.

[0034] In this embodiment, the signal processing module uses an MCU as the core processing unit, integrating a CAN / LIN bus controller and an Ethernet controller to achieve unified access and parsing of multi-protocol signals. The specific parsing rules are as follows: The CAN signals of the high-voltage energy storage system and steering control system adopt a standard frame format with a data length of 8 bytes. For example, the battery pack fusion temperature signal corresponds to ID0x001, with bytes 0-1 representing the inlet temperature; the battery voltage signal corresponds to ID0x002, with bytes 0-2 representing the voltage value. The CANopen signal of the electric drive system adopts a PDO mapping method. The drive motor speed corresponds to COB-ID0x180 + slave address, with a data length of 4 bytes; the torque signal corresponds to COB-ID0x280 + slave address, with a data length of 4 bytes. LIN signal parsing: The LIN signal frame header of the implement operation system contains a synchronization field and a PID field, with a data length of 8 bytes. The implement resistance signal corresponds to PID0x01, with bytes 0-1 representing the resistance value. The Ethernet signal of the attitude perception system adopts the UDP protocol, port number 5000, and the data frame format is a custom structure, including timestamp, latitude, longitude, and vehicle pitch angle. The parsed data is stored in the SRAM built into the MCU, using a circular buffer to avoid data overflow. The buffer read address is indexed by the synchronization timestamp to ensure the sequential reading of data.

[0035] After the electric conversion of electric tractors, the original sensor signals are often accompanied by noise and interference, and need to be cleaned and standardized first.

[0036] Furthermore, the "Kalman filter + 3σ criterion" is used to eliminate signal noise and outliers. The specific parameters and process are as follows: For continuously varying signals such as voltage, current, and rotational speed, a discrete Kalman filter is used, and the state equation and observation equation are defined as follows: State equation: X(k) = A·X(k-1) + B·U(k) + W(k); Where k is the discrete-time index, representing the k-th sampling time; X(k) is the system state vector at time k, containing the physical quantities to be estimated (such as voltage, current, speed, etc.); A is the state transition matrix (value 1, assuming the signal is short-term stationary); X(k-1) is the state vector at the previous time (k-1); B is the control matrix (value 0, no external control input); U(k) is the control vector; W(k) is the process noise vector (Gaussian white noise, variance Q=0.01). This state equation is one of the fundamental models of the Discrete Kalman Filter (DKF), used for modeling the prediction stage of continuously changing signals such as voltage, current, and rotational speed. Its physical meaning is: The current state X(k) is mainly determined by the previous state X(k-1), plus a random process noise W(k). Since B=0, there is no known external control input to the system, and the state change is entirely driven by internal dynamics and noise. In the actual filtering process, Kalman filtering predicts the state through the state equation, and then updates the measured values ​​by combining them with the observation equation (described below), thereby achieving the optimal estimation of the signal (minimum mean square error).

[0037] Observation equation: Z(k) = H·X(k) + V(k); Where Z(k) is the observation vector at time k, i.e. the actual collected signals such as voltage, current, and rotation speed; H is the observation matrix, which maps the state vector X(k) to the observation space; X(k) is the state vector at time k, which is defined in the state equation; V(k) is the observation noise vector, which is assumed to be Gaussian white noise with a mean of 0 and a variance (covariance matrix) of R. The physical meaning of the observation equation is as follows: the actual measured value Z(k) is obtained by linearly transforming the true system state X(k) through the observation matrix H and then superimposing the observation noise V(k). Due to the presence of observation noise, the measured value is not completely accurate. Kalman filtering utilizes this equation to fuse the predicted state with the measured value in the update step, and obtains the optimal state estimate by balancing the confidence of the prediction and the observation through Kalman gain.

[0038] Furthermore, the Kalman filter process formula is defined as follows: Initialization: X(0) = Z(0): The initial state estimate is set to the first observation. Since there is no prior state information at the beginning of filtering, the first measurement is directly used as the initial state estimate. It is assumed that the first observation Z(0) is a reasonable approximation of the true state, even though it contains observation noise; P(0) = 1: The initial estimation error covariance matrix (if the state is scalar, it is scalar 1; if it is multidimensional, it is usually the identity matrix multiplied by scalar 1). Since the uncertainty of the initial state estimate is large (variance is 1), the confidence of the initial estimate X(0) is not high. If the state is multidimensional, the covariance matrix is ​​initialized to the identity matrix, indicating that the estimation errors between the states are independent and the initial uncertainty is the same. Such initialization is beneficial for the filter to quickly use the observation to correct the state in subsequent steps. Where X(0) is the initial state estimation vector; Z(0) is the initial observation vector; P(0) is the initial estimation error covariance matrix; 1 is the initial scalar value, specifically meaning that: if the state is a scalar, then P(0)=1 indicates that the initial estimation error variance is 1; if the state is a vector (dimension n), then P(0)=In (identity matrix), that is, the initial estimation error variance of each state is 1, and they are independent of each other.

[0039] Prediction Step: The prediction step uses the state equation mentioned above to make prior estimates of the current state and uncertainty, that is, to make predictions before obtaining the current observation value Z(k). The prediction step is the first major stage of Kalman filtering and includes the following two key calculations: State prediction: X pred (k)=A·X(k−1); Where k is the discrete-time index, representing the k-th sampling time; X pred (k) is the predicted state vector; A is the state transition matrix; X(k−1) is the state estimate at time (k-1) of the previous time step; The physical meaning of state prediction is to predict the state at the current moment based on the optimal estimate X(k−1) of the previous moment and the system's state transition matrix (A). Since the paper assumes that the signal is short-term stationary (A=1 or identity matrix), the predicted state is equal to the state estimate of the previous moment, that is, the signal changes slowly in a short time.

[0040] Covariance prediction: P pred (k)=A·P(k−1)·A T +Q; Among them, P pred (k) is the prediction error covariance matrix; P(k−1) is the estimation error covariance matrix at the previous time step (k−1); A T Let Q be the transpose of the state transition matrix A; let Q be the process noise covariance matrix; and let A be the state transition matrix. The physical meaning of covariance prediction is to predict the uncertainty of the current state estimate. It consists of the following two parts: The uncertainty in model propagation means that the estimation error P(k−1) from the previous time step propagates to the current time step A·P(k−1)·A through the state transition matrix A. T The increased uncertainty due to process noise, coupled with the process noise covariance Q, represents the new random disturbance introduced during state transitions. The prediction covariance P... pred (k) is always greater than or equal to the covariance P(k−1) of the previous time step, reflecting that the prediction step increases uncertainty (because the prediction is based on imperfect old information).

