A method for roll-over prevention control of a counterbalanced fork truck

Abstract

The present application relates to a kind of anti-rollover control method for counterbalanced fork truck, collection fork truck dynamic and load data, calculate and normalize roll risk, energy consumption and smoothness index, construct multi-objective fitness function, adopt offline RBF neural network and multi-objective genetic algorithm optimization to generate MAP table, online table lookup obtains optimal control parameter, through CAN bus is sent to fork truck controller.The anti-rollover control method for counterbalanced fork truck, by adopting offline RBF and genetic algorithm to generate MAP table realizes online quick table lookup, and four types of working conditions are adjusted weight, introduce subsection dynamic adjustment weight to enhance transient response, introduce weight self-adapting re-optimization to ensure safety boundary, dynamically adjust crossover mutation probability in re-optimization and introduce annealing factor and dynamic retry number, balance exploration and development, solve the problems of poor real-time, insufficient scene adaptation, low robustness in extreme conditions, missing safety boundary and low optimization efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for forklifts, specifically to an anti-tipping control method for counterbalance forklifts. Background Technology

[0002] Forklift intelligent control technology is a core branch of industrial vehicle control technology. It focuses on achieving autonomous perception, precise decision-making, efficient execution, and safety protection of forklifts through multi-sensor fusion, AI algorithms, automated control, and intelligent scheduling. It is widely used in warehousing and logistics, intelligent manufacturing, port terminals, and other scenarios. Anti-rollover control is one of its core sub-directions. Its purpose is to monitor forklift load, vehicle posture, travel speed, and mast movement status in real time, dynamically predict the risk of rollover instability, and improve the driving and operation stability of forklifts under heavy load, steering, slope, and mast movement conditions by actively intervening in operating parameters such as vehicle speed, steering, and lifting actions, thereby avoiding rollover accidents.

[0003] In current technologies, anti-tipping control of counterbalance forklifts is based on assessing stability through lateral load transfer rate, and then optimizing through fuzzy PID, chain increment or quantum genetic algorithm to adjust the cylinder, steering or torque distribution, thereby achieving intelligent anti-tipping control of counterbalance forklifts. In practical use, the driving environment of counterbalance forklifts is one of the important factors affecting their stability. However, most existing technologies ignore the influence of external environmental factors such as road conditions and terrain slope, resulting in deviations in the calculation of core parameters such as lateral acceleration, insufficient accuracy in stability judgment, and inability to adapt to complex actual operation scenarios. Moreover, the simplified dynamic model or single data-driven model is difficult to adapt to the strong nonlinear and time-varying characteristics of forklifts, and lacks robustness to extreme conditions such as high-speed steering, heavy lifting, sudden load changes, and unstructured road surfaces, which can easily lead to problems such as control lag, over-control, or control failure. Therefore, an anti-rollover control method for counterbalance forklifts is proposed to solve the above problems. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an anti-tipping control method for counterbalance forklifts, which has the advantages of environmental adaptability and working condition self-correction. It solves the problem that in actual use, the driving environment of a counterbalance forklift is one of the important factors affecting its stability. Existing technologies mostly ignore the influence of external environmental factors such as road conditions and terrain slope, leading to deviations in the calculation of core parameters such as lateral acceleration, insufficient accuracy in stability judgment, and inability to adapt to complex actual operating scenarios. Furthermore, the simplified dynamic models or single data-driven models used are difficult to adapt to the strong nonlinear and time-varying characteristics of forklifts, and lack robustness to extreme working conditions such as high-speed steering, heavy lifting, sudden load changes, and unstructured road surfaces, easily resulting in problems such as control lag, over-control, or control failure.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for preventing tipping over of a counterbalance forklift, comprising the following steps: S1. Collect forklift speed V and longitudinal acceleration. Lateral acceleration Vertical acceleration Unloaded tare weight Goods quality Wheel angle δ, mast tilt angle θ, wheel angle change rate gantry tilt angle change rate And preprocessing is performed to obtain the basic dataset; S2. Based on the basic dataset and the inherent parameters of the forklift, the roll risk index R, control energy consumption index E, and ride comfort index S are calculated, and then normalized to obtain... , , ; S3. Normalized Side Roll Risk Indicator Controlling energy consumption indicators Smoothness index Construct a multi-objective fitness function F; S4. Optimal control parameters are obtained by combining offline optimization with online table lookup: In the offline phase, an RBF neural network is trained using simulation data. A multi-objective genetic algorithm is used to randomly generate candidate solutions, which are then input into the trained RBF network to obtain predictions. , , Then, the multi-objective fitness function F is calculated, and the population evolution is driven by minimizing F. Finally, a non-dominated optimal solution set is generated, and the optimal control parameter MAP table is obtained based on the non-dominated optimal solution set. During the online phase, the forklift controller directly retrieves the optimal control parameter MAP table to obtain the optimal control parameters for that operating condition. S5. Send the specific instructions in the optimal control parameters to the corresponding forklift pull-out controller via the CAN bus.

