Sugarcane harvester feed quantity monitoring method and system using particle swarm optimization algorithm to optimize BP neural network

Abstract

The invention relates to a sugarcane harvester feed quantity monitoring method based on a BP neural network optimized by a particle swarm algorithm. The method comprises the following steps: collecting working data of key parts of a sugarcane harvester; filtering the data; constructing a sugarcane harvester feed quantity monitoring model of the BP neural network optimized by the particle swarm optimization; the data acquisition comprises the pressure of an oil outlet of a root cutter hydraulic motor, the pressure of an oil outlet of a lower conveying roller hydraulic motor, the displacement of an upper conveying roller and the flow of a cutting roller hydraulic motor; and the data filtering processing is to carry out noise reduction on the data acquired by the system by using an iterative wavelet threshold noise reduction method. According to the invention, real-time monitoring of the feeding amount during working of the sugarcane harvester is realized, and technical support is provided for guiding a driver to adjust the working state of the harvester, real-time adjustment of the working parameters of the harvester based on the feeding amount and improvement of the harvesting quality and efficiency.

Description

Technical Field

[0001] The invention relates to the technical field of agricultural machinery and equipment, and in particular to a sugarcane harvester feeding amount monitoring method and a system thereof using a particle swarm algorithm to optimize a BP neural network. Background Art

[0002] Sugarcane is an important cash crop in my country and the main raw material for producing sucrose. At present, the domestic industry mainly relies on manual sugarcane cutting, which has low harvesting efficiency and high cost. Promoting the mechanization of sugarcane harvesting can reduce the labor intensity of sugarcane farmers and increase their income. As an important parameter of the combine harvester, the feed rate refers to the mass of materials fed into the harvester per unit time, and its unit is kg / s. The loss rate, impurity rate and work efficiency of the sugarcane harvester harvesting process are closely related to the feed rate. If the feed rate is too large during the harvesting process, the working parts of the harvester's root cutting, transportation and impurity removal systems will be overloaded and blocked, and the harvesting loss rate and impurity rate will increase. At the same time, if the feed rate is too small, the harvester's operating efficiency will be reduced and the harvesting progress will be affected. At present, in the process of harvesting, the driver's experience is often used to estimate the working condition of the harvester and adjust the working state of the harvester, which is greatly affected by the driver's subjective factors, and the work intensity of long-term observation and adjustment is also high. Therefore, it is of great significance to guide the driver to complete the harvesting operation, improve the harvesting quality and work efficiency through accurate monitoring of the feed rate.

[0003] Research on feed amount monitoring methods at home and abroad mainly focuses on harvesting small-grain crops such as corn, wheat, and rice. Feed amount monitoring methods can be divided into pressure-based, torque-based, power-based, etc. For example, a feed amount measuring device based on the pressure difference of the feed drive hydraulic motor and the engine speed has been verified in wheat, corn, and soybeans. For example, a real-time feed amount measurement device measures the pressure of the crop on the inclined conveyor by installing a pressure sensor under the outlet of the inclined conveyor, thereby reflecting the change in feed amount. However, due to the complex terrain of the field operation environment, the tilt of the combine harvester body will cause a large measurement error of the pressure sensor. For example, by monitoring the torque and speed of the threshing drum power shaft, the power consumption of the cutting flow drum during threshing operation is calculated, and the relationship between threshing power consumption and feed amount is fitted to obtain a feed amount monitoring model.

[0004] For example, by installing a torque sensor and a speed sensor at the header screw conveyor, the screw conveyor power can be monitored, and the harvester feed amount can be calculated based on the screw conveyor power.

