A material weighing and conveying control method, device and big data system

Abstract

The present invention discloses a material weighing and conveying control method, device and big data system, wherein the material weighing and conveying control big data system comprises a material information collection terminal 1-n, a material information transmission terminal, a material monitoring terminal and a cloud server, wherein the material information collection terminal is used for real-time detection of material information; the material monitoring terminal is configured with a material weighing and conveying control model to achieve accurate weighing of materials, material conveying control and management and monitoring of material parameters; the cloud server is used for centralized storage and management of material information data; the material information transmission terminal is used for information interaction between the material information collection terminal, the material monitoring terminal and the cloud server. The present invention can improve the accuracy of material weight detection and ensure the accuracy of material weighing and conveying control.

Description

Technical Field

[0001] The invention relates to a material weighing and conveying control method, device and big data system. Background Art

[0002] The material conveying system generally includes a storage hopper and a weighing hopper arranged above and below. The material in the storage hopper falls into the weighing hopper through the storage door for weighing, and then is output to the material transport device through the discharge door at the outlet of the weighing hopper. The purpose of weighing is to accurately control the material conveying amount. At present, material weighing and conveying control mainly use weight sensors set on the weighing hopper to weigh materials, and control the opening and closing of the storage door and the discharge door based on the pre-set weight threshold. Due to the influence of factors such as the weight sensor's own accuracy and environmental interference, it is difficult to ensure the accuracy of material weight detection, and the threshold-based opening and closing control scheme is also difficult to ensure the accuracy of material weighing and conveying control. Summary of the invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a material weighing and conveying control method, device and big data system that can achieve accurate weighing.

[0004] Technical solution: The material weighing and conveying control method of the present invention comprises:

[0005] Obtain the output of position sensor 1, position sensor 2 and weight sensor. Position sensor 1 is used to measure the opening of the discharge door, position sensor 2 is used to measure the opening of the storage door, and the weight sensor is used to measure the weight of the material in the weighing bucket.

[0006] Construct material weighing and conveying control model:

[0007] The material weighing and conveying control model includes a material weighing parameter prediction model 1-3, a parameter controller 1-3, an Elman neural network model 1-2 of a Transformer neural network model-GWO, and a fuzzy recursive neural network model of a Transformer neural network model-GWO;

[0008] The error and error change rate between the given weight value of the material and the output of the material weighing parameter prediction model 1 are used as the input of the parameter controller 1, the given weight value of the material is used as the corresponding input of the fuzzy recursive neural network model of the Transformer neural network model-GWO, and the cumulative sum of the disturbance control amount of the material weight output by the parameter controller 1 and the fuzzy recursive neural network model of the Transformer neural network model-GWO is used as the corresponding input of the Elman neural network model 1 of the Transformer neural network model-GWO and the Elman neural network model 2 of the Transformer neural network model-GWO respectively;

[0009] The output of position sensor 2 is respectively used as the input of material weighing parameter prediction model 3, the corresponding input of Elman neural network model 1 of Transformer neural network model-GWO, and the corresponding input of fuzzy recursive neural network model of Transformer neural network model-GWO; the output of position sensor 1 is respectively used as the input of material weighing parameter prediction model 2, the corresponding input of Elman neural network model 2 of Transformer neural network model-GWO, and the corresponding input of fuzzy recursive neural network model of Transformer neural network model-GWO; the output of weight sensor is respectively used as the input of material weighing parameter prediction model 1, the corresponding input of Elman neural network model 1 of Transformer neural network model-GWO, the corresponding input of Elman neural network model 2 of Transformer neural network model-GWO, and the corresponding input of fuzzy recursive neural network model of Transformer neural network model-GWO;

[0010] The error and error change rate of the output of Elman neural network model 1 of Transformer neural network model-GWO and the output of material weighing parameter prediction model 3 are used as the input of parameter controller 3, and the cumulative sum of the output of parameter controller 3 and the disturbance control amount of the storage door output by the fuzzy recursive neural network model of Transformer neural network model-GWO is used as the control amount of the storage door; the error and error change rate of the output of Elman neural network model 2 of Transformer neural network model-GWO and the output of material weighing parameter prediction model 2 are used as the input of parameter controller 2, and the cumulative sum of the disturbance control amount of the unloading door output by the fuzzy recursive neural network model of Transformer neural network model-GWO is used as the control amount of the unloading door.

[0011] Further, the material weighing parameter prediction model includes a variational mode decomposition model, a Transformer neural network model-ARIMA model 1-2, a Transformer neural network model-GWO's Elman neural network model 1-2, a GWO's BiGRU neural network model-GWO's fuzzy recursive neural network model and a TDL beat delay;

[0012] The parameter sensor outputs are used as inputs to the variational mode decomposition model and the Transformer neural network model-ARIMA model 1 respectively. The variational mode decomposition model outputs parameter trend values, low-frequency components and high-frequency components. The parameter trend values ​​are used as inputs to the Transformer neural network model-ARIMA model 2, the low-frequency components are used as inputs to the Transformer neural network model-GWO's Elman neural network model 1, and the high-frequency components are used as inputs to the Transformer neural network model-GWO's Elman neural network model 2; Transformer neural network model-ARIMA model 1, Transformer neural network model-GWO The outputs of the network model-ARIMA model 2, Transformer neural network model-GWO's Elman neural network model 1, Transformer neural network model-GWO's Elman neural network model 2 and TDL beat delay device are respectively used as the corresponding inputs of GWO's BiGRU neural network model-GWO's fuzzy recursive neural network model, the output of GWO's BiGRU neural network model-GWO's fuzzy recursive neural network model is used as the input of TDL beat delay device, and GWO's BiGRU neural network model-GWO's fuzzy recursive neural network model outputs the parameter sensor output prediction value as the material weighing parameter prediction model output value.

