Intelligent anti-blocking method and system for material conveying

Abstract

The application relates to the technical field of material conveying equipment, and discloses a material conveying intelligent anti-blocking method and system, which comprises the following steps: acquiring a motor current sequence and extracting a feature vector of a current blocking state; initializing a particle swarm and performing particle swarm iterative optimization; in each iteration: calculating swarm distribution entropy based on the current positions of all particles; generating a disturbance term when stagnation degree exceeds a stagnation degree threshold; calculating a blocking risk penalty term for each particle; calculating an inertia weight for each particle, wherein the inertia weight is obtained by subtracting the blocking risk penalty term from a basic weight after the basic weight is adjusted by the swarm distribution entropy, and then adding the disturbance term; updating the speed and position of each particle by using the inertia weight; and outputting the control parameters corresponding to the global optimal particle and adjusting the material conveying system after iteration termination. The application can not only reduce the influence of the defects that the standard particle swarm algorithm is prone to premature convergence and falls into local optimization, but also improve the stability of the whole system operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of material conveying equipment, and particularly relates to a material conveying intelligent anti-blocking method and system. BACKGROUND

[0002] In the process of material conveying, due to the fluctuation of material characteristics such as humidity, particle size, viscosity, uneven feeding rate or changes in the state of the conveying equipment, blockage often occurs. Material blockage not only interrupts production, but also can damage key equipment such as motors and conveyor belts due to a sudden increase in load, and even cause safety accidents.

[0003] The simple logic control anti-blocking method based on fixed threshold in the prior art, for example, by monitoring the motor current, when the current exceeds a certain set value, the feeding rate is reduced or the machine is stopped for processing. This anti-blocking method often leads to a too conservative compromise between safety and blockage in the efficiency of material conveying, and cannot optimize the performance of the conveying system.

[0004] Particle swarm optimization algorithm can be applied to intelligent anti-blocking of material conveying due to its simple concept, fast convergence speed and easy implementation. However, in the scene where the solution space is complex and there are multiple potential blockage risk areas, the population diversity of the particle swarm optimization algorithm will quickly lose with iteration, leading to premature convergence and failing to find the globally optimal combination of conveying parameters.

[0005] Moreover, the application of the particle swarm optimization algorithm in the prior art often separates the optimization process from the actual physical risk of the system, and fails to integrate real-time blockage risk into the iteration process of the algorithm, which may lead to the optimal solution found by the particle swarm optimization algorithm being very efficient in theory, but being on the critical edge of blockage in actual production, resulting in poor stability of system operation. SUMMARY

[0006] The present application provides a material conveying intelligent anti-blocking method and system to solve the problems of easy premature convergence into local optimal solution and poor stability of system operation in the prior art.

[0007] In a first aspect, the material conveying intelligent anti-blocking method of the present application comprises the following steps:

[0008] Obtain the real-time motor current sequence of the material conveying system, extract the energy of a specific frequency band to obtain the feature vector of the current blockage state; initialize the particle swarm, wherein each particle represents a set of control parameters including the target feeding rate and the target conveyor belt speed;

[0009] Particle swarm optimization is performed iteratively until the termination condition is met. In each iteration: based on the current position of all particles, the population distribution entropy representing the diversity of the particle swarm is calculated by kernel density estimation; the stagnation degree is obtained by counting the number of consecutive unupdated times of the global optimal solution. When the stagnation degree exceeds the stagnation degree threshold, a perturbation term is generated; otherwise, the perturbation term is zero; for each particle, based on the control parameters represented by the particle and the feature vector, a blocking risk penalty term is calculated by a risk assessment model.

[0010] The inertial weight is calculated for each particle, and the velocity and position of each particle are updated using the inertial weight. The calculation process of the inertial weight is as follows: the basic weight is adjusted by the population distribution entropy, then the blockage risk penalty term is subtracted, and then the perturbation term is added. The basic weight decreases with the number of iterations. When the population distribution entropy is lower than the preset entropy threshold, the calculated weight of the blockage risk penalty term is increased. After the iteration terminates, the control parameters corresponding to the globally optimal particle are output, and the material conveying system is adjusted.