[0041] Provide the prior state estimate X at the current moment. pred (k) and prior uncertainty P pred (k) This provides the basis for the Kalman gain calculation in the update step, thus determining how to fuse predictions and observations. Due to the presence of process noise Q, prediction uncertainty gradually accumulates, prompting the filter to place more trust in the new observations in the update step. This prediction step, which estimates the prior prediction of the current state and its uncertainty, is a key step in Kalman filtering for time updates using a system model. It provides necessary prior information for subsequent update steps.

[0042] Update steps: The update step is a crucial stage in Kalman filtering that uses current observations to correct the predicted state, and it includes the following three core equations: 1. Kalman gain: ; Where K(k) is the Kalman gain matrix (or a vector if the observations are scalars), used to balance the weights of predictions and observations; P pred (k) is the prediction error covariance matrix, representing the uncertainty of the predicted state; H is the observation matrix, mapping the state vector to the observation space; H T Let H be the transpose of H; R is the observation noise covariance matrix, representing the variance of the observation noise.

[0043] 2. State update: X(k) = X pred (k)+K(k)·(Z(k)−H·X pred (k)); Where X(k) is the updated state estimation vector, i.e., the optimal estimate at the current time; X pred Z(k) is the predicted state vector, derived from the prediction step; Z(k) is the observation vector, i.e., the actual measurement value at the current moment; H·X pred (k) represents the predicted observation, i.e., the projection of the predicted state onto the observation space; Wherein, (Z(k)−H·X predThe physical meaning of (k) is to calculate the degree of "surprise" between the actual observation and the model prediction, and use this difference as a feedback signal to correct the state prediction in the optimal way (by Kalman gain weighting) so that the estimated value continuously approaches the true state of the system.

[0044] 3. Covariance Update: P(k) = (I − K(k)·H)·P pred (k); Where P(k) is the updated estimation error covariance matrix, representing the uncertainty of the state estimation; I is the identity matrix, with the same dimension as the state vector; K(k)·H is the product of the Kalman gain and the observation matrix, which acts on the prediction covariance to reduce uncertainty; P pred (k) is the prediction error covariance matrix; The above update steps use observed values ​​to correct predicted values, thereby reducing the uncertainty of state estimation. Its physical meaning and computational logic are as follows: The purpose of calculating the Kalman gain is to determine the relative confidence level between the predicted and observed values ​​and to calculate the optimal weight matrix K(k). Wherein, the numerator P... pred (k)·H T Projecting the uncertainty of the state prediction onto the observation space. Denominator H·P pred (k)·H T +R represents the uncertainty in the predicted observations (the representation of the prediction covariance in the observation space) plus observation noise. If the observation noise R is large, the denominator is large, the gain is small, and the filter trusts the predicted values ​​more; if the prediction uncertainty P... pred A larger (k) value results in a larger numerator and a larger gain, leading to greater trust in the observed values ​​during filtering.

[0045] The purpose of state update is to correct the predicted state by weighting the observation residuals (news) to obtain the optimal estimate X(k). Wherein, the observation residual Z(k) − H·X pred (k) The difference between the actual and predicted observations reflects the prediction bias. The residual is multiplied by the gain K(k) and added to the predicted state: the correction magnitude is determined by the gain; a larger gain results in a larger correction magnitude. The resulting state estimate, after fusing observation information, is closer to the true state than a simple prediction.

[0046] The purpose of covariance update is to update the uncertainty of the state estimate, reflecting the reduction in uncertainty after fusing observation information. The update factor I−K(k)·H acts on the predicted covariance P. pred (k), which "compresses" it. Due to the effect of gain K(k), the updated covariance P(k) is usually smaller than the predicted covariance P. pred (k) indicates a reduction in uncertainty. The covariance matrix P(k) quantifies the accuracy of the state estimate; the smaller the value, the more reliable the estimate.

[0047] The update step is the correction stage of the Kalman filter, and its function can be summarized as follows: Integrating observational information: Incorporating actual measurements into state estimation to compensate for deficiencies in forecasts.

[0048] Optimize estimation accuracy: Minimize the covariance of the estimation error by using the optimal weights (Kalman gain) to obtain the estimate with the minimum mean square error.

[0049] Reduce uncertainty: Utilize observational information to reduce the uncertainty of state estimation and improve the reliability of the estimation.

[0050] The filtering frequency is kept consistent with the signal acquisition frequency to ensure real-time processing: voltage, current, and attitude data: 100Hz (filtered every 10ms); rotational speed: 50Hz (filtered every 20ms). This ensures that the filtered estimate can be obtained in a timely manner at each sampling point, meeting the real-time requirements.

[0051] Furthermore, for the filtered signal, a sliding window (window size N=50 sampling points) is used to calculate the mean μ and standard deviation σ, and extreme outliers exceeding the range of [μ-3σ, μ+3σ] are removed. Specific steps include: ① Sort the signal data within the sliding window, remove the maximum and minimum values, and then calculate the mean μ; ② Calculate the sum of squares of the deviations of the data within the window from μ to obtain the standard deviation. ; in, x represents the standard deviation of the window data, reflecting the dispersion of the data after removing extreme values; i Let be the i-th filtered signal data point within the sliding window, where i = 1, 2, ..., N; μ is the mean of the window data; N is the sliding window size, i.e., the number of consecutive sampling points used to calculate the statistic, which is taken as N = 50 in this paper; the standard deviation is calculated only for the N-2 data points after removing the maximum and minimum values, and the sum of squared deviations reflects the overall fluctuation of the data relative to the mean, serving as a quantitative indicator of dispersion. The denominator N-2, since the two extreme values ​​are removed, reduces the effective sample size to N-2, making the standard deviation estimate closer to the unbiased estimate of the population standard deviation. A two-stage filtering method based on the sliding window and the 3σ criterion is used to further remove extreme outliers after Kalman filtering, improving signal quality.

[0052] ③ Determine whether the current sampling point x is within [μ-3σ, μ±3σ]. If it is, retain it; otherwise, replace it with the valid data from the previous time step and mark the number of anomalies. Wherein, [μ-3σ,μ±3σ] is the outlier detection interval.

[0053] ④ If abnormal values ​​occur three times consecutively, a sensor fault warning will be triggered and sent to the health status assessment module.

[0054] Finally, the Z-score normalization is performed on the analytically filtered multi-source heterogeneous signal to eliminate dimensional differences. The normalization formula is as follows: ; Among them, X std The normalized heterogeneous signal value; X is the current heterogeneous signal value; μ ref The mean of the reference dataset; σ ref The standard deviation of the reference dataset; Furthermore, the system eliminates the differences in units and orders of magnitude between different signals (such as voltage, temperature, and speed), allowing them to be compared and calculated on the same scale. During initial no-load operation, the system collects baseline values ​​(mean and standard deviation) for each signal under normal conditions. In subsequent operation, each real-time signal value is subtracted from its baseline mean and then divided by its baseline standard deviation to obtain a dimensionless standard score. The standard score is limited to a certain range (e.g., -3 to ±3), with values ​​exceeding this range treated as boundary values ​​to avoid interference from extreme values. During the system's power-on initialization phase, the tractor runs unloaded for 5 minutes, collecting stable data for each signal (1000 sampling points per signal), and calculating the μ value of each signal. ref With σ ref It is stored in non-volatile memory (Flash), and specific reference values ​​are as follows (example): Battery pack fusion temperature: μ ref =25℃, σ ref =5℃; Battery voltage: μ ref =380V, σ ref =20V; Drive motor speed: μ ref =1500rpm, σ ref =500rpm; Agricultural implement resistance: μ ref =500N, σ ref =200N; Vehicle pitch angle: μ ref =0°,σ ref =2°.

