Intelligent transverse and longitudinal prediction control system and method applied to electric forklift

Abstract

The invention discloses an intelligent transverse and longitudinal prediction control system applied to an electric forklift, and the system comprises a longitudinal driving intention analysis and prediction subsystem which is used for analyzing and predicting a longitudinal driving intention and predicting an optimal longitudinal driving intention sequence; the transverse driving intention analysis and prediction subsystem is used for analyzing and predicting a transverse driving intention and predicting to obtain an optimal transverse driving intention sequence; the traction motor rotating speed prediction control system is used for controlling the rotating speed of the traction motor; according to the method, parameter variables such as the rotating speed of the traction motor and the rotating speed of the oil pump motor can be predicted and adjusted in a mode of predicting transverse and longitudinal driving intentions, so that stability and smoothness of transverse and longitudinal control of a vehicle are guaranteed; and the conditions of longitudinal instability, jerking and poor micro-motion caused by untimely response of the traction motor and unsmooth steering, abnormal sound of the oil pump and the like caused by unreasonable control of the transverse oil pump motor are avoided.

Description

Technical Field

[0001] The present invention relates to the field of motion control of electric forklifts, and in particular to an intelligent lateral and longitudinal prediction control system and method applied to electric forklifts. Background Art

[0002] With the increasing global awareness of environmental protection and the requirements of various countries to reduce carbon emissions, electric forklifts have become an ideal substitute for traditional internal combustion forklifts due to their zero emissions and low noise, with low maintenance costs and energy consumption; the application scenarios of electric forklifts have expanded from traditional warehouse handling to more industries, such as ports, automobiles, steel and other industries. Technological advances in batteries, motors, electronic control systems, etc. have made electric forklifts not only more reliable and durable, but also more efficient and energy-saving, enhancing the competitiveness of the products.

[0003] Although electric forklifts have adopted advanced technologies and control systems to optimize their operating performance during the design and manufacturing process, there may still be some challenges and problems in actual applications. For example, in the longitudinal control and lateral control of electric forklifts, in the longitudinal control of electric forklifts, due to the failure of the vehicle action to match the driving intention in time, 1) Unsmooth acceleration and deceleration control: If the accelerator pedal responds too quickly or too slowly, it may cause a sudden acceleration change when the forklift starts or stops, affecting driving comfort and possibly causing damage to the cargo; 2) Delayed braking response: The braking system responds slowly or the braking force does not meet the driver's expectations, which may result in a long parking distance in an emergency and increase the risk of accidents.

[0004] In the lateral control of electric forklifts, the lateral movement of the vehicle cannot be matched with the driving intention in time, resulting in: 1) Inconsistent steering response: When the driver is operating the steering, the actual steering angle of the electric forklift cannot be consistent with the driver's expectations in real time, which may cause the lateral movement of the electric forklift to be not smooth; 2) Poor lateral stability: If the electric forklift is loaded with high or unbalanced cargo, lateral movement may increase the risk of rollover. If the design of the forklift does not fully consider lateral stability, the driver's operating intention may not be safely realized; 3) Unclear steering force feedback: On manually controlled electric forklifts, if the force feedback provided by the steering mechanism is not obvious or does not conform to conventional driving habits, the driver may be confused, which will affect the accuracy of the operation. Different drivers have different driving habits. If the control system is too rigid and cannot adapt to personalized needs, it may cause some drivers to feel uncomfortable, which will affect operational efficiency and safety. Summary of the invention

[0005] In view of the problems existing in the above-mentioned prior art, the present invention provides a crusher lubricating oil replacement reminder alarm system and a working method thereof, which are used to solve the problems of acceleration and deceleration, starting, uneven braking, and uneven steering in longitudinal and lateral control of electric forklifts.

[0006] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is: an intelligent lateral and longitudinal prediction control system applied to an electric forklift, comprising:

[0007] Longitudinal driving intention analysis and prediction subsystem: used to analyze and predict longitudinal driving intention and predict the optimal longitudinal driving intention sequence;

[0008] Lateral driving intention analysis and prediction subsystem: used to analyze and predict lateral driving intentions and predict the optimal lateral driving intention sequence;

[0009] Traction motor speed prediction control system: used for linear matching table lookup between the traction motor and the longitudinal driving intention, and the speed of the traction motor is controlled by smoothly outputting the speed of the traction motor during the transition period of each intention; oil pump motor speed prediction control system: used for linear matching table lookup between the oil pump motor and the lateral driving intention, and the speed of the oil pump motor is controlled by smoothly outputting the speed of the oil pump motor during the transition period of each intention.

