AGV relative position control method based on kernel function prediction

Abstract

The invention relates to the field of control, in particular to an AGV relative position control method based on kernel function prediction, and the method comprises the steps: collecting the relative motion state characteristics of an AGV, and constructing a time sequence and an expected time sequence; for each feature, calculating a bandwidth adjustment coefficient based on multi-cycle length analysis, performing feature adaptive bandwidth adjustment on a Gaussian kernel function, and calculating an adaptive kernel similarity between the current moment and the historical moment; based on the similarity and a Gaussian process regression model, obtaining initial prediction values under each period length, and fusing the initial prediction values into a comprehensive prediction value through confidence weight fusion; and finally, carrying out closed-loop control on the relative position of the AGV according to the deviation between the comprehensive predicted value and the expected value. According to the method, fusion of feature adaptive kernel similarity calculation and multi-cycle prediction is realized, and the control precision and robustness under disturbance of tire slipping, load change and the like are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of control. More particularly, the present application relates to an AGV relative position control method based on kernel function prediction. BACKGROUND

[0002] AGV (Automated Guided Vehicle) is a key equipment for realizing automatic material handling, commonly used in multi-level garage, port and other scenes, and its high-precision relative position control is crucial.

[0003] Gaussian kernel function is a technique that maps input data to a high-dimensional feature space, and processes nonlinear problems in the original space by constructing linear relationships in the high-dimensional space. In AGV relative position control, kernel function can be used to process nonlinear features in sensor data, such as path curvature and obstacle shape, to more accurately describe the relationship between AGV and the surrounding environment.

[0004] However, the existing Gaussian kernel function assumes that the contribution weights of each position feature dimension (such as lateral and longitudinal deviation) are the same when calculating similarity. This ignores the different influences of actual disturbances (such as sudden increase in longitudinal error caused by inertia) on the reliability of different dimensions. This defect causes the similarity measure to be inaccurate, resulting in control prediction deviation, which can easily cause position error or even collision risk in dense multi-vehicle scenarios. SUMMARY

[0005] To solve the above technical problem of insufficient AGV control accuracy, the present application provides the following technical solutions.

[0006] An AGV relative position control method based on kernel function prediction, comprising: marking the AGV to be controlled as a target AGV, collecting a plurality of state features of the relative motion between the target AGV and a reference AGV, constructing a feature vector at each time, and obtaining a time sequence corresponding to each feature and an expected time sequence; For each feature, the following operations are performed to obtain the comprehensive prediction value of the feature: based on a plurality of preset cycle lengths, calculating the bandwidth adjustment coefficient of the feature under any cycle length according to the difference between the time sequence and the expected time sequence of the feature; adjusting the bandwidth of the Gaussian kernel function according to the bandwidth adjustment coefficient to calculate the adaptive kernel similarity of the feature vector between the current time and the historical time; based on the adaptive kernel similarity and the Gaussian process regression model, obtaining the initial prediction value of the feature under each cycle length, and weighting and fusing each initial prediction value to obtain the comprehensive prediction value of the feature; controlling the relative position of the target AGV based on the deviation between the comprehensive prediction value of each feature and its expected value.

[0007] Optionally, the feature vector comprises: relative position information, relative heading angle information, and respective linear and angular velocity information between the target AGV and the reference AGV.

[0008] Optionally, the bandwidth adjustment coefficient comprises: obtaining a feature error sequence of the feature according to a time sequence of the feature and an expected time sequence, taking a mean value of absolute values of elements in the feature error sequence as an average error, and calculating a first degree of drift based on a ratio of each absolute value to the average error; calculating a standard deviation of the feature error sequence, dividing the feature error sequence into sliding windows based on a period length, and taking a standard deviation of all elements in a window as a window standard deviation; calculating a ratio of each window standard deviation to the standard deviation of the feature error sequence, summing all the ratios, and taking an average to obtain a second degree of drift; adding the first degree of drift and the second degree of drift and taking an average to obtain the bandwidth adjustment coefficient.