[0005] However, small-grain crops such as corn, wheat, and rice are small in size and relatively consistent in size after maturity, and the material transportation is relatively continuous and uniform, so it is relatively easy to conduct relevant inspections. When harvesting sugarcane, the sugarcane is longer and the diameter varies greatly, and the harvesting environment is more severe, requiring more powerful and complex monitoring technologies and systems. Most of the previous feed rate monitoring technologies use a single parameter to respond to changes in feed rate, and lack feed rate calculation models with multiple input parameters, making it difficult to adapt to the changing field harvesting environment. Summary of the invention

[0006] In view of the defects in the prior art, the present invention provides a sugarcane harvester feeding amount monitoring method using a particle swarm algorithm to optimize a BP neural network, the method comprising: Working data collection of key components of sugarcane harvester; Data filtering processing; Construct a sugarcane harvester feeding amount monitoring model based on particle swarm optimization and BP neural network; The data collection includes the oil outlet pressure of the root cutter hydraulic motor, the oil outlet pressure of the lower conveying roller hydraulic motor, the displacement of the upper conveying roller, and the flow rate of the segmenting roller hydraulic motor; The data filtering process is to reduce the noise of the data collected by the system using an iterative wavelet threshold noise reduction method.

[0007] Furthermore, the iterative wavelet threshold denoising method specifically includes: Get a noisy signal After that, the signal is subjected to wavelet transform, and the signal is represented as a linear combination of a series of wavelet functions of different scales and time shifts, where the coefficient of each item is called the wavelet coefficient; the wavelet transform formula is as follows, ; is the noise signal, Wavelet mother function, is the scale factor, is the translation factor, is the wavelet coefficient; Wavelet decomposition is used to divide wavelet coefficients into approximate components, i.e., low-frequency parts, and detail components, i.e., high-frequency parts. The low-frequency parts usually contain the main information of the signal, and the high-frequency parts are associated with noise and disturbance. According to the needs of analysis, the low-frequency parts are further decomposed in multiple layers. The expression after decomposition is: ; is the number of wavelet decomposition layers, Approximate coefficient, Detail coefficient, is the scale space orthonormal basis, is the standard orthogonal basis of the wavelet space; The high-frequency wavelet coefficients of each scale are subjected to threshold processing. The algorithm introduces a threshold factor. Before soft threshold processing, the detail coefficients are first subjected to an iterative deletion process to remove data edges and mutation points. The threshold factor calculation method is as follows: ; is the control coefficient, ; After threshold processing, the updated detail coefficient is obtained, and the noise reduction signal is obtained after wavelet reconstruction; the wavelet reconstruction calculation method is as follows: ; is the noise reduction signal, is a constant, is the updated wavelet coefficient.

[0008] Furthermore, the particle swarm construction The sugarcane harvester feeding amount monitoring model of algorithm-optimized BP neural network specifically includes: using particle swarm algorithm to optimize the weights and thresholds of BP neural network, training BP neural network, and monitoring the feeding amount according to the trained BP neural network model; the particle swarm algorithm-optimized BP neural network includes a particle swarm algorithm part and a BP neural network part.

[0009] Furthermore, the method of optimizing the BP neural network using the particle swarm algorithm specifically includes: Determine the topological structure of the neural network, including the number of input layer nodes, hidden layer nodes, and output layer nodes of the neural network; the number of hidden layer nodes is determined by an empirical formula method, and then the optimal number of nodes is determined by trial and error; the empirical formula is as follows: ; is the number of input layers, is the number of output layers, is a constant; Optimize the weights and thresholds of BP neural network through particle swarm algorithm; The weights and thresholds obtained by the particle swarm algorithm are applied to the neural network, and the data set is divided into a training set and a test set to train the BP neural network model; By using the trained particle swarm algorithm to optimize the BP neural network sugarcane harvester feeding amount monitoring model, the feeding amount can be monitored and the real-time feeding amount of the sugarcane harvester during operation can be output.