[0010] Furthermore, the longitudinal driving intention analysis and prediction subsystem includes: longitudinal driving intention data processing, longitudinal driver intention cluster analysis, and longitudinal motion parameter adaptive adjustment module.

[0011] Furthermore, the lateral driving intention analysis and prediction subsystem includes: lateral driving intention data processing, hidden Markov model HMM construction, HMM model evaluation, and driver lateral driving intention prediction module.

[0012] Furthermore, when the vehicle is in action, the longitudinal driving intention analysis and prediction subsystem needs to process the detected longitudinal driving related data, and perform cluster analysis and longitudinal driving intention prediction on the processed data; the traction motor speed prediction control system adjusts the acceleration gradient, braking rate, and traction motor speed according to the monitored longitudinal driving intention data to achieve data prediction and control processing for longitudinal driving; the lateral driving intention analysis and prediction subsystem processes the lateral action related data generated by the vehicle, and performs HMM model construction and prediction of the optimal lateral driving intention sequence at the next moment on the processed data; the oil pump motor speed prediction control system adjusts the oil pump motor speed in advance according to the predicted lateral driving intention sequence to achieve speed control of the oil pump motor.

[0013] An intelligent lateral and longitudinal predictive control method for an electric forklift, comprising: a longitudinal predictive control method and a lateral predictive control method,

[0014] Longitudinal predictive control method:

[0015] When the driver controls the motion of the electric forklift, the longitudinal driving intention analysis and prediction are first performed: the longitudinal driving intention data of the electric forklift is collected in real time, and the feature data set is obtained after preliminary filtering. The longitudinal driving intention data is respectively transformed and processed by wavelet function conversion and wavelet soft threshold denoising method, and then the longitudinal driving intention feature data is clustered and predicted by K-means clustering analysis method. After the clustering prediction is completed, a first-order Markov model is built for state prediction. The longitudinal driving intention prediction data obtained by the first-order Markov model prediction analysis is input into the traction motor speed prediction control system, and the longitudinal motion parameters are adaptively adjusted. The speed of the traction motor is predicted and controlled to ensure the stability and smoothness of the vehicle's longitudinal driving.

[0016] Lateral predictive control method:

[0017] After collecting and processing the lateral driving intention data of the electric forklift, an HMM model is built, and a parameter set of the lateral driving intention is constructed. The parameter set is input into the HMM for model solving, and the lateral driving intention prediction at the next moment is realized. The driving intention prediction set is obtained, and the Baum-Welch algorithm is called to obtain the lateral optimal driving intention prediction sequence; the lateral driving intention prediction sequence is output to the oil pump motor speed prediction control system. The oil pump motor speed prediction control system realizes the adaptive adjustment of the oil pump motor speed of the electric forklift during lateral control according to the lateral driving intention prediction sequence, so as to ensure the lateral control smoothness of the electric forklift.

[0018] Furthermore, the longitudinal predictive control method is specifically as follows:

[0019] First, longitudinal driving intention data is collected, mainly including: accelerator pedal opening, vehicle speed, brake pedal opening data, and the feature data set Data is obtained after preliminary filtering of the data L1 ; Using discrete wavelet function to analyze Data L1 The data in the time-frequency domain are respectively transformed into discrete wavelet functions as shown in Formula 1 and Formula 2:

[0020] φ(t)=∑ k g(k)φ(2t-k) (1)

[0021] ψ(t)=∑ k h(k)φ(2t-k) (2)

[0022] Where: g(k), g(k) are filter coefficients; ψ(t) is the wavelet function; φ(t) is the scale transformation function; k≥0 indicates the translation of the function in the time domain;

[0023] By performing wavelet decomposition and reconstruction on the longitudinal driving intention data set DataL1, we can initially obtain the high-frequency and low-frequency components of each driving data set and obtain the data set Data L2 Since the low-frequency component data is close to the driving intention and the high-frequency component data is a mutation data, it is necessary to process the data after wavelet processing. L2 , perform wavelet soft threshold denoising: for each wavelet coefficient, if its absolute value exceeds a preset threshold, then retain the coefficient but reduce its amplitude to the threshold; if the absolute value is less than or equal to the threshold, then set the coefficient to zero, thereby ensuring that the smaller coefficients caused by noise are removed by the soft threshold processing method, as shown in Formula 3:

[0024]

[0025] c'=sign(c).max(|c|-T,0)(4)where: c is the wavelet coefficient; T is the threshold; sign(c) is the sign function of the wavelet coefficient c; max(|c|-T,0) is the larger of |c|-T and 0; when |c| is greater than the threshold T, the output is |c|-T and multiplied by its sign; c' is the new wavelet coefficient;