[0009] Optionally, the adaptive kernel similarity comprises: obtaining a pre-sequence window containing the current time, and selecting any time except the current time in the pre-sequence window as a target time; assigning a comprehensive weight to each feature according to statistical information of a feature error sequence of each feature in the pre-sequence window; the comprehensive weight is an average of a first similarity weight and a second similarity weight; calculating an adaptive kernel similarity between feature vectors of the current time and the target time according to the comprehensive weight of each feature, a reciprocal of the bandwidth adjustment coefficient, and an initial bandwidth value of the Gaussian kernel function.

[0010] Optionally, the adaptive kernel similarity comprises: obtaining a pre-sequence window containing the current time, and selecting any time except the current time in the pre-sequence window as a target time; adjusting an initial bandwidth of the Gaussian kernel function based on a reciprocal of the bandwidth adjustment coefficient corresponding to each feature to obtain an adjusted Gaussian kernel function bandwidth; calculating distance components of each feature between the current time and the target time according to the adjusted Gaussian kernel function bandwidth; calculating a negative exponential function value of a sum of distance components of all features to obtain an adaptive kernel similarity between feature vectors of the current time and the target time.

[0011] Optionally, the weight in the weighted fusion comprises: For each period length, a preceding window containing the current time is obtained. The mean of all adaptive kernel similarities within the preceding window at the current time is calculated. The ratio of the mean of the adaptive kernel similarities under a single period length to the sum of the mean of the adaptive kernel similarities under all period lengths is used as the weight of the prediction confidence corresponding to that period length.

[0012] Optionally, obtaining the first similarity weight includes: Calculate the standard deviation of the feature error sequence of each feature within the preceding window, and use the ratio of the standard deviation of the feature error sequence of each feature to the average of the standard deviations of the feature error sequences of all features as the first similarity weight of that feature.

[0013] Optionally, obtaining the second similarity weight includes: Calculate the cumulative absolute value of the feature error sequence of each feature within the preceding window, and use the ratio of the cumulative value of each feature to the mean of the cumulative values ​​of all features as the second similarity weight of that feature.

[0014] Optionally, the adjusted Gaussian kernel function bandwidth is the product of the reciprocal of the bandwidth adjustment coefficient and the initial bandwidth.

[0015] The beneficial effects of this invention are: This invention introduces a bandwidth adjustment coefficient to adaptively adjust the bandwidth of the Gaussian kernel function. Combined with a multi-cycle length prediction mechanism, this effectively improves the accuracy and adaptability of AGV relative position control. Specifically, for each feature, the bandwidth adjustment coefficient is dynamically calculated based on the difference between its time series and the expected time series, enabling the Gaussian kernel function to more accurately reflect the local characteristics of feature changes. Using a Gaussian process regression model based on adaptive kernel similarity, initial predicted values ​​for different cycle lengths are obtained, and a weighted fusion is used to obtain a comprehensive predicted value, thereby enhancing the robustness of the prediction and its ability to characterize complex motion patterns. Finally, control decisions are made based on the deviation between the comprehensive predicted value and the expected value for each feature, achieving more accurate and stable tracking and control of the AGV's relative position, reducing control errors, and improving the system's response performance and reliability in dynamic environments. Attached Figure Description

[0016] Figure 1 This is a flowchart of steps S1-S5 in an AGV relative position control method based on kernel function prediction according to an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0018] The application scenario of the present application is: the relative position control environment of AGVs such as stereo garage, port heavy load AGV, etc. which exist strong nonlinear disturbance (such as tire slip, wet ground, load inertia, etc.), and at least includes two AGVs.

[0019] Reference Figure 1 A kind of AGV relative position control method based on kernel function prediction includes steps S1-S5, as follows: S1: the AGV needing to be controlled is marked as target AGV, the relative motion state features between target AGV and reference AGV are collected, the feature vector of each time is constructed, and the time sequence corresponding to each feature and the expected time sequence are obtained.

