A method for agricultural machinery to travel on multiple terrains

Abstract

The invention discloses a multi-terrain traveling method for agricultural machinery, belonging to the technical field of agricultural machinery control. The invention monitors the inclination angle and Z-axis acceleration of the agricultural machinery in real time, constructs an inclination angle feature sequence and a Z-axis acceleration feature sequence, reflects the change of the inclination angle and the Z-axis acceleration within a period of time, processes the four sequences by using a dual-channel fusion recursive neural network, predicts the resistance of the current road surface, calculates the target power according to the real-time resistance, and dynamically adjusts the target power, so that the agricultural machinery can achieve adaptive control and adapt to different terrains when facing complex terrain, which not only optimizes energy utilization efficiency but also reduces energy waste.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural machinery control, and in particular to a method for agricultural machinery to travel on multiple terrains. Background Art

[0002] With the rapid development of agricultural modernization, the global demand for high efficiency and high quality in agricultural production is increasing. Especially in the context of climate change and tight land resources, the demand for efficient operation of agricultural machinery in complex and changeable terrains is becoming increasingly urgent. Modern agriculture not only emphasizes large-scale production capacity, but also pays attention to the adaptability of agricultural machinery in various terrains. However, traditional agricultural machinery often finds it difficult to adapt to different terrain conditions (such as hills, mountains, and wetlands), resulting in low energy efficiency. Traditional agricultural machinery often cannot accurately estimate the required power when climbing slopes or muddy roads, and needs to consume more energy to achieve the target acceleration or target speed, resulting in energy waste. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for agricultural machinery to travel on multiple terrains, which solves the problem that the existing agricultural machinery cannot adaptively adjust the required power and wastes energy.

[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a method for agricultural machinery to travel on multiple terrains, comprising the following steps:

[0005] S1. Constructing a tilt angle sequence and a tilt angle feature sequence according to the tilt angle of the agricultural machinery during operation;

[0006] S2. Constructing a Z-axis acceleration sequence and a Z-axis acceleration feature sequence according to the Z-axis acceleration of the agricultural machinery during operation;

[0007] S3, predicting the resistance of the agricultural machinery based on the dual-channel fusion recursive neural network according to the tilt angle sequence, the tilt angle feature sequence, the Z-axis acceleration sequence and the Z-axis acceleration feature sequence;

[0008] S4. Calculate the target power of the agricultural machinery according to the resistance, apply the target power to the agricultural machinery, and adjust the target power based on the difference between the actual forward acceleration of the agricultural machinery and the target forward acceleration of the agricultural machinery.

[0009] Furthermore, the tilt angle sequence in S1 is composed of tilt angles at multiple moments, and the Z-axis acceleration sequence in S2 is composed of Z-axis accelerations at multiple moments;

[0010] The elements in the tilt angle characteristic sequence in S1 include: angle fluctuation coefficient and angle variation coefficient;

[0011] The elements in the Z-axis acceleration characteristic sequence in S2 include: Z-axis acceleration fluctuation coefficient and Z-axis acceleration variation coefficient.

[0012] Further, the method for obtaining the angle fluctuation coefficient is as follows: take the tilt angles at N moments, take the square of each tilt angle to obtain N angle square values, subtract the square value of each angle from the square value of the tilt angle mean to obtain N angle differences, add the absolute values ​​of the N angle differences, and normalize the addition result to obtain the angle fluctuation coefficient, wherein the tilt angle mean is the average value of the tilt angles at N moments;

[0013] The method for obtaining the Z-axis acceleration fluctuation coefficient is as follows: take the Z-axis acceleration at N moments, take the square of each Z-axis acceleration to obtain N acceleration square values, subtract each acceleration square value from the square value of the Z-axis acceleration mean to obtain N acceleration differences, add the absolute values ​​of the N acceleration differences, and normalize the addition result to obtain the Z-axis acceleration fluctuation coefficient, where the Z-axis acceleration mean is the average value of the Z-axis acceleration at N moments.