[0013] Further, the parameter controller includes a Transformer neural network model-PID controller, a Transformer neural network model-GWO's BiGRU neural network model, a Transformer neural network model-GWO's Elman neural network model, a Transformer neural network model-GWO's fuzzy recursive neural network model, and a GWO's BiGRU neural network model-PID controller;

[0014] The error and error change rate between the given parameter target value and the parameter feedback value are taken as the input of the Transformer neural network model-PID controller, the output of the Transformer neural network model-PID controller and the output of the Elman neural network model of the Transformer neural network model-GWO are taken as the corresponding inputs of the Elman neural network model of the Transformer neural network model-GWO, the cumulative sum of the output of the Elman neural network model of the Transformer neural network model-GWO and the output of the Transformer neural network model-PID controller is taken as the input of the fuzzy recurrent neural network model of the Transformer neural network model-GWO, the parameter feedback value is taken as the input of the BiGRU neural network model of the Transformer neural network model-GWO, the difference and the change rate of the difference between the output of the fuzzy recurrent neural network model of the Transformer neural network model-GWO and the output of the BiGRU neural network model of the Transformer neural network model-GWO are taken as the input of the BiGRU neural network model-PID controller of GWO, and the output of the BiGRU neural network model-PID controller of GWO is taken as the output of the parameter controller.

[0015] The material weighing and conveying control device described in the present invention includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the material weighing and conveying control method described is implemented.

[0016] The material weighing and conveying control big data system described in the present invention includes material information collection terminals 1-n, material information transmission terminals, material monitoring terminals and cloud servers. The material information collection terminals are used to perform real-time detection of material information; the material monitoring terminals are configured with material weighing and conveying control models to achieve precise weighing of materials, material conveying control, and management and monitoring of material parameters; the cloud server is used to realize centralized storage and management of material information data; the material information transmission terminals are used to realize information interaction among the material information collection terminals, the material monitoring terminals and the cloud servers.

[0017] Furthermore, the material information collection terminal includes an STM32 single-chip microcomputer, a temperature sensor, a wind speed sensor, a position sensor, a weight sensor, a camera and a LoRa communication module. Each sensor is connected to the STM32 single-chip microcomputer through a corresponding signal conditioning circuit, and the camera and the LoRa communication module are respectively connected to the STM32 single-chip microcomputer.

[0018] Furthermore, the material transport device driving mechanism, the discharge door actuating mechanism and the storage door actuating mechanism are respectively connected to the STM32 single chip microcomputer.

[0019] Furthermore, the material information transmission end includes an STM32 microcontroller, and a LoRa communication module, an NB-IoT module and a USB interface respectively connected to the STM32 microcontroller; the NB-IoT module realizes two-way data interaction between the gateway and the cloud server; the USB interface is connected to the material monitoring end to realize information interaction between the material information transmission end and the material monitoring end; the LoRa communication module is used to realize information interaction between the material information collection terminal and the material information transmission end.

[0020] Furthermore, the material information collection terminal and the material information transmission terminal build a self-organizing communication network based on the LoRa communication module.

[0021] Furthermore, the material monitoring end adopts an industrial control computer, and the software functions of the material monitoring end include communication parameter setting, material information data analysis and data management, and material weighing and conveying control model.

[0022] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0023] (1) The present invention realizes accurate and reliable prediction of the output of the storage door, discharge door and material weighing position sensor and weight sensor in the material weighing process through the material weighing parameter prediction model, thereby improving the accuracy, rapidity and robustness of the material weighing process of the entire system; the parameter controller realizes accurate control of the storage door, discharge door and material given weight in the material weighing process, thereby improving the accuracy and reliability of the material weighing and transportation of the entire system.

[0024] (2) Transformer neural network model - GWO's Elman neural network model can provide more attention features when predicting input data. The input data is processed in parallel by the Transformer neural network model, and the dot product operation of the encoding layer is performed to generate attention to the input data. The Transformer neural network model's encoding layer multi-head attention mechanism and feedforward layer realize multi-attention prediction of the input data. After each multi-head attention mechanism and feedforward layer, a layer normalization operation is added. The input data processed by the layer normalization can reduce problems such as gradient explosion and improve the stability of the Transformer neural network model. The original data features are retained to a certain extent during the input data processing. Two linear addition layers are added to the encoding layer to achieve the effect of improving the model feature learning. Since the receiving layer in the Elman neural network model has a time delay characteristic, it can store the weights and thresholds output by the hidden layer and provide feedback. Compared with the BP neural network model and the radial basis neural network, the receiving layer structure can effectively increase the global stability of the Elman neural network. The multi-head attention and time delay characteristics in the Elman neural network model of Transformer neural network model - GWO improve the robustness, accuracy and reliability of the model's prediction.