[0010] Further, the operation steps of the particle swarm algorithm to optimize the BP neural network are: setting algorithm parameters, initializing the speed and position of particles in the particle swarm, and randomly generating the initial speed and initial position of particles within an interval; that is, each initial position of each randomly generated particle is assigned to the weight and threshold in the first model; each particle in the particle swarm algorithm represents a potential solution, and the position of the particle corresponds to the set of weights and thresholds in the BP neural network; the speed of the particle determines the moving direction and distance of the particle in the search space, affecting the update of the weight and threshold; the optimal solution found by each particle in the search process by calculating the fitness value through the fitness function is recorded as the individual optimal position; in the entire particle swarm, the best solution among the individual optimal positions of all particles is selected as the global optimal position to guide the entire group to move towards the optimal solution; the mean square error between the actual output value and the expected value of the neural network is selected as the fitness function; the particles dynamically adjust their positions and speeds by tracking the individual optimal position and the global optimal position; the particle position and speed update formulas are respectively: ; ; For the The sequence number in the iteration is The speed of the individual, For the The sequence number in the iteration is The location of the individual; is the inertia weight; and is a constant with a value between 0 and 1; No. The sequence number in the iteration is The local optimum of an individual; The global optimum value of the kth iteration; and acceleration factor; inertia weight It plays the role of keeping the individual inertia of movement, enabling it to have the ability to expand the search outward from its own position and maintain the exploration of unknown space; by calculating the current particle fitness value and comparing it with the individual's historical best position and the global best position; updating the particle's speed and position through the nonlinear decreasing speed inertia weight method, when the maximum number of iterations is reached or the preset conditions are met, the optimization result is used as the initial weight and threshold of the neural network; the weight and threshold of the BP neural network are optimized through the particle swarm algorithm.

[0011] Furthermore, the nonlinear decreasing speed inertia weight method specifically includes using a larger , so that the algorithm has a strong search ability in the entire space, and a smaller , to accelerate convergence; the calculation method is as follows ; and are the current and maximum number of iterations; and is the inertia weight at the beginning and end of the iteration; is the weight reduction adjustment factor.

[0012] Furthermore, the acceleration factor is adjusted nonlinearly according to the acceleration factor of the algorithm iteration, specifically including: the acceleration coefficient is adjusted as follows ; ; and are the initial value and the end value of the acceleration factor respectively; Substituting equations (8)-(10) into equation 6, we can obtain the nonlinear adjustment strategy and the particle velocity update equation of the improved particle swarm algorithm as follows: ; The improved particle swarm algorithm was used to optimize the BP neural network method.

[0013] Furthermore, the data set is divided into a training set and a test set in a ratio of 8:2 to train the BP neural network model.

[0014] Furthermore, the method includes four input layers, namely, the oil outlet pressure of the root cutter hydraulic motor, the oil outlet pressure of the lower conveyor roller hydraulic motor, the upper conveyor roller displacement and the cutting roller hydraulic motor flow, each layer containing N data; the method includes one output layer, namely, the feed amount of the sugarcane harvester; the input amount is collected and denoised through the working data collection of the key components of the sugarcane harvester and the data filtering processing, and then the influence of the dimension on the model training is removed through standardization processing.

[0015] On the other hand, the present invention provides a sugarcane harvester feeding amount monitoring system using a particle swarm algorithm to optimize a BP neural network, the system comprising: Data collection module: using calibrated sensors to collect data in plots with different sugarcane growth conditions; the different sugarcane growth conditions include good sugarcane growth, poor sugarcane growth and severe sugarcane lodging; Data signal denoising module: used to build and run iterative wavelet threshold denoising method to denoise the data collected by the system; Neural network module: used to build and run the BP neural network model and monitor the feeding amount based on the collected data; Optimization module: used to build and run the particle swarm algorithm to optimize the initial weights and thresholds of the BP neural network training model; Test module: used to make predictions on test data.

[0016] Beneficial effects of the present invention: 1. The iterative wavelet threshold denoising method provided by the present invention reduces the noise of the data collected by the system, improves the validity of the data, improves the data quality, and reduces the problems of baseline drift and noise in the signal acquisition process.