[0020] In AGV relative position control, it is difficult to accurately describe the dynamic behavior of the system only by relying on single sensor or single time data. The traditional method directly uses fixed features for similarity calculation, ignoring the actual contribution difference of different features in the movement process. Therefore, multi-source time series data need to be collected, and the complete feature vector and expected sequence are constructed in combination with the expected motion trajectory of the system.

[0021] Specifically, the AGV needing to be controlled is marked as target AGV, another AGV is used as reference AGV, the relative motion state features between the two are collected, including: relative lateral position, relative longitudinal position, relative heading angle, target AGV linear velocity, reference AGV linear velocity, target AGV angular velocity and reference AGV angular velocity.

[0022] After the above-mentioned 7 feature values of each time are subjected to conventional preprocessing operations, the feature vector of this time is formed.

[0023] In addition, from the system startup to the current time, all historical values of each feature constitute the time sequence of the feature, and at the same time, the expected time sequence of each feature from the system startup to the current time is obtained from the control system, which is derived from the upper path planning and motion control system.

[0024] S2: based on the difference between the time sequence and the expected time sequence of each feature, the bandwidth adjustment coefficient of the feature under any cycle length is calculated based on the preset multiple cycle lengths; the bandwidth of the Gaussian kernel function is adaptively adjusted according to the bandwidth adjustment coefficient.

[0025] The traditional Gaussian kernel function uses fixed bandwidth, considering that all features contribute equally to similarity. However, in actual movement, different features are affected differently by disturbance, for example, the error of lateral position is larger when turning, and the longitudinal position fluctuates significantly when starting and stopping. Therefore, the bandwidth of each feature needs to be dynamically adjusted according to the error size and fluctuation degree.

[0026] In addition, the AGV will experience different time scale motion modes during operation, such as short start-stop jitter, medium-term turning adjustment, long-term trajectory tracking, etc. In order to capture these multi-time scale dynamic characteristics, the present application presets multiple cycle lengths (such as 5, 7, 10 time points, etc.), and analyzes based on different cycles respectively, to achieve comprehensive modeling of motion details and trends.

[0027] Specifically, for any feature, the following operations are performed: Calculate the feature error sequence of the feature, that is, the difference between the time sequence of the feature and the corresponding elements of the expected time sequence.

[0028] Calculate the mean of the absolute values of all elements in the above feature error sequence as the average error; calculate the ratio of the absolute value of each element to the average error, sum all the ratios and take the average to get the first slip degree of the feature.

[0029] Calculate the standard deviation of the feature error sequence, select any one of the above cycle lengths such as 5 time points, divide the feature error sequence into multiple windows by sliding window, get the window standard deviation of all elements in the window; calculate the ratio of each window standard deviation to the standard deviation of the feature error sequence, and take the average of all the ratios to get the second slip degree.

[0030] Add the first slip degree and the second slip degree to get the average bandwidth adjustment coefficient of the feature under this cycle length.

[0031] Further, the bandwidth adjustment coefficient of all features under each cycle length is obtained.

[0032] The larger the above first slip degree, the higher the overall error level of the feature, and the bandwidth should be reduced to enhance the discrimination ability of the feature; the larger the above second slip degree, the stronger the local fluctuation of the feature, and the bandwidth should be reduced to suppress noise and transient disturbance.

[0033] By combining the first slip degree and the second slip degree, the larger the bandwidth adjustment coefficient, the stronger the overall deviation and the fluctuation of the feature, and the kernel bandwidth should be significantly reduced, so that the Gaussian kernel function is more sensitive to the feature, and the adaptability to nonlinear disturbance is improved.

[0034] S3: Calculate the adaptive kernel similarity of the feature vector between the current time and the historical time according to the adjusted bandwidth of the Gaussian kernel function.

[0035] In the prior art, the calculation formula of the standard Gaussian kernel function is: However, the present application makes adaptive improvements to the above formula in view of the significant difference in feature error of AGV in strong disturbance scenarios.