[0055] The standardized data range is controlled within the interval [-3, 3]. Values ​​outside this range are forcibly truncated to ±3. The truncation rule is as follows: if X... std If X <−3, then let X std =−3; if X =−3; std >3, then let X std =3, in order to avoid the impact of extreme values ​​on subsequent models.

[0056] When a health status indicator is generated or a fault is triggered, the indicator generation time, triggering rules, and corresponding parameter values ​​are stored in the Flash non-volatile memory. At the same time, the execution module and the optimization module record control commands, execution feedback information, and model optimization records to achieve accurate tracing of the cause of the fault, which facilitates subsequent fault investigation and maintenance.

[0057] In another embodiment, a spatiotemporal fusion strategy is employed to achieve the coordinated integration of multi-source heterogeneous signals: Time fusion: Using 1PPS to synchronize the timestamps of each sensor, data with different sampling frequencies and different communication protocols are unified into a 100Hz equally spaced time series through linear interpolation, filling in the data loss caused by transmission delay or different sampling rates, and forming a time-aligned multi-source signal sequence. Spatial fusion: When multiple sensors measure the same physical quantity (such as the battery pack fusion temperature), a weighted average is used to fuse the data from each sensor to obtain a more reliable and accurate estimate. The weights are allocated based on the sensor's accuracy; the higher the accuracy, the greater the weight. The spatial fusion formula is as follows: T bat =(T in ×0.5+T out ×0.5); Among them, T bat T is the battery pack fusion temperature. in The measured value is from the battery pack inlet temperature sensor; T out The reading is the battery pack outlet temperature sensor measurement; 0.5 is the weighting coefficient, representing the contribution ratio of the two sensors in the fusion. It is assumed here that the inlet and outlet temperature sensors have the same accuracy, so both have a weight of 0.5. Spatiotemporal fusion strategies aim to address the temporal and spatial inconsistencies of multi-source heterogeneous signals, achieving collaborative data integration: temporal fusion eliminates time asynchrony and sampling rate differences, ensuring all signals share the same time reference and are sampled at equal intervals, facilitating subsequent synchronization processing and feature extraction. Spatial fusion integrates multi-sensor measurements under the same conditions, utilizing redundant information to improve data reliability and accuracy, and reducing the impact of individual sensor failures or noise.

[0058] In this example, the following key safety parameters are extracted from the fused and standardized multi-source heterogeneous signals as the basis for health assessment, as shown below: High-voltage electrical system parameters: battery pack fusion temperature (value after fusion, unit °C); battery cell voltage balance (difference between maximum and minimum values, unit V, extraction method: read the voltage of 16 cells from the battery management system (BMS) and calculate the maximum and minimum values); high-voltage circuit insulation impedance (directly acquired by sensors); battery state of charge (SOC) (calculated and output by BMS, unit %), based on ampere-hour integration method, initial SOC=100%, charge and discharge current integral correction.

[0059] Key actuator parameters: drive motor stator temperature (filtered value, unit °C); steering motor current (collected by steering controller, unit A, conversion formula is the same as high-pressure current); hydraulic valve group pressure (collected by hydraulic system sensors); actuator response deviation (time difference between the issuance of control command and the completion of actuator action, unit ms, extraction method: record the command issuance timestamp and the sensor feedback action completion timestamp, calculate the difference).

[0060] Operating load parameters: implement resistance (filtered value, unit N); power output shaft torque (filtered value, unit N·m).

[0061] In some embodiments, the health status assessment module includes a security rule unit, which includes at least one of the following rules: If the high-voltage battery pack fusion temperature exceeds the first threshold and the implement is under heavy load, a performance-limited flag will be generated. If a critical actuator response deviation is detected to continuously exceed the tolerance range, a protective intervention flag is generated, and an action is taken to switch the vehicle to a safe state.

[0062] In some embodiments, the health status identifier is based on a security rule unit, which performs a weighted evaluation of the multi-source heterogeneous signal parameters to generate a three-level health status identifier: normal status, warning status, and fault status. Among them, the warning state triggers power reduction control, and the fault state triggers emergency shutdown protection.

[0063] In this embodiment, by extracting time-domain and frequency-domain features from key safety parameters, a quantitative basis for health status assessment is provided: Among them, the temporal feature extraction uses a sliding window (window size N=100 sampling points, step size 50 sampling points) to extract the following temporal features: mean, variance, maximum value, minimum value, and rate of change; The frequency domain feature extraction process employs Fast Fourier Transform (FFT) to extract frequency domain features from periodic fluctuation signals such as agricultural implement resistance and motor speed. The specific steps include: The data of N sampling points in the window are padded with zeros to expand to 256 points (to meet the efficiency of FFT operation). The Hanning window is used to suppress spectral leakage; Perform an FFT operation to obtain the frequency spectrum; Extract the fundamental frequency, fundamental amplitude, and total harmonic distortion (THD). Frequency domain characteristics are used to identify the periodic patterns of load fluctuations, aiding in the assessment of operational stability.

[0064] In this embodiment, the safety rule unit of the health status assessment module integrates expert rules and fuzzy inference logic. All rules clearly define thresholds, judgment conditions, and execution actions, and can be directly implemented through programming, as detailed below: Rule 1 (Battery overheating + load linkage protection): If the average temperature of the battery pack is >55℃ (first threshold) and the resistance of the implement is >1500N (heavy load judgment threshold) and the implement displacement sensor feedback is in the lifting state (displacement >300mm), then a "performance limited" health status mark is generated, triggering power reduction control (drive motor power is reduced to 50% of rated power, hydraulic motor power is reduced to 30% of rated power).

[0065] Rule 2 (Actuator Failure Protection): If the response deviation of any actuator (drive motor, steering motor, hydraulic valve group) is >100ms (tolerance range) and occurs 5 times consecutively (anti-shake count), a "protective intervention required" health status indicator will be generated, and the following safety actions will be performed: stop implement operation (hydraulic valve group depressurizes, implements are lowered to the lowest position), limit the tractor speed to below 5km / h, and issue an audible and visual alarm (buzzer frequency 2kHz, red alarm light flashing, cycle 1s).