[0011] Preferably, the step of acquiring the real-time motor current sequence of the material conveying system and extracting the energy of a specific frequency band to obtain the feature vector of the current blockage state includes: performing a fast Fourier transform on the acquired motor current sequence with a time length of T; calculating the first preset frequency band corresponding to normal operation. The energy E1 within, and the second preset frequency band corresponding to the initial blockage. , The energy E2 within the blockage is calculated; after normalizing the energy E1 and E2, V1 and V2 are obtained. V1 and V2 constitute the feature vector V=[V1, V2] of the current blockage state.

[0012] Preferably, the step of calculating the population distribution entropy representing particle swarm diversity through kernel density estimation includes: selecting a Gaussian kernel function as the kernel function for kernel density estimation; and calculating based on the current positions of all N particles. Calculate at any position probability density function at Where K is the Gaussian kernel function and h is the bandwidth. For the first The current position of each particle; based on the calculated probability density function. Using Shannon's entropy formula The population distribution entropy was calculated.

[0013] Preferably, the congestion risk penalty term is generated by a pre-trained multilayer perceptron risk assessment model. The input of the multilayer perceptron risk assessment model is a concatenated vector of the feature vector of the current congestion state and the control parameters represented by the particles, and the output is a congestion risk value in the range of [0, 1].

[0014] Preferably, the inertia weight is calculated for each particle, the inertia weight being obtained by subtracting the jam risk penalty term and adding the perturbation term from the base weight which decreases linearly with iteration number, after the population distribution entropy is adjusted, comprising: calculating the base weight which decreases linearly with iteration number t , wherein and are preset maximum and minimum base weights, is the maximum iteration number, and t is the current iteration number;

[0015] Logistic mapping is used when the stagnation degree exceeds the stagnation degree threshold the perturbation term is generated, wherein, when k is 0, is a random initial value in (0, 1), is the parameter of the Logistic mapping; z(k) is the iteration value of the Logistic mapping at the kth step, and z(k+1) is the iteration value of the Logistic mapping at the k+1th step; when the population distribution entropy H(t) is lower than the preset entropy threshold, the jam risk penalty term is multiplied by an amplification coefficient to obtain the adjusted jam risk penalty term , otherwise the adjusted jam risk penalty term is equal to ;

[0016] the inertia weight wherein, is the inertia weight at the current iteration number t, is the base weight which decreases linearly with iteration number t, is the value of the current population distribution entropy H(t) after normalization processing, is an entropy adjustment coefficient, is a risk penalty coefficient, is the adjusted jam risk penalty term, is the perturbation term.

[0017] Preferably, the velocity and position of each particle are updated using the inertia weight, comprising: updating the particle velocity according to the following formula,

[0018]

[0019] updating the particle position according to the following formula,

[0020]

[0021] wherein: is the velocity of the ith particle in the dth dimension at the t+1th iteration, and w(t) is the inertia weight, is the velocity of the i-th particle in the d-th dimension at the t-th iteration, c1 and c2 are learning factors, and r1 and r2 are random numbers uniformly distributed between 0 and 1, is the value of the individual optimal solution found by the i-th particle in the d-th dimension at the t-th iteration, is the position of the i-th particle in the d-th dimension at the t-th iteration, is the position of the i-th particle in the d-th dimension at the (t+1)-th iteration, is the value of the global optimal solution found by the entire particle swarm in the d-th dimension at the t-th iteration.

[0022] In a second aspect, the material conveying intelligent anti-blocking system comprises the following units:

[0023] A vector extraction unit acquires a real-time motor current sequence of the material conveying system, extracts energy in a specific frequency band to obtain a feature vector of the current blocking state, and initializes a particle swarm, wherein each particle represents a set of control parameters including a target feeding rate and a target conveying belt speed.

[0024] An iterative optimization unit performs particle swarm iterative optimization until a termination condition is met, and in each iteration: based on the current positions of all particles, a population distribution entropy representing the diversity of the particle swarm is calculated by kernel density estimation; the stagnation degree is obtained by counting the number of consecutive non-updates of the global optimal solution, and when the stagnation degree exceeds a stagnation degree threshold, a disturbance term is generated, otherwise the disturbance term is zero; for each particle, based on the control parameters represented by the particle and the feature vector, a blocking risk penalty term is calculated by a risk assessment model.

[0025] A weight calculation unit calculates an inertia weight for each particle and updates the velocity and position of each particle using the inertia weight; the calculation process of the inertia weight is: the base weight is adjusted by the population distribution entropy, then the blocking risk penalty term is subtracted, and then the disturbance term is added; the base weight decreases with the number of iterations, and when the population distribution entropy is lower than a preset entropy threshold, the calculation weight of the blocking risk penalty term increases; an output unit outputs the control parameters corresponding to the global optimal particle after iteration termination, and adjusts the material conveying system.