[0017] 2. The present invention provides a method for nonlinearly decreasing the inertia weight of the particle swarm algorithm based on the number of population iterations. A larger inertia weight is used in the initial calculation. , so that the algorithm has a strong search ability in the entire space, and a smaller inertia weight is used in the later stage , to accelerate convergence.

[0018] 3. A particle swarm algorithm acceleration factor based on algorithm iteration provided by the present invention The nonlinear adjustment strategy realizes the idea of ​​"first big then small, first small then big", which makes it easier for the algorithm to seek the local optimum of each particle in the initial stage and converge to the global optimum at the end of the algorithm, so as to improve the convergence speed and accuracy.

[0019] 4. The present invention provides a sugarcane harvester feeding amount monitoring method and system based on an improved particle swarm algorithm optimized BP neural network, which realizes real-time monitoring of the feeding amount of the sugarcane harvester when it is working. It provides technical support for guiding the driver to adjust the working state of the harvester and for real-time adjustment of the working parameters of the harvester based on the feeding amount, thereby improving the harvesting quality and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The present invention has the following accompanying drawings: Figure 1 It is a flow chart of the iterative wavelet threshold denoising method of the present invention; Figure 2 This is the effect of the iterative wavelet threshold denoising method of the present invention; Figure 3 The particle swarm algorithm of the present invention optimizes the BP neural network process; Figure 4 It is the improved particle swarm algorithm to optimize the BP neural network test set verification effect. DETAILED DESCRIPTION

[0021] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0022] Method Example: The present invention provides a method for monitoring the feeding amount of a sugarcane harvester by improving the particle swarm algorithm to optimize the BP neural network. Step 1: Collect working data of key components of sugarcane harvester. The data include: oil outlet pressure of root cutter hydraulic motor, oil outlet pressure of lower conveyor roller hydraulic motor, displacement of upper conveyor roller, flow of hydraulic motor of cutting roller. In plots with different sugarcane growth conditions (good growth of sugarcane, poor growth of sugarcane, severe lodging of sugarcane), use calibrated sensors to collect data within a certain distance (for example, 100m, etc.), and carry out at least 3 straight-line harvests for each plot. The operating speed range is 0.5-1.5m / s.

[0023] Step 2: Data filtering. The present invention provides an iterative wavelet threshold denoising method to reduce the noise of the data collected by the system and improve the validity of the data. The iterative wavelet threshold denoising method process is as follows: Figure 1 The iterative wavelet threshold denoising method is as follows: (1) Obtaining a noisy signal After that, the signal is subjected to wavelet transform, and the signal is represented as a linear combination of a series of wavelet functions of different scales and time shifts, where the coefficient of each item is called the wavelet coefficient. The wavelet transform formula is as follows: ; is the noise signal, Wavelet mother function, is the scale factor, is the translation factor, is the wavelet coefficient.

[0024] (2) Wavelet decomposition can be used to divide wavelet coefficients into approximate components (low-frequency parts) and detail components (high-frequency parts). The low-frequency parts usually contain the main information of the signal, while the high-frequency parts are associated with noise and disturbances. According to the needs of analysis, the obtained low-frequency parts can be further decomposed into multiple layers. The expression after decomposition is: ; is the number of wavelet decomposition layers, Approximate coefficient, Detail coefficient, is the scale space orthonormal basis, is the standard orthogonal basis of the wavelet space.

[0025] (3) After that, it is necessary to perform threshold processing on the high-frequency wavelet coefficients of each scale. The traditional threshold processing method is to retain the wavelet coefficients greater than the threshold, and delete and return to zero the wavelet coefficients less than the threshold. This will cause the denoising effect to depend on the selection of the threshold. If the threshold is too small, the denoising effect will be unsatisfactory. If the threshold is too large, some effective signals will be eliminated. Therefore, the present invention proposes an iterative wavelet threshold method. The algorithm introduces a threshold factor. Before the soft threshold processing, the detail coefficients are first iteratively deleted to remove data edges and mutation points. The denoising algorithm of the present invention combines the multi-scale analysis capability of the wavelet transform, and can adaptively adjust the threshold and detail coefficient according to the local characteristics of the signal, which enables it to better process signals with non-uniform noise levels. The threshold factor calculation method is as follows, ; is the control coefficient, . (4) After threshold processing, the updated detail coefficients are obtained, and the noise reduction signal is obtained after wavelet reconstruction. The wavelet reconstruction calculation method is as follows.