[0036] In one embodiment, a bandwidth adjustment coefficient is introduced to achieve feature adaptation: In the formula, The adaptive kernel similarity between the feature vectors at the current time and the target time. For the number of features, For the current moment The value of each feature, For the target time The value of each feature, For the initial bandwidth, For the first The reciprocal of the bandwidth adjustment coefficient for each feature. It is an exponential function with the natural number e as its base. It is the sum of the distance components of all features.

[0037] Specifically, a preceding window containing the current time is obtained, meaning the endpoint of the preceding window is the current time, and any time within the preceding window other than the current time is selected as the target time.

[0038] pass The bandwidth of each feature is adjusted independently; the larger the error, the more... The smaller the value, the smaller the denominator, and the greater the contribution of this feature to the similarity. This allows the Gaussian kernel function to adaptively focus on high-error features when calculating similarity, thereby improving its ability to identify key perturbations.

[0039] Although the above operations have achieved bandwidth adaptation, the difference terms of each feature are still assumed by default. The contribution to overall similarity is adjusted solely by bandwidth. In actual motion, fluctuations or accumulated errors in certain features (such as lateral position) have a greater impact on safety and control and should be given higher weight.

[0040] In another embodiment, a preceding window containing the current time is obtained, and any time other than the current time within the preceding window is selected as the target time.

[0041] Calculate the standard deviation of the feature error sequence of each feature within the preceding window, and use the ratio of the standard deviation of the feature error sequence of each feature to the average of the standard deviations of the feature error sequences of all features as the first similarity weight of that feature; calculate the cumulative value of the absolute error of the feature error sequence of each feature within the preceding window, and use the ratio of the calculated cumulative value of each feature to the average of the cumulative values ​​of all features as the second similarity weight of that feature.

[0042] Furthermore, the average of the sum of the first similarity weight and the second similarity weight is used as the comprehensive weight of this feature.

[0043] Based on the combined weights of each feature, the reciprocal of the bandwidth adjustment coefficient, and the initial bandwidth value of the Gaussian kernel function, the adaptive kernel similarity between the feature vectors at the current time and the target time is calculated. That is, it satisfies the following relationship: In the formula, The adaptive kernel similarity between the feature vectors at the current time and the target time. For the number of features, For the first The combined weight of each feature, For the current moment The value of each feature, For the target time The value of each feature, For the initial bandwidth, For the first The reciprocal of the bandwidth adjustment coefficient for each feature. It is an exponential function with the natural number e as its base.

[0044] A higher first similarity weight indicates greater volatility of the feature; a higher second similarity weight indicates greater cumulative error of the feature. The combined weight integrates both, further emphasizing the focus on highly volatile and error-prone features during kernel similarity calculation. Increasing the combined weight enhances the feature's contribution to similarity calculation, thus more accurately reflecting key state changes in prediction.

[0045] It should be noted that S3 above is performed with a certain period length. By repeating the above operation for each period length, the adaptive kernel similarity of the feature vectors between the current time and the historical time can be obtained for each period length.

[0046] S4: Based on the adaptive kernel similarity and Gaussian process regression model, the initial predicted values ​​of each feature at each period length are obtained, and the initial predicted values ​​are weighted and fused to obtain the comprehensive predicted value of the feature.

[0047] In S2 above, multiple period lengths were introduced to capture motion characteristics at different time scales, and in S3 above, adaptive kernel similarity calculation based on each period length was completed. However, if prediction is based solely on a single period length, multi-scale information cannot be fully utilized, and inappropriate period selection may affect prediction accuracy. For example: If only short periods are used, although they can respond to instantaneous disturbances, they are easily affected by noise and the prediction results fluctuate greatly; if only long periods are used, although the trend is stable, they cannot capture sudden changes in the state in time.

[0048] Therefore, in order to realize comprehensive and robust prediction of the motion state of the AGV, a multi-cycle prediction fusion mechanism is proposed, that is, prediction is performed under each cycle length, and then the prediction results under each cycle length are weighted and fused according to the confidence of the prediction results, so as to comprehensively integrate the detail sensitivity of the short cycle and the trend stability of the long cycle.