[0026] The threshold T can balance the relationship between denoising effect and retaining signal features. The selection of the threshold T is determined using the Donoho and Johnstone empirical formula, as shown in Formula 5:

[0027]

[0028] Where: σ is the noise standard deviation; N is the signal length;

[0029] After processing by wavelet soft threshold denoising method, Data L3 ; Use K-means clustering analysis method to process the noise-processed Data L3 The data set is processed and clustered into five types of driving intention feature data: speed maintenance, vehicle acceleration, deceleration, and parking. First, five initial center points are set according to the five types of driving intention data: speed maintenance, vehicle acceleration, deceleration, and parking. The point with the smallest Euclidean distance from each data point to the center point is assigned to the center point, as shown in Formula 6:

[0030]

[0031] Where: d is the Euclidean distance; x i is the i-th data point; c i is the position of the i-th center point;

[0032] According to the average value of the coordinates of all data points in the cluster, the center point position of each cluster is updated and calculated, as shown in formula (7):

[0033]

[0034] Where: C j is the set of all data points assigned to the jth center point; |C j | is a set |C j |The number of elements; According to the optimization target of the K-means algorithm, the sum of square errors within the cluster is minimized, and the algorithm is updated and iterated, as shown in formula (8):

[0035]

[0036] Where: J is the total intra-cluster error; k is the number of clusters; C j is the set of all data points assigned to the jth center point;

[0037] After real-time intention clustering analysis of the longitudinal driving intention data, a first-order Markov model is built to predict the state of the longitudinal driving intention. The longitudinal driving intention feature data at the next moment after state prediction by the first-order Markov model is transmitted to the traction motor speed prediction control system for subsequent traction motor speed prediction control.

[0038] Furthermore, a first-order Markov model is constructed to predict the state of longitudinal driving intention: First, a Markov state set S is constructed, as shown in formula (9):

[0039] S={s2 s2...s k}(9)

[0040] Where: s1, s2, s k The longitudinal driving intention data obtained by real-time cluster analysis,

[0041] Substitute the longitudinal driving intention data into the state transfer matrix, as shown in formula (10):

[0042] A=[ξ ij ](10)

[0043] Where: A is the N×N state transition probability matrix, ξ ij is the longitudinal driving intention state s i To state s j According to the state prediction formula, the longitudinal driving intention state data is predicted, and the state prediction is shown in formula (11):

[0044] p(t)=p(t-1)·A(11)

[0045] Where: p(t) and p(t-1) are 1×N-dimensional row vectors, representing the probability distribution at time t and t-1.

[0046] Furthermore, the longitudinal motion parameters are adaptively adjusted, mainly including: acceleration gradient adjustment, braking rate adjustment, so as to achieve smooth processing of the rotation speed of the traction motor involved in longitudinal driving.

[0047] Furthermore, the lateral predictive control method is specifically as follows:

[0048] First, the lateral driving intention data is collected, mainly including: steering angle, vehicle speed and other data. After data filtering, the feature data set DataT1 is obtained; the HMM model is constructed by inputting the HMM parameter set, and the parameter set includes: a finite set of driving intentions, a driving intention feature sequence, a driving intention transfer probability, and a driving intention initial distribution probability. According to the set emergency steering and general steering, the emergency steering is set to a steering wheel angular velocity greater than β1° / s when the speed is α1km / h, and the general steering is set to a steering wheel angular velocity greater than β2° / s when the speed is α2km / h; the emergency steering and general steering are set values; the driver's steering behavior is divided into two actions, and the emergency steering action and the general steering action are used as the two implicit states of the lateral driving model to construct the turning angle observation value set Cr={Cr1,Cr2,...,Cr n}, combined with the Baum-Welch algorithm to obtain the optimal driving intention sequence, the corresponding parameter model of the HMM model, is composed of vector Description, as shown in formula (12):

[0049]

[0050] Where: Q = [Q1, Q2] is the driving intention set, Cr is the turning angle observation value set, α ij is the transition probability from driving intention i to driving intention j, where i,j∈[1,n], b j (k) is the probability of driving operation k observed at driving intention j, P is the distribution probability of the driver's initial driving intention, and the driving intention transfer probability is shown in formula (13):

[0051]

[0052] Where: α t (i) is the probability of observing the operation before time t at driving intention i, β t (i) The probability of observing the operation after time t when driving intention i; b j (Cr t+1 ) is the driving operation Cr observed at driving intention j t+1 Probability

[0053] According to the Baum-Welch algorithm re-estimation formula, the estimated probability of driving intention transfer can be obtained, as shown in formulas (14), (15), and (16):