[0006] Furthermore, in step S1, the fork height is further collected. , terrain slope angle α, yaw rate γ; Lateral acceleration corrected based on terrain slope angle α To obtain the equivalent lateral acceleration in the actual road surface coordinate system The corrected value is then substituted into the calculation of the roll risk index R; At the same time, according to vehicle speed Equivalent lateral acceleration and the quality of goods The current operating conditions are categorized online into four types of operating condition labels: low speed light load, low speed heavy load, high speed light load, and high speed heavy load. when and At that time, it is classified as low speed and light load; when and At that time, it is classified as low-speed heavy-load; when and At that time, it is classified as high-speed light-load; when and At that time, it is classified as high-speed heavy-load; in, To preset the vehicle speed threshold, Fork height is set to a preset quality threshold. Used to combine unloaded self-weight Goods quality The weighted calculation of the mast tilt angle θ is used to determine the real-time center of gravity height h of the vehicle, and the forklift yaw rate γ is used to characterize the vehicle's inertial state. The expression for the multi-objective fitness function F in S3 is: ; Wherein, the weight coefficients satisfy Weighting coefficients , , Adjust according to the operating condition label.

[0007] Furthermore, the weighting coefficients , , Further, dynamic adjustment is introduced based on working condition labels, yaw rate γ, and yaw rate of change, through segmented condition rules; Let the default weight vectors corresponding to the four working condition labels be... And satisfy ; The first set of conditional rules is as follows: When the yaw rate If so, the default weight of the corresponding working condition label will be maintained; When the yaw rate ,but ,in , ,and ; When the yaw rate ,but ,in , , By normalization conditions Sure; The second set of conditional rules is as follows: When the rate of change of yaw angular velocity Then, further adjustments will be made based on the current weights: ; in, , And renormalize to make ; when If so, the current weight remains unchanged.

[0008] in, , These are the first and second threshold values ​​for the yaw rate, respectively. The threshold for the rate of change of yaw angular velocity. This is the gain coefficient for the effect of the rate of change of yaw rate.

[0009] Furthermore, in the S4 offline stage, a multi-objective genetic algorithm is executed to optimize according to different working condition labels to obtain the optimal control parameter MAP table under each working condition label; During the online phase, the forklift controller directly retrieves the optimal control parameter MAP table under the corresponding operating condition label based on the real-time identified operating condition label, and obtains the optimal control parameters under that operating condition label. In the S4 offline phase, candidate solutions are randomly generated within a set range, including the target pressure of the hydraulic support cylinder. Displacement of dynamic balance block Rear wheel active steering angle .

[0010] Furthermore, after generating the non-dominated optimal solution set, a standard-setting judgment for the non-dominated optimal solution set is introduced. Based on forklift speed V and cargo weight The terrain slope angle α is used to calculate the roll risk index under the current working conditions through a pre-calibrated mapping relationship. Controlling energy consumption indicators Smoothness index and set ; Among them, the weighting coefficient , , The value is consistent with that in S3; Calculate the normalized index for each solution in the non-dominated optimal solution set. , , And the multi-objective fitness function value F; Judgment Formula 1: , , ; If there exists a solution that satisfies criterion one, then select the solution with the smallest F from these solutions as the optimal control parameter for that working condition; If no solution satisfies criterion one, then adjust the weights of the multi-objective fitness function F and re-execute the optimization. in, This is a preset threshold for achieving the target.

[0011] Furthermore, the weights of the multi-objective fitness function F in the re-execution of optimization are further dynamically adjusted based on the non-compliance items and the deviation magnitude; Based on the actual tilt risk indicators achieved in the current candidate solution set Controlling energy consumption indicators Smoothness index Each with corresponding limit , , The comparison results are dynamically updated with weights as follows: like Then let , ; like If the weight remains unchanged, then the weight remains unchanged. like Then let , ; like If the weight remains unchanged, then the weight remains unchanged. like Then let , ; like If the weight remains unchanged, then the weight remains unchanged. in, , , All are positive coefficients. , , These represent the deviation magnitudes of the corresponding items.

[0012] Then to , , Normalization is performed so that the sum of the normalized weights is 1, and the normalized weights are used as the weight coefficients of the fitness function F in the next round of optimization.

