AGV pallet fork speed control method based on CatBoost
The actual lifting and lowering speed of the AGV fork is predicted through the CatBoost method, which solves the problem of insufficient speed linearization meter and lack of correspondence between hydraulic systems in the prior art, and realizes precise speed control in various states.
Patent Information
- Application Number
- CN202311850174.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-29
- Publication Date
- 2024-11-12
AI Technical Summary
The existing AGV fork speed control method requires the production of a priori speed linearization table, insufficient control refinement, and lack of correspondence between the speed linearization output and the actual proportional valve given to the hydraulic system, resulting in an increase in error.
Using CatBoost-based method, data is collected by installing sensors, Pearson's correlation coefficient preferred variables, random search to optimize hyperparameters, Catboost gradient lift training, predict fork lift speed, and build a model to minimize the expected lift speed and actual lift The difference between speeds.
Accurately predicting the actual lifting and lowering speed of forks under various states, improving the accuracy and stability of fork speed control, and solving the problem of insufficient speed linearization meter and the lack of correspondence between proportional valves in the hydraulic system.
Smart Images

Figure CN118929522A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of AGV (automatic guided vehicle). More specifically, the present invention relates to a CatBoost-based AGV fork speed control method. Background Art
[0002] 1. Introduction to the development background of related technologies:
[0003] AGV (Automated Guided Vehicle) is a kind of transportation tool equipped with electromagnetic or optical automatic navigation equipment, which can travel along the specified navigation path, with safety protection and various handling functions. This kind of car is used for automated handling tasks in the industrial field, without the need for a driver, and uses rechargeable batteries as the power source.
[0004] The lifting drive mechanism of the fork in the AGV is similar to that of an ordinary forklift, and both use a hydraulic pump mechanism to control the lifting and lowering movement of the fork. However, unlike a traditional forklift, the control system of the AGV forklift needs to have autonomous decision-making capabilities and be able to intelligently control the movement of the fork according to the task and environmental requirements.
[0005] In traditional forklifts, the driver can achieve the appropriate lifting speed through manual control, while the controller of the AGV forklift can only receive simple commands based on the preset output proportional valve. Due to the nonlinear characteristics of the hydraulic system, there may be a complex nonlinear relationship between the output of the proportional valve and the actual fork lifting speed. Therefore, directly using a simple linear conversion may introduce errors. When using PID for speed control, if the controller parameters are not set correctly, it may cause the system to respond excessively or insufficiently. Another possible method is to control by querying the speed linear table, but this requires a large amount of experimental data to form a speed table in advance. However, due to various uncertainties in actual operation, as well as differences in the working environment and hydraulic system maintenance of different AGVs, the error may gradually increase.
[0006] 2. Existing technology:
[0007] After searching the existing relevant technical literature, the following search results were obtained:
[0008] (1) Chinese patent document: "Method for automatic adjustment of PID parameters for AGV speed regulation based on genetic algorithm convolutional neural network", patent (application) number is 202310441748.4, and the technical solution recorded therein is:
[0009] The method comprises the following steps: "S10: installing sensors on the AGV to collect on-site environmental data; S20: processing the data collected in step S10 and preprocessing the data; S30: initializing the convolutional neural network; S40: initializing various parameters of the genetic algorithm; S50: real number encoding the population, and taking the error between the predicted data and the expected data as the fitness function; S60: model training, adjusting the parameters of the network model according to the genetic algorithm; S70: testing deployment and putting it into use".
[0010] The technical effects recorded are:
[0011] "When deploying AGVs, it is necessary to manually adjust PID parameters, which consumes a lot of manpower costs. When the environment changes significantly, manual adjustments need to be made again."
[0012] This technical solution uses a convolutional neural network based on a genetic algorithm to output PID parameters, thereby realizing an automatic PID tuning method for AGV.
[0013] (2) Chinese patent document: "Forklift-type AGV fork lifting speed control method and device", patent (application) number is 201811601214.9, the technical solution recorded is:
[0014] The control method comprises the following steps: "Step S1, obtaining an instruction to lift the fork; Step S2, determining whether the fork has cargo; Step S3, determining whether the fork has risen; Step S4, querying a speed linearization table to obtain a speed output value";
[0015] The technical effects recorded are:
[0016] "By preforming a speed linearization table of the controller's speed output value and the actual lifting speed of the fork, when the fork is rising and falling, the corresponding speed linearization table is queried according to no-load and loaded, and the corresponding rising and falling speed output values are obtained, so that the speed control is very accurate. There is no difference in the actual lifting speed of the fork when it is empty and loaded, thereby improving the accuracy of the fork lifting height";
[0017] The closest existing technology and its technical problems:
[0018] Comparative document (2) is the prior art closest to the present invention. This solution determines whether the fork has cargo based on the cargo sensor, determines whether the fork has risen based on the target height value and the current height value, queries the speed linearization table, obtains the speed output value, and outputs it to the motor or proportional valve of the driving pump of the hydraulic system, wherein the controller queries the speed linearization table to obtain the speed output value. This technical solution has the following two disadvantages:
[0019] (1) It is necessary to make a priori speed linearization table, but the fork speed control is not refined enough.
