CN107506862B - Online real-time grinding particle size prediction system and method based on Internet of things - Google Patents

Abstract

The invention relates to an online real-time prediction system and method of grinding particle size based on the Internet of Things. The system includes a data acquisition unit, a wireless data transmission unit, a data management unit, an MES data reading unit, a grinding particle size prediction unit and a BP neural network model The evaluation unit; the grinding particle size prediction unit is used to use the historical value of the grinding parameter as the input of the BP neural network to train the BP neural network model, and input the real-time value of the grinding parameter to the trained BP neural network model for grinding Granular predictions. The prediction method of the present invention collects the grinding parameters in real time through the Internet of Things, and optimizes the production process; the online real-time prediction of the grinding particle size is performed through the BP neural network model, and the real-time prediction of the grinding particle size is improved without the establishment of a mechanism model. sex and precision.

Description

An online real-time prediction system and method for grinding particle size based on the Internet of Things

technical field

The present invention relates to the field of Internet of Things and grinding technology, in particular to an online real-time prediction method and system for grinding particle size based on Internet of Things.

Background technique

Grinding and grading operation is a crucial link in the process of beneficiation. The purpose of grinding is to crush large-grained ore to a certain extent, so that useful minerals and gangue minerals are separated into a single dissociated state, so as to facilitate the separation of useful minerals. Sorting. Grinding particle size is not only the most important production index of grinding operation, but also a key factor affecting the concentrate grade and metal recovery rate of subsequent sorting operations. Accurate and real-time acquisition of particle size information is the key to controlling the grinding process and improving grinding efficiency and product quality. Therefore, it is of great practical significance to realize the detection of particle size.

In the actual industrial production process, the ore dressing plant uses manual off-line testing to obtain the grinding particle size data, but it cannot meet the real-time requirements of control. Therefore, it is of great significance to realize the online detection of particle size. At present, the patented method for on-line detection of particle size mainly includes "201310349946.4 (an online prediction system and method for ore grinding particle size)" to conduct online real-time prediction of ore grinding particle size by solving the mechanism model for predicting ore grinding particle size. The mechanism model for predicting grinding particle size is complex, with many influencing factors, and the influence of each factor changes in real time, so it is difficult to accurately establish a mechanism model for grinding particle size prediction.

Contents of the invention

The embodiment of the present invention provides an online real-time prediction system and method of grinding particle size based on the Internet of Things, which can improve the real-time performance and accuracy of grinding particle size prediction without establishing a mechanism model

The present invention provides an online real-time prediction system for grinding particle size based on the Internet of Things, including:

The data acquisition unit is used to simulate and generate the real-time values of the grinding parameters and the real-time values of the grinding particle size, and perform data processing on the real-time values of the grinding parameters to obtain standard voltage data;

The wireless data transmission unit is used to send the standard voltage data to the data management unit;

The data management unit is used to receive the standard voltage data, restore the standard voltage data to the real-time value of the grinding parameter and store it;

The MES data reading unit is used to read the real-time value of the grinding parameter stored in the data management unit, and store the historical value of the grinding parameter;

The grinding particle size prediction unit is used to train the BP neural network model with the historical value of the grinding parameter as the input of the BP neural network, and input the real-time value of the grinding parameter into the trained BP neural network model for grinding particle size prediction. predict;

The BP neural network model evaluation unit is used to compare the real-time value of the predicted grinding particle size with the real-time value of the simulated grinding particle size to evaluate the BP neural network model.

In the online real-time prediction system of grinding particle size based on the Internet of Things of the present invention, the data acquisition unit includes:

The real-time data generation module is used to establish a grinding virtual object model to simulate the industrial site data collection situation, and generate a set of real-time values of grinding parameters at intervals of 5S;

The data processing module is used to standardize the real-time values of the generated grinding parameters and convert them into standard voltage data of [0,5V].

In the online real-time prediction system of grinding particle size based on the Internet of Things of the present invention, the data management unit includes:

The data display module is used to convert the received standard voltage data into a standard voltage signal label to display;

The data conversion module is used to restore the received standard voltage quantity to the real-time value of the grinding parameter;

The data storage module is used to store the real-time values of the grinding parameters.