[0066] Rule 3 (Emergency Stop for Insulation Fault): If the insulation impedance of the high-voltage circuit is <10MΩ (safety threshold), a "fault" sign will be generated immediately, triggering an emergency stop: the main switch of the high-voltage energy storage system will be cut off, all motor control circuits will be disconnected, the implements will be lowered urgently, the tractor will stop all actions, and the audible and visual alarms will be continuously output (the buzzer will sound continuously, and the red alarm light will be constantly on).

[0067] Rule 4 (Battery Voltage Equalization Warning): If the voltage equalization of a single battery cell is greater than 0.5V (warning threshold), a "warning" flag is generated, triggering equalization control (the BMS initiates passive equalization to discharge cells with excessively high voltage), while limiting the charging current to ≤10A.

[0068] Rule 5 (Motor Over-Temperature Protection): If the stator temperature of the drive motor is >120℃ (first threshold), a "warning" sign is generated, and the power is reduced to 70% of the rated power; if it is >150℃ (fault threshold), a "fault" sign is generated, and the drive motor stops running.

[0069] Furthermore, for the coupled "temperature-load-attitude" scenario, fuzzy reasoning is used to supplement the limitations of expert rules, as follows: Input quantities: battery pack fusion temperature (fuzzy subset: low temperature [0~20℃], normal temperature [20~40℃], high temperature [40~60℃], ultra-high temperature [60℃+]), implement resistance (fuzzy subset: light load [0~500N], medium load [500~1000N], heavy load [1000~2000N], extremely heavy load [2000N+]), vehicle pitch angle (fuzzy subset: small tilt angle [-2°~2°], medium tilt angle [2°~5°], large tilt angle [5°+]); Output: Health score (fuzzy subset: Good [80~100 points], Average [60~80 points], Poor [40~60 points], Terrible [below 40 points]); Fuzzy rules (example): If the battery pack fusion temperature is room temperature, the agricultural implement resistance is medium load, and the vehicle pitch angle is small tilt angle, then the health status is good. If the battery pack fusion temperature is high, the agricultural implement resistance is heavy load, and the vehicle pitch angle is medium tilt angle, then the health status is poor. If the battery pack fusion temperature is extremely high, the agricultural implement resistance is extremely heavy, and the vehicle pitch angle is large, then the health status is poor. Finally, using the centroid method of defuzzification, the fuzzy inference results are converted into specific scores. Scores ≥80 indicate "normal," 60-79 indicate "warning," and <60 indicate "fault." These scores are then weighted and fused with expert rule results (expert rule weight 0.7, fuzzy inference weight 0.3) to generate a three-level health status identifier. Based on the aforementioned expert rules and fuzzy inference, three-level identifiers—"normal," "warning," and "fault"—are generated, represented by 8-bit binary codes (0x01 = normal, 0x02 = warning, 0x03 = fault). These are sent to the control decision module via the CAN bus with a transmission cycle of 100ms and simultaneously stored in Flash memory, recording the identifier generation time, triggering rules, and corresponding parameter values ​​for easy fault tracing later. In another embodiment, assessing system health includes the following: Let the input feature vector be F, and the set of key security parameters be P; Wherein, F represents the feature vector extracted from the original multi-source heterogeneous signal for advanced evaluation. F includes: time-domain features such as mean, variance, and rate of change; and frequency-domain features such as fundamental frequency and harmonic distortion rate. P represents the direct safety parameter value after being cleaned and standardized by the signal processing module. For example: P[0] = battery pack fusion temperature; P[1] = high-voltage circuit insulation impedance; P[2] = actuator response deviation (others such as SOC, voltage balance, motor temperature, implement resistance, etc.).

[0070] Furthermore, the expert rule engine and fuzzy reasoning engine are evaluated in parallel; Expert rule engine output: H rule =Rules(F,P)∈{normal, warning, fault}; Where Rules(F, P) is the expert rule evaluation function, a decision function based on deterministic logic; F is the feature vector; P is the set of key security parameters; H rule This is a health status indicator output by the expert rule engine, a discrete variable with values ​​ranging from {normal, warning, fault}. Normal: indicates that all monitored key parameters are within safe thresholds and there are no abnormalities. Warning: indicates that one or more parameters deviate from the normal range but have not reached an immediate dangerous level, requiring performance limiting or operator attention. Fault: indicates that a serious abnormality has been detected, and protective actions (such as shutdown) must be taken immediately. The expert rule engine processes the data as follows: it makes judgments based on a series of predefined "IF-THEN" rules with clear thresholds (e.g., "if the insulation resistance is <10MΩ, then it is determined to be a fault"). Rules are assigned priorities (fault > warning > normal).

[0071] Fuzzy inference engine output: S fuzzy =FuzzyInference(F,P)∈[0,100], and map to H fuzzy ∈{normal, warning, fault}; FuzzyInference(F,P) is the fuzzy comprehensive evaluation function, a decision function based on fuzzy logic; S fuzzy The health score output by the fuzzy inference engine is a continuous variable, ranging from [0, 100]. A higher score indicates a better overall system health. fuzzy The health status identifier mapped by the fuzzy score is obtained by discretizing the Sfuzzy score by setting a threshold: if S fuzzy ≥80, then H fuzzy =Normal; if 60≤S fuzzy <80, then H fuzzy =Warning; if S fuzzy <60, then H fuzzy = Fault; The fuzzy inference engine processes the following steps: Precise input (e.g., temperature = 58℃) ​​is transformed into a fuzzy linguistic value (e.g., the membership degree of "high temperature" is 0.8) through a membership function. A fuzzy rule base (e.g., "If 'temperature' is high and 'load' is heavy, then 'health' is poor") is applied for calculation. The fuzzy output set is then aggregated into a single precise value.

[0072] Furthermore, a fusion decision-making process is implemented between the expert rule engine and the fuzzy inference engine (defining a Fusion function). The algorithm logic prioritizes safety. When the expert system detects a warning or fault (in a fast and deterministic manner), its conclusion is adopted to ensure rapid response to explicit risks. Only when the expert system deems everything normal is the incremental evaluation result of the fuzzy inference engine used to handle complex coupled operating conditions. The mathematical expression for the fusion decision is as follows: ; Among them, H final This is the final health status identifier determined by the system, i.e., the authoritative status sent to the control decision module after fusion decision-making. Its value is one of three: normal, warning, or fault, and is encoded according to a convention (e.g., 0x01, 0x02, 0x03); H rule Health status identifiers output by the expert rule engine; H fuzzy This represents the health status identifier mapped from the fuzzy rating; if indicates the start of a conditional branch. It means "is an element in the set of".