[0026] Preferably, the acquisition of the real-time motor current sequence of the material conveying system and the extraction of the feature vector of the current blocking state in the specific frequency band comprise: performing fast Fourier transform on the collected motor current sequence with a time length of T; calculating the energy E1 in a first preset frequency band corresponding to the normal operation, and the energy E2 in a second preset frequency band corresponding to the initial blocking, , ​​The energy E2 within the blockage is calculated; after normalizing the energy E1 and E2, V1 and V2 are obtained. V1 and V2 constitute the feature vector V=[V1, V2] of the current blockage state.

[0027] Preferably, the step of calculating the population distribution entropy representing particle swarm diversity through kernel density estimation includes: selecting a Gaussian kernel function as the kernel function for kernel density estimation; and calculating based on the current positions of all N particles. Calculate at any position probability density function at Where K is the Gaussian kernel function and h is the bandwidth. For the first The current position of each particle; based on the calculated probability density function. Using Shannon's entropy formula The population distribution entropy was calculated.

[0028] Preferably, the congestion risk penalty term is generated by a pre-trained multilayer perceptron risk assessment model. The input of the multilayer perceptron risk assessment model is a concatenated vector of the feature vector of the current congestion state and the control parameters represented by the particles, and the output is a congestion risk value in the range of [0, 1].

[0029] The beneficial effects of this invention are as follows: By integrating real-time characteristics of the congestion state, population distribution entropy (a measure of particle swarm diversity), and perturbation terms to address search stagnation into the construction process of inertia weights, this invention reduces the impact of the standard particle swarm optimization algorithm's tendency to premature convergence and get trapped in local optima. By applying congestion risk as a penalty to the particle iteration process, the search direction is guided to avoid high-risk conditions. The risk penalty is strengthened in the later stages of convergence, when population diversity is low, ensuring that the found material conveying control parameters not only optimize conveying efficiency but also improve the stability of the entire system. Attached Figure Description

[0030] Figure 1 A schematic flowchart of the intelligent anti-blocking method for material conveying provided in an embodiment of the present invention. Detailed Implementation

[0031] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0032] like Figure 1 As shown, an embodiment of the intelligent anti-blocking method for material conveying provided by the present invention includes the following steps:

[0033] S1, obtaining real-time motor current sequence of the material conveying system, extracting energy in a specific frequency band to obtain a feature vector of the current jam state; initializing a particle swarm, wherein each particle represents a set of control parameters including target feed rate and target conveyor speed.

[0034] Specifically, in the material conveying process, real-time current information of the motor of the material conveying system is collected, fast Fourier transform is performed on the current sequence, the total energy in a specific frequency band such as 0 Hz to 10 Hz is calculated, and the energy values of other frequency bands are combined to obtain a multi-dimensional feature vector. The change of the material flow state can be obtained through the feature vector. The size of the particle swarm is set, for example, containing fifty particles, and the search range of each control parameter is set, for example, the target feed rate is between 10 tons per hour and 50 tons per hour, and the target conveyor speed is between 1 meter per second and 2.5 meters per second. A set of initial control parameters is randomly generated as the initial position for each of the 50 particles in these ranges, and a random initial speed is assigned.

[0035] S2, performing particle swarm iteration optimization until the termination condition is met, in each iteration: based on the current position of all particles, the population distribution entropy representing the diversity of the particle swarm is calculated by kernel density estimation; the stagnation degree is obtained by counting the number of consecutive non-updated global optimal solutions, and when the stagnation degree exceeds the stagnation degree threshold, a disturbance term is generated, otherwise the disturbance term is zero; for each particle, based on the control parameters represented by the particle and the feature vector, the jam risk penalty term is calculated by the risk assessment model.