[0014] Further, the method for obtaining the angle variation coefficient is as follows: take the tilt angles at N moments, calculate the difference between the tilt angles at two adjacent moments, obtain N-1 adjacent angle differences, take the square of each adjacent angle difference, obtain N-1 angle difference square values, take the average of the N-1 angle difference square values, obtain the angle difference average, subtract each angle difference square value from the angle difference average, obtain N-1 angle deviations, add the absolute values ​​of the N-1 angle deviations, and normalize the addition result to obtain the angle variation coefficient;

[0015] The method for obtaining the Z-axis acceleration variation coefficient is as follows: take the Z-axis acceleration at N moments, calculate the difference between the Z-axis acceleration at two adjacent moments, obtain N-1 adjacent acceleration differences, take the square of each adjacent acceleration difference, obtain N-1 acceleration difference square values, take the average of the N-1 acceleration difference square values, obtain the acceleration difference mean, subtract each acceleration difference square value from the acceleration difference mean, obtain N-1 acceleration deviations, add the absolute values ​​of the N-1 acceleration deviations, and normalize the addition result to obtain the acceleration variation coefficient.

[0016] Furthermore, the dual-channel fusion recursive neural network in S3 includes: an angle tilt calculation channel, a Z-axis acceleration calculation channel, a fusion layer, and a recursive output layer;

[0017] The angle tilt operation channel is used to process the tilt angle sequence and the tilt angle feature sequence to obtain the first channel output value;

[0018] The Z-axis acceleration operation channel is used to process the Z-axis acceleration sequence and the Z-axis acceleration characteristic sequence to obtain the second channel output value;

[0019] The fusion layer is used to fuse the output value of the first channel and the output value of the second channel to obtain a fused feature value;

[0020] The recursive output layer is used to predict the resistance of agricultural machinery based on the current fused feature values ​​and historical predicted resistance.

[0021] Furthermore, the expression of the angle tilt operation channel is: , where G A,t is the output value of the first channel of the angle tilt calculation channel at time t, A t-n is the tilt angle at the tnth moment in the tilt angle sequence, w A,t-n A t-n The weight, θ Aw is the angular fluctuation coefficient, w Aw is θ Aw The weight, θ Ac is the angle variation coefficient, w Ac is θ Ac The weight of b A is the offset of the angle tilt calculation channel, t is the current time, n is the subscript of the historical time, and N is the length of the statistical data;

[0022] The expression of the Z-axis acceleration operation channel is: , where G Z,t is the output value of the second channel output by the Z-axis acceleration operation channel at time t, Z t-n is the Z-axis acceleration at the tnth moment in the Z-axis acceleration sequence, w Z,t-n Z t-n The weight, θ Zw is the Z-axis acceleration fluctuation coefficient, w Zw is θ Zw The weight, θ Zc is the Z-axis acceleration variation coefficient, w Zc is θ Zc The weight of b Z It is the offset of the Z-axis acceleration calculation channel.

[0023] Furthermore, the expression of the fusion layer is: , where H t is the fusion feature value output by the fusion layer at the tth moment, w A,t G A,t The weight, w Z,t G Z,t The weight of b G is the bias of the fusion layer.

[0024] Furthermore, the expression of the recursive output layer is: , where F t is the resistance at time t, wH,t H t The weight of F t-n is the resistance at the tnth moment, w F,t-n F t-n The weight of b F is the bias of the recurrent output layer, and M is the length of the historical resistance.

[0025] Further, S4 includes the following sub-steps:

[0026] S41. Calculate the target power of the agricultural machinery based on the resistance: , where P target is the target power, a target is the target forward acceleration of the agricultural machinery, m is the mass of the agricultural machinery, F t is the resistance at the tth moment, V target is the target speed of the agricultural machinery;

[0027] S42, after applying the target power to the agricultural machine, collecting the actual forward acceleration of the agricultural machine;

[0028] S43, calculating the difference between the actual forward acceleration and the target forward acceleration of the agricultural machinery to obtain the forward acceleration difference, when the absolute value of the forward acceleration difference is less than the threshold value, the target power is not adjusted, and the sub-step is ended; when the absolute value of the forward acceleration difference is greater than the threshold value, the resistance of the agricultural machinery is compensated, and the process proceeds to step S44;

[0029] S44. Calculate the new target power of the agricultural machinery according to the compensated resistance: , where F t,ew To compensate for the resistance at the tth moment, P target,new is the new target power.