[0049] Specifically, for each cycle length, the feature vectors at each time in the above-obtained pre-window, the feature vector at the current time, and the adaptive kernel similarity between the current time and each time in the pre-window are input into the trained Gaussian process regression model, and the model outputs the initial prediction value of each feature at the next time of the current time.

[0050] Further, if the prediction results under each cycle length are simply averaged, the reliability difference of different cycles at different times will be ignored. Therefore, it is also necessary to calculate the prediction confidence weight of the cycle length to evaluate the reliability of the prediction.

[0051] For each cycle length, the mean of all adaptive kernel similarities in the pre-window at the current time is calculated, and the ratio of the mean of the adaptive kernel similarity under a single cycle length to the sum of the means of the adaptive kernel similarities under all cycle lengths is taken as the prediction confidence weight corresponding to the cycle length.

[0052] The higher the prediction confidence weight is, the more consistent the historical features under the cycle length are with the current state, and the more reliable the prediction result is.

[0053] Further, the initial prediction values under each cycle length are weighted and summed according to the prediction confidence weight, to obtain the comprehensive prediction value of the feature at the next time.

[0054] S5: Based on the deviation between the comprehensive prediction value of each feature and the expected value, the relative position of the target AGV is controlled.

[0055] After obtaining the comprehensive prediction value of each feature, it needs to be compared with the expected value, and a control instruction is generated according to the deviation to realize accurate relative position tracking of the AGV. The traditional method mainly relies on feedback control of the current error, and there is a problem of response lag. The present application introduces feedforward prediction information, which can compensate for the expected deviation in advance, thereby improving the dynamic performance of the system.

[0056] Specifically, the comprehensive prediction value of each feature and the expected value are input into the position loop and speed loop controller, and the controller generates a control signal according to the deviation to drive the target AGV to perform corresponding position and speed adjustment.

[0057] The above controller can be one or a combination of common closed-loop controllers such as a PID controller, a fuzzy controller, and a model predictive controller, and the present application does not make a specific limitation on the type of controller.

[0058] The application solves the problems of large error and easy collision of the traditional AGV relative position control under strong nonlinear disturbance by the triple mechanism of feature adaptive bandwidth adjustment, multi-period similarity fusion and feedforward-feedback compound control. Each step is linked together to form a full-process adaptive closed loop from data acquisition, feature analysis, prediction modeling to control execution, thereby improving the reliability, accuracy and adaptability of the system in complex scenarios.

[0059] It should be noted that the basic form and calculation method of the Gaussian kernel function used in the application belong to the known technology, and the specific implementation can refer to the related classic literature or open source tool library.

[0060] The training process of the Gaussian process regression model includes steps such as kernel function selection, hyperparameter optimization (such as bandwidth, noise variance), model fitting based on historical data, which are all standard processes of Gaussian process regression.

[0061] It should be pointed out that for those skilled in the art, without departing from the concept of the application, a number of modifications and improvements can be made, which are all within the protection scope of the application. Therefore, the protection scope of the application patent should be subject to the appended claims.

Claims

1. An AGV relative position control method based on kernel function prediction, characterized in that, The method comprises the following steps: Mark the AGV to be controlled as a target AGV, collect a plurality of state characteristics of the relative motion between the target AGV and a reference AGV, construct a feature vector at each time, and obtain a time sequence corresponding to each feature and an expected time sequence; For each feature, the following operations are performed to obtain a comprehensive prediction value of the feature: Based on a plurality of preset cycle lengths, calculate a bandwidth adjustment coefficient of the feature under any cycle length according to the difference between the time sequence and the expected time sequence of the feature; adjust the bandwidth of the Gaussian kernel function according to the bandwidth adjustment coefficient, calculate the adaptive kernel similarity between the feature vectors between the current time and the historical time, obtain the initial prediction value of the feature under each cycle length based on the adaptive kernel similarity and the Gaussian process regression model, and weight and fuse each initial prediction value to obtain the comprehensive prediction value of the feature; Control the relative position of the target AGV based on the deviation between the comprehensive prediction value of each feature and the expected value.