[0036] Specifically, the termination condition of iteration is set as reaching the maximum number of iterations, such as 200 times, or the global optimal solution is not improved for 30 consecutive iterations. For the position of all particles in the current particle swarm, the position of each particle is kernel density estimated by using Gaussian kernel function, and all Gaussian kernel functions are superimposed to form the probability density function of the entire particle swarm in the solution space. Based on the probability density function, the value of the population distribution entropy is calculated by the Shannon entropy formula, and the larger the value, the more dispersed the particle distribution and the better the diversity. A stagnation counter is set in advance, and the initial value is zero. After each iteration, the newly generated global optimal solution is compared with the global optimal solution of the last generation. If the newly generated global optimal solution is better, the stagnation counter is cleared; if the global optimal solution of the last generation is better, the stagnation counter is increased by 1. When the value of the stagnation counter exceeds the stagnation degree threshold, for example, the stagnation degree threshold is 20, a chaotic variable fluctuating between 0 and 1 is generated by a chaotic system such as Logistic mapping as a disturbance term; otherwise, the value of the disturbance term is zero.

[0037] Specifically, the neural network is used as the risk assessment model, and the target feed rate and the target conveyor speed represented by each particle, together with the feature vector of the current jam state extracted from the motor current, are used as inputs of the neural network. The risk assessment model outputs a value between 0 and 1, which is the predicted probability of jam under the control parameters, serving as the jam risk penalty term.

[0038] S3, calculate the inertia weight for each particle, which is obtained by subtracting the jam risk penalty term from the base weight that decreases with the iteration number, and then adding the disturbance term after adjusting the base weight by the population distribution entropy; when the population distribution entropy is lower than a preset entropy threshold, the calculation weight of the jam risk penalty term is increased; and the inertia weight is used to update the speed and position of each particle.

[0039] For example, assume that a base weight is set to decrease linearly from 0.9 to 0.4 with the iteration number. The calculated population distribution entropy is normalized to correct the base weight, and the weight is correspondingly reduced when the entropy value is large. It is determined whether the population distribution entropy is lower than a preset entropy threshold, such as 0.5. If yes, the jam risk penalty term calculated in the previous step is multiplied by an amplification coefficient, such as 1.5, and then the adjusted jam risk penalty term is subtracted from the corrected base weight. Then, the disturbance term calculated in the foregoing step is added to obtain the inertia weight.

[0040] For each particle, the new speed of the particle is obtained by adding three parts: the first part is the product of the old speed of the particle and the inertia weight; the second part is the difference between the historical optimal position of the particle and the current position, multiplied by a learning factor and a random number; and the third part is the difference between the global historical optimal position and the current position, multiplied by another learning factor and a random number. The new position of the particle is equal to the old position plus the newly calculated speed.

[0041] S4, after the iteration is terminated, the control parameters corresponding to the global optimal particle are output, and the material conveying system is adjusted.

[0042] When the iteration process meets the termination condition, the position corresponding to the optimal fitness value is found from the historical optimal positions experienced by all particles. The control parameters represented by this position are the optimal solution obtained by the present optimization. The set of parameters is sent to the programmable logic controller or the distributed control system of the material conveying system, so that the running parameters of the system are consistent with the optimal solution.

[0043] In an optional embodiment, the method of obtaining the real-time motor current sequence of the material conveying system and extracting the feature vector of the current jam state from the energy of a specific frequency band comprises: performing fast Fourier transform on the collected motor current sequence with a time length of T; calculating a first preset frequency band corresponding to the normal operation The energy E1 within, and the second preset frequency band corresponding to the initial blockage. , The energy E2 within the blockage is calculated; after normalizing the energy E1 and E2, V1 and V2 are obtained. V1 and V2 constitute the feature vector V=[V1, V2] of the current blockage state.

[0044] To assess the operating status of the material conveying system, this embodiment converts the time-domain motor current signal into a frequency-domain signal to analyze its energy distribution. For example, motor current data is continuously collected for 10 seconds at a sampling rate of 1000 Hz, resulting in a current sequence containing 10,000 data points. Performing a Fast Fourier Transform on the current sequence yields the energy intensity spectrum of the current signal at different frequencies, representing the vibration and load characteristics caused by different operating states. Based on the analysis of historical data, under normal conveying conditions, the motor load is stable, and the energy is mainly concentrated in a lower frequency range, such as the first preset frequency band set to 5 to 15 Hz. However, when materials begin to accumulate and show a tendency to blockage, the motor load fluctuation intensifies, generating abnormal energy peaks in another higher frequency range. In this case, the second preset frequency band can be set to 30 to 50 Hz. By integrating or summing the spectral data within these two frequency bands respectively, energy values ​​E1 and E2 can be quantified, reflecting the intensity of normal operation and blockage trends. To eliminate the differences in the dimensions and numerical range of the original energy values, energy values ​​E1 and E2 are further normalized. For example, by using the max-min normalization method, both values ​​are scaled to the range of 0 to 1 to obtain normalized values ​​V1 and V2. These two values ​​together constitute the feature vector V of the current congestion state.