[0013] Furthermore, in the re-execution of the optimization process, a maximum number of retries is set. ; If the number of consecutive re-executions of the optimization method reaches After that, it was still impossible to find a solution that simultaneously satisfies... , , If the solution is not found, the re-optimization process is terminated, and the solution with the smallest multi-objective fitness function F value in the current non-dominated solution set is selected as the optimal control parameter for that operating condition.

[0014] Furthermore, the re-execution optimization process further incorporates dynamic adjustment of the crossover probability based on the working condition label. and mutation probability ; If the identified operating condition label is low speed and light load, then maintain... ; If the identified operating condition label is low-speed heavy-load, then a load factor is introduced. That is, to adjust to ; If the identified operating condition label is high speed and light load, then a vehicle speed deviation is introduced. That is, to adjust to ; If the identified operating condition label is high-speed heavy load, then the crossover probability should be adjusted according to the above rules. and mutation probability ; in, As the baseline crossover probability, As the baseline mutation probability, , , , It is a positive coefficient.

[0015] Furthermore, the dynamic adjustment of the crossover probability and mutation probability At this time, an annealing factor that decreases with the number of re-executions of the optimization process is introduced. ; Let the current iteration be the t-th iteration of the optimization process, and let the annealing factor be... , This is the annealing rate coefficient; Then the adjustment coefficient in the actual adjustment formula is multiplied by Taking the operating condition label as low speed and heavy load as an example, the adjustment formula is as follows: ; If the identified operating condition label is high speed light load, high speed heavy load, or low speed light load, then the above rules shall be applied simultaneously.

[0016] Furthermore, the maximum number of retries for re-performing the optimization process. Furthermore, a dynamic adjustment mechanism based on the comprehensive deviation of the current solution set is introduced; Let the overall deviation be defined. ,in , , As defined in claim 6; but ,in Based on the minimum number of retries, It is a positive coefficient. To round down; Annealing factor The current iteration number t in the dynamic The upper limit is defined as the value when t reaches the maximum value. If it is still not possible to find a solution that simultaneously satisfies... , , If the solution is not found, the re-optimization process is terminated, and the solution with the smallest multi-objective fitness function F value in the current non-dominated solution set is selected as the optimal control parameter for that operating condition.

[0017] Compared with the prior art, the present invention provides an anti-tipping control method for counterbalance forklifts, which has the following beneficial effects: 1. This anti-tipping control method for counterbalance forklifts collects multi-source dynamic data of the forklift, constructs a fitness function with tilt risk, control energy consumption, and ride comfort as objectives, and generates an optimal control parameter MAP table by combining an offline RBF neural network surrogate model with a multi-objective genetic algorithm. During online operation, the table is directly looked up to output instructions, which significantly improves the real-time performance and optimization accuracy of control decisions. It solves the problems of large computational load and difficulty in engineering implementation of online real-time optimization in existing technologies, as well as the inability of a single dynamic model to take into account the needs of multi-objective optimization, and avoids control lag. 2. This anti-rollover control method for counterbalance forklifts corrects lateral acceleration by adjusting the terrain slope angle, eliminating the interference of road slope on sensor readings and making the calculation of rollover risk indicators more realistic. At the same time, it subdivides the working conditions into four categories: low speed light load, low speed heavy load, high speed light load, and high speed heavy load according to vehicle speed and load, and adjusts the fitness function weights according to the working condition labels, thereby adapting to complex and ever-changing working scenarios. It solves the problems of insufficient stability judgment accuracy due to ignoring road conditions in existing technologies, and the inability of fixed weights to adapt to the differences in sensitivity of different working conditions, thus improving the scenario adaptability of anti-rollover control. 3. The anti-rollover control method for counterbalance forklifts introduces yaw rate and its rate of change, and dynamically adjusts the weight coefficients in the fitness function through piecewise conditional rules. When the vehicle experiences high yaw rate or sudden changes, the penalty for rollover risk is automatically increased, enabling the control strategy to respond promptly to transient instability trends. This solves the problems of insufficient robustness and lag response of existing simplified models to extreme conditions such as high-speed steering and sudden load changes, and enhances the transient safety of active anti-rollover. 4. The anti-tipping control method for counterbalance forklifts introduces a compliance judgment after generating the non-dominated solution set. If there is no solution that satisfies all three indicators as being better than the limit, the fitness function weights are dynamically adjusted according to the non-compliant items and deviation magnitudes, and the optimization is re-executed. This ensures that the finally selected control parameters meet the engineering safety boundary and accelerates convergence to a feasible solution. It solves the problems that traditional multi-objective optimization results may not meet the safety threshold and the original weights cannot focus on the out-of-limit indicators during re-optimization. It avoids outputting unusable control parameters and improves the engineering reliability of the control strategy. 5. The anti-tipping control method for counterbalanced forklifts dynamically adjusts the crossover and mutation probabilities based on the working condition labels during the re-optimization process, introduces an annealing factor that decreases with the number of retries, and dynamically sets the maximum number of retries based on the comprehensive deviation. This achieves a balance between exploration and development in the genetic algorithm, optimizes resource utilization and convergence stability, and solves the problems that fixed genetic operator parameters cannot adapt to the search space characteristics of different working conditions, fixed number of retries leads to wasted computing resources or insufficient optimization, and difficulty in convergence due to later oscillations. This improves the optimization efficiency and robustness of the algorithm under extreme working conditions. Attached Figure Description