[0025] (3) Transformer neural network model - GWO's BiGRU neural network model, a bidirectional propagation mechanism is added to the GRU neural network model to form a BiGRU neural network model. Compared with the GRU neural network model, the bidirectional gated recurrent unit of the BiGRU neural network model can connect the past and future information output by the parameter sensor, make full use of the time series information output by the parameter sensor, and use forward and backward to sequentially process the parameter feature sequence output by the parameter sensor, thereby improving the learning ability of the BiGRU neural network model. The multi-head attention mechanism of the Transformer neural network model and the bidirectional information collection of the GWO BiGRU neural network model improve the accuracy, robustness and reliability of the predicted input information.

[0026] (4) Transformer neural network model - GWO's fuzzy recursive neural network model. The fuzzy recursive neural network model has good structural knowledge expression fault tolerance and natural language processing capabilities. It has good adaptability to fuzzy systems with complex nonlinear relationships and unclear relationships. The fuzzy recursive neural network model will fuzzify the monitoring data when there is coupling between the input information and the state information of the monitored object cannot be directly quantified and judged. The multi-head attention mechanism of the Transformer neural network model and the fuzzy reasoning of the fuzzy recursive neural network model of GWO improve the accuracy, robustness and reliability of the predicted input information.

[0027] (5) Transformer neural network model-PID controller is the output of the Transformer neural network model as the PID controller K P , K I and K D The important basis of PID controller is proportional control, while integral control can effectively reduce the steady-state error of the controlled quantity. Differential control can promote the faster response of the inertial system of the controlled object. The multi-head attention mechanism of the Transformer neural network model combined with the PID controller improves the accuracy, rapid response and precision of the PID controller. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a structural schematic diagram of a material weighing and conveying control model in an embodiment of the present invention;

[0029] Figure 2 It is a structural schematic diagram of a material weighing parameter prediction model in an embodiment of the present invention;

[0030] Figure 3 is a schematic diagram of the structure of a parameter controller in an embodiment of the present invention;

[0031] Figure 4 Schematic diagram of the structure of the material weighing and conveying control big data system in an embodiment of the present invention;

[0032] Figure 5 It is a structural schematic diagram of a material information collection terminal in an embodiment of the present invention;

[0033] Figure 6 It is a structural schematic diagram of a material information transmission terminal in an embodiment of the present invention;

[0034] Figure 7 It is a functional diagram of the material monitoring end software in an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The present invention will be further described below in conjunction with the accompanying drawings.

[0036] The embodiment of the present invention provides a material weighing and conveying control method, which specifically includes the following steps:

[0037] (1) Obtain the output of position sensor 1, position sensor 2, and weight sensor, where position sensor 1 is used to measure the opening of the discharge door, position sensor 2 is used to measure the opening of the storage door, and the weight sensor is used to measure the weight of the material in the weighing hopper. The material in the storage hopper enters the weighing hopper through the storage door, and the material in the weighing hopper enters the material transport device through the discharge door, and is transported to the subsequent process section by the material transport device.

[0038] (2) Constructing material weighing and conveying control model

[0039] like Figure 1 As shown, the material weighing and conveying control model includes a material weighing parameter prediction model 1-3, a parameter controller 1-3, a Transformer neural network model-GWO's Elman neural network model 1-2 and a Transformer neural network model-GWO's fuzzy recursive neural network model.

[0040] The error and error change rate between the given weight value of the material and the output of the material weighing parameter prediction model 1 are used as the input of the parameter controller 1, the given weight value of the material is used as the corresponding input of the fuzzy recursive neural network model of the Transformer neural network model-GWO, and the cumulative sum of the disturbance control amount of the material weight output by the parameter controller 1 and the fuzzy recursive neural network model of the Transformer neural network model-GWO is used as the corresponding input of the Elman neural network model 1 of the Transformer neural network model-GWO and the Elman neural network model 2 of the Transformer neural network model-GWO respectively.

[0041] The output of position sensor 2 is respectively used as the input of material weighing parameter prediction model 3, the corresponding input of Elman neural network model 1 of Transformer neural network model-GWO, and the corresponding input of fuzzy recursive neural network model of Transformer neural network model-GWO; the output of position sensor 1 is respectively used as the input of material weighing parameter prediction model 2, the corresponding input of Elman neural network model 2 of Transformer neural network model-GWO, and the corresponding input of fuzzy recursive neural network model of Transformer neural network model-GWO; the output of weight sensor is respectively used as the input of material weighing parameter prediction model 1, the corresponding input of Elman neural network model 1 of Transformer neural network model-GWO, the corresponding input of Elman neural network model 2 of Transformer neural network model-GWO, and the corresponding input of fuzzy recursive neural network model of Transformer neural network model-GWO.

[0042] The error and error change rate of the output of Elman neural network model 1 of Transformer neural network model-GWO and the output of material weighing parameter prediction model 3 are used as the input of parameter controller 3, and the cumulative sum of the output of parameter controller 3 and the disturbance control amount of the storage door output by the fuzzy recursive neural network model of Transformer neural network model-GWO is used as the control amount of the storage door; the error and error change rate of the output of Elman neural network model 2 of Transformer neural network model-GWO and the output of material weighing parameter prediction model 2 are used as the input of parameter controller 2, and the cumulative sum of the disturbance control amount of the unloading door output by the fuzzy recursive neural network model of Transformer neural network model-GWO is used as the control amount of the unloading door.