[0054]

[0055] P i =χ1(i) (16)

[0056] is the estimated probability of driving intention transferring from i to driving intention j, The estimated probability of observing driving action k at time j is P i is the distribution estimate probability of the initial driving intention, δ(o t ,v k ) is the optimal driving intention operation sequence; t (i) is the probability of observing a complete driving operation sequence with driving intention i at time t;

[0057] The maximum likelihood estimation function is used to estimate the initial parameter vector of the HMM model, as shown in Formula 17:

[0058]

[0059] Where: X is the observation sequence; L(θ; X) is the likelihood function; P(x i |θ) is the probability mass function; θ is the target parameter;

[0060] Get the initial parameter vector of the HMM model Then, according to the HMM initial model parameter vector , and according to the set of observation values ​​of the corner experimental data Cr, it can be estimated that the HMM model parameter vector , combined with the state transition probability Realize the driving intention prediction at the next moment and obtain the driving intention prediction set Q f ; Predict the lateral driving intention set Q f , according to Q f After predicting the steering wheel angle value, the steering angle value prediction sequence Cr is obtained f = {Cr 1f ,Cr 2f ,...,Cr nf}, the steering angle prediction sequence is transmitted to the oil pump motor speed prediction control system, and the oil pump motor speed is predicted according to the steering angle value prediction sequence, thereby achieving smooth steering control during the steering process.

[0061] The beneficial effects of the present invention are as follows: by predicting the lateral and longitudinal driving intentions, parameter variables such as the traction motor speed and the oil pump motor speed are predicted and adjusted to ensure the stability and smoothness of the vehicle in the lateral and longitudinal control, and to avoid longitudinal instability, setbacks, poor micro-mobility caused by the untimely response of the traction motor, and uneven steering, abnormal noise from the oil pump caused by unreasonable control of the lateral oil pump motor. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is the control principle diagram of the present invention;

[0063] Figure 2 This is the overall flow chart of the longitudinal driving intention analysis and prediction subsystem;

[0064] Figure 3 This is the overall flow chart of the lateral driving intention analysis and prediction subsystem. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solution and advantages of the present invention more clear, the present invention is further described in detail below through the accompanying drawings and embodiments. However, it should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the scope of the present invention.

[0066] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which the present invention belongs. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0067] like Figure 1 As shown, an intelligent lateral and longitudinal predictive control system applied to an electric forklift is divided into four parts: a longitudinal driving intention analysis and prediction subsystem, a lateral driving intention analysis and prediction subsystem, a traction motor speed predictive control system, and an oil pump motor speed predictive control system.

[0068] When the vehicle is in action, the longitudinal driving intention analysis and prediction subsystem needs to process the detected longitudinal driving related data, and perform cluster analysis and longitudinal driving intention prediction on the processed data; the traction motor speed prediction control system adjusts the acceleration gradient, braking rate, and traction motor speed according to the monitored longitudinal driving intention data to achieve data prediction and control processing for longitudinal driving; the lateral driving intention analysis and prediction subsystem processes the lateral action related data generated by the vehicle, and builds an HMM model for the processed data and predicts the optimal lateral driving intention sequence at the next moment; the oil pump motor speed prediction control system adjusts the oil pump motor speed in advance according to the predicted lateral driving intention sequence to achieve speed control of the oil pump motor.

[0069] The longitudinal driving intention analysis and prediction subsystem includes modules such as longitudinal driving intention data processing, longitudinal driver intention cluster analysis, and longitudinal motion parameter adaptive adjustment.

[0070] like Figure 2 As shown, the longitudinal predictive control method is as follows:

[0071] First, longitudinal driving intention data is collected, mainly including: accelerator pedal opening, vehicle speed, brake pedal opening and other data. After preliminary filtering of the data, the feature data set Data is obtained. L1 ; Using discrete wavelet function to analyze Data L1 The data in the time-frequency domain are transformed into discrete wavelet functions as shown in Formula 1 and Formula 2:

[0072] φ(t)=∑ k g(k)φ(2t-k) (1)

[0073] ψ(t)=∑ k h(k)φ(2t-k) (2)

[0074] Where: g(k), g(k) are filter coefficients; ψ(t) is the wavelet function; φ(t) is the scale transformation function; k≥0 indicates the translation of the function in the time domain.