In the online real-time prediction system of grinding particle size based on the Internet of Things of the present invention, the grinding particle size prediction unit includes:

Initialization module, for setting up the initial model based on BP neural network including input layer, hidden layer and output layer;

The parameter setting module is used for setting the parameters of the initial model;

The learning module uses the historical value of the grinding parameter as the training set of the initial model of the BP neural network to train the initial model of the BP neural network to obtain the trained BP neural network model;

The grinding particle size prediction module uses the real-time value of the grinding parameters as the test set of the trained BP neural network model, and the output is the predicted value of the grinding particle size.

The present invention also provides a method for online real-time prediction of grinding particle size based on the Internet of Things, comprising the following steps:

Step 1: Simulate and generate real-time values of grinding parameters and grinding particle size, and perform data processing on the real-time values of grinding parameters to obtain standard voltage data;

Step 2: Send the standard voltage data to the data management unit;

Step 3: Receive the standard voltage data, restore the standard voltage data to the real-time value of the grinding parameters and store it;

Step 4: read the real-time value of the grinding parameter stored in the data management unit, and store the historical value of the grinding parameter;

Step 5: use the historical value of the grinding parameter as the input of the BP neural network to train the BP neural network model, and input the real-time value of the grinding parameter into the trained BP neural network model to predict the grinding particle size;

Step 6: Compare the real-time value of the predicted grinding particle size with the real-time value of the simulated grinding particle size to evaluate the BP neural network model.

In the online real-time prediction method of grinding particle size based on the Internet of Things of the present invention, said step 1 includes:

Step 1.1: Establish a grinding virtual object model to simulate industrial field data collection, and generate a set of real-time values of grinding parameters at intervals of 5 seconds;

Step 1.2: The data processing module standardizes the real-time values of the generated grinding parameters and converts them into standard voltage data of [0,5V].

In the online real-time prediction method of grinding particle size based on the Internet of Things of the present invention, said step 3 includes:

Step 3.1: Convert the received standard voltage data into a standard voltage signal label and display it;

Step 3.2: Restore the received standard voltage quantity to the real-time value of the grinding parameter;

Step 3.3: Store real-time values of grinding parameters.

In the online real-time prediction method of grinding particle size based on the Internet of Things of the present invention, said step 5 includes:

Step 5.1: Establish an initial model based on BP neural network including input layer, hidden layer and output layer;

Step 5.2: Set parameters for the initial model;

Step 5.3: use the historical value of the grinding parameter as the training set of the initial model of the BP neural network to train the initial model of the BP neural network, constantly adjust the weights and thresholds of each layer, and obtain the trained BP neural network model;

Step 5.4: The real-time value of the grinding parameters is used as the test set of the trained BP neural network model, and the output is the predicted value of the grinding particle size.

In the online real-time prediction method of grinding particle size based on the Internet of Things of the present invention, said step 5.1 includes:

Step 5.1.1: Set the number of nodes in the input layer to 4, and the nodes correspond to 4 grinding parameters, including: ball mill feed volume, ball mill inlet water feed volume, pump pool supplementary water volume, and cyclone feed concentration;

Step 5.1.2: Set the number of hidden layers to 2, and the number of hidden layer neurons to 30 and 20 respectively.

In the online real-time prediction method of grinding particle size based on the Internet of Things of the present invention, said step 5.2 includes:

Step 5.2.1: Set hidden layer weight w _ij , hidden layer threshold θ _i , output layer weight w _ki and output layer threshold a _k ;

Step 5.2.2: Set the minimum expected target error to 1e-5, and the learning rate to 0.05;

其中，n为4个s维的输入列向量。Step 5.2.3: The activation function of the hidden layer and the output layer of the BP neural network is selected as an S-type logarithmic function, namely

Among them, n is 4 s-dimensional input column vectors.

The present invention proposes an online real-time prediction system and method for grinding particle size based on the Internet of Things. The grinding parameters are collected in real time through the Internet of Things, and the production process is optimized; the online real-time prediction of the grinding particle size is carried out through the BP neural network model, and the real-time and accuracy of the grinding particle size prediction are improved without the establishment of a mechanism model.