[0020] (2) There is a lack of correspondence between the speed value output by speed linearization and the speed value actually given to the proportional valve of the hydraulic system. Summary of the invention
[0021] The present invention provides an AGV fork speed control method based on CatBoost, the purpose of which is to predict the actual lifting speed of the fork teeth of a forklift AGV under various states.
[0022] In order to achieve the above object, the technical solution adopted by the present invention is:
[0023] The CatBoost-based AGV fork speed control method of the present invention includes the following process:
[0024] Step 1, install the sensor: the sensor includes a piezoelectric sensor, a photoelectric encoder, and a wire encoder;
[0025] Step 2: Data collection and preprocessing: The data includes the load F of the fork tine. w , the vehicle's speed V, the current height of the fork tine F h , expected lifting speed S, actual lifting speed S real The internal and external information of the vehicle is selected using the Pearson correlation coefficient;
[0026] Step 3: Use random search to optimize Catboost hyperparameters.
[0027] Step 4: Catboost gradient boosting training;
[0028] Step 5: Use Catboost to complete the training model F(x) to predict the fork lifting speed.
[0029] In the step 1, a piezoelectric sensor is installed on the AGV fork to obtain the load L; a photoelectric encoder is installed on the AGV driving wheel to obtain the driving speed of the AGV; a pull-wire encoder is installed between the AGV fork lifting mechanism and the chassis to obtain the height information of the fork.
[0030] In step 2, various data during the operation of the AGV are collected, and the internal and external data of the vehicle are recorded as the independent variable X, including the body power, fork load, vehicle linear speed and angular speed, current height and target height of the fork; the actual lifting speed of the fork is recorded as the dependent variable Y; the Pearson correlation coefficient is used to find the variable with the highest correlation; the calculation formula is as follows:
[0031]
[0032] Among them, xi and yi are the observed values of variables x and y, and is the average value;
[0033] Retention correlation coefficient ρ x,y Variables > 0.6;
[0034] The processed data were divided into training set, validation set and test set according to the ratio of 7:2:1.
[0035] The step 3 comprises the following specific steps:
[0036] Step 3.1: Define the Catboost parameter search space θ random , where depth random Tree depth (1 to 10), learning rate learning_rate random (0.01 to 0.3), number of iterations itrationrandom random (100 to 1000), L2 regularization parameter (1 to 30);
[0037] θ random ={depth random , learning_rate random ,…} Formula 2
[0038] Step 3.2: Build a network using randomly selected parameter combinations:
[0039]
[0040] Step 3.3, set the Catboost error calculation method to the average error RMSE between the predicted value and the actual value. The calculation formula is as follows:
[0041]
[0042] Where n is the number of samples, and the input feature of each sample is the AGV fork speed control method Xi based on CatBoost AGV fork speed control method based on CatBoost (F w , V, F h , S), y i is the actual value of the ith sample, i.e., the lifting speed label is the predicted value of the i-th sample;
[0043] Step 3.4: Update the optimal parameter θ optimal , if the current parameter combination is CatBoost-based AGV fork speed control method θ randomIf the model of the AGV fork speed control method based on CatBoost performs better than the previous optimal model, the optimal parameters are updated;
[0044] Step 3.5: Repeat steps 3.2 to 3.4 until the preset number of iterations is reached and the optimal hyperparameter h is obtained. * :
[0045]
[0046] Among them, D val represents the validation set.
[0047] In step 4, the input of the model is the independent variable X in step 2, and the output value is the actual lifting speed of the fork tine; the neural network is trained using the collected data; the training goal is to minimize the expected lifting speed S and the actual lifting speed S real The best hyperparameter h* obtained in step 3 is used during training.