[0043] like Figure 2 As shown, the material weighing parameter prediction model includes a variational mode decomposition model, a Transformer neural network model-ARIMA model 1-2, a Transformer neural network model-GWO's Elman neural network model 1-2, a GWO's BiGRU neural network model-GWO's fuzzy recursive neural network model and a TDL beat delay device.

[0044] The parameter sensor outputs are used as inputs to the variational mode decomposition model and the Transformer neural network model-ARIMA model 1 respectively. The variational mode decomposition model outputs parameter trend values, low-frequency components and high-frequency components. The parameter trend values ​​are used as inputs to the Transformer neural network model-ARIMA model 2, the low-frequency components are used as inputs to the Transformer neural network model-GWO's Elman neural network model 1, and the high-frequency components are used as inputs to the Transformer neural network model-GWO's Elman neural network model 2; Transformer neural network model-ARIMA model 1, Transformer neural network model-GWO The outputs of the network model-ARIMA model 2, Transformer neural network model-GWO's Elman neural network model 1, Transformer neural network model-GWO's Elman neural network model 2 and TDL beat delay device are respectively used as the corresponding inputs of GWO's BiGRU neural network model-GWO's fuzzy recursive neural network model, the output of GWO's BiGRU neural network model-GWO's fuzzy recursive neural network model is used as the input of TDL beat delay device, and GWO's BiGRU neural network model-GWO's fuzzy recursive neural network model outputs the parameter sensor output prediction value as the material weighing parameter prediction model output value.

[0045] like Figure 3As shown, the parameter controller includes a Transformer neural network model-PID controller, a Transformer neural network model-GWO's BiGRU neural network model, a Transformer neural network model-GWO's Elman neural network model, a Transformer neural network model-GWO's fuzzy recursive neural network model and a GWO's BiGRU neural network model-PID controller.

[0046] The error and error change rate between the given parameter target value and the parameter feedback value are taken as the input of the Transformer neural network model-PID controller, the output of the Transformer neural network model-PID controller and the output of the Elman neural network model of the Transformer neural network model-GWO are taken as the corresponding input of the Elman neural network model of the Transformer neural network model-GWO, and the cumulative sum of the output of the Elman neural network model of the Transformer neural network model-GWO and the output of the Transformer neural network model-PID controller is taken as the input of the fuzzy recursive neural network model of the Transformer neural network model-GWO; the parameter feedback value is taken as the input of the BiGRU neural network model of the Transformer neural network model-GWO, the difference and the change rate of the difference between the output of the fuzzy recursive neural network model of the Transformer neural network model-GWO and the output of the BiGRU neural network model of the Transformer neural network model-GWO are taken as the input of the BiGRU neural network model-PID controller of GWO, and the output of the BiGRU neural network model-PID controller of GWO is taken as the output of the parameter controller.

[0047] 1. Design of prediction model for material weighing parameters

[0048] The material weighing parameter prediction models include variational mode decomposition model, Transformer neural network model-ARIMA model, Transformer neural network model-GWO's Elman neural network model, TDL beat delay device and GWO's BiGRU neural network model-GWO's fuzzy recursive neural network model.

[0049] 1) Variational mode decomposition model design

[0050] The variational mode decomposition model (VMD) can obtain the changes in the output information of short-time parameter sensors on different time scales and reduce the non-stationarity of the time series parameters of the output information of short-time parameter sensors. The VMD model is an adaptive, non-recursive parameter sensor output information processing method that can decompose the unstable and nonlinear short-time parameter sensor output information into multiple relatively stable subsequences. The variational mode decomposition model (VMD) is a decomposition method for processing non-stationary parameter sensor output information, which can solve the problem of modal aliasing. After the parameter sensor output information is decomposed by the VMD model, K VIMF components are obtained, and the frequency modulation-amplitude modulation signal of each component is shown in formula (1):

[0051]

[0052] Among them, u k (t) is the kth VIMF component obtained by decomposition, k = 1, 2, ..., K; A k (t) is the instantaneous amplitude, is the phase; t is the time.

[0053] The specific steps of VMD model decomposition are as follows:

[0054] Step 1: Calculate the relevant analytical signal and construct the spectrum through Hilbert transform; Step 2: Estimate the respective center frequencies and transfer the modal spectrum to the baseband; Step 3: Estimate the bandwidth through the gradient norm. After the parameter sensor output information is decomposed by the VMD model, a series of VIMF components can be obtained, and the parameter sensor output information is decomposed by VMD model to perform VMD denoising.

[0055] 2) Transformer neural network model-ARIMA model design

[0056] The Transformer neural network model-ARIMA model is a series connection of the Transformer neural network model and the ARIMA model.