Description of drawings

Fig. 1 is a kind of block diagram of the online real-time prediction system of grinding granularity based on Internet of Things of the present invention;

Fig. 2 is a flow chart of an online real-time prediction method for grinding particle size based on the Internet of Things of the present invention.

Detailed ways

An online real-time prediction system and method for grinding particle size based on the Internet of Things of the present invention will be described in detail below in conjunction with the accompanying drawings.

The invention performs online real-time prediction on the grinding particle size based on the Internet of Things and BP neural network. The BP neural network can learn and store a large number of input-output pattern mapping relationships without revealing the mathematical equations describing the mapping relationship in advance. Mathematical theory has proved that it has the function of realizing any complex nonlinear mapping, which makes it particularly suitable for solving problems with complex internal mechanisms. Its learning rule is to use the steepest descent method to continuously adjust the weights and thresholds of the network through backpropagation to minimize the sum of squared errors of the network.

As shown in Figure 1, an online real-time prediction system for grinding particle size based on the Internet of Things of the present invention includes: data acquisition unit 1, wireless data transmission unit 2, data management unit 3, MES data reading unit 4, grinding particle size prediction Unit 5 and BP neural network model evaluation unit 6.

The data acquisition unit 1 is used for simulating and generating real-time values of grinding parameters and grinding particle size, and performing data processing on the real-time values of grinding parameters to obtain standard voltage data. The wireless data transmission unit 2 is used to send the standard voltage data to the data management unit. The data management unit 3 is used to receive the standard voltage data, restore the standard voltage data to the real-time value of the grinding parameter and then store it. The MES data reading unit 4 is used to read the real-time value of the grinding parameter stored by the data management unit, and store the historical value of the grinding parameter. Grinding particle size prediction unit 5 is used to train the BP neural network model with the historical value of the grinding parameter as the input of the BP neural network, and inputs the real-time value of the grinding parameter into the trained BP neural network model to determine the grinding particle size. predict. The BP neural network model evaluation unit 6 is used to compare the real-time value of the predicted grinding particle size with the simulated real-time value of the grinding particle size to evaluate the BP neural network model.

The data acquisition unit 1 includes: a real-time data generation module 11 and a data processing module 12 . The real-time data generation module 11 is used to establish a grinding virtual object model to simulate the industrial field data collection situation, and generate a set of real-time values of grinding parameters at intervals of 5S. Grinding parameters include: ball mill feeding amount, ball mill inlet water feeding amount, pump pool replenishment water amount and cyclone feeding concentration. Grinding parameters are stored in the oracle local database. The data processing module 12 is used to standardize the acquired real-time values of the grinding parameters and convert them into standard voltage data of [0,5V].

The data management unit 3 includes: a data display module 31 , a data conversion module 32 and a data storage module 33 . The data display module 31 is used to display the standard voltage signal label converted from the received standard voltage data, so as to confirm the successful reception of the wirelessly transmitted data. The data conversion module 32 is used to restore the received standard voltage quantity to the real-time value of the grinding parameter. The data storage module 33 is used for storing real-time values of grinding parameters. During specific implementation, the data management unit 3 is an industrial PC.

The grinding particle size prediction unit 5 includes: an initialization module 51 , a parameter setting module 52 , a learning module 53 and a grinding particle size prediction module 54 . The initialization module 51 is used to establish an initial model based on a BP neural network including an input layer, a hidden layer and an output layer. The parameter setting module 52 is used for setting parameters of the initial model. The learning module 53 uses the historical value of the grinding parameter as the training set of the initial model of the BP neural network to train the initial model of the BP neural network to obtain a trained BP neural network model. The grinding particle size prediction module 54 takes the real-time value of the grinding parameter as the test set of the trained BP neural network model, and outputs the predicted value of the grinding particle size.