[0018] Figure 1 This is a flowchart of an anti-tipping control method for a counterbalance forklift proposed in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1: Please refer to Figure 1 The anti-tipping control method for a counterbalance forklift in this embodiment includes the following steps: S1. Collect forklift speed V and longitudinal acceleration. Lateral acceleration Vertical acceleration Unloaded tare weight Goods quality Wheel angle δ, mast tilt angle θ, wheel angle change rate gantry tilt angle change rate And preprocessing is performed to obtain the basic dataset; S2. Based on the basic dataset and the inherent parameters of the forklift, the roll risk index R, control energy consumption index E, and ride comfort index S are calculated, and then normalized to obtain... , , ; S3. Normalized Side Roll Risk Indicator Controlling energy consumption indicators Smoothness index Construct a multi-objective fitness function F; S4. Optimal control parameters are obtained by combining offline optimization with online table lookup: In the offline phase, an RBF neural network is trained using simulation data. A multi-objective genetic algorithm is used to randomly generate candidate solutions, which are then input into the trained RBF network to obtain predictions. , , Then, the multi-objective fitness function F is calculated, and the population evolution is driven by minimizing F. Finally, a non-dominated optimal solution set is generated, and the optimal control parameter MAP table is obtained based on the non-dominated optimal solution set. During the online phase, the forklift controller directly retrieves the optimal control parameter MAP table to obtain the optimal control parameters for that operating condition. S5. Send the specific instructions in the optimal control parameters to the corresponding forklift pull-out controller via the CAN bus.

[0021] It should be noted that this claim defines the basic steps of the anti-rollover control method. Preprocessing includes conventional operations such as filtering and noise reduction and outlier removal. The specific calculation formulas for the rollover risk index R, control energy consumption index E, and ride comfort index S can be derived from the forklift dynamics model. In this basic embodiment, the multi-objective fitness function F adopts a normalized index. , , The linear weighted sum form, i.e. Among them, the weighting coefficient , , With a preset fixed value, the population evolution is driven by minimizing F. Through iterative optimization using a multi-objective genetic algorithm, a non-dominated optimal solution set is finally generated. Each candidate solution in the solution set corresponds to a set of control parameters. Based on this solution set, an offline optimal control parameter MAP table can be constructed. During online operation, the forklift controller directly looks up the table to obtain the optimal control parameters corresponding to the current working condition and sends them to each actuator ECU via the CAN bus to achieve anti-rollover control.

[0022] Example 2: Please refer to Figure 1 Based on Example 1, a method for preventing tipping over of a counterbalanced forklift is proposed, in step S1, to further collect the fork height. , terrain slope angle α, yaw rate γ; Lateral acceleration corrected based on terrain slope angle α To obtain the equivalent lateral acceleration in the actual road surface coordinate system The corrected value is then substituted into the calculation of the roll risk index R; At the same time, according to vehicle speed Equivalent lateral acceleration and the quality of goods The current operating conditions are categorized online into four types of operating condition labels: low speed light load, low speed heavy load, high speed light load, and high speed heavy load. when and At that time, it is classified as low speed and light load; when and At that time, it is classified as low-speed heavy-load; when and At that time, it is classified as high-speed light-load; when and At that time, it is classified as high-speed heavy-load; in, To preset the vehicle speed threshold, Fork height is set to a preset quality threshold. Used to combine unloaded self-weight Goods quality The weighted calculation of the mast tilt angle θ is used to determine the real-time center of gravity height h of the vehicle, and the forklift yaw rate γ is used to characterize the vehicle's inertial state. The expression for the multi-objective fitness function F in S3 is: ; Wherein, the weight coefficients satisfy Weighting coefficients , , Adjust according to the operating condition label.