[0057] The Transformer neural network model is a model that is completely based on the self-attention mechanism. Its original model consists of two parts: Encoder and Decoder. The position encoding is generated by adding the homogeneous matrix generated by the encoding function to the original data. For data at different positions, historical data with different contents and lengths are used. Therefore, it carries unique position information, which enables the Transformer neural network model to have the ability to distinguish the order of input data. The Transformer neural network model structure consists of multi-head self-attention layers, layer normalization, feedforward neural network layers, position encoding, and residual connections. The most important part is the self-attention mechanism, which is also called the internal attention mechanism. It can dynamically generate weights for different connections. The multi-head self-attention mechanism used in the Transformer neural network model has the following calculation process:

[0058] MSA(Q,K,V)=δ(h 0 ,h 1 ,…,h n )W 0 (2)

[0059] Among them, MSA represents multi-head self-attention calculation, Q, K, V represent query, key, and value respectively, δ represents concatenation operation, and h i represents the i-th self-attention head, n represents the number of self-attention heads, and W is the weight.

[0060] The multi-head self-attention mechanism can connect different positions in the sequence to calculate the long-term relationship representation at different time points, and capture the correlation coefficients of the data in multiple dimensions, so that the Transformer neural network model has stronger global feature extraction and modeling capabilities. The output of the multi-head attention mechanism is layer-normalized, and then passes through a feedforward neural network layer consisting of two fully connected layers and a ReLU activation function, and then undergoes a layer normalization process, thus realizing the output of the Encoder of a Transformer neural network model.

[0061] The full name of the ARIMA model is the Autoregressive Integrated Moving Average Model, also known as the Autoregressive Integrated Moving Average Model (moving can also be called sliding). It is a common statistical model for time series prediction. The basic idea of ​​the ARIMA model is to treat the data sequence with time characteristic attributes as a random sequence, use this random sequence as a sample to reasonably sample, use it as the training data of the model to train and fit the model, and then use this model to predict future values.

[0062] The process of establishing an ARIMA model usually includes: Step 1: Obtain the original time series data. Step 2: Data preprocessing, including filling in missing values, data type conversion, checking and processing of outliers, etc. Step 3: Data desharpening sampling. Step 4: Use the drawing function to draw the sample time series data, and observe the drawing to determine whether the data is stable. If the data is non-stationary data, it is necessary to perform a d-order difference operation to obtain the d value. Step 5: For the stable sample time series data, apply the autocorrelation coefficient ACF and the partial autocorrelation coefficient PACF respectively, and obtain the autoregressive coefficient p and the number of sliding average items q by comparison. Step 6: Use the three key parameters p, d, and q to establish the ARIMA model. Step 7: Fit the created ARIMA model. Step 8: Test the ARIMA model.

[0063] 3) Transformer neural network model - GWO's Elman neural network model design

[0064] The Transformer neural network model-GWO's Elman neural network model is a series connection of the Transformer neural network model and the GWO's Elman neural network model. The Transformer neural network model refers to the relevant design method in the embodiment of the present invention.

[0065] The Elman neural network model of GWO is the Elman neural network model optimized by GWO. The GWO algorithm is a heuristic optimization method that simulates the exploration and hunting cooperation behavior of gray wolf groups to optimize the objective function and solve multiple parameter optimization problems. In the social hierarchy of gray wolves, the first level is the leader of the gray wolf population, called α wolf. α wolf is the candidate for the optimal solution in the group. This individual represents the individual with the highest fitness in the group, that is, the individual with the smallest objective function value; the second level is β wolf in the gray wolf population. β wolf is the candidate for the second optimal solution in the group. This individual represents the individual with the second highest fitness in the group, ranking after α wolf; followed by δ wolf as the candidate for the third optimal solution in the group, ranking after α and β wolf. The worst solution candidate wolf group ω at the bottom, led by the α, β, and δ wolf groups, enters the favorable area of ​​the prey search area to complete the hunting and predation process, and updates its position according to the position of the α, β, and δ wolf individuals. ω wolf enters the favorable area of ​​the prey search area towards the three wolf individuals with the best fitness, α, β, and δ, to find a better solution. During the hunting stage of the gray wolf group, according to the cooperation and competition behavior among the gray wolves, the three wolf individuals α, β, and δ are initialized as the current optimal solution, and the prey position is judged according to the current optimal solution. At the same time, the other gray wolf groups are guided to update their positions according to the current optimal solution gray wolf position, so as to gradually approach the prey and complete the hunting process.

[0066] The Elman neural network model consists of an input layer, an output layer, a hidden layer, and a receiving layer. The receiving layer in the Elman neural network model has a time delay characteristic, which can store the weights and thresholds of the hidden layer output and provide feedback. Compared with the BP neural network model and the radial basis neural network, the receiving layer structure can effectively increase the global stability of the Elman neural network. The input-output function relationship between the internal layers of the Elman neural network model can be expressed as:

[0067]

[0068] Among them, y(k) is the output of the neural network, w is the inter-layer connection weight of the neural network, x(k) is the output of the hidden layer, and x c (k) is the input from the hidden layer to the receiving layer, k is the number of iterations; u(k-1) is the input of the neural network, f 1 (·) and f 2 (·) is a nonlinear function.

[0069] The steps of using GWO to optimize the weights and thresholds of the Elman neural network model are as follows:

[0070] Step 1: Algorithm parameter settings, such as wolf pack size N, maximum number of iterations t max , the weights and thresholds of the Elman neural network model. Step 2: Initialize the gray wolf position vector. Step 3: Calculate the individual fitness value. Step 4: Save the current optimal individual parameters in the position vector. Step 5: Update the optimization control factor and the position vector of the candidate wolf. Step 6: Determine the iterative termination condition. If t<t max , then repeat steps 3-5; otherwise, stop the iteration. Step 7: Output the weights and thresholds of the Elman neural network model and substitute them into the Elman neural network model.