[0048] The step 4 comprises the following specific steps:
[0049] Step 4.1, Initialize the model: First initialize a constant model:
[0050]
[0051] Among them, F0(x) is the initial model, L(y i ,γ) is the loss function, which is the RMSE in step 3.3;
[0052] Step 4.2, iterative training: Start iterative training; in each round of iteration, a new model is trained to fit the residual of the current model:
[0053]
[0054]
[0055]
[0056] F t (x) = F t-1 (x)+γ t h(x) Formula 10
[0057] Among them, γ it is the residual of the i-th sample in the t-th iteration, h t (x) is the model trained in the tth round of iteration, γ t is the learning rate of the tth iteration, f t (x) is the model after the tth iteration;
[0058] Step 4.3, output the final model: The final model is the sum of all iteratively trained models:
[0059]
[0060] Among them, the AGV fork speed control method F(x) based on CatBoost is the final model, and m is the number of iterations.
[0061] The step 5 comprises the following specific steps:
[0062] Step 5.1: Integrate the trained Catboost model into the controller of the AGV fork;
[0063] Step 5.2: Start the AGV forklift and input all the optimized parameters of the Catboost model, including fork load, vehicle speed, fork height, expected speed, and expected lifting speed, which are recorded as Xi(F w , V, F h , S), and get the predicted value y = F(X i ), y is the actual lifting speed of the fork tine.
[0064] The present invention adopts the above technical solution to predict the actual lifting speed by integrating the forklift AGV sensor data and the input expected speed, so that the expected speed is close to the actual fork tine movement speed under various conditions. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] Figure 1 It is a schematic diagram of the installation structure of the sensor of the present invention;
[0066] Figure 2 for Figure 1 A schematic top view of the structure shown;
[0067] Figure 3 The figure is a flow chart of the fork speed control method of the present invention. DETAILED DESCRIPTION
[0068] The specific implementation modes of the present invention are further explained in detail below by describing the embodiments with reference to the accompanying drawings, so as to help those skilled in the art to have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention.
[0069] like Figure 1 , Figure 2 The figure shows a schematic diagram of the structure of the present invention. The present invention is a CatBoost-based AGV fork speed control method. The present invention is mainly used for the fork speed control of AGV (automatic guided vehicle), especially in the occasions where the fork lifting speed needs to be accurately controlled, such as warehousing logistics, manufacturing industry, etc.
[0070] In order to solve the problems existing in the prior art and overcome its defects, and to achieve the invention purpose of predicting the actual lifting speed of the fork teeth of the forklift AGV in various states, the technical solution adopted by the present invention is:
[0071] like Figure 3 As shown, the AGV fork speed control method of the present invention is as follows:
[0072] Step 1: Install the sensor. Install a piezoelectric sensor on the AGV fork to obtain the load L. Install a photoelectric encoder on the AGV drive wheel to obtain the AGV's travel speed. Install a wire encoder between the AGV fork lifting mechanism and the chassis to obtain the fork height information.
[0073] Installation diagram Figure 1 , Figure 2 shown.
[0074] Step 2, data collection and preprocessing: collect various data during AGV operation, record the internal and external data of the vehicle as the independent variable X, the internal and external data of the vehicle include the body power, fork load, vehicle linear speed and angular velocity, current height and target height of the fork; record the actual lifting speed of the fork as the dependent variable Y; use the Pearson correlation coefficient to find the variable with the highest correlation; the calculation formula is as follows:
[0075]
[0076] Among them, xi and yi are the observed values of variables x and y, and is the average value. Variables with correlation coefficient ρx,y>0.6 are retained. The processed data are divided into training set, validation set and test set according to 7∶2∶1;
[0077] Step 3: Use random search to optimize Catboost hyperparameters, including the following steps:
[0078] Step 3.1: Define the Catboost parameter search space θ random , where depth random Tree depth (1 to 10), learning rate learning_rate random (0.01 to 0.3), number of iterations itrationrandom random (100 to 1000), L2 regularization parameter (1 to 30).
[0079] θ random ={depth random ,learning_rate random ,...} Formula 2
[0080] Step 3.2: Build a network using randomly selected parameter combinations:
[0081]
[0082] Step 3.3, set the Catboost error calculation method to the average error RMSE between the predicted value and the actual value. The calculation formula is as follows:
[0083]
[0084] Where n is the number of samples, and the input feature of each sample is the AGV fork speed control method Xi based on CatBoost AGV fork speed control method based on CatBoost (F w , V, F h , S), y i is the actual value of the ith sample, i.e., the lifting speed label is the predicted value of the sample.
[0085] Step 3.4: Update the optimal parameter θ optimal , if the current parameter combination is CatBoost-based AGV fork speed control method θ random If the model performance of the AGV fork speed control method based on CatBoost is better than the previous optimal model, the optimal parameters are updated.