[0075] By performing wavelet decomposition and reconstruction on the longitudinal driving intention data set DataL1, we can initially obtain the high-frequency and low-frequency components of each driving data set and obtain the data set Data L2 Since the low-frequency component data is close to the driving intention and the high-frequency component data is a mutation data, it is necessary to process the data after wavelet processing. L2 , perform wavelet soft threshold denoising: for each wavelet coefficient, if its absolute value exceeds a preset threshold, then retain the coefficient but reduce its amplitude to the threshold; if the absolute value is less than or equal to the threshold, then set the coefficient to zero, thereby ensuring that the smaller coefficients caused by noise are removed by the soft threshold processing method, as shown in Formula 3:

[0076]

[0077] c'=sign(c).max(|c|-T,0) (4)

[0078] Where: c is the wavelet coefficient; T is the threshold; sign(c) is the sign function of the wavelet coefficient c; max(|c|-T,0) is the larger of |c|-T and 0; when |c| is greater than the threshold T, the output is |c|-T and multiplied by its sign; c' is the new wavelet coefficient.

[0079] The threshold T can balance the relationship between denoising effect and retaining signal features. The selection of the threshold T is determined using the Donoho and Johnstone empirical formula, as shown in Formula 5:

[0080]

[0081] Where: σ is the noise standard deviation; N is the signal length;

[0082] After processing by wavelet soft threshold denoising method, Data L3 ; Use K-means clustering analysis method to process the noise-processed Data L3 The data set is processed and clustered into five types of driving intention feature data: speed maintenance, vehicle acceleration, deceleration, and parking. First, five initial center points are set according to the five types of driving intention data: speed maintenance, vehicle acceleration, deceleration, and parking. The point with the smallest Euclidean distance from each data point to the center point is assigned to the center point, as shown in Formula 6:

[0083]

[0084] Where: d is the Euclidean distance; x i is the i-th data point; c i is the position of the i-th center point.

[0085] According to the average value of the coordinates of all data points in the cluster, the center point position of each cluster is updated and calculated, as shown in formula (7):

[0086]

[0087] Where: C j is the set of all data points assigned to the jth center point; |C j | is a set |C j |The number of elements;

[0088] According to the optimization target of the K-means algorithm, the sum of square errors within the cluster is minimized, and the algorithm is updated and iterated, as shown in formula (8):

[0089]

[0090] Where: J is the total intra-cluster error; k is the number of clusters; C j is the set of all data points assigned to the jth center point;

[0091] After real-time intention clustering analysis of longitudinal driving intention data, a first-order Markov model is built to predict the state of longitudinal driving intention;

[0092] First, construct the Markov state set S, as shown in formula (9):

[0093] S={s2 s2 ... s k} (9)

[0094] Where: s1, s2, s k The longitudinal driving intention data obtained by real-time cluster analysis,

[0095] Substitute the longitudinal driving intention data into the state transfer matrix, as shown in formula (10):

[0096] A=[ξ ij ] (10)

[0097] Where: A is the N×N state transition probability matrix, ξ ij is the longitudinal driving intention state s i To state s j The probability of

[0098] According to the state prediction formula, the longitudinal driving intention state data is predicted. The state prediction is shown in formula (11):

[0099] p(t)=p(t-1)·A (11)where: p(t) and p(t-1) are 1×N-dimensional row vectors, representing the probability distribution at time t and t-1;

[0100] The longitudinal driving intention feature data at the next moment after state prediction by the first-order Markov model is transmitted to the traction motor speed prediction control system accordingly, so as to perform subsequent speed prediction control of the traction motor; in the traction motor prediction control system monitoring module, the longitudinal driving intention sequence and the first-order Markov state prediction sequence obtained by clustering are monitored to perform adaptive adjustment of the longitudinal motion parameters, mainly including: acceleration gradient adjustment and braking rate adjustment, so as to achieve speed smoothing of the traction motor involved in longitudinal driving.

[0101] The lateral driving intention analysis and prediction subsystem includes modules such as lateral driving intention data processing, Hidden Markov Model (HMM) construction, HMM model evaluation, and driver lateral driving intention prediction.

[0102] like Figure 3 As shown, the lateral predictive control method is as follows:

[0103] First, the lateral driving intention data is collected, mainly including: steering angle, vehicle speed and other data. After data filtering, the feature data set DataT1 is obtained; the HMM model is constructed by inputting the HMM parameter set, and the parameter set includes: a finite set of driving intentions, a driving intention feature sequence, a driving intention transfer probability, and a driving intention initial distribution probability. According to the set emergency steering and general steering, the emergency steering is set to a steering wheel angular velocity greater than β1° / s when the speed is α1km / h, and the general steering is set to a steering wheel angular velocity greater than β2° / s when the speed is α2km / h; the emergency steering and general steering are set values; the driver's steering behavior is divided into two actions, and the emergency steering action and the general steering action are used as the two implicit states of the lateral driving model to construct the turning angle observation value set Cr={Cr1,Cr2,...,Cr n}, combined with the Baum-Welch algorithm to obtain the optimal driving intention sequence, the corresponding parameter model of the HMM model, is composed of vector Description, as shown in formula (12):