[0045] In an optional embodiment, the calculation of the population distribution entropy representing particle swarm diversity through kernel density estimation includes: selecting a Gaussian kernel function as the kernel function for kernel density estimation; and calculating based on the current positions of all N particles. Calculate at any position probability density function at Where K is the Gaussian kernel function and h is the bandwidth. For the first The current position of each particle; based on the calculated probability density function. Using Shannon's entropy formula The population distribution entropy was calculated.

[0046] To calculate the dispersion of particle distribution, i.e., particle swarm diversity, in the particle swarm algorithm, this embodiment employs kernel density estimation. During calculation, it is assumed that the particle swarm size N is 100, each particle represents a set of control parameters, and its position xi is a vector. Bandwidth h is a key smoothing parameter, which can be set to, for example, 0.5. For any point in the search space... , calculate its distance to each particle , weight by Gaussian kernel function K, average the contribution of all 100 particles to get the probability density function of point . If particles are dense in a certain area, the probability density function value of this area is high; otherwise, if the particle distribution is sparse, the probability density function value is low. After obtaining the probability density function of the entire search space, the population distribution entropy H is calculated using the Shannon entropy formula. If the population distribution entropy H is high, it means that the particle distribution is extensive and the diversity is high, which is in the exploration stage. If the population distribution entropy H is low, it means that the particles have been highly concentrated in one or a few areas, and the diversity is low, tending to converge.

[0047] In an optional embodiment, the congestion risk penalty term is generated by a pre-trained multi-layer perception risk assessment model, the input of the multi-layer perception risk assessment model is the feature vector of the current congestion state and the splicing vector of the control parameters represented by the particle, and the output is the congestion risk value in the range of [0, 1].

[0048] The evaluation of congestion risk is realized by a data-driven deep learning model, i.e. a multi-layer perception risk assessment model. The multi-layer perception risk assessment model is trained using a large amount of historical data in the offline stage, which includes motor current features and corresponding control parameters under various working conditions, as well as labels of whether congestion occurs. Through training, the multi-layer perception risk assessment model learns the complex nonlinear mapping relationship from system state features and control strategies to congestion risk.

[0049] When online optimization, the multi-layer perception risk assessment model receives the input vector formed by splicing. For example, the feature vector V of the current congestion state extracted from the motor current is [0.2, 0.7], and the control parameters represented by a particle in the particle swarm are the conveying speed of 1.5 meters per second and the gate opening of 80%, i.e. [1.5, 0.8]. Then, the spliced input vector is [0.2, 0.7, 1.5, 0.8]. The spliced input vector is input into the multi-layer perception risk assessment model, and a series of weighted summations and nonlinear activation function calculations are performed on the input vector by multiple neuron layers to generate a single numerical value in the output layer, i.e. the congestion risk value . For example, output a value of 0.92, indicating that the possibility of congestion is very high under the current system state using the control parameters represented by the particle. The congestion risk value is used to punish the corresponding particle and guide the entire particle swarm to optimize to a safer operating area.

[0050] In an alternative embodiment, the inertia weight is calculated for each particle, the inertia weight being obtained by subtracting the jam risk penalty term from a base weight decreasing linearly with the iteration number, then adding the perturbation term, after the population distribution entropy adjustment, comprising: calculating a base weight decreasing linearly with the iteration number t wherein and are preset maximum and minimum base weights, is the maximum iteration number, and t is the current iteration number;

[0051] Logistic mapping is used when the stagnation degree exceeds the stagnation degree threshold The perturbation term is generated, wherein, when k is 0, is a random initial value within (0, 1), is a parameter of the Logistic mapping; z(k) is the iteration value of the Logistic mapping at the kth step, and z(k+1) is the iteration value of the Logistic mapping at the (k+1)th step; when the population distribution entropy H(t) is lower than a preset entropy threshold, the jam risk penalty term is multiplied by an amplification coefficient to obtain an adjusted jam risk penalty term otherwise, the adjusted jam risk penalty term is equal to ;

[0052] The inertia weight wherein, is the inertia weight at the current iteration number t, is a base weight decreasing linearly with the iteration number t, is a value of the current population distribution entropy H(t) after normalization processing, is an entropy adjustment coefficient, is a risk penalty coefficient, is the adjusted jam risk penalty term, is the perturbation term.