[0073] Meanwhile, the final health score is S final for: ; Among them, S final The health score is given in the final system report; H rule Health status identifiers output by the expert rule engine; H fuzzy This represents the health status identifier mapped from the fuzzy rating; if indicates the start of a conditional branch. It means "is an element in the set of..."; When H final By H rule When making the decision, S final A default mapping value corresponding to this state (e.g., a fault mapping of 30 points and a warning mapping of 65 points) is used for quantified display. When H final By H fuzzy When making the decision, S final It is equivalent to the original S fuzzy .

[0074] Among them, ifH rule ∈{Warning, Fault}: Defines the system's hard safety decision. It assigns the highest priority to expert rule conclusions under abnormal conditions. As soon as an expert rule reports a problem (warning or fault), its conclusion is immediately taken into account, regardless of the results of fuzzy reasoning. This ensures that the system's response to explicit risks is unconditional, immediate, and highly deterministic.

[0075] Among them, ifH rule=Normal: Defines the soft optimization region for system performance. Only when expert rules fail to identify any explicit problems does the system enter refined evaluation mode, allowing fuzzy inference to provide a more detailed and adaptive conclusion (H). fuzzy This allows the system to perform better power allocation and performance adjustments based on complex operating conditions within safety boundaries.

[0076] The final output is achieved through the final action (A). final Output, where A final By H final The only certainty is H final Transmitted via CAN bus according to agreed-upon encoding. For example: H final =Normal→A final =Run at full power; H final =Warning → A final = Reduce power to 70%; H final =Fault→A final =Emergency shutdown; The above definition of the Fusion function as a priority-based conditional selection, along with the mathematical logic of the entire evaluation process from input, parallel computation to fusion output, constitutes a computable and implementable health assessment model. Among them, the output triplet (H) final A final S final ): Represents the complete output of one work cycle (e.g., 100ms) of the health assessment model. H final and A final For real-time control, S final It can be used for human-machine interface display or historical data analysis.

[0077] In some embodiments, the prediction model includes a working condition identification unit, a power distribution unit, and an instruction generation unit; The working condition identification unit extracts time-domain and frequency-domain features based on the standardized multi-source heterogeneous signals, and identifies the current tractor working condition by combining historical working condition samples. The power distribution unit dynamically allocates the power ratio of each motor based on the identified operating conditions and health status indicators to ensure that the power output matches the operating conditions. The instruction generation unit converts the power allocation result into specific control signals, generates coordinated control instructions, and sends them to the execution module.

[0078] In this embodiment, key information is extracted from the standardized signal (multi-source heterogeneous signal) to determine the device status. This is achieved through the extraction of time-domain and frequency-domain features, including but not limited to generating the following feature vectors: Time-domain features (8-dimensional, F1−F8) are extracted from the signal waveform itself, including average value, fluctuation magnitude, maximum value, minimum value, rate of change, energy intensity, and impulse characteristics. These features describe the basic statistical behavior of the signal. Let the standardized original signal be X. std =[X std ,1,X std ,2,...,X std , N (N is the number of signal sampling points), the core time-domain feature formulas include: Mean: ; Among them, X std The original signal sequence is standardized; N is the number of signal sampling points; X std,i Let X be a vector std The i-th element in the summation; i=1 is the lower bound index of the summation; N is the upper bound index of the summation; F1 is the first time-domain feature extracted, i.e., the arithmetic mean of the signal sequence; The process of calculating the mean is as follows: Take the values ​​X of all N data points in the sequence... std,i Add them together to get the sum. Divide the sum by the number of data points N to get the average value F1. The average value F1 reflects the center value around which the signal fluctuates within the selected time window. For example: the average battery voltage reflects the average output voltage level of the battery. The average motor speed reflects the average operating speed of the motor. As a benchmark for condition assessment: deviations from the historical normal average are early signs of condition changes. For example, a continuous increase in the average motor current indicates increased load or decreased efficiency. The object of the average value processing is X. std It is a clean signal (multi-source heterogeneous signal) that has been denoised by Kalman filtering, removed outliers by the 3σ criterion, and normalized by Z-score.

[0079] The mean F1 is the first of eight time-domain features (F1-F8), which together constitute the part describing the time-domain characteristics of the signal. These eight time-domain features, together with the subsequently extracted eight frequency-domain features, form a 16-dimensional feature vector. This feature vector will be used as input for subsequent health status assessment.

[0080] variance: ; Where F2 is the second time-domain feature, representing the sample variance of the signal sequence; N is the number of signal sampling points, i.e., the total number of data points within the current analysis window; X std,i The i-th data point of the standardized signal sequence; F1 is the first time-domain feature; This indicates that the summation is performed on all terms from 1 to N for i; (X std,i -F1) 2The square of the difference between each data point and the mean reflects the degree of deviation of the individual data point; N-1 is used as the denominator instead of N, which is an unbiased estimate of the sample variance, especially when N is small, it can better approximate the population variance; The physical meaning of F² is that variance measures the dispersion or fluctuation range of a signal around its mean. A larger variance indicates more drastic signal fluctuations; a smaller variance indicates a more stable signal. For example, a sudden increase in the variance of motor current may indicate load fluctuations or mechanical loosening; an increase in the variance of battery voltage may reflect instability during charging and discharging. Therefore, F² is a key indicator for evaluating the operational stability of a system.

[0081] The remaining six dimensions—maximum, minimum, range (maximum minus minimum), average rate of change, root mean square, and peak factor—are extracted using single-value statistical logic, resulting in an 8-dimensional time-domain feature vector [F1, F2, ..., F8]. T From central tendency (mean), dispersion (variance, range), extreme values ​​(maximum / minimum), rate of change (average rate of change), energy (root mean square) to impulse characteristics (peak factor), the statistical behavior of the signal within the time window is characterized from multiple perspectives. This 8-dimensional vector, together with the subsequently extracted 8-dimensional frequency domain features, forms a 16-dimensional feature vector, which is input into a health status assessment model (such as a classifier, regression model, or neural network) to achieve functions such as fault diagnosis, life prediction, or condition classification. These features are simple to calculate and have clear physical meaning, making them suitable for real-time extraction in embedded systems, providing key feature inputs for online monitoring and intelligent maintenance of tractor electric drive systems.