[0030] Furthermore, the formula for compensating the resistance of the predicted agricultural machinery in S43 is: , where a act is the actual forward acceleration, μ is the proportional coefficient.

[0031] The beneficial effects of the present invention are as follows: the present invention monitors the inclination angle and Z-axis acceleration of agricultural machinery in real time, constructs an inclination angle feature sequence and a Z-axis acceleration feature sequence, reflects the changes in the inclination angle and the Z-axis acceleration over a period of time, processes the four sequences using a dual-channel fusion recursive neural network, predicts the resistance of the current road surface, calculates the target power based on the real-time resistance, and dynamically adjusts the target power, so that the agricultural machinery can achieve adaptive control and adapt to different terrains when facing complex terrain, which not only optimizes energy utilization efficiency, but also reduces energy waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1A flowchart of a method for agricultural machinery to travel on multiple terrains;

[0033] Figure 2 Schematic diagram of the structure of a dual-channel fusion recurrent neural network. DETAILED DESCRIPTION

[0034] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0035] like Figure 1 As shown, a method for agricultural machinery to travel on multiple terrains comprises the following steps:

[0036] S1. Constructing a tilt angle sequence and a tilt angle feature sequence according to the tilt angle of the agricultural machinery during operation;

[0037] S2. Constructing a Z-axis acceleration sequence and a Z-axis acceleration feature sequence according to the Z-axis acceleration of the agricultural machinery during operation;

[0038] S3, predicting the resistance of the agricultural machinery based on the dual-channel fusion recursive neural network according to the tilt angle sequence, the tilt angle feature sequence, the Z-axis acceleration sequence and the Z-axis acceleration feature sequence;

[0039] S4. Calculate the target power of the agricultural machinery according to the resistance, apply the target power to the agricultural machinery, and adjust the target power based on the difference between the actual forward acceleration of the agricultural machinery and the target forward acceleration of the agricultural machinery.

[0040] The tilt angle sequence in S1 is composed of tilt angles at multiple moments, and the Z-axis acceleration sequence in S2 is composed of Z-axis accelerations at multiple moments;

[0041] The elements in the tilt angle characteristic sequence in S1 include: angle fluctuation coefficient and angle variation coefficient;

[0042] The elements in the Z-axis acceleration characteristic sequence in S2 include: Z-axis acceleration fluctuation coefficient and Z-axis acceleration variation coefficient.

[0043] The inclination angle directly reflects the slope of the terrain where the agricultural machinery is located. When going uphill, the inclination angle increases, which means that the agricultural machinery needs to overcome a greater gravity component, so the resistance increases. The Z-axis acceleration can reflect the movement change of the agricultural machinery in the vertical direction, which is directly related to the unevenness of the terrain.

[0044] In this embodiment, the Z-axis acceleration is collected by a three-axis accelerometer, and the tilt angle is collected by a pitch angle sensor.

[0045] The method for obtaining the angle fluctuation coefficient is as follows: take the tilt angles at N moments, take the square of each tilt angle to obtain N angle square values, subtract the square of each angle value from the square of the tilt angle mean value to obtain N angle differences, add the absolute values ​​of the N angle differences, and normalize the addition result to obtain the angle fluctuation coefficient, wherein the tilt angle mean value is the average value of the tilt angles at N moments;

[0046] The method for obtaining the Z-axis acceleration fluctuation coefficient is as follows: take the Z-axis acceleration at N moments, take the square of each Z-axis acceleration to obtain N acceleration square values, subtract each acceleration square value from the square value of the Z-axis acceleration mean to obtain N acceleration differences, add the absolute values ​​of the N acceleration differences, and normalize the addition result to obtain the Z-axis acceleration fluctuation coefficient, where the Z-axis acceleration mean is the average value of the Z-axis acceleration at N moments.

[0047] The present invention enhances each tilt angle by taking the square of each tilt angle, then subtracts the square of each angle from the square of the mean tilt angle to reflect the difference from the mean, and adds each difference to reflect the change of N tilt angles.