[0104]

[0105] Where: Q = [Q1, Q2] is the driving intention set, Cr is the turning angle observation value set, α ij is the transition probability from driving intention i to driving intention j, where i,j∈[1,n], b j (k) is the probability of driving operation k observed at driving intention j, P is the distribution probability of the driver's initial driving intention, and the driving intention transfer probability is shown in formula (13):

[0106]

[0107] Where: α t (i) is the probability of observing the operation before time t at driving intention i, β t (i) The probability of observing the operation after time t when driving intention i; b j (Cr t+1 ) is the driving operation Cr observed at driving intention j t+1 Probability

[0108] According to the Baum-Welch algorithm re-estimation formula, the estimated probability of driving intention transfer can be obtained, as shown in formulas (14), (15), and (16):

[0109]

[0110]

[0111] P i =χ1(i) (16)

[0112] is the estimated probability of driving intention transferring from i to driving intention j, The estimated probability of observing driving action k at time j is P i is the distribution estimate probability of the initial driving intention, δ(o t ,v k ) is the optimal driving intention operation sequence; t (i) is the probability of observing a complete driving operation sequence with driving intention i at time t;

[0113] The maximum likelihood estimation function is used to estimate the initial parameter vector of the HMM model, as shown in Formula 17:

[0114]

[0115] Where: X is the observation sequence; L(θ; X) is the likelihood function; P(x i |θ) is the probability mass function; θ is the target parameter;

[0116] Get the initial parameter vector of the HMM model Then, according to the HMM initial model parameter vector According to the set of observation values ​​of the corner experimental data, Cr, it can be estimated that the HMM model parameter vector Combined state transition probability Realize the driving intention prediction at the next moment and obtain the driving intention prediction set Q f ;

[0117] The predicted lateral driving intention set Q f , according to Q f After predicting the steering wheel angle value, the steering angle value prediction sequence Cr is obtained f = {Cr 1f ,Cr 2f ,...,Cr nf}, the steering angle prediction sequence is transmitted to the oil pump motor speed prediction control system, and the oil pump motor speed is predicted according to the steering angle value prediction sequence, thereby achieving smooth steering control during the steering process.

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification, equivalent substitution or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent lateral and longitudinal predictive control system for an electric forklift, comprising: Longitudinal driving intention analysis and prediction subsystem: used to analyze and predict longitudinal driving intention and predict the optimal longitudinal driving intention sequence; Lateral driving intention analysis and prediction subsystem: used to analyze and predict lateral driving intentions and predict the optimal lateral driving intention sequence; Traction motor speed prediction control system: used for linear matching table lookup between the traction motor and the longitudinal driving intention, and controls the speed of the traction motor by smoothly outputting the speed of the traction motor during each intention transition period; Oil pump motor speed prediction control system: used for linear matching lookup table between oil pump motor and lateral driving intention, and controls the speed of oil pump motor by smoothly outputting the speed of oil pump motor during each intention transition period.

2. The intelligent lateral and longitudinal predictive control system for electric forklifts according to claim 1, characterized in that: The longitudinal driving intention analysis and prediction subsystem includes: longitudinal driving intention data processing, longitudinal driver intention cluster analysis, and longitudinal motion parameter adaptive adjustment module.

3. The intelligent lateral and longitudinal predictive control system for electric forklifts according to claim 1, characterized in that: The lateral driving intention analysis and prediction subsystem includes: lateral driving intention data processing, hidden Markov model HMM construction, HMM model evaluation, and driver lateral driving intention prediction module.

4. The intelligent lateral and longitudinal predictive control system for electric forklifts according to claim 1, characterized in that: When the vehicle is in action, the longitudinal driving intention analysis and prediction subsystem needs to process the detected longitudinal driving related data, and perform cluster analysis and longitudinal driving intention prediction on the processed data; the traction motor speed prediction control system adjusts the acceleration gradient, braking rate, and traction motor speed according to the monitored longitudinal driving intention data to achieve data prediction and control processing for longitudinal driving; the lateral driving intention analysis and prediction subsystem processes the lateral action related data generated by the vehicle, and builds an HMM model for the processed data and predicts the optimal lateral driving intention sequence at the next moment; the oil pump motor speed prediction control system adjusts the oil pump motor speed in advance according to the predicted lateral driving intention sequence to achieve speed control of the oil pump motor.