[0026] ; is the noise reduction signal, is a constant, is the updated wavelet coefficient. Signal noise reduction effect Figure 2 shown.

[0027] Step 3: Construct a sugarcane harvester feeding amount monitoring model based on the BP neural network optimized by the improved particle swarm algorithm. Use the improved particle swarm algorithm to optimize the weights and thresholds of the BP neural network, train the BP neural network, and monitor the feeding amount based on the trained BP neural network model. The improved particle swarm algorithm to optimize the BP neural network consists of two parts: the improved particle swarm algorithm part and the BP neural network part. The method for constructing the improved particle swarm algorithm to optimize the BP neural network is as follows: (1) Determine the topological structure of the neural network, including the number of input layer nodes, hidden layer nodes, and output layer nodes of the network. The improved particle swarm algorithm optimized BP neural network sugarcane harvester feed monitoring method provided by the present invention has 4 input layers (root cutter hydraulic motor oil outlet pressure, lower conveyor roller hydraulic motor oil outlet pressure, upper conveyor roller displacement, cutting roller hydraulic motor flow), each layer contains N data. 1 output layer (sugarcane harvester feed amount). The input quantity is collected and denoised through steps 1 and 2, and then standardized to remove the influence of the dimension on model training. The number of hidden layer nodes is determined by the empirical formula method, and then the optimal number of nodes is determined by trial and error. The empirical formula is as follows: ; is the number of input layers, is the number of output layers, is a constant.

[0028] (2) Optimizing the weights and thresholds of the BP neural network by improving the particle swarm algorithm. When constructing a BP neural network, the corresponding weights and thresholds are randomly initialized according to the corresponding number of neurons in the input layer, hidden layer, and output layer. The initialization result directly affects the network convergence speed and model accuracy, causing the model training results to fall into a local minimum. Therefore, it is very important to choose a reasonable method to determine the initial weights and initial thresholds in the BP neural network. The specific process of optimizing the BP neural network by the improved particle swarm algorithm provided by the present invention is as follows: Figure 3 , the steps are as follows: 1. Set the algorithm parameters, initialize the speed and position of the particles in the particle swarm, and randomly generate the initial speed and initial position of the particles within a range. Each initial position of each randomly generated particle is assigned to the weight and threshold in the first model. Each particle in the particle swarm algorithm represents a potential solution, and the position of the particle corresponds to the set of weights and thresholds in the BP neural network. The speed of the particle determines the direction and distance of the particle's movement in the search space, which can affect the update of the weight and threshold. The optimal solution (i.e., the set corresponding to the optimal weight and threshold) found by each particle in the search process through the fitness function to calculate the fitness value is recorded as the individual optimal position. This is an important indicator for evaluating the performance of the particle swarm algorithm. In the entire particle swarm, the best solution among the individual optimal positions of all particles is selected as the global optimal position to guide the entire group to move towards the optimal solution. The mean square error between the actual output value and the expected value of the neural network is selected as the fitness function. The smaller the fitness value, the better the effect.

[0029] Particles dynamically adjust their positions and velocities by tracking their individual optimal positions and global optimal positions. The particle position and velocity update formulas are: ; ; For the The sequence number in the iteration is The speed of the individual, For the The sequence number in the iteration is The location of the individual; is the inertia weight; and is a constant with a value between 0 and 1; No. The sequence number in the iteration is The local optimum of an individual; No. The global optimal value of the iteration; and Acceleration factor.