[0082] Furthermore, frequency domain features (8-dimensional, F9−F) 16 The signal is converted into a frequency spectrum, from which the intensity and frequency of the most important frequency components, the degree of harmonic distortion, and the concentration of energy in the spectrum are extracted. These characteristics can effectively reflect the periodicity and potential faults of rotating machinery. Based on Fast Fourier Transform (FFT) for normalized signal X std The processing and frequency domain characteristic formula is as follows: Fundamental amplitude: F9 = max(|FFT(X)) std )∣); Where F9 is the 9th feature, i.e., the fundamental amplitude; max is the maximum value function; FFT(X) std ) for X std The complex spectrum sequence obtained after performing FFT; FFT is Fast Fourier Transform; X std The standardized signal sequence; The fundamental amplitude traditionally refers to the amplitude of the fundamental frequency (lowest frequency) component in a periodic signal. However, in this formula, it is defined as the maximum value of the amplitude spectrum, that is, the amplitude corresponding to the frequency component with the strongest energy in the signal. For signals with obvious periodic characteristics (such as motor speed and vibration signals), this maximum value often corresponds to the fundamental frequency or major harmonics, reflecting the intensity of that frequency component. For example, an abnormal increase in the fundamental amplitude in the motor current spectrum may indicate a sudden load change or electrical imbalance; an increase in the fundamental amplitude in the vibration spectrum may reflect mechanical misalignment or looseness. By performing an FFT on the standardized signal, the signal is transformed from the time domain to the frequency domain, and a series of statistical features (such as amplitude, centroid frequency, bandwidth, entropy, etc.) are extracted to reveal the frequency structure information of the signal. Frequency domain features are particularly sensitive to periodic and impulsive faults and can effectively supplement the deficiencies of time domain features.

[0083] Fundamental frequency: F 10 =f max (Calculated from the FFT frequency resolution fs / N); Among them, F 10 The 10th characteristic is the fundamental frequency; f max The frequency point corresponding to the largest amplitude in the spectrum; FFT stands for Fast Fourier Transform; fs is the sampling frequency; N is the number of points in the FFT. The fundamental frequency represents the repetition frequency of the dominant periodic component in a signal. For example, for a motor speed / vibration signal: F 10 This typically corresponds to the motor's rotational frequency (RPM) or its multiples (such as the fundamental electrical frequency in the motor current). For agricultural implement resistance signals: F 10 It reflects the periodic load fluctuations caused by uneven ground or crop distribution during cultivation. Monitoring F 10 Changes in frequency can help detect changes in the system's operating status in a timely manner. For example, a decrease in the fundamental frequency of a tractor drive motor may indicate that an increased load has led to a decrease in speed; a sudden change in the fundamental frequency of a hydraulic system pressure signal may indicate abnormal valve operation or leakage.

[0084] The remaining 6 dimensions—Total Harmonic Distortion (THD), amplitude percentage of each harmonic (2nd-5th), and frequency domain centroid—are extracted using frequency domain statistics and harmonic analysis logic. The final frequency domain feature is an 8-dimensional vector [F9, F 10 ,...,F 16 ] T .

[0085] In this embodiment, the above 16-dimensional feature vectors are fused together to form a comprehensive feature list for evaluating the health status of the component. Based on the above, time-domain features and frequency-domain features are concatenated sequentially to generate the feature vector input to the prediction model: F=[F1,F2,...,F8,F9,...,F 16 ] T ∈R16×1; Where F is the final eigenvector, which is a column vector; ∈R16×1 represents a 16-row, 1-column real vector space; T is the transpose symbol; This vector integrates the statistical properties and structural information of the signal in the time and frequency domains, forming a low-dimensional feature representation that comprehensively describes the current signal state.

[0086] In another embodiment, the work condition identification and determination includes four types of work conditions (plowing / sowing / harvesting / transportation). For example, the confidence threshold is set to 0.7. A simplified neural network (such as a feature vector layer input + four types of work conditions layer) is used. The feature vector is used as input. After linear transformation by the weight matrix (dimension R4×16) and the bias vector (dimension R4×1), it is normalized by the Softmax function and outputs the probability distribution (dimension R4×1) of the four types of work conditions (plowing / sowing / harvesting / transportation). The core formula is: P=Softmax(W·F+b). Where P is the output probability vector; Softmax(·) is the function that converts the scores of the four working conditions into a probability distribution; W·F+b is the output of the linear transformation, i.e. the score of each category; W·F is matrix multiplication; b is the bias vector; W is the weight matrix; and F is the input feature vector. Specifically, a 3D feature vector is obtained from front-end signal processing. A linear transformation is performed using a weight matrix and bias to obtain scores for four operating conditions. Each score reflects the correlation between the feature and the corresponding operating condition category. The scores are converted into probabilities using a Softmax function, making the output easily interpretable (probabilities for each category, summing to 1). After identifying the current operating condition, control strategies (such as power distribution and operating parameters) can be adjusted to optimize performance or save energy. The load and stress on components differ under different operating conditions; the operating condition identification unit helps to more accurately assess the health status of components (e.g., allowing higher temperatures under heavy load conditions). Finally, the running time for each operating condition is recorded, providing a basis for maintenance planning. Reliable identification of four operating conditions—plowing, sowing, harvesting, and transportation—is achieved. This provides crucial operating condition information for subsequent adaptive control and health status management.

[0087] The working condition determination rule sets a confidence threshold of 0.7 and uses a dual-condition approach of maximum probability index and maximum probability value to determine the final working condition. The formula for the final working condition is as follows: ; Where T represents the final working condition; a rgmax (P) is the index corresponding to the maximum value; max(P) is the maximum value of the probability vector; Let P be the confidence threshold, a scalar; and let P be the probability vector. The rule formula for the final working condition is the decision layer of the working condition identification unit. Based on the output probability vector distribution and the preset confidence threshold, it determines the final working condition category.

[0088] In this embodiment, health status is divided into three levels: normal, warning, and fault. Only the warning status needs correction. Normal / fault is a special value. The correction rules are strictly quantified (abnormal motors are downweighted, normal motors are upweighted, and the total weight is kept to 1). Based on the component health status, the health status of each key motor (drive motor, PTO motor, hydraulic motor) is determined: During tillage, the drive motor and PTO motor each account for 40%, and the hydraulic motor accounts for 20%; during transportation, the drive motor accounts for 70%. Under normal conditions, power is allocated strictly according to the baseline weighting. During a warning, while protecting abnormal components, the system strives to maintain total output power. The weighting of the abnormal motor is reduced (to 75% of its original weight), and the remaining weight is proportionally increased for other normal motors, ensuring a total weighting of 100%. In case of a fault, the fault is isolated to prevent further damage. The power of the faulty motor is immediately cut off (weight reduced to 0), and the remaining normal motors are redistributed their power according to their original baseline weighting. The system then calculates the final weights, combines them with the total power demand, and converts them into specific power control commands for each motor.

[0089] Finally, the entire system works in concert to achieve a real-time closed loop from raw signals to motor power control commands.

[0090] In some embodiments, the coordinated control commands are used to coordinate a composite command set of the electric drive system, the implement operation system, and the steering control system to achieve smooth switching of power modes, adaptive adjustment of steering assist, and linkage control between implement actions and vehicle movement.

[0091] In this embodiment, the coordinated control command is a composite command set that enables the linkage of electric drive, agricultural implement operation, and steering control. The specific logic is as follows: When switching from transportation mode to farmland mode, the drive motor power linearly increases from 80% of the rated power to 40% of the rated power (with corresponding weight adjustment), and the PTO motor power linearly increases from 10% to 40%, with a switching time of 2 seconds to avoid the impact caused by sudden power changes; conversely, the power decreases linearly during the switch, with a time of 1.5 seconds.