[0023] It should be noted that the preset vehicle speed threshold The value ranges from 5 to 15 km / h, and the preset mass threshold ranges from 0.3 to 0.5 times the rated load capacity. Additional data collection includes fork height, terrain slope angle, and yaw rate. Lateral acceleration is corrected using slope. Four working condition labels are defined based on vehicle speed and load. Real-time center of gravity height is calculated using fork height and other parameters. The fitness function weights are adjusted according to the working condition. The corrected equivalent lateral acceleration is more realistic. The four working condition labels enable detailed working condition analysis. Real-time center of gravity height improves model accuracy. Weighted working condition adaptive analysis enhances optimization targeting. It solves problems such as road slope causing accelerometer reading distortion, leading to misjudgment of tilt risk, different sensitivity to tilt, energy consumption, and ride comfort under different working conditions, the inability of fixed weights to adapt, and the need for real-time calculation of center of gravity height changes with fork lifting.

[0024] Among them, the weighting coefficient , , Further, dynamic adjustment is introduced based on working condition labels, yaw rate γ, and yaw rate of change, through segmented condition rules; Let the default weight vectors corresponding to the four working condition labels be... And satisfy ; The first set of conditional rules is as follows: When the yaw rate If so, the default weight of the corresponding working condition label will be maintained; When the yaw rate ,but ,in , ,and ; When the yaw rate ,but ,in , , By normalization conditions Sure; The second set of conditional rules is as follows: When the rate of change of yaw angular velocity Then, further adjustments will be made based on the current weights: ; in, , And renormalize to make ; when If so, the current weight remains unchanged.

[0025] in, , These are the first and second threshold values ​​for the yaw rate, respectively. The threshold for the rate of change of yaw angular velocity. This is the gain coefficient for the effect of the rate of change of yaw rate.

[0026] It should be noted that this claim introduces a method based on the yaw rate γ and its rate of change. Segmented dynamic weight adjustment, The value range is 0.1-0.3 rad / s, with 0.2 rad / s being preferred. The value range is 0.4-0.8 rad / s, with 0.6 rad / s being preferred. The value range is 0.5-1.5 rad / s 2 1.0 rad / s is preferred. 2 , The value range is 0.5-2.0, with 1.0 being preferred; This solves the problem that relying solely on operating condition labels cannot respond to the transient stability of a vehicle. It enables the weights to be automatically adjusted according to the yaw rate and rate of change, increasing the penalty for roll risk during sharp turns or fishtailing tendencies, thereby improving the active safety of the control.

[0027] In the S4 offline phase, a multi-objective genetic algorithm is further executed to find the optimal control parameters MAP table under each working condition label. During the online phase, the forklift controller directly retrieves the optimal control parameter MAP table under the corresponding operating condition label based on the real-time identified operating condition label, and obtains the optimal control parameters under that operating condition label. In the S4 offline phase, candidate solutions are randomly generated within a set range, including the target pressure of the hydraulic support cylinder. Displacement of dynamic balance block Rear wheel active steering angle .

[0028] It should be noted that the offline phase specifies that a multi-objective genetic algorithm is executed to optimize different working condition labels to generate a MAP table of optimal control parameters for each working condition. In the online phase, the control parameters are obtained directly by looking up the table based on the real-time identified working condition labels. This solves the problems of large differences in optimal parameters for different working conditions, inability to take into account uniform parameters, and large computational load for online real-time optimization. It achieves accurate optimization for different working conditions and millisecond-level online response, and clarifies the collaborative control variables of the three actuators.

[0029] Example 3: Please refer to Figure 1 Based on Example 2, a method for preventing rollover of counterbalance forklifts further introduces a non-dominated optimal solution set compliance judgment after generating the non-dominated optimal solution set. Based on forklift speed V and cargo weight The terrain slope angle α is used to calculate the roll risk index under the current working conditions through a pre-calibrated mapping relationship. Controlling energy consumption indicators Smoothness index and set ; Among them, the weighting coefficient , , The value is consistent with that in S3; Calculate the normalized index for each solution in the non-dominated optimal solution set. , , And the multi-objective fitness function value F; Judgment Formula 1: , , ; If there exists a solution that satisfies criterion one, then select the solution with the smallest F from these solutions as the optimal control parameter for that working condition; If no solution satisfies criterion one, then adjust the weights of the multi-objective fitness function F and re-execute the optimization. in, This is a preset threshold for achieving the target.

[0030] It should be noted that the tilt risk indicator Controlling energy consumption indicators Smoothness index Calculations are performed using pre-defined mapping relationships, such as based on forklift speed V and cargo mass. The terrain slope angle α is obtained by looking up a table or calculating using a function; Normalized typical value: tilt risk indicator Controlling energy consumption indicators Smoothness index , The value range is usually 0.6-0.9; It solves the problem that multi-objective optimization solutions may not meet engineering safety boundaries, ensures that the final selected control parameters meet basic safety, energy consumption, and smoothness requirements, and automatically triggers the re-optimization process to avoid outputting unusable solutions.