[0071] 4) BiGRU neural network model of GWO - Fuzzy recurrent neural network model design of GWO

[0072] The BiGRU neural network model of GWO-the fuzzy recursive neural network model of GWO is a series connection of the BiGRU neural network model of GWO and the fuzzy recursive neural network model of GWO.

[0073] The BiGRU neural network model of GWO is the BiGRU neural network model optimized by GWO. The GRU neural network model is a special type of RNN. The GRU neural network model more completely describes the long-term and short-term temporal correlation of the parameter time series output by the parameter sensor. The GRU neural network model is mainly composed of a reset gate and an update gate. The reset gate determines the degree to which the parameter sensor output information is forgotten, and the update gate determines how much parameter sensor output information is passed from the previous hidden state to the current hidden layer. The number of hidden neurons n and the initial learning rate r of the GRU neural network model are important parameters in the GRU neural network model, which have a vital impact on the construction of the GRU neural network model and its prediction accuracy. At present, the parameters of the GRU neural network model are generally set manually, which is very random and it is difficult to achieve the best prediction effect of the parameter sensor output value. In order to make the learning ability of the GRU neural network model more powerful, the present invention adds a bidirectional propagation mechanism to the GRU neural network model to form a BiGRU neural network model. Compared with the GRU neural network model, the bidirectional gated recurrent unit of the BiGRU neural network model can connect the past and future information output by the parameter sensor, make full use of the time series information output by the parameter sensor, and use forward and backward to sequentially process the parameter feature sequence output by the parameter sensor, thereby improving the learning ability of the BiGRU neural network model.

[0074] The process of GWO optimizing the BiGRU neural network model refers to the design method of GWO optimizing the Elman neural network model in the embodiment of the present invention.

[0075] The fuzzy recursive neural network model has good structural knowledge expression fault tolerance and natural language processing capabilities, and has good adaptability to fuzzy systems with complex nonlinear relationships that are unclear. The fuzzy recursive neural network model includes an input layer, an output layer, and one or more hidden layers. The training process of the fuzzy recursive neural network model is to continuously correct the network weights and thresholds so that the error between the actual output and the expected output gradually decreases and reaches the expected error. The fuzzy recursive neural network model requires fuzzification of the monitoring data when there is a coupling correlation between the input information and the state information of the monitored object cannot be directly quantified and judged. The membership vector is obtained through fuzzy membership function processing, and then the obtained vector is input into the hidden layer to obtain a fuzzy output vector, and the final input information fusion value is obtained through defuzzification processing. The first layer is the input layer, which is connected to the input signal; the second layer is the fuzzification layer, which is used to fuzzify the input signal according to the membership function, with local feedback connection, and the output of this layer is used as part of the input of the nodes of this layer; the third layer is the hidden layer, with local feedback connection, and the output of this layer is used as part of the input of the nodes of this layer, which is used to realize the mapping from the fuzzification layer to the fuzzy output layer, and the number of its nodes can be finally determined according to the number of input data; the fourth layer is the fuzzy output layer, with local feedback connection, and the output of this layer is used as part of the input of the nodes of this layer, and its output fuzzy value represents the possible fusion value of multiple input values; the fifth layer is the clarification layer, which obtains the fusion value of the input information through membership defuzzification processing.

[0076] The process of GWO optimizing the fuzzy recursive neural network model refers to the design method of GWO optimizing the Elman neural network model in the embodiment of the present invention.

[0077] 2. Parameter Controller Design

[0078] The parameter controllers include Transformer neural network model-PID controller, Transformer neural network model-GWO's BiGRU neural network model, Transformer neural network model-GWO's Elman neural network model, Transformer neural network model-GWO's fuzzy recursive neural network model and GWO's BiGRU neural network model-PID controller.

[0079] The Transformer neural network model-GWO's BiGRU neural network model is a Transformer neural network model and a BiGRU neural network model of GWO connected in series, the Transformer neural network model-GWO's fuzzy recursive neural network model is a Transformer neural network model and a fuzzy recursive neural network model of GWO connected in series, and the design methods of the Transformer neural network model, the fuzzy recursive neural network model of GWO and the Elman neural network model of GWO refer to the relevant design methods of the relevant models in the embodiments of the present invention.

[0080] 1) Transformer neural network model-PID controller design

[0081] Transformer neural network model-PID controller is the K of the Transformer neural network model output as the PID controller P , K I and K D The important basis of PID controller is proportional control, while integral control can effectively reduce the steady-state error of the control quantity, but it is very likely to cause an increase in overshoot. Differential control can promote the response speed of the inertial system of the controlled object and effectively reduce overshoot. The relationship between the input and output of the PID controller is:

[0082]

[0083] Among them, u(t) represents output; e(t) represents input; K P Represents the proportionality coefficient; K I Represents the integral coefficient; K D represents the differential coefficient.

[0084] 2) BiGRU neural network model of GWO-PID controller design

[0085] GWO's BiGRU neural network model - PID controller is the output of GWO's BiGRU neural network model as the K of the PID controller P , K I and K D Value, PID controller refers to the design method of Transformer neural network model-PID controller in the embodiment of the present invention.