[0086] Step 3.5: Repeat steps 3.2 to 3.4 until the preset number of iterations is reached and the optimal hyperparameter h* is obtained:
[0087]
[0088] Among them, D val represents the validation set.
[0089] Step 4: Catboost gradient boosting training: The input of the model is the independent variable X in step 2, and the output value is the actual lifting speed of the fork tine;
[0090] The neural network is trained using the collected data; the training goal is to minimize the expected lift speed S and the actual lift speed S real The difference between the two; the optimal hyperparameter h obtained in step 3 is used during training * ; Specifically, gradient boosting training includes the following steps:
[0091] Step 4.1, Initialize the model: First initialize a constant model:
[0092]
[0093] Among them, F0(x) is the initial model, L(y i,γ) is the loss function, which is the RMSE in step 3.3;
[0094] Step 4.2, iterative training: Start iterative training; in each round of iteration, a new model is trained to fit the residual of the current model:
[0095]
[0096]
[0097]
[0098] F t (x) = F t-1 (x)+γ t h t (x) Formula 10
[0099] Among them, γ it is the residual of the i-th sample in the t-th iteration, h t (x) is the model trained in the tth round of iteration, γ t is the learning rate of the tth iteration, f t (x) is the model after the tth iteration.
[0100] Step 4.3, output the final model: The final model is the sum of all iteratively trained models:
[0101]
[0102] Among them, the AGV fork speed control method F(x) based on CatBoost is the final model, and m is the number of iterations.
[0103] Step 5: Use the trained model F(x) using Catboost to predict the fork tine lifting speed. Specifically, the following steps are included:
[0104] Step 5.1: Integrate the trained Catboost model into the controller of the AGV fork;
[0105] Step 5.2: Start the AGV forklift and input all parameters of the Catboost model, including fork load, vehicle speed, fork height, expected speed, and expected lifting speed, which are recorded as Xi(F w , V, F h , S), and get the predicted value y = F(X i ), y is the actual lifting speed of the fork tine.
[0106] The present invention proposes the above-mentioned technical solution of the AGV fork speed control method based on CatBoost, which solves the problem of the correspondence between the fork-tine AGV hydraulic proportional valve and the actual fork lifting speed. Its innovation and beneficial effects are: by integrating the forklift AGV sensor data and the input expected speed to predict the actual lifting speed, the expected speed is close to the actual fork movement speed under various conditions.
[0107] Comparative analysis of the present invention and prior art solutions:
[0108] The technical solution of the Chinese patent document (1) introduced in the background technology is similar to the technical solution of the present invention in that both use model prediction to control the AGV; different from the technical solution of the present invention in that this technical solution uses a convolutional neural network to predict PID parameters.
[0109] The technical solution of Chinese patent document (2) is the same as the technical solution of the present invention in that the lifting speed of the AGV fork is controlled by an algorithm; the difference from the technical solution of the present invention is that this technical solution uses a speed linearization table to obtain the speed output value under different load conditions and lifting conditions.
[0110] Therefore, the present invention solves the problems and defects in the prior art and achieves significant technical progress.
[0111] When implementing the technical solution of the present invention, the following issues should be noted:
[0112] The present invention requires that a forklift AGV equipped with various sensors perform random operations and collect a certain amount of data to train CatBoost regression. This method predicts the actual lifting speed by integrating sensor data and inputting the expected speed. Overall, it is a process that includes data collection, model selection, training, and optimization parameters. Training may require a certain amount of time and computing resources, especially when the dimension of the input data is high or the training data set is large.
[0113] If the training data is insufficient or the model is too complex, the neural network may suffer from the problem of overfitting, that is, the model performs well on the training data but performs poorly on new data.
[0114] When the hardware computing power is sufficient, use neural networks or other machine learning methods to automatically learn and adjust the relationship between the output proportional valve and the actual lifting speed based on real-time sensor data to improve the control accuracy and stability; or use fuzzy control or other intelligent control methods to automatically select appropriate controller parameters according to different tasks and environmental requirements to achieve adaptive lifting control.
[0115] The present invention is described above by way of example in conjunction with the accompanying drawings. It is obvious that the specific implementation of the present invention is not limited to the above-mentioned method. As long as various non-substantial improvements are made using the method concept and technical solution of the present invention, or the concept and technical solution of the present invention are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.
Claims
1. A CatBoost-based AGV fork speed control method, characterized in that: The control method includes the following steps: Step 1, install the sensor: the sensor includes a piezoelectric sensor, a photoelectric encoder, and a wire encoder; Step 2: Data collection and preprocessing: The data is the internal and external information of the vehicle, including the load F of the fork tine. w , the vehicle's speed V, the current height of the fork tine F h , expected lifting speed S, actual lifting speed S real ; and use the Pearson correlation coefficient for optimization; Step 3: Use random search to optimize Catboost hyperparameters. Step 4: Catboost gradient boosting training; Step 5: Use Catboost to complete the training model F(x) to predict the fork lifting speed.