[0031] Among them, the weights of the multi-objective fitness function F are further dynamically adjusted based on the non-compliance items and deviation magnitude during the re-execution of the optimization process; Based on the actual tilt risk indicators achieved in the current candidate solution set Controlling energy consumption indicators Smoothness index Each with corresponding limit , , The comparison results are dynamically updated with weights as follows: like Then let , ; like If the weight remains unchanged, then the weight remains unchanged. like Then let , ; like If the weight remains unchanged, then the weight remains unchanged. like Then let , ; like If the weight remains unchanged, then the weight remains unchanged. in, , , All are positive coefficients. , , These represent the deviation magnitudes of the corresponding items.

[0032] Then to , , Normalization is performed so that the sum of the normalized weights is 1, and the normalized weights are used as the weight coefficients of the fitness function F in the next round of optimization.

[0033] It should be noted that this solves the problem that the original weights cannot effectively improve the out-of-target indicators during re-optimization, making the algorithm focus on the worst indicator at present, and the greater the deviation, the heavier the penalty, thus accelerating convergence to a solution that satisfies all limits.

[0034] Additionally, during the re-execution of the optimization process, a maximum number of retries is set. ; If the number of consecutive re-executions of the optimization method reaches After that, it was still impossible to find a solution that simultaneously satisfies... , , If the solution is not found, the re-optimization process is terminated, and the solution with the smallest multi-objective fitness function F value in the current non-dominated solution set is selected as the optimal control parameter for that operating condition.

[0035] It should be noted that this claim sets a fixed maximum number of retries. The value ranges from 3 to 10, and the optimization is performed 5 times. If the number of consecutive re-executions of the optimization reaches a certain threshold... If a solution that simultaneously satisfies criterion one is still not found, the re-optimization process is terminated, and the solution with the smallest multi-objective fitness function F value in the current non-dominated solution set is selected as the optimal control parameter under this working condition. It solves the problem that re-optimization may get stuck in an infinite loop or converge very slowly, and provides a mandatory termination condition in engineering to ensure that the algorithm ends within a finite number of steps, avoiding infinite optimization due to extreme conditions.

[0036] Furthermore, the optimization process is further refined by dynamically adjusting the crossover probability based on the operating condition label. and mutation probability ; If the identified operating condition label is low speed and light load, then maintain... ; If the identified operating condition label is low-speed heavy-load, then a load factor is introduced. That is, to adjust to ; If the identified operating condition label is high speed and light load, then a vehicle speed deviation is introduced. That is, to adjust to ; If the identified operating condition label is high-speed heavy load, then the crossover probability should be adjusted according to the above rules. and mutation probability ; in, As the baseline crossover probability, As the baseline mutation probability, , , , It is a positive coefficient.

[0037] It should be noted that the baseline crossover probability ranges from 0.6 to 0.9, preferably 0.8, and the baseline mutation probability ranges from 0.01 to 0.1, preferably 0.05. The loading factor... The value range is 0-1; in This is the high-speed threshold, with a value ranging from 10-15 km / h. It can take negative values ​​(for low speeds) or positive values ​​(for high speeds). , , , The value range is 0.5-2.0, with 1.0 being preferred; This solves the problem that fixed genetic operator parameters cannot adapt to the search space characteristics under different working conditions. It increases the mutation probability under heavy load to escape local optima and adjusts the crossover probability to explore new combinations under high speed, thereby improving search efficiency and solution quality.

[0038] Among them, dynamically adjusting the crossover probability and mutation probability At this time, an annealing factor that decreases with the number of re-executions of the optimization process is introduced. ; Let the current iteration be the t-th iteration of the optimization process, and let the annealing factor be... , This is the annealing rate coefficient; Then the adjustment coefficient in the actual adjustment formula is multiplied by Taking the operating condition label as low speed and heavy load as an example, the adjustment formula is as follows: ; If the identified operating condition label is high speed light load, high speed heavy load, or low speed light load, then the above rules shall be applied simultaneously.

[0039] It should be noted that the annealing rate coefficient ranges from 0.1 to 0.5, preferably 0.3. In actual adjustment, the adjustment coefficient in claim 8 should be multiplied by... This approach resolves the contradiction between the need for strong perturbations to explore the global optimum in the early stages and the need for stable convergence in the later stages, achieving a balance between exploration and development, improving convergence stability, and avoiding premature entrapment in local optima or later oscillations.