2. The AGV relative position control method based on kernel function prediction according to claim 1, characterized in that, The feature vector comprises: The relative position information, the relative heading angle information, and the linear velocity and angular velocity information of the target AGV and the reference AGV. 3.The AGV relative position control method based on kernel function prediction according to claim 1, wherein, The acquisition of the bandwidth adjustment coefficient comprises: Obtain the feature error sequence of the feature according to the time sequence and the expected time sequence of the feature, take the mean of the absolute values of the elements in the feature error sequence as the average error, and calculate the first degree of drift based on the ratio of each element absolute value to the average error; Calculate the standard deviation of the feature error sequence, divide the feature error sequence into sliding windows based on a cycle length, and take the standard deviation of all elements in the window as the window standard deviation; Calculate the ratio of each window standard deviation to the standard deviation of the feature error sequence, sum all the ratios and take the average to obtain the second degree of drift; Add the first degree of drift and the second degree of drift to obtain the bandwidth adjustment coefficient.

4. The AGV relative position control method based on kernel function prediction according to claim 3, characterized in that, The acquisition of the adaptive kernel similarity comprises: Obtain a pre-window containing the current time, and select any time in the pre-window except the current time as a target time; According to the statistical information of the feature error sequence of each feature in the pre-window, a comprehensive weight is assigned to each feature; the comprehensive weight is the average of the sum of the first similarity weight and the second similarity weight; According to the comprehensive weight of each feature, the reciprocal of the bandwidth adjustment coefficient, and the initial bandwidth value of the Gaussian kernel function, the adaptive kernel similarity between the feature vectors of the current time and the target time is calculated.

5. The AGV relative position control method based on kernel function prediction according to claim 3, characterized in that, The acquisition of the adaptive kernel similarity comprises: Obtain a pre-window containing the current time, and select any time in the pre-window except the current time as a target time; Adjust the initial bandwidth of the Gaussian kernel function based on the reciprocal of the bandwidth adjustment coefficient of each feature to obtain the adjusted Gaussian kernel function bandwidth; According to the adjusted Gaussian kernel function bandwidth, calculate the distance components of each feature between the current time and the target time; Calculate the negative exponential function value of the sum of the distance components of all features to obtain the adaptive kernel similarity between the feature vectors of the current time and the target time.

6. The AGV relative position control method based on kernel function prediction according to claim 1, characterized in that, The acquisition of the weight in the weighted fusion comprises: For each cycle length, a pre-window containing the current time is obtained, the mean of all adaptive kernel similarities in the pre-window of the current time is calculated, and the ratio of the mean of the adaptive kernel similarity under the single cycle length to the sum of the means of the adaptive kernel similarities under all cycle lengths is taken as the prediction confidence weight corresponding to the cycle length.

7. The AGV relative position control method based on kernel function prediction according to claim 4, characterized in that, The obtaining of the first similarity weight comprises: The standard deviation of the feature error sequence of each feature in the pre-window is calculated, and the ratio of the standard deviation of the feature error sequence of each feature to the average of the standard deviations of the feature error sequences of all features is taken as the first similarity weight of the feature.

8. The AGV relative position control method based on kernel function prediction according to claim 4, characterized in that, The obtaining of the second similarity weight comprises: The cumulative value of the absolute error value of the feature error sequence of each feature in the pre-window is calculated, and the ratio of the cumulative value of each feature to the average of the cumulative values of all features is taken as the second similarity weight of the feature.

9. The AGV relative position control method based on kernel function prediction according to claim 5, characterized in that, The adjusted Gaussian kernel function bandwidth is the product of the inverse of the bandwidth adjustment coefficient and the initial bandwidth.

Images