5. An intelligent lateral and longitudinal predictive control method for an electric forklift, characterized in that: include: Longitudinal predictive control method and lateral predictive control method, Longitudinal predictive control method: When the driver controls the motion of the electric forklift, the longitudinal driving intention analysis and prediction are first performed: the longitudinal driving intention data of the electric forklift is collected in real time, and the feature data set is obtained after preliminary filtering. The longitudinal driving intention data is respectively transformed and processed by wavelet function conversion and wavelet soft threshold denoising method, and then the longitudinal driving intention feature data is clustered and predicted by K-means clustering analysis method. After the clustering prediction is completed, a first-order Markov model is built for state prediction. The longitudinal driving intention prediction data obtained by the first-order Markov model prediction analysis is input into the traction motor speed prediction control system, and the longitudinal motion parameters are adaptively adjusted. The speed of the traction motor is predicted and controlled to ensure the stability and smoothness of the vehicle's longitudinal driving. Lateral predictive control method: After collecting and processing the lateral driving intention data of the electric forklift, an HMM model is built, and a parameter set of the lateral driving intention is constructed. The parameter set is input into the HMM for model solving, and the lateral driving intention prediction at the next moment is realized. The driving intention prediction set is obtained, and the Baum-Welch algorithm is called to obtain the lateral optimal driving intention prediction sequence; the lateral driving intention prediction sequence is output to the oil pump motor speed prediction control system. The oil pump motor speed prediction control system realizes the adaptive adjustment of the oil pump motor speed of the electric forklift during lateral control according to the lateral driving intention prediction sequence, so as to ensure the lateral control smoothness of the electric forklift.

6. The intelligent lateral and longitudinal predictive control method for an electric forklift according to claim 5, characterized in that: The longitudinal predictive control method is specifically as follows: First, longitudinal driving intention data is collected, mainly including: accelerator pedal opening, vehicle speed, brake pedal opening data, and the feature data set Data is obtained after preliminary filtering of the data L1 ; Using discrete wavelet function to analyze Data L1 The data in the time-frequency domain are respectively transformed into discrete wavelet functions as shown in Formula 1 and Formula 2: φ(t)=∑ k g(k)φ(2t-k) (1) ψ(t)=∑ k h(k)φ(2t-k) (2) Where: g(k), g(k) are filter coefficients; ψ(t) is the wavelet function; φ(t) is the scale transformation function; k≥0 indicates the translation of the function in the time domain; By performing wavelet decomposition and reconstruction on the longitudinal driving intention data set DataL1, we can initially obtain the high-frequency and low-frequency components of each driving data set and obtain the data set Data L2 Since the low-frequency component data is close to the driving intention and the high-frequency component data is a mutation data, it is necessary to process the data after wavelet processing. L2 , perform wavelet soft threshold denoising: for each wavelet coefficient, if its absolute value exceeds a preset threshold, then retain the coefficient but reduce its amplitude to the threshold; if the absolute value is less than or equal to the threshold, then set the coefficient to zero, thereby ensuring that the smaller coefficients caused by noise are removed by the soft threshold processing method, as shown in Formula 3: c'=sign(c).max(|c|-T,0) (4) Where: c is the wavelet coefficient; T is the threshold; sign(c) is the sign function of the wavelet coefficient c; max(|c|-T,0) is the larger of |c|-T and 0; when |c| is greater than the threshold T, the output is |c|-T and multiplied by its sign; c' is the new wavelet coefficient; The threshold T can balance the relationship between denoising effect and retaining signal features. The selection of the threshold T is determined using the Donoho and Johnstone empirical formula, as shown in Formula 5: Where: σ is the noise standard deviation; N is the signal length; After processing by wavelet soft threshold denoising method, Data L3 ; Use K-means clustering analysis method to process the noise-processed Data L3 The data set is processed and clustered into five types of driving intention feature data: speed maintenance, vehicle acceleration, deceleration, and parking. First, five initial center points are set according to the five types of driving intention data: speed maintenance, vehicle acceleration, deceleration, and parking. The point with the smallest Euclidean distance from each data point to the center point is assigned to the center point, as shown in Formula 6: Where: d is the Euclidean distance; x i is the i-th data point; c i is the position of the i-th center point; According to the average value of the coordinates of all data points in the cluster, the center point position of each cluster is updated and calculated, as shown in formula (7): Where: C j is the set of all data points assigned to the jth center point; |C j | is a set |C j |The number of elements; According to the optimization target of the K-means algorithm, the sum of square errors within the cluster is minimized, and the algorithm is updated and iterated, as shown in formula (8): Where: J is the total intra-cluster error; k is the number of clusters; C j is the set of all data points assigned to the jth center point; After real-time intention clustering analysis of the longitudinal driving intention data, a first-order Markov model is built to predict the state of the longitudinal driving intention. The longitudinal driving intention feature data at the next moment after state prediction by the first-order Markov model is transmitted to the traction motor speed prediction control system for subsequent traction motor speed prediction control.