[0040] Finally, retry the optimization process up to the maximum number of times. Furthermore, a dynamic adjustment mechanism based on the comprehensive deviation of the current solution set is introduced; Let the overall deviation be defined. ,in , , As defined in claim 6; but ,in Based on the minimum number of retries, It is a positive coefficient. To round down; Annealing factor The current iteration number t in the dynamic The upper limit is defined as the value when t reaches the maximum value. If it is still not possible to find a solution that simultaneously satisfies... , , If the solution is not found, the re-optimization process is terminated, and the solution with the smallest multi-objective fitness function F value in the current non-dominated solution set is selected as the optimal control parameter for that operating condition.

[0041] It should be noted that, if ,but When t reaches If no solution satisfying criterion one is found, the re-optimization terminates and the solution with the smallest current F is output, thus solving the fixed-valued problem. To address issues of wasted computing resources or insufficient optimization, the number of retries is made adaptive to the problem difficulty. At the same time, the decay of the annealing factor is bounded by a dynamic upper limit to avoid premature truncation, thereby improving the overall optimization resource utilization.

[0042] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0043] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for preventing tipping over of a counterbalance forklift, characterized in that, Includes the following steps: S1. Collect forklift speed V and longitudinal acceleration. Lateral acceleration Vertical acceleration Unloaded tare weight Goods quality Wheel angle δ, mast tilt angle θ, wheel angle change rate gantry tilt angle change rate And preprocessing is performed to obtain the basic dataset; S2. Based on the basic dataset and the inherent parameters of the forklift, the roll risk index R, control energy consumption index E, and ride comfort index S are calculated, and then normalized to obtain... , , ; S3. Normalized Side Roll Risk Indicator Controlling energy consumption indicators Smoothness index Construct a multi-objective fitness function F; S4. Optimal control parameters are obtained by combining offline optimization with online table lookup: In the offline phase, an RBF neural network is trained using simulation data. A multi-objective genetic algorithm is used to randomly generate candidate solutions, which are then input into the trained RBF network to obtain predictions. , , Then, the multi-objective fitness function F is calculated, and the population evolution is driven by minimizing F. Finally, a non-dominated optimal solution set is generated, and the optimal control parameter MAP table is obtained based on the non-dominated optimal solution set. During the online phase, the forklift controller directly retrieves the optimal control parameter MAP table to obtain the optimal control parameters for that operating condition. S5. Send the specific instructions in the optimal control parameters to the corresponding forklift controllers via the CAN bus.

2. The anti-tipping control method for a counterbalance forklift according to claim 1, characterized in that: The fork height is further collected in S1. , terrain slope angle α, yaw rate γ; Lateral acceleration corrected based on terrain slope angle α To obtain the equivalent lateral acceleration in the actual road surface coordinate system The corrected value is then substituted into the calculation of the roll risk index R; At the same time, according to vehicle speed Equivalent lateral acceleration and the quality of goods The current operating conditions are categorized online into four types of operating condition labels: low speed light load, low speed heavy load, high speed light load, and high speed heavy load. when and At that time, it is classified as low speed and light load; when and When this occurs, it is classified as low-speed heavy-load; when and At that time, it is classified as high-speed light-load; when and At that time, it is classified as high-speed heavy-load; in, To preset the vehicle speed threshold, Fork height is set to a preset quality threshold. Used to combine unloaded self-weight Goods quality The weighted calculation of the mast tilt angle θ is used to determine the real-time center of gravity height h of the vehicle, and the forklift yaw rate γ is used to characterize the vehicle's inertial state. The expression for the multi-objective fitness function F in S3 is: ； Wherein, the weight coefficients satisfy Weighting coefficients , , Adjust according to the operating condition label.

3. The anti-tipping control method for a counterbalance forklift according to claim 2, characterized in that: The weighting coefficient , , Further, dynamic adjustment is introduced based on working condition labels, yaw rate γ, and yaw rate of change, through segmented condition rules; Let the default weight vectors corresponding to the four working condition labels be... And satisfy ; The first set of conditional rules is as follows: When the yaw rate If so, the default weight of the corresponding working condition label will be maintained; When the yaw rate ,but ; in , ,and ; When the yaw rate ,but ,in , , By normalization conditions Sure; The second set of conditional rules is as follows: When the rate of change of yaw angular velocity Then, further adjustments will be made based on the current weights: ； in, , And renormalize to make ; when If so, the current weight remains unchanged; in, , These are the first and second threshold values ​​for the yaw rate, respectively. The threshold for the rate of change of yaw angular velocity. This is the gain coefficient for the effect of the rate of change of yaw rate.