2. The CatBoost-based AGV fork speed control method according to claim 1, characterized in that: In the step 1, a piezoelectric sensor is installed on the AGV fork to obtain the load L; a photoelectric encoder is installed on the AGV driving wheel to obtain the driving speed of the AGV; a pull-wire encoder is installed between the AGV fork lifting mechanism and the chassis to obtain the height information of the fork.
3. The CatBoost-based AGV fork speed control method according to claim 1, characterized in that: In step 2, various data during the operation of the AGV are collected, and the internal and external data of the vehicle are recorded as the independent variable X, including the body power, fork load, vehicle linear speed and angular speed, current height and target height of the fork; the actual lifting speed of the fork is recorded as the dependent variable Y; the Pearson correlation coefficient is used to find the variable with the highest correlation; the calculation formula is as follows: Among them, x i and i are the observed values of variables x and y, and is the average value; Retention correlation coefficient ρ x,y Variables > 0.6; The processed data were divided into training set, validation set and test set according to the ratio of 7:2:
1.
4. The CatBoost-based AGV fork speed control method according to claim 1, characterized in that: The step 3 comprises the following specific steps: Step 3.1: Define the Catboost parameter search space θ random , where depth random Tree depth (1 to 10), learning rate learning_rate random (0.01 to 0.3), number of iterations itrationrandom random (100 to 1000), L2 regularization parameter (1 to 30); θ random ={depth random , learning_rate random , ...} Formula 2 Step 3.2: Build a network using randomly selected parameter combinations: Step 3.3, set the Catboost error calculation method to the average error RMSE between the predicted value and the actual value. The calculation formula is as follows: Where n is the number of samples, and the input feature of each sample is the AGV fork speed control method Xi based on CatBoost AGV fork speed control method based on CatBoost (F w , V, F h , S), y i is the actual value of the ith sample, i.e., the lifting speed label is the predicted value of the i-th sample; Step 3.4: Update the optimal parameter θ optimal , if the current parameter combination is CatBoost-based AGV fork speed control method θ random If the model performance of the AGV fork speed control method based on CatBoost is better than the previous optimal model, the optimal parameters are updated.
5. The CatBoost-based AGV fork speed control method according to claim 4, characterized in that: After step 3.4, proceed to step 3.5; Step 3.5: Repeat steps 3.2 to 3.4 until the preset number of iterations is reached and the optimal hyperparameter h is obtained. * : Among them, D val represents the validation set.
6. The CatBoost-based AGV fork speed control method according to claim 1, characterized in that: In step 4, the input of the model is the independent variable X in step 2, and the output value is the actual lifting speed of the fork tine; the collected data is used to train the neural network.
7. The CatBoost-based AGV fork speed control method according to claim 6, characterized in that: The goal of the training neural network is to minimize the expected lifting speed S and the actual lifting speed S real The difference between the two; the optimal hyperparameter h obtained in step 3 is used during training * .
8. The CatBoost-based AGV fork speed control method according to claim 6 or 7, characterized in that: The step 4 comprises the following specific steps: Step 4.1, Initialize the model: First initialize a constant model: Among them, F0(x) is the initial model, L(y i , γ) is the loss function, which is the RMSE in step 3.3; Step 4.2, iterative training: Start iterative training; in each round of iteration, a new model is trained to fit the residual of the current model: F t F(x) = t-1 F(x) + γ t ht(x) Equation 10 Among them, γ it is the residual of the i-th sample in the t-th iteration, h t (x) is the model trained in the tth round of iteration, γ t is the learning rate of the tth iteration, f t (x) is the model after the tth iteration; Step 4.3, output the final model: The final model is the sum of all iteratively trained models: Among them, the AGV fork speed control method F(x) based on CatBoost is the final model, and m is the number of iterations.
9. The CatBoost-based AGV fork speed control method according to claim 1, characterized in that: The step 5 comprises the following specific steps: Step 5.1: Integrate the trained Catboost model into the controller of the AGV fork; Step 5.2: Start the AGV forklift and input all the optimized parameters of the Catboost model, including fork load, vehicle speed, fork height, expected speed, and expected lifting speed, which are recorded as Xi(F w , V, F h , S), and get the predicted value y = F(X i ), y is the actual lifting speed of the fork tine.