7. The intelligent lateral and longitudinal predictive control method for an electric forklift according to claim 6, characterized in that: Build a first-order Markov model to predict the state of longitudinal driving intention: First, construct the Markov state set S, as shown in formula (9): S={s2 s2...s k } (9) Where: s1, s2, s k The longitudinal driving intention data obtained by real-time cluster analysis, Substitute the longitudinal driving intention data into the state transfer matrix, as shown in formula (10): A=[ξ ij ] (10)Where A is the N×N state transition probability matrix, ξ ij is the longitudinal driving intention state s i To state s j probability; According to the state prediction formula, the longitudinal driving intention state data is predicted. The state prediction is shown in formula (11): p(t)=p(t-1)·A (11) Wherein: p(t) and p(t-1) are 1×N-dimensional row vectors, representing the probability distribution at time t and t-1.

8. The intelligent lateral and longitudinal predictive control method for an electric forklift according to claim 5, characterized in that: The adaptive adjustment of longitudinal motion parameters mainly includes: acceleration gradient adjustment and braking rate adjustment, so as to achieve smooth processing of the rotation speed of the traction motor involved in longitudinal driving.

9. The intelligent lateral and longitudinal predictive control method for an electric forklift according to claim 5, characterized in that: The lateral predictive control method is specifically as follows: First, the lateral driving intention data is collected, mainly including: steering angle, vehicle speed and other data. After data filtering, the feature data set DataT1 is obtained; the HMM model is constructed by inputting the HMM parameter set, and the parameter set includes: a finite set of driving intentions, a driving intention feature sequence, a driving intention transfer probability, and a driving intention initial distribution probability. According to the set emergency steering and general steering, the emergency steering is set to a steering wheel angular velocity greater than β1° / s when the speed is α1km / h, and the general steering is set to a steering wheel angular velocity greater than β2° / s when the speed is α2km / h; the emergency steering and general steering are set values; the driver's steering behavior is divided into two actions, and the emergency steering action and the general steering action are used as the two implicit states of the lateral driving model to construct the turning angle observation value set Cr={Cr1,Cr2,...,Cr n }, combined with the Baum-Welch algorithm to obtain the optimal driving intention sequence, the corresponding parameter model of the HMM model, is composed of vector Description, as shown in formula (12): Where: Q = [Q1, Q2] is the driving intention set, Cr is the turning angle observation value set, α ij is the transition probability from driving intention i to driving intention j, where i,j∈[1,n], b j (k) is the probability of driving operation k observed at driving intention j, P is the distribution probability of the driver's initial driving intention, and the driving intention transfer probability is shown in formula (13): Where: α t (i) is the probability of observing the operation before time t at driving intention i, β t (i) The probability of observing the operation after time t when driving intention i; b j (Cr t+1 ) is the driving operation Cr observed at driving intention j t+1 Probability According to the Baum-Welch algorithm re-estimation formula, the estimated probability of driving intention transfer can be obtained, as shown in formulas (14), (15), and (16): P i =χ1(i) (16) is the estimated probability of driving intention transferring from i to driving intention j, The estimated probability of observing driving action k at time j is P i is the distribution estimate probability of the initial driving intention, δ(o t ,v k ) is the optimal driving intention operation sequence; t (i) is the probability of observing a complete driving operation sequence with driving intention i at time t; The maximum likelihood estimation function is used to estimate the initial parameter vector of the HMM model, as shown in Formula 17: Where: X is the observation sequence; L(θ; X) is the likelihood function; P(x i |θ) is the probability mass function; θ is the target parameter; Get the initial parameter vector of the HMM model Then, according to the HMM initial model parameter vector According to the set of observation values ​​of the corner experimental data, Cr, it can be estimated that the HMM model parameter vector Combined state transition probability Realize the driving intention prediction at the next moment and obtain the driving intention prediction set Q f ; The predicted lateral driving intention set Q f , according to Q f After predicting the steering wheel angle value, the steering angle value prediction sequence Cr is obtained f = {Cr 1f ,Cr 2f ,...,Cr nf }, the steering angle prediction sequence is transmitted to the oil pump motor speed prediction control system, and the oil pump motor speed is predicted according to the steering angle value prediction sequence, thereby achieving smooth steering control during the steering process.

Images