4. The anti-tipping control method for a counterbalance forklift according to claim 2, characterized in that: In the S4 offline stage, a multi-objective genetic algorithm is further executed according to different working condition labels to obtain the optimal control parameter MAP table under each working condition label. During the online phase, the forklift controller directly retrieves the optimal control parameter MAP table under the corresponding operating condition label based on the real-time identified operating condition label, and obtains the optimal control parameters under that operating condition label. In the S4 offline phase, candidate solutions are randomly generated within a set range, including the target pressure of the hydraulic support cylinder. Displacement of dynamic balance block Rear wheel active steering angle .

5. The anti-tipping control method for a counterbalance forklift according to claim 4, characterized in that: After generating the non-dominated optimal solution set, a standard-setting judgment for the non-dominated optimal solution set is further introduced. Based on forklift speed V and cargo weight The terrain slope angle α is used to calculate the roll risk index under the current working conditions through a pre-calibrated mapping relationship. Controlling energy consumption indicators Smoothness index and set ; Among them, the weighting coefficient , , The value is consistent with that in S3; Calculate the normalized index for each solution in the non-dominated optimal solution set. , , And the multi-objective fitness function value F; Judgment Formula 1: , , ; If there exists a solution that satisfies criterion one, then select the solution with the smallest F from these solutions as the optimal control parameter for that working condition; If no solution satisfies criterion one, then adjust the weights of the multi-objective fitness function F and re-execute the optimization. in, This is a preset threshold for achieving the target.

6. The anti-tipping control method for a counterbalance forklift according to claim 5, characterized in that: The weights of the multi-objective fitness function F, which are adjusted during the re-execution of the optimization process, are further dynamically adjusted based on the non-compliance items and the deviation magnitude. Based on the actual tilt risk indicators achieved in the current candidate solution set Controlling energy consumption indicators Smoothness index Each with corresponding limit , , The comparison results are dynamically updated with weights as follows: like Then let , ; like If the weight remains unchanged, then the weight remains unchanged. like Then let , ; like If the weight remains unchanged, then the weight remains unchanged. like Then let , ; like If the weight remains unchanged, then the weight remains unchanged. in, , , All are positive coefficients. , , These represent the deviation magnitudes of the corresponding items; Then to , , Normalization is performed so that the sum of the normalized weights is 1, and the normalized weights are used as the weight coefficients of the fitness function F in the next round of optimization.

7. The anti-tipping control method for a counterbalance forklift according to claim 5, characterized in that: In the re-execution of the optimization process, a maximum number of retries is set. ; If the number of consecutive re-executions of the optimization method reaches After that, it was still impossible to find a solution that simultaneously satisfies... , , If the solution is not found, the re-optimization process is terminated, and the solution with the smallest multi-objective fitness function F value in the current non-dominated solution set is selected as the optimal control parameter for that operating condition.

8. The anti-tipping control method for a counterbalance forklift according to claim 7, characterized in that: The re-execution optimization process further incorporates dynamic adjustment of crossover probability based on operating condition labels. and mutation probability ; If the identified operating condition label is low speed and light load, then maintain... ; If the identified operating condition label is low-speed heavy load, then a load factor is introduced. That is, to adjust to ; If the identified operating condition label is high speed and light load, then a vehicle speed deviation is introduced. That is, to adjust to ; If the identified operating condition label is high-speed heavy load, then the crossover probability should be adjusted according to the above rules. and mutation probability ; in, As the baseline crossover probability, As the baseline mutation probability, , , , It is a positive coefficient.

9. A method for preventing tipping over of a counterbalance forklift according to claim 8, characterized in that: The dynamic adjustment of crossover probability and mutation probability At this time, an annealing factor that decreases with the number of re-executions of the optimization process is introduced. ; Let the current iteration be the t-th iteration of the optimization process, and let the annealing factor be... , This is the annealing rate coefficient; Then the adjustment coefficient in the actual adjustment formula is multiplied by Taking the operating condition label as low speed and heavy load as an example, the adjustment formula is as follows: ； If the identified operating condition label is high speed light load, high speed heavy load, or low speed light load, then the above rules shall be applied simultaneously.

10. A method for preventing tipping over of a counterbalance forklift according to claim 9, characterized in that: The maximum number of retries to re-execute the optimization search Furthermore, a dynamic adjustment mechanism based on the comprehensive deviation of the current solution set is introduced; Let the overall deviation be defined. ,in , , As defined in claim 6; but ,in Based on the minimum number of retries, It is a positive coefficient. To round down; Annealing factor The current iteration number t in the dynamic The upper limit is defined as the value when t reaches the maximum value. If it is still not possible to find a solution that simultaneously satisfies... , , If the solution is not found, the re-optimization process is terminated, and the solution with the smallest multi-objective fitness function F value in the current non-dominated solution set is selected as the optimal control parameter for that operating condition.

Images