[0053] Exemplarily, the base weight is linearly decreased from a larger initial value such as 0.9, to a smaller value such as 0.4, with the increase of the iteration number t. In a total of equal to 200 iterations, the particle swarm has a stronger global exploration ability in the early stage, and then focuses more on detailed search near the existing optimal solution in the later stage. When falling into a local optimal solution, the perturbation term is activated , a small perturbation term is added to the inertia weight by Logistic mapping, which helps particles jump out of local optimal solution. Meanwhile, when the population distribution entropy H(t) is lower than the preset entropy threshold , it indicates that the particle swarm diversity is insufficient and there is a risk of premature convergence. At this time, the congestion risk penalty term is multiplied by an amplification factor α greater than 1, such as 1.8, to obtain the adjusted congestion risk penalty term , which punishes particles that are superior in performance but high-risk and lead to population aggregation.

[0054] In an optional embodiment, the updating of the speed and position of each particle using the inertia weight comprises: updating the particle speed according to the following formula,

[0055]

[0056] updating the particle position according to the formula , wherein: is the speed of the i-th particle in the d-th dimension at the t+1-th iteration, w(t) is the inertia weight, is the speed of the i-th particle in the d-th dimension at the t-th iteration, c1 and c2 are learning factors, r1 and r2 are random numbers uniformly distributed between 0 and 1, is the value of the individual optimal solution found by the i-th particle in the d-th dimension at the t-th iteration, is the position of the i-th particle in the d-th dimension at the t-th iteration, is the position of the i-th particle in the d-th dimension at the t+1-th iteration, is the value of the global optimal solution found by the entire particle swarm in the d-th dimension at the t-th iteration.

[0057] The updating of the motion trajectory of each particle is divided into two steps of speed updating and position updating, and the new speed of the particle is determined by weighting three parts, the first part is the speed of the particle itself at the last time multiplied by the inertia weight w(t), which represents the inertia of the particle motion; the second part is the difference between the current position of the particle and the optimal position it has experienced , which guides the particle to move towards the historical optimal position. The third part is the difference between the current position of the particle and the global optimal position found by the entire particle swarm so far , which makes the particle move towards the optimal position in the group. Random numbers r1 and r2 increase randomness in the search process, avoiding all particles flying towards the optimal solution in the same way.

[0058] The implementation principle of the material conveying intelligent anti-blocking method is as follows: the real-time characteristics of the blocking state, the population distribution entropy for measuring the diversity of the particle group and the disturbance term for coping with search stagnation are jointly integrated into the construction process of the inertia weight, so that the influence of the local optimal defect of the standard particle swarm algorithm prone to premature convergence is reduced. In addition, the blocking risk is used as a penalty term acting on the particle iteration process, which guides the search direction to avoid high-risk working conditions, and the risk penalty is increased in the late convergence stage, so that the final obtained material conveying control parameters not only realize the optimization of conveying efficiency, but also improve the stability of the whole system operation.

[0059] The embodiment of the material conveying intelligent anti-blocking system provided by the application comprises the following units:

[0060] A vector extraction unit acquires a real-time motor current sequence of the material conveying system, extracts energy of a specific frequency band to obtain a feature vector of the current blocking state, and initializes a particle swarm, wherein each particle represents a set of control parameters comprising a target feeding rate and a target conveying belt speed.

[0061] An iteration optimization unit performs particle swarm iteration optimization until a termination condition is met, and in each iteration: based on the current position of all particles, a population distribution entropy representing the diversity of the particle swarm is calculated through kernel density estimation; the stagnation degree is obtained by counting the number of consecutive non-updated global optimal solutions, and when the stagnation degree exceeds a stagnation degree threshold, a disturbance term is generated, otherwise the disturbance term is zero; for each particle, based on the control parameters represented by the particle and the feature vector, a blocking risk penalty term is calculated through a risk assessment model.

[0062] A weight calculation unit calculates an inertia weight for each particle, which is obtained by subtracting the blocking risk penalty term from the basic weight that decreases with the number of iterations after adjusting the population distribution entropy, and then adding the disturbance term; when the population distribution entropy is lower than a preset entropy threshold, the calculation weight of the blocking risk penalty term is increased; the speed and position of each particle are updated using the inertia weight.