[0048] The present invention realizes enhanced processing of each Z-axis acceleration by taking the square of each Z-axis acceleration, and then subtracts the square of each acceleration value from the square of the mean Z-axis acceleration to reflect the difference with the mean, and adds each difference to reflect the change of N Z-axis accelerations.

[0049] The method for obtaining the angle variation coefficient is as follows: take the tilt angles at N moments, calculate the difference between the tilt angles at two adjacent moments, obtain N-1 adjacent angle differences, take the square of each adjacent angle difference, obtain N-1 angle difference square values, take the average of the N-1 angle difference square values, obtain the angle difference average, subtract each angle difference square value from the angle difference average, obtain N-1 angle deviations, add the absolute values ​​of the N-1 angle deviations, and normalize the addition result to obtain the angle variation coefficient;

[0050] The method for obtaining the Z-axis acceleration variation coefficient is as follows: take the Z-axis acceleration at N moments, calculate the difference between the Z-axis acceleration at two adjacent moments, obtain N-1 adjacent acceleration differences, take the square of each adjacent acceleration difference, obtain N-1 acceleration difference square values, take the average of the N-1 acceleration difference square values, obtain the acceleration difference mean, subtract each acceleration difference square value from the acceleration difference mean, obtain N-1 acceleration deviations, add the absolute values ​​of the N-1 acceleration deviations, and normalize the addition result to obtain the acceleration variation coefficient.

[0051] In the present invention, N-1 adjacent angle differences are obtained by sequentially subtracting the tilt angle at the previous moment from the tilt angle at the next moment, so that the tilt angles at every two adjacent moments will produce an adjacent angle difference; N-1 adjacent acceleration differences are obtained by sequentially subtracting the Z-axis acceleration at the previous moment from the Z-axis acceleration at the next moment, so that the Z-axis accelerations at every two adjacent moments will produce an adjacent acceleration difference.

[0052] The present invention subtracts two adjacent inclination angles to obtain multiple adjacent angle differences to reflect the change of the angle, then takes the square of each adjacent angle difference to enhance the adjacent angle difference, and subtracts the square of each angle difference from the angle difference mean to reflect the irregularity of the change of the inclination angle during the operation of the agricultural machinery.

[0053] The present invention subtracts two adjacent Z-axis accelerations to obtain a plurality of adjacent acceleration differences, which reflect the change of the Z-axis acceleration, and then takes the square of each adjacent acceleration difference to enhance the adjacent acceleration difference. The square value of each acceleration difference is subtracted from the mean value of the acceleration difference to reflect the irregularity of the acceleration change during the operation of the agricultural machinery.

[0054] like Figure 2 As shown, the dual-channel fusion recursive neural network in S3 includes: an angle tilt calculation channel, a Z-axis acceleration calculation channel, a fusion layer and a recursive output layer;

[0055] The angle tilt operation channel is used to process the tilt angle sequence and the tilt angle feature sequence to obtain the first channel output value;

[0056] The Z-axis acceleration operation channel is used to process the Z-axis acceleration sequence and the Z-axis acceleration characteristic sequence to obtain the second channel output value;

[0057] The fusion layer is used to fuse the output value of the first channel and the output value of the second channel to obtain a fused feature value;

[0058] The recursive output layer is used to predict the resistance of agricultural machinery based on the current fused feature values ​​and historical predicted resistance.

[0059] The present invention can process multi-dimensional data simultaneously and fully capture terrain features by separately setting angle tilt and Z-axis acceleration calculation channels. The fusion layer integrates the outputs of the two channels to provide more comprehensive terrain analysis results. The recursive output layer simultaneously considers the current fusion feature value and historical predicted resistance to predict the current resistance of agricultural machinery and improve the prediction accuracy.