[0086] 3. Material weighing and conveying control model design

[0087] The material weighing and conveying control model includes a parameter controller, a material weighing parameter prediction model, a Transformer neural network model-GWO's Elman neural network model and a Transformer neural network model-GWO's fuzzy recursive neural network model. Their design methods refer to the design methods and steps of the relevant models in the embodiments of the present invention.

[0088] The embodiment of the present invention further provides a material weighing and conveying control device, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the material weighing and conveying control method described in the embodiment of the present invention is implemented. The material weighing and conveying control device has all the advantages of the material weighing and conveying control method.

[0089] like Figure 4 As shown, an embodiment of the present invention also provides a material weighing and conveying control big data system, including material information collection terminals 1-n, material information transmission terminals, material monitoring terminals and cloud servers, wherein the material information collection terminals are used to perform real-time detection of material information; the material monitoring terminal is configured with a material weighing and conveying control model to achieve precise weighing of materials, material conveying control, and management and monitoring of material parameters; the cloud server is used to realize centralized storage and management of material information data; the material information transmission terminal is used to realize information interaction among the material information collection terminals, the material monitoring terminal and the cloud server.

[0090] like Figure 5 As shown in the figure, the material information collection terminal is used as a material information and environmental parameter perception terminal, including an STM32 single-chip microcomputer, a temperature sensor, a wind speed sensor, a position sensor, a weight sensor, a camera and a LoRa communication module. Each sensor is connected to the STM32 single-chip microcomputer through a corresponding signal conditioning circuit, and the camera and the LoRa communication module are connected to the STM32 single-chip microcomputer respectively. The material transport device driving mechanism, the unloading door action mechanism and the storage door action mechanism are connected to the STM32 single-chip microcomputer respectively. The software of the material information collection terminal mainly realizes LoRa communication, environmental parameters, material information collection, material weighing and material transportation. The software is designed in C language with high compatibility, which greatly improves the work efficiency of software design and development and enhances the reliability, readability and portability of the program code.

[0091] like Figure 6As shown, the material information transmission end includes an STM32 microcontroller, and a LoRa communication module, an NB-IoT module and a USB interface respectively connected to the STM32 microcontroller; the NB-IoT module realizes two-way data interaction between the gateway and the cloud server; the USB interface is connected to the material monitoring end to realize information interaction between the material information transmission end and the material monitoring end; the material information collection terminal and the material information transmission end build a self-organizing communication network based on the LoRa communication module to realize information interaction between the material information collection terminal and the material information transmission end.

[0092] The material monitoring end uses industrial control computers, such as Figure 7 As shown in the figure, the functions of the material monitoring software include communication parameter setting, material information data analysis and data management, material weighing and conveying control model. The software uses Microsoft Visual++ 6.0 as the development tool.

Claims

1. A material weighing and conveying control method, characterized in that: include: Obtain the output of position sensor 1, position sensor 2 and weight sensor. Position sensor 1 is used to measure the opening of the discharge door, position sensor 2 is used to measure the opening of the storage door, and the weight sensor is used to measure the weight of the material in the weighing bucket. Construct material weighing and conveying control model: The material weighing and conveying control model includes a material weighing parameter prediction model 1-3, a parameter controller 1-3, an Elman neural network model 1-2 of a Transformer neural network model-GWO, and a fuzzy recursive neural network model of a Transformer neural network model-GWO; The error and error change rate between the given weight value of the material and the output of the material weighing parameter prediction model 1 are used as the input of the parameter controller 1, the given weight value of the material is used as the corresponding input of the fuzzy recursive neural network model of the Transformer neural network model-GWO, and the cumulative sum of the disturbance control amount of the material weight output by the parameter controller 1 and the fuzzy recursive neural network model of the Transformer neural network model-GWO is used as the corresponding input of the Elman neural network model 1 of the Transformer neural network model-GWO and the Elman neural network model 2 of the Transformer neural network model-GWO respectively; The output of position sensor 2 is respectively used as the input of material weighing parameter prediction model 3, the corresponding input of Elman neural network model 1 of Transformer neural network model-GWO, and the corresponding input of fuzzy recursive neural network model of Transformer neural network model-GWO; the output of position sensor 1 is respectively used as the input of material weighing parameter prediction model 2, the corresponding input of Elman neural network model 2 of Transformer neural network model-GWO, and the corresponding input of fuzzy recursive neural network model of Transformer neural network model-GWO; the output of weight sensor is respectively used as the input of material weighing parameter prediction model 1, the corresponding input of Elman neural network model 1 of Transformer neural network model-GWO, the corresponding input of Elman neural network model 2 of Transformer neural network model-GWO, and the corresponding input of fuzzy recursive neural network model of Transformer neural network model-GWO; The error and error change rate of the output of Elman neural network model 1 of Transformer neural network model-GWO and the output of material weighing parameter prediction model 3 are used as the input of parameter controller 3, and the cumulative sum of the output of parameter controller 3 and the disturbance control amount of the storage door output by the fuzzy recursive neural network model of Transformer neural network model-GWO is used as the control amount of the storage door; the error and error change rate of the output of Elman neural network model 2 of Transformer neural network model-GWO and the output of material weighing parameter prediction model 2 are used as the input of parameter controller 2, and the cumulative sum of the disturbance control amount of the unloading door output by the fuzzy recursive neural network model of Transformer neural network model-GWO is used as the control amount of the unloading door.