[0030] Inertia Weight It plays a role in keeping the individual inertia of movement, enabling it to have the ability to expand outward from its own position and keep exploring the unknown space. Value. Larger The smaller the value, the more likely it is that the individual will be freed from the constraints of the local minimum, but the global optimal solution may be ignored. The value is conducive to approaching the optimal solution and making the algorithm converge quickly in the global scope, but it will also miss the opportunity to find a better solution. In general, particles should have a strong global optimization ability in the early stage of the particle optimization process, and can quickly and accurately determine the approximate range of feasible solutions, and then further determine the optimal solution through local optimization. Therefore, the present invention proposes a method based on the number of population iterations to The method of nonlinear decreasing adjustment is to use a larger , so that the algorithm has a strong search ability in the entire space, and a smaller , to accelerate convergence. The calculation method is as follows ; and are the current and maximum number of iterations; and is the inertia weight at the beginning and end of the iteration; is the weight reduction adjustment factor.

[0031] At the same time, traditional PSO balances local and global search capabilities by setting a fixed acceleration factor, which makes it difficult to adjust the search capabilities of particles in time when needed. Therefore, based on the influence mechanism of the acceleration factor on PSO, this paper proposes a nonlinear adjustment strategy of the acceleration factor according to algorithm iteration, and its acceleration coefficient is adjusted as follows ; ; and are the initial value and the end value of the acceleration factor respectively. This strategy realizes the First big then small, The idea of ​​"starting small and then growing" makes it easier for the algorithm to seek the local optimum of each particle in the initial stage and converge to the global optimum at the end of the algorithm, so as to improve the convergence speed and accuracy.

[0032] Substituting equations (8)-(10) into equation 6, we can obtain the nonlinear adjustment strategy and the particle velocity update equation of the improved particle swarm algorithm as follows.

[0033] ; So far, the improved particle swarm algorithm to optimize the BP neural network method can be obtained.

[0034] 2. Calculate the current particle fitness value and compare it with the individual historical best position and the global best position. Update the particle speed and position through the improved nonlinear decreasing speed inertia weight. When the maximum number of iterations is reached or the preset conditions are met, the optimization result is used as the initial weight and threshold of the neural network. So far, the weight and threshold of the BP neural network are optimized by the improved particle swarm algorithm.

[0035] (3) The weights and thresholds obtained by the improved particle swarm algorithm are applied to the neural network, and the data set is divided into a training set and a test set in a certain ratio (8:2) to train the BP neural network model.

[0036] During training, the expected value is calculated based on the weights and thresholds, and the error between the expected value and the true value is calculated. The error is compared with the set expected error.

[0037] When the error is greater than the expected error, the reverse error is calculated, the weights and thresholds are updated, and the expected value is recalculated to obtain the error between the expected value and the true value. The error is compared with the expected error.

[0038] When the error is greater than the expected error or reaches the maximum number of iterations, the model training is completed. The sugarcane harvester feeding amount monitoring model based on the improved particle swarm algorithm and BP neural network is obtained.

[0039] The test results of the test set are as follows Figure 4 shown.

[0040] (4) The sugarcane harvester feeding amount monitoring model is optimized by using the trained improved particle swarm algorithm based on the BP neural network to monitor the feeding amount and output the real-time feeding amount of the sugarcane harvester during operation.

[0041] System Example: The present invention provides a sugarcane harvester feeding amount monitoring system based on an improved particle swarm algorithm and optimized BP neural network, comprising the following modules.

[0042] Data acquisition module: In plots with different sugarcane growth conditions (good sugarcane growth, poor sugarcane growth, severe sugarcane lodging), calibrated sensors are used to collect data within a certain distance (for example, 100m, etc.).

[0043] Data signal denoising module: used to build and run an iterative wavelet threshold denoising method to denoise the data collected by the system and improve the validity of the data.

[0044] Neural network module: used to build and run the BP neural network model to monitor the feeding amount based on the collected data.