[0060] The expression of the angle tilt operation channel is: , where G A,tis the output value of the first channel of the angle tilt calculation channel at time t, A t-n is the tilt angle at the tnth moment in the tilt angle sequence, w A,t-n A t-n The weight, θ Aw is the angular fluctuation coefficient, w Aw is θ Aw The weight, θ Ac is the angle variation coefficient, w Ac is θ Ac The weight of b A is the offset of the angle tilt calculation channel, t is the current time, n is the subscript of the historical time, and N is the length of the statistical data;

[0061] The expression of the Z-axis acceleration operation channel is: , where G Z,t is the output value of the second channel output by the Z-axis acceleration operation channel at time t, Z t-n is the Z-axis acceleration at the tnth moment in the Z-axis acceleration sequence, w Z,t-n Z t-n The weight, θ Zw is the Z-axis acceleration fluctuation coefficient, w Zw is θ Zw The weight, θ Zc is the Z-axis acceleration variation coefficient, w Zc is θ Zc The weight of b Z It is the offset of the Z-axis acceleration calculation channel.

[0062] The present invention comprehensively considers the tilt angle data or Z-axis acceleration data of N-1 historical moments to capture the continuity and trend of terrain changes, and then combines the tilt angle feature information or Z-axis acceleration feature information to enhance the expression ability of the model.

[0063] The expression of the fusion layer is: , where H t is the fusion feature value output by the fusion layer at the tth moment, w A,t G A,t The weight, w Z,t G Z,t The weight of b G is the bias of the fusion layer.

[0064] The expression of the recursive output layer is: , where F t is the resistance at time t, w H,t H t The weight of F t-n is the resistance at the tnth moment, w F,t-n F t-nThe weight of b F is the bias of the recursive output layer, and M is the length of the historical resistance.

[0065] The present invention effectively integrates the information of the two channels of angle tilt and Z-axis acceleration through the fusion layer, provides a more comprehensive representation of terrain features, considers the fusion features of the current moment through the recursive output layer, and also introduces historical resistance information, which can better capture the time dependence of resistance changes.

[0066] S4 includes the following sub-steps:

[0067] S41. Calculate the target power of the agricultural machinery based on the resistance: , where P target is the target power, a target is the target forward acceleration of the agricultural machinery, m is the mass of the agricultural machinery, F t is the resistance at the tth moment, V target is the target speed of the agricultural machinery;

[0068] S42, after applying the target power to the agricultural machine, collecting the actual forward acceleration of the agricultural machine;

[0069] S43, calculating the difference between the actual forward acceleration and the target forward acceleration of the agricultural machinery to obtain the forward acceleration difference, when the absolute value of the forward acceleration difference is less than the threshold value, the target power is not adjusted, and the sub-step is ended; when the absolute value of the forward acceleration difference is greater than the threshold value, the resistance of the agricultural machinery is compensated, and the process proceeds to step S44;

[0070] S44. Calculate the new target power of the agricultural machinery according to the compensated resistance: , where F t,ew To compensate for the resistance at the tth moment, P target,new is the new target power.

[0071] In this embodiment, after obtaining the new target power in step S44, the process may jump to step S42 until a target power is found that satisfies the condition that the absolute value of the forward acceleration difference is less than a threshold value.

[0072] In this embodiment, the threshold is specifically set based on experiments or experience.

[0073] In this embodiment, after the resistance is predicted by the dual-channel fusion recursive neural network, it is only an estimated basic resistance. An estimated target power is first calculated, and then the target power is adjusted based on the feedback of the agricultural machinery, so that the estimation of the resistance is more accurate. When the dual-channel fusion recursive neural network predicts that the resistance is similar (that is, the predicted resistance within a period of time is similar), there is no need to recalculate the target power in S41. When the dual-channel fusion recursive neural network predicts that the resistance is greatly different (that is, the predicted resistance within a period of time has undergone a large fluctuation or mutation), enter S41 to calculate the target power and adjust the target power.

[0074] The present invention calculates the target power first based on the predicted resistance, and after applying the target power, based on the feedback of the agricultural machinery, that is, the actual forward acceleration generated, when the difference between the actual forward acceleration and the target forward acceleration of the agricultural machinery is small, the resistance is more accurate and the target power is not adjusted. When the difference between the two is large, the resistance predicted to be encountered by the agricultural machinery is compensated to achieve readjustment of the target power to make it more suitable for the power required by the terrain at that time. On different terrains, the controlled power is adjusted in real time to reduce energy waste.