2. The material weighing and conveying control method according to claim 1, characterized in that: The material weighing parameter prediction models include variational mode decomposition model, Transformer neural network model-ARIMA model 1-2, Transformer neural network model-GWO's Elman neural network model 1-2, GWO's BiGRU neural network model-GWO's fuzzy recursive neural network model and TDL beat delay; The parameter sensor outputs are used as inputs to the variational mode decomposition model and the Transformer neural network model-ARIMA model 1 respectively. The variational mode decomposition model outputs parameter trend values, low-frequency components and high-frequency components. The parameter trend values ​​are used as inputs to the Transformer neural network model-ARIMA model 2, the low-frequency components are used as inputs to the Transformer neural network model-GWO's Elman neural network model 1, and the high-frequency components are used as inputs to the Transformer neural network model-GWO's Elman neural network model 2; Transformer neural network model-ARIMA model 1, Transformer neural network model-GWO The outputs of the network model-ARIMA model 2, Transformer neural network model-GWO's Elman neural network model 1, Transformer neural network model-GWO's Elman neural network model 2 and TDL beat delay device are respectively used as the corresponding inputs of GWO's BiGRU neural network model-GWO's fuzzy recursive neural network model, the output of GWO's BiGRU neural network model-GWO's fuzzy recursive neural network model is used as the input of TDL beat delay device, and GWO's BiGRU neural network model-GWO's fuzzy recursive neural network model outputs the parameter sensor output prediction value as the material weighing parameter prediction model output value.

3. The material weighing and conveying control method according to claim 1, characterized in that: The parameter controllers include Transformer neural network model-PID controller, Transformer neural network model-GWO's BiGRU neural network model, Transformer neural network model-GWO's Elman neural network model, Transformer neural network model-GWO's fuzzy recursive neural network model and GWO's BiGRU neural network model-PID controller; The error and error change rate between the given parameter target value and the parameter feedback value are taken as the input of the Transformer neural network model-PID controller, the output of the Transformer neural network model-PID controller and the output of the Elman neural network model of the Transformer neural network model-GWO are taken as the corresponding inputs of the Elman neural network model of the Transformer neural network model-GWO, the cumulative sum of the output of the Elman neural network model of the Transformer neural network model-GWO and the output of the Transformer neural network model-PID controller is taken as the input of the fuzzy recurrent neural network model of the Transformer neural network model-GWO, the parameter feedback value is taken as the input of the BiGRU neural network model of the Transformer neural network model-GWO, the difference and the change rate of the difference between the output of the fuzzy recurrent neural network model of the Transformer neural network model-GWO and the output of the BiGRU neural network model of the Transformer neural network model-GWO are taken as the input of the BiGRU neural network model-PID controller of GWO, and the output of the BiGRU neural network model-PID controller of GWO is taken as the output of the parameter controller.

4. A material weighing and conveying control device, comprising a memory and a processor, wherein a computer program is stored in the memory, characterized in that: When the computer program is executed by a processor, the material weighing and conveying control method according to any one of claims 1 to 3 is implemented.

5. A material weighing and conveying control big data system, used to implement the material weighing and conveying control method according to any one of claims 1 to 3, characterized in that: It includes material information collection terminals 1-n, material information transmission terminals, material monitoring terminals and cloud servers. The material information collection terminals are used to detect material information in real time. The material monitoring terminals are equipped with material weighing and conveying control models to achieve precise weighing of materials, material conveying control, and management and monitoring of material parameters. The cloud server is used to realize centralized storage and management of material information data. The material information transmission terminal is used to realize information interaction among the material information collection terminals, material monitoring terminals and cloud servers.

6. The material weighing and conveying control big data system according to claim 5 is characterized in that: The material information collection terminal includes an STM32 microcontroller, a temperature sensor, a wind speed sensor, a position sensor, a weight sensor, a camera and a LoRa communication module. Each sensor is connected to the STM32 microcontroller through a corresponding signal conditioning circuit, and the camera and the LoRa communication module are respectively connected to the STM32 microcontroller.

7. The material weighing and conveying control big data system according to claim 6 is characterized in that: The driving mechanism of the material transport device, the action mechanism of the unloading door and the action mechanism of the material storage door are respectively connected to the STM32 single chip microcomputer.

8. The material weighing and conveying control big data system according to claim 6, characterized in that: The material information transmission end includes an STM32 microcontroller, and a LoRa communication module, an NB-IoT module and a USB interface respectively connected to the STM32 microcontroller; the NB-IoT module realizes two-way data interaction between the gateway and the cloud server; the USB interface is connected to the material monitoring end to realize information interaction between the material information transmission end and the material monitoring end; the LoRa communication module is used to realize information interaction between the material information collection terminal and the material information transmission end.

9. The material weighing and conveying control big data system according to claim 8, characterized in that: The material information collection terminal and the material information transmission terminal build a self-organizing communication network based on the LoRa communication module.

10. The material weighing and conveying control big data system according to claim 5, characterized in that: The material monitoring end uses an industrial control computer, and the software functions of the material monitoring end include communication parameter settings, material information data analysis and data management, and material weighing and conveying control models.

Images