[0045] Optimization module: used to build and run the improved particle swarm algorithm to optimize the initial weights and thresholds of the BP neural network training model.

[0046] Test module: used to make predictions on test data.

[0047] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the present invention. Those skilled in the art should understand that the technical solutions implemented in the present invention can be modified or replaced by equivalents without departing from the design spirit and scope of the present invention, which should be included in the scope of the claims of the present invention.

[0048] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

Claims

1. A sugarcane harvester feeding amount monitoring method based on particle swarm algorithm optimized BP neural network, characterized in that: The method comprises: Working data collection of key components of sugarcane harvester; Data filtering processing; Construct a sugarcane harvester feeding amount monitoring model based on particle swarm optimization and BP neural network; The data collection includes the oil outlet pressure of the root cutter hydraulic motor, the oil outlet pressure of the lower conveying roller hydraulic motor, the displacement of the upper conveying roller, and the flow rate of the segmenting roller hydraulic motor; The data filtering process is to reduce the noise of the data collected by the system using an iterative wavelet threshold noise reduction method.

2. The monitoring method according to claim 1, characterized in that: The iterative wavelet threshold denoising method specifically comprises: Get a noisy signal After that, the signal is subjected to wavelet transform, and the signal is represented as a linear combination of a series of wavelet functions of different scales and time shifts, where the coefficient of each item is called the wavelet coefficient; the wavelet transform formula is as follows, ； is the noise signal, Wavelet mother function, is the scale factor, is the translation factor, is the wavelet coefficient; Wavelet decomposition is used to divide wavelet coefficients into approximate components, i.e., low-frequency parts, and detail components, i.e., high-frequency parts. The low-frequency parts usually contain the main information of the signal, and the high-frequency parts are associated with noise and disturbance. According to the needs of analysis, the low-frequency parts are further decomposed in multiple layers. The expression after decomposition is: ； is the number of wavelet decomposition layers, Approximate coefficient, Detail coefficient, is the scale space orthonormal basis, is the standard orthogonal basis of the wavelet space; The high-frequency wavelet coefficients of each scale are subjected to threshold processing. The algorithm introduces a threshold factor. Before soft threshold processing, the detail coefficients are first subjected to an iterative deletion process to remove data edges and mutation points. The threshold factor calculation method is as follows: ； is the control coefficient, ; After threshold processing, the updated detail coefficient is obtained, and the noise reduction signal is obtained after wavelet reconstruction; the wavelet reconstruction calculation method is as follows: ； is the noise reduction signal, is a constant, is the updated wavelet coefficient.

3. The monitoring method according to claim 2, characterized in that: The construction of a sugarcane harvester feeding amount monitoring model using a particle swarm algorithm to optimize a BP neural network specifically includes: using a particle swarm algorithm to optimize the weights and thresholds of the BP neural network, training the BP neural network, and monitoring the feeding amount according to the trained BP neural network model; the particle swarm algorithm to optimize the BP neural network includes a particle swarm algorithm part and a BP neural network part.

4. The monitoring method according to claim 3, characterized in that: The method for optimizing BP neural network by particle swarm algorithm specifically includes: Determine the topological structure of the neural network, including the number of input layer nodes, hidden layer nodes, and output layer nodes of the neural network; the number of hidden layer nodes is determined by an empirical formula method, and then the optimal number of nodes is determined by trial and error; the empirical formula is as follows: ； is the number of input layers, is the number of output layers, is a constant; Optimize the weights and thresholds of BP neural network through particle swarm algorithm; The weights and thresholds obtained by the particle swarm algorithm are applied to the neural network, and the data set is divided into a training set and a test set to train the BP neural network model; By using the trained particle swarm algorithm to optimize the BP neural network sugarcane harvester feeding amount monitoring model, the feeding amount can be monitored and the real-time feeding amount of the sugarcane harvester during operation can be output.