[0092] When the resistance of the implement exceeds 1500N in tillage conditions, the driving speed is automatically reduced (from 5km / h to 3km / h), while the PTO torque is increased (by 10%). When the implement is lifted, the PTO motor stops running, the drive motor power is increased by 20%, the driving speed is increased, and the work efficiency is improved.

[0093] Figure 2 This is an architecture diagram of the execution module of the integrated control system for the electrification retrofit of electric tractors provided in this application embodiment.

[0094] In some embodiments, the execution module includes a control drive unit, a data acquisition unit, and a data output unit; The control and drive unit is used to execute vehicle power distribution strategy, steering control, and implement start-stop and lifting actions on the high-voltage energy storage system, electric drive system, implement operation system, steering control system, and attitude perception system. The data acquisition unit is used to collect the execution parameters of the high-voltage energy storage system, electric drive system, agricultural implement operation system, steering control system, and attitude perception system in real time. The data output unit is used to send the collected execution parameters as execution feedback information to the optimization module.

[0095] In this embodiment, the execution module includes a control drive unit, a data acquisition unit, and a data output unit. The hardware interface and communication protocol are clearly defined to ensure compatibility with other modules, as detailed below: The control drive unit integrates the drive motor controller, PTO motor controller, hydraulic valve group actuator, and steering controller.

[0096] Controller interface: The drive motor controller uses a three-phase AC output (U / V / W) to connect to the stator winding of the drive motor; The hydraulic valve group driver uses PWM output and is connected to the electro-hydraulic proportional valve; the steering controller uses DC output and is connected to the steering motor.

[0097] The data acquisition unit integrates multiple analog and digital signal acquisition modules to collect parameters such as current, voltage, temperature, and position of the actuator. Interfaces: The analog signal acquisition modules communicate with the MCU via an SPI interface, while the digital signal acquisition modules connect to the MCU via a GPIO interface.

[0098] The data output unit uses a CAN / LIN bus and an Ethernet interface to communicate with the optimization module and health status assessment information. The CAN bus baud rate is 250kbps (status feedback) and 500kbps (control commands). The Ethernet interface transmits large amounts of feedback information such as attitude and positioning, with port number 5001.

[0099] In this embodiment: the control decision module issues control commands (CAN / Ethernet) → the execution module data input unit receives and verifies them → the control drive unit converts them into execution signals (PWM / analog / three-phase power) → drives each actuator to move → the data acquisition unit collects execution parameters (temperature, current, position, etc.) → after filtering and encoding, they are sent to the optimization module through the data output unit → the optimization module adjusts the prediction model based on feedback to form a closed loop.

[0100] In some embodiments, the adaptive optimization process of the optimization module includes: The execution feedback information is compared with the expected results to generate performance deviation data; If the effect deviation data value exceeds the preset threshold, the feature weights of the prediction model are adjusted based on the correlation between the deviation value and the multi-source heterogeneous signal. If the deviation value is within the preset threshold, the current working condition data, control commands, and execution feedback information will be used as new samples to incrementally train the prediction model, thereby improving the prediction model's adaptability to complex working conditions.

[0101] In this embodiment, the optimization module is implemented based on the MCU and shares NPU resources with the control decision module. The adaptive optimization process is defined as follows: The execution feedback information (actual power, torque, temperature, etc.) is compared with the reference values ​​of the MPC algorithm to calculate the absolute deviation and relative deviation; The formula for calculating absolute deviation is as follows: ΔX abs =|X actual -X ref |; Where, ΔX abs X represents the absolute deviation. actual This is the actual measured value; X ref For reference only; The physical meaning of the absolute deviation calculation formula is that subtraction is the most direct way to measure the difference, and it can give a specific value, representing the absolute gap between the actual output and the expected target.

[0102] The formula for calculating relative deviation is as follows: ; Where, ΔX rel For relative deviation; X ref For reference value; ΔX abs This represents the absolute deviation; multiplication × 100% converts the decimal to a percentage. The purpose of division is to normalize, eliminating the influence of dimensions and orders of magnitude. Division converts absolute deviation into a ratio relative to a reference value. This allows deviations of different physical quantities (such as power vs. temperature) or different orders of magnitude (such as large motors vs. small motors) to be compared on the same scale. For example: ΔX abs =8 N⋅m,X ref =100 N⋅m, then 8 ÷ 100 = 0.08. This means the deviation accounts for 8% of the target value. Multiplication by 100% converts the proportion into an intuitive percentage. Percentages are a standard way to express relative magnitude in engineering, facilitating the setting and understanding of thresholds (e.g., deviation must not exceed 5%). For example, 0.08 × 100% = 8%, directly indicating a relative deviation of 8%. The system obtains the absolute deviation through subtraction, and then converts it into a relative deviation (percentage) through division and multiplication, thus obtaining a standardized, dimensionless control effect evaluation index. This index is compared with the dynamic operating condition threshold to drive subsequent optimization decisions (adjusting the prediction model or accumulating learning). This design ensures the fairness and adaptability of the evaluation, which is the foundation for the intelligent system to achieve accurate self-optimization.

[0103] Specifically, for example, relative deviation can be used as the effect deviation data; Further preset deviation thresholds (dynamically adjusted based on operating conditions): torque deviation threshold of 5% for farmland operating conditions, speed deviation threshold of 3% for transportation operating conditions, and motor temperature deviation threshold of 2℃.

[0104] If ΔX rel > Threshold: Based on the Pearson correlation analysis between the deviation value and the multi-source heterogeneous signal, adjust the feature weights of the prediction model (the higher the correlation, the greater the weight; for example, the torque deviation is highly correlated with the PTO motor power, so the PTO power feature weight is increased). The weight adjustment step size is 0.01. After adjustment, recalculate the objective function until the deviation is ≤ threshold. If ΔX rel ≤Threshold: The current working condition data (feature vector), control commands, and execution feedback information are added as new samples to the model training set. Incremental training is triggered every 100 samples (the convolutional and LSTM layers are frozen, only the fully connected layers are trained, the number of iterations is 20, and the learning rate is 0.0001). The model parameters are updated and stored in Flash to improve the predictive model's adaptability to complex working conditions.

[0105] The adjusted predictive model (feature weights or network parameters) will be used for the next round of condition identification or health assessment, thereby improving the overall system's perception accuracy. It can not only quickly correct for large errors, but also silently accumulate data and optimize the predictive model during daily operation, enabling the tractor to maintain excellent operational performance and reliability even when facing unseen complex operating conditions, component aging, or environmental changes.