[0063] An output unit outputs the control parameters corresponding to the global optimal particle after iteration termination, and adjusts the material conveying system.

[0064] Specifically, the acquisition of the real-time motor current sequence of the material conveying system and the extraction of the energy of the specific frequency band to obtain the feature vector of the current blocking state comprises: performing fast Fourier transform on the collected motor current sequence with a time length of T; calculating the energy E1 in a first preset frequency band corresponding to the normal operation, and the energy E2 in a second preset frequency band corresponding to the initial blocking. ] and the energy E2 in a second preset frequency band[ , The energy E2 within the blockage is calculated; after normalizing the energy E1 and E2, V1 and V2 are obtained. V1 and V2 constitute the feature vector V=[V1, V2] of the current blockage state.

[0065] The calculation of the population distribution entropy representing particle swarm diversity through kernel density estimation includes: selecting a Gaussian kernel function as the kernel function for kernel density estimation; and calculating based on the current positions of all N particles. Calculate at any position probability density function at Where K is the Gaussian kernel function and h is the bandwidth. For the first The current position of each particle; based on the calculated probability density function. Using Shannon's entropy formula The population distribution entropy was calculated.

[0066] The congestion risk penalty term is generated by a pre-trained multilayer perceptron risk assessment model. The input of the multilayer perceptron risk assessment model is the concatenated vector of the feature vector of the current congestion state and the control parameters represented by the particles, and the output is the congestion risk value in the range of [0, 1].

[0067] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A material conveying intelligent anti-blocking method, characterized in that, The process includes the following steps: acquiring the real-time motor current sequence of the material conveying system, extracting energy from a specific frequency band to obtain the feature vector of the current congestion state, including: performing a fast Fourier transform on the acquired motor current sequence with a time length of T; calculating the first preset frequency band corresponding to normal operation. The energy E1 within, and the second preset frequency band corresponding to the initial blockage. , The energy E2 within the blockage is calculated; after normalizing the energy E1 and E2, V1 and V2 are obtained. V1 and V2 constitute the feature vector V=[V1, V2] of the current blockage state; the particle swarm is initialized, where each particle represents a set of control parameters including the target feed rate and the target conveyor belt speed. performing particle swarm iterative optimization until a termination condition is met, in each iteration: calculating a population distribution entropy representing diversity of the particle swarm by kernel density estimation based on current positions of all particles; obtaining a stagnation degree by counting a continuous number of times that a global optimal solution is not updated, and generating a perturbation term when the stagnation degree exceeds a stagnation degree threshold, otherwise the perturbation term is zero; for each particle, calculating a congestion risk penalty term by a risk assessment model based on control parameters represented by the particle and the feature vector; calculating an inertia weight for each particle and updating a speed and a position of each particle by using the inertia weight; the inertia weight is calculated by adjusting a base weight by the population distribution entropy, then subtracting the congestion risk penalty term, and then adding the perturbation term; the base weight decreases with the number of iterations, and a calculation weight of the congestion risk penalty term increases when the population distribution entropy is lower than a preset entropy threshold; after the iteration is terminated, outputting control parameters corresponding to a global optimal particle, and adjusting the material conveying system.

2. The material conveying intelligent anti-blocking method of claim 1, wherein, the population distribution entropy representing diversity of the particle swarm is calculated by kernel density estimation, including: a Gaussian kernel function is selected as a kernel function of the kernel density estimation; According to the current positions of all N particles , the probability density function at any position is calculated , where K is a Gaussian kernel function, h is a bandwidth, is the current position of the th particle; based on the computed probability density function The population distribution entropy is calculated by the Shannon entropy formula .

3. The material conveying intelligent anti-blocking method of claim 1, wherein, the congestion risk penalty term is generated by a pre-trained multilayer perceptron risk assessment model, an input of the multilayer perceptron risk assessment model is a splicing vector of the current congestion state feature vector and the control parameters represented by the particle, and an output of the multilayer perceptron risk assessment model is the congestion risk penalty term in the range of [0, 1].