5. The monitoring method according to claim 4, characterized in that: The operation steps of the particle swarm algorithm for optimizing the BP neural network are as follows: setting algorithm parameters, initializing the speed and position of particles in the particle swarm, and randomly generating the initial speed and initial position of particles within an interval; that is, assigning each initial position of each randomly generated particle to the weight and threshold in the first model; each particle in the particle swarm algorithm represents a potential solution, and the position of the particle corresponds to the set of weights and thresholds in the BP neural network; the speed of the particle determines the moving direction and distance of the particle in the search space, affecting the update of the weight and threshold; the optimal solution found by each particle in the search process by calculating the fitness value through the fitness function is recorded as the individual optimal position; in the entire particle swarm, the best solution among the individual optimal positions of all particles is selected as the global optimal position to guide the entire group to move towards the optimal solution; the mean square error between the actual output value and the expected value of the neural network is selected as the fitness function; the particles dynamically adjust their positions and speeds by tracking the individual optimal position and the global optimal position; the particle position and speed update formulas are respectively: ； ； For the The sequence number in the iteration is The speed of an individual, For the The sequence number in the iteration is The location of the individual; is the inertia weight; and is a constant with a value between 0 and 1; No. The sequence number in the iteration is The local optimum of an individual; No. The global optimal value of the iteration; and acceleration factor; inertia weight It plays the role of keeping the individual inertia of movement, enabling it to have the ability to expand the search outward from its own position and maintain the exploration of unknown space; by calculating the current particle fitness value and comparing it with the individual's historical best position and the global best position; updating the particle's speed and position through the nonlinear decreasing speed inertia weight method, when the maximum number of iterations is reached or the preset conditions are met, the optimization result is used as the initial weight and threshold of the neural network; the weight and threshold of the BP neural network are optimized through the particle swarm algorithm.

6. The monitoring method according to claim 5, characterized in that: The nonlinear decreasing speed inertia weight method specifically includes using a larger , so that the algorithm has a strong search ability in the entire space, and a smaller , to accelerate convergence; the calculation method is as follows ； and are the current and maximum number of iterations; and is the inertia weight at the beginning and end of the iteration; is the weight reduction adjustment factor.

7. The monitoring method according to claim 6, characterized in that: The acceleration factor is adjusted nonlinearly according to the acceleration factor of the algorithm iteration, specifically including: the acceleration coefficient is adjusted as follows ； ； and are the initial value and the end value of the acceleration factor respectively; Substituting equations (8)-(10) into equation 6, we can obtain the nonlinear adjustment strategy and the particle velocity update equation of the improved particle swarm algorithm as follows: ； The improved particle swarm algorithm was used to optimize the BP neural network method.

8. The monitoring method according to claim 7, characterized in that: The data set is divided into a training set and a test set in a ratio of 8:2 to train the BP neural network model.

9. The monitoring method according to claim 7, characterized in that: The method comprises four input layers, namely, the oil outlet pressure of the root cutter hydraulic motor, the oil outlet pressure of the lower conveying roller hydraulic motor, the upper conveying roller displacement and the segmenting roller hydraulic motor flow, each layer containing N data; the method comprises one output layer, namely, the feed amount of the sugarcane harvester; the input amount is collected and denoised through the working data collection of the key components of the sugarcane harvester and the data filtering processing, and then the influence of the dimension on the model training is removed through the standardization processing.

10. A sugarcane harvester feeding amount monitoring system based on particle swarm algorithm optimized BP neural network, characterized in that: The system comprises: Data collection module: using calibrated sensors to collect data in plots with different sugarcane growth conditions; the different sugarcane growth conditions include good sugarcane growth, poor sugarcane growth and severe sugarcane lodging; Data signal denoising module: used to build and run iterative wavelet threshold denoising method to denoise the data collected by the system; Neural network module: used to build and run the BP neural network model and monitor the feeding amount based on the collected data; Optimization module: used to build and run the particle swarm algorithm to optimize the initial weights and thresholds of the BP neural network training model; Test module: used to make predictions on test data.

Images