[0106] The optimization module constructs an adaptive optimization process that includes deviation calculation, threshold judgment, and dynamic optimization. It compares the execution feedback information with the expected value, calculates the absolute and relative deviations, and adopts different optimization strategies based on whether the deviation exceeds the threshold: when the deviation exceeds the threshold, it adjusts the feature weights of the prediction model based on Pearson correlation analysis to quickly correct the model deviation; when the deviation is within the threshold, it uses operating data, control commands, and execution feedback information as new samples for incremental model training, freezes the deep network, and trains only the fully connected layers, thereby improving the model's adaptability to complex operating conditions, component aging, and environmental changes while ensuring optimization efficiency. The optimization module and the control decision module share hardware resources to achieve real-time optimization and updates of the model, enabling the control system to maintain high accuracy and high adaptability throughout its entire life cycle.

[0107] The following section provides a detailed description and explanation of the solutions in this application, using a specific scenario of integrated control for the electrification retrofitting of electric tractors: In this embodiment of the application, an integrated control system for the electrification conversion of electric tractors is provided. This system is designed specifically for the electrification upgrade of traditional fuel tractors and can be adapted to the conversion needs of fuel tractors of different tonnages and different types of operations. It does not require reconstruction of the original chassis mechanical structure of the tractor, but only replaces the original fuel power and control components through modular integration. It realizes the coordinated control of multiple systems such as high-voltage energy storage, electric drive, and agricultural implement operation, taking into account the convenience of conversion, control precision and operational safety.

[0108] Compared to traditional fuel-powered tractors, this system achieves zero emissions during operation through electric drive, aligning with the trend of green agricultural development. At the same time, it avoids ineffective power consumption through adaptive power allocation based on working conditions. For example, the drive motor accounts for 70% of the power in transportation conditions, while the drive and PTO motors are allocated as needed in tillage conditions, achieving precise power supply, improving energy efficiency, and reducing operating energy consumption.

[0109] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0110] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0111] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0112] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0113] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0114] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0115] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0116] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0117] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0118] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0119] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. An integrated control system for the electrification conversion of electric tractors, characterized in that, The system includes: The signal processing module is used to acquire and process the multi-source heterogeneous signals output by the modified tractor; A health status assessment module is used to assess the system health and generate a health status identifier based on the multi-source heterogeneous signals. The control decision module is used to generate a prediction model based on the health status identifier and the processed multi-source heterogeneous signal, and output cooperative control commands based on the prediction model. The execution module is used to respond to the coordinated control command, execute the corresponding vehicle control actions, and generate execution feedback information; An optimization module is used to optimize the prediction model based on the execution feedback information.

2. The integrated control system for the electrification conversion of electric tractors according to claim 1, characterized in that, The modified tractor includes: High-voltage energy storage system, electric drive system, agricultural implement operation system, steering control system, attitude sensing system; The high-voltage energy storage system is used to provide high-voltage power to the vehicle. The electric drive system is used to drive the tractor and output power. The agricultural implement system is used for tilling, sowing, and harvesting operations. The steering control system is used for vehicle steering adjustment; The attitude perception system is used to collect the vehicle's operating attitude and position information.

3. The integrated control system for the electrification conversion of electric tractors according to claim 2, characterized in that, The multi-source heterogeneous signal includes at least one of the following: Heterogeneous signals of the vehicle's high-voltage battery status, power output shaft speed and torque, implement resistance, steering motor speed and steering angle, and vehicle attitude and position.

4. The integrated control system for the electrification conversion of electric tractors according to claim 3, characterized in that, The processing of the multi-source heterogeneous signals includes: For the aforementioned multi-source heterogeneous signals, a bus protocol is used for protocol conversion and format parsing to extract valid multi-source heterogeneous signal data; For the multi-source heterogeneous signal, the Kalman filter algorithm is used to remove high-frequency noise and abnormal pulses in the multi-source heterogeneous signal, and extreme outliers are removed by the 3σ criterion to ensure the stability of the multi-source heterogeneous signal. For the multi-source heterogeneous signals, the Z-score normalization method is used to convert the parsed and eliminated multi-source heterogeneous signals into signals of a unified dimension, eliminating the difference in dimensions and providing a standardized data foundation for subsequent modules.

5. The integrated control system for the electrification conversion of electric tractors according to claim 4, characterized in that, The health status assessment module includes a security rule unit, which includes at least one of the following rules: If the high-voltage battery pack fusion temperature exceeds the first threshold and the implement is under heavy load, a performance-limited flag will be generated. If a critical actuator response deviation is detected to continuously exceed the tolerance range, a protective intervention flag is generated, and an action is taken to switch the vehicle to a safe state.

6. The integrated control system for the electrification conversion of electric tractors according to claim 5, characterized in that, The health status identifier is based on a safety rule unit, which performs a weighted evaluation of the multi-source heterogeneous signal parameters to generate a three-level health status identifier: normal status, warning status, and fault status. Among them, the warning state triggers power reduction control, and the fault state triggers emergency shutdown protection.

7. The integrated control system for the electrification conversion of electric tractors according to claim 6, characterized in that, The prediction model includes a working condition identification unit, a power distribution unit, and a command generation unit; The working condition identification unit extracts time-domain and frequency-domain features based on the standardized multi-source heterogeneous signals, and identifies the current tractor working condition by combining historical working condition samples. The power distribution unit dynamically allocates the power ratio of each motor based on the identified operating conditions and health status indicators to ensure that the power output matches the operating conditions. The instruction generation unit converts the power allocation result into specific control signals, generates coordinated control instructions, and sends them to the execution module.

8. The integrated control system for the electrification conversion of electric tractors according to claim 7, characterized in that, The coordinated control commands are used to coordinate the composite command set of the electric drive system, the implement operation system, and the steering control system to achieve smooth switching of power modes, adaptive adjustment of steering assist, and linkage control between implement actions and vehicle movement.

9. The integrated control system for the electrification conversion of electric tractors according to claim 8, characterized in that, The execution module includes a control drive unit, a data acquisition unit, and a data output unit; The control and drive unit is used to execute vehicle power distribution strategy, steering control, and implement start-stop and lifting actions on the high-voltage energy storage system, electric drive system, implement operation system, steering control system, and attitude perception system. The data acquisition unit is used to collect the execution parameters of the high-voltage energy storage system, electric drive system, agricultural implement operation system, steering control system, and attitude perception system in real time. The data output unit is used to send the collected execution parameters as execution feedback information to the optimization module.

10. The integrated control system for the electrification conversion of electric tractors according to claim 9, characterized in that, The adaptive optimization process of the optimization module includes: The execution feedback information is compared with the expected results to generate performance deviation data; If the effect deviation data value exceeds the preset threshold, the feature weights of the prediction model are adjusted based on the correlation between the deviation value and the multi-source heterogeneous signal. If the deviation value is within the preset threshold, the current working condition data, control commands, and execution feedback information will be used as new samples to incrementally train the prediction model.

Images