4. The material conveying intelligent anti-blocking method of claim 1, wherein, the inertia weight for each particle is calculated by adjusting a base weight that decreases with the number of iterations by the population distribution entropy, then subtracting the congestion risk penalty term, and then adding the perturbation term, including: computing basis weights that linearly decrease with iteration number t , wherein and are preset maximum and minimum base weights, is the maximum number of iterations, and t is the current iteration number. when the stagnation exceeds the stagnation threshold value, a Logistic map is adopted generating a perturbation term, wherein, when k is 0, is a random initial value within (0, 1), is a parameter of the Logistic map; z(k) is an iterative value of the Logistic map at the kth step, and z(k+1) is an iterative value of the Logistic map at the (k+1)th step. when the population distribution entropy H(t) is lower than a preset entropy threshold, multiplying the jam risk penalty term by an amplification coefficient to obtain an adjusted jam risk penalty term , otherwise the adjusted jam risk penalty term is equal to ; inertia weight wherein, is the inertia weight at the current iteration number t, is a base weight that linearly decreases with the iteration number t, is the value of the current population distribution entropy H(t) after normalization processing, is an entropy adjustment coefficient, is a risk penalty coefficient, is the adjusted congestion risk penalty term, is a perturbation term.

5. The material conveying intelligent anti-blocking method of claim 1, wherein, the speed and the position of each particle are updated by using the inertia weight, including: updating the particle speed according to the following formula, updating the particle position according to the following formula, where: is the velocity of the i-th particle in the d-th dimension at the t+1-th iteration, w(t) is the inertia weight, is the velocity of the i-th particle in the d-th dimension at the t-th iteration, c1 and c2 are learning factors, r1 and r2 are random numbers uniformly distributed between 0 and 1, is the value of the individual optimum found by the i-th particle in the d-th dimension at the t-th iteration, is the position of the i-th particle in the d-th dimension at the t-th iteration, is the position of the i-th particle in the d-th dimension at the t+1-th iteration, is the value of the global optimum found by the entire swarm in the d-th dimension at the t-th iteration.

6. A material conveying intelligent anti-blocking system, characterized in that, including the following units: The vector extraction unit acquires the real-time motor current sequence of the material conveying system, extracts the energy of a specific frequency band to obtain the feature vector of the current blockage state, including: performing a fast Fourier transform on the acquired motor current sequence with a time length of T; calculating the first preset frequency band corresponding to normal operation. The energy E1 within, and the second preset frequency band corresponding to the initial blockage. , The energy E2 within the blockage is calculated; after normalizing the energy E1 and E2, V1 and V2 are obtained. V1 and V2 constitute the feature vector V=[V1, V2] of the current blockage state; the particle swarm is initialized, where each particle represents a set of control parameters including the target feed rate and the target conveyor belt speed. an iterative optimization unit, performing particle swarm iterative optimization until a termination condition is met, in each iteration: calculating a population distribution entropy representing diversity of the particle swarm by kernel density estimation based on current positions of all particles; obtaining a stagnation degree by counting a continuous number of times that a global optimal solution is not updated, and generating a perturbation term when the stagnation degree exceeds a stagnation degree threshold, otherwise the perturbation term is zero; for each particle, calculating a congestion risk penalty term by a risk assessment model based on control parameters represented by the particle and the feature vector; a weight calculation unit, calculating an inertia weight for each particle and updating a speed and a position of each particle by using the inertia weight; the inertia weight is calculated by adjusting a base weight that decreases with the number of iterations by the population distribution entropy, then subtracting the congestion risk penalty term, and then adding the perturbation term; the base weight decreases with the number of iterations, and a calculation weight of the congestion risk penalty term increases when the population distribution entropy is lower than a preset entropy threshold; an output unit, after the iteration is terminated, outputting control parameters corresponding to a global optimal particle, and adjusting the material conveying system.

7. The material conveying intelligent anti-blocking system of claim 6, wherein, the population distribution entropy representing diversity of the particle swarm is calculated by kernel density estimation, including: a Gaussian kernel function is selected as a kernel function of the kernel density estimation; According to the current positions of all N particles , the probability density function at an arbitrary position is calculated where K is a Gaussian kernel function, h is a bandwidth, is the current position of the th particle; based on the computed probability density function The population distribution entropy is calculated by the Shannon entropy formula ​ 8. The material conveying intelligent anti-blocking system of claim 6, wherein, The jam risk penalty term is generated by a pre-trained multi-layer perception risk assessment model, an input of the multi-layer perception risk assessment model is a feature vector of the current jam state and a spliced vector of control parameters represented by the particle, and an output of the multi-layer perception risk assessment model is the jam risk penalty term in the range of [0, 1].

Images