[0075] The formula for compensating the resistance of the predicted agricultural machinery in S43 is: , where a act is the actual forward acceleration, μ is the proportional coefficient.

[0076] In this embodiment, the range of the proportionality coefficient is 0-1.

[0077] When the actual resistance is large, the calculated target power is small, which makes the actual forward acceleration small. Therefore, the resistance is increased to increase the target power. When the actual resistance is small, the calculated target power is large, which makes the actual forward acceleration large. Therefore, the resistance is reduced to reduce the target power.

[0078] The present invention monitors the inclination angle and Z-axis acceleration of agricultural machinery in real time, constructs inclination angle feature sequence and Z-axis acceleration feature sequence, reflects the change of inclination angle and Z-axis acceleration within a period of time, processes the four sequences using a dual-channel fusion recursive neural network, predicts the resistance of the current road surface, calculates the target power according to the real-time resistance, and dynamically adjusts the target power, so that the agricultural machinery can achieve adaptive control and adapt to different terrains when facing complex terrain, which not only optimizes energy utilization efficiency, but also reduces energy waste.

[0079] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for agricultural machinery to travel on multiple terrains, characterized in that: The following steps are involved: S1. Constructing a tilt angle sequence and a tilt angle feature sequence according to the tilt angle of the agricultural machinery during operation; S2. Constructing a Z-axis acceleration sequence and a Z-axis acceleration feature sequence according to the Z-axis acceleration of the agricultural machinery during operation; S3, predicting the resistance of the agricultural machinery based on the dual-channel fusion recursive neural network according to the tilt angle sequence, the tilt angle feature sequence, the Z-axis acceleration sequence and the Z-axis acceleration feature sequence; S4. Calculate the target power of the agricultural machine according to the resistance, apply the target power to the agricultural machine, and adjust the target power based on the difference between the actual forward acceleration of the agricultural machine and the target forward acceleration of the agricultural machine; The S4 comprises the following sub-steps: S41. Calculate the target power of the agricultural machinery based on the resistance: , where P target is the target power, a target is the target forward acceleration of the agricultural machinery, m is the mass of the agricultural machinery, F t is the resistance at the tth moment, V target is the target speed of the agricultural machinery; S42, after applying the target power to the agricultural machine, collecting the actual forward acceleration of the agricultural machine; S43, calculating the difference between the actual forward acceleration and the target forward acceleration of the agricultural machinery to obtain the forward acceleration difference, when the absolute value of the forward acceleration difference is less than the threshold value, the target power is not adjusted, and the sub-step ends; when the absolute value of the forward acceleration difference is greater than the threshold value, the resistance of the agricultural machinery is compensated, and the process proceeds to step S44; S44. Calculate the new target power of the agricultural machinery according to the compensated resistance: , where F t,ew To compensate for the resistance at the tth moment, P target,new is the new target power.

2. The agricultural machinery multi-terrain traveling method according to claim 1, characterized in that: The tilt angle sequence in S1 is composed of tilt angles at multiple moments, and the Z-axis acceleration sequence in S2 is composed of Z-axis accelerations at multiple moments; The elements in the tilt angle characteristic sequence in S1 include: angle fluctuation coefficient and angle variation coefficient; The elements in the Z-axis acceleration characteristic sequence in S2 include: a Z-axis acceleration fluctuation coefficient and a Z-axis acceleration variation coefficient.

3. The agricultural machinery multi-terrain traveling method according to claim 2, characterized in that: The method for obtaining the angle fluctuation coefficient is as follows: taking the tilt angles at N moments, taking the square of each tilt angle to obtain N angle square values, subtracting the square value of each angle from the square value of the tilt angle mean to obtain N angle differences, adding the absolute values ​​of the N angle differences, and normalizing the addition result to obtain the angle fluctuation coefficient, wherein the tilt angle mean is the average value of the tilt angles at N moments, and N is the length of the statistical data; The method for obtaining the Z-axis acceleration fluctuation coefficient is as follows: take the Z-axis acceleration at N moments, take the square of each Z-axis acceleration to obtain N acceleration square values, subtract each acceleration square value from the square value of the Z-axis acceleration mean to obtain N acceleration differences, add the absolute values ​​of the N acceleration differences, and normalize the addition result to obtain the Z-axis acceleration fluctuation coefficient, where the Z-axis acceleration mean is the average value of the Z-axis acceleration at N moments.

4. The agricultural machinery multi-terrain traveling method according to claim 2, characterized in that: The method for obtaining the angle variation coefficient is as follows: taking the tilt angles at N moments, calculating the difference between the tilt angles at two adjacent moments, obtaining N-1 adjacent angle differences, taking the square of each adjacent angle difference, obtaining N-1 angle difference square values, taking the average of the N-1 angle difference square values, obtaining the angle difference average, subtracting each angle difference square value from the angle difference average, obtaining N-1 angle deviations, adding the absolute values ​​of the N-1 angle deviations, and normalizing the addition result to obtain the angle variation coefficient, where N is the length of the statistical data; The method for obtaining the Z-axis acceleration variation coefficient is as follows: taking the Z-axis acceleration at N moments, calculating the difference between the Z-axis acceleration at two adjacent moments, obtaining N-1 adjacent acceleration differences, taking the square of each adjacent acceleration difference, obtaining N-1 acceleration difference square values, taking the average of the N-1 acceleration difference square values, obtaining the acceleration difference average, subtracting each acceleration difference square value from the acceleration difference average, obtaining N-1 acceleration deviations, adding the absolute values ​​of the N-1 acceleration deviations, and normalizing the addition result to obtain the acceleration variation coefficient.

5. The agricultural machinery multi-terrain traveling method according to claim 1, characterized in that: The dual-channel fusion recursive neural network in S3 includes: an angle tilt calculation channel, a Z-axis acceleration calculation channel, a fusion layer and a recursive output layer; The angle tilt operation channel is used to process the tilt angle sequence and the tilt angle feature sequence to obtain a first channel output value; The Z-axis acceleration operation channel is used to process the Z-axis acceleration sequence and the Z-axis acceleration characteristic sequence to obtain a second channel output value; The fusion layer is used to fuse the output value of the first channel and the output value of the second channel to obtain a fusion feature value; The recursive output layer is used to predict the resistance of the agricultural machinery according to the current fused feature value and the historical predicted resistance.

6. The agricultural machine multi-terrain travel method according to claim 5, characterized in that: The expression of the angle tilt operation channel is: , where G A,t is the output value of the first channel of the angle tilt calculation channel at time t, A t-n is the tilt angle at the tnth moment in the tilt angle sequence, w A,t-n A t-n The weight, θ Aw is the angular fluctuation coefficient, w Aw is θ Aw The weight, θ Ac is the angle variation coefficient, w Ac is θ Ac The weight of b A is the offset of the angle tilt calculation channel, t is the current time, n is the subscript of the historical time, and N is the length of the statistical data; The expression of the Z-axis acceleration operation channel is: , where G Z,t is the output value of the second channel output by the Z-axis acceleration operation channel at time t, Z t-n is the Z-axis acceleration at the tnth moment in the Z-axis acceleration sequence, w Z,t-n Z t-n The weight, θ Zw is the Z-axis acceleration fluctuation coefficient, w Zw is θ Zw The weight, θ Zc is the Z-axis acceleration variation coefficient, w Zc is θ Zc The weight of b Z It is the offset of the Z-axis acceleration calculation channel.

7. The agricultural machine multi-terrain travel method according to claim 6, characterized in that: The expression of the fusion layer is: , where H t is the fusion feature value output by the fusion layer at the tth moment, w A,t G A,t The weight, w Z,t G Z,t The weight of b G is the bias of the fusion layer.

8. The agricultural machine multi-terrain travel method according to claim 7, characterized in that: The expression of the recursive output layer is: , where F t is the resistance at time t, w H,t H t The weight of F t-n is the resistance at the tnth moment, w F,t-n F t-n The weight of b F is the bias of the recursive output layer, and M is the length of the historical resistance.

9. The agricultural machine multi-terrain travel method according to claim 1, characterized in that: The formula for compensating the resistance of the predicted agricultural machinery in S43 is: , where a act is the actual forward acceleration, μ is the proportional coefficient.

Images