As shown in Figure 2, it is a kind of online real-time prediction method of grinding granularity based on the Internet of Things of the present invention, comprising the steps:

Step 1: Simulate and generate real-time values of grinding parameters and grinding particle size, and perform data processing on the real-time values of grinding parameters to obtain standard voltage data;

Step 2: Send the standard voltage data to the data management unit;

Step 3: Receive the standard voltage data, restore the standard voltage data to the real-time value of the grinding parameters and store it;

Step 4: read the real-time value of the grinding parameter stored in the data management unit, and store the historical value of the grinding parameter;

Step 5: use the historical value of the grinding parameter as the input of the BP neural network to train the BP neural network model, and input the real-time value of the grinding parameter into the trained BP neural network model to predict the grinding particle size;

Step 6: Compare the real-time value of the predicted grinding particle size with the real-time value of the simulated grinding particle size to evaluate the BP neural network model.

Step 1 includes:

Step 1.1: Establish a grinding virtual object model to simulate industrial field data collection, and generate a set of real-time values of grinding parameters at intervals of 5 seconds.

During the specific implementation, the virtual object model of grinding is established by using matlab. Grinding parameters include: ball mill feeding amount, ball mill inlet water feeding amount, pump pool replenishment water amount and cyclone feeding concentration. Grinding parameters are stored in the oracle local database.

Step 1.2: The data processing module standardizes the real-time values of the generated grinding parameters, converts them into standard voltage data of [0,5V], and uploads the standard voltage data to the wireless node.

Step 2: Send the standard voltage data to the data management unit, specifically:

Step 2.1: The wireless node uploads the standard voltage data to the smart wireless gateway through wireless transmission of data;

Step 2.2: After receiving the standard voltage data, the intelligent wireless gateway synchronizes the real-time data to the industrial PC through the LAN.

Step 3 includes:

Step 3.1: The industrial control PC converts the received standard voltage data into a standard voltage signal label and displays it to confirm that the wirelessly transmitted data is successfully received;

Step 3.2: Restore the received standard voltage data of [0,5V] to the real-time value of the grinding parameters;

Step 3.3: Store the real-time values of the grinding parameters in the oracle database of the industrial PC.

Step 4: read the real-time value of the grinding parameter stored in the industrial control PC, and store the historical value of the grinding parameter, specifically: use MES (manufacturing execution system) to read the real-time value of the grinding parameter on the industrial control PC, include:

Step 4.1: Connect the PC where the MES system is located with the industrial control PC through the LAN segment setting;

Step 4.2: Configure ODBC (data source manager) on the PC where the MES system is located, so that the MES system can read the data of the oracle database of the industrial PC, including: real-time values of grinding parameters and historical values of grinding parameters.

Step 5 includes:

Step 5.1: set up the initial model based on BP neural network including input layer, hidden layer and output layer, step 5.1 further includes:

Step 5.1.1: Set the number of nodes in the input layer to 4, and the nodes correspond to 4 grinding parameters, including: ball mill feed volume, ball mill inlet water feed volume, pump pool supplementary water volume, and cyclone feed concentration;

Step 5.1.2: Set the number of hidden layers to 2, and the number of hidden layer neurons to 30 and 20 respectively.

Step 5.2: Perform parameter setting on the initial model, further including:

Step 5.2.1: Set hidden layer weight w _ij , hidden layer threshold θ _i , output layer weight w _ki , and output layer threshold a _k ; where j represents the jth input sample, i represents the ith intermediate hidden contains layer nodes, k represents the kth output layer node;

Step 5.2.2: Set the minimum expected target error ε to 1e-5, and the learning rate to 0.05;

其中，n为4个s维的输入列向量；Step 5.2.3: The activation function of the hidden layer and the output layer of the BP neural network is selected as an S-type logarithmic function, namely

Among them, n is 4 s-dimensional input column vectors;

Step 5.3: use the historical value of the grinding parameter as the training set of the initial model of the BP neural network to train the initial model of the BP neural network, constantly adjust the weights and thresholds of each layer, and obtain the trained BP neural network model; further include:

Step 5.3.1: Obtain the historical values of the grinding parameters as the training set of the initial model of the BP neural network, including: ball mill feed volume, ball mill inlet feed water volume, pump pool supplementary water volume, and cyclone feed concentration historical values ;

Step 5.3.2: Normalize the training data of the training set;

计算隐含层第i个节点的输入，通过公式

计算隐含层第i个节点的输出；M为输入样本维度；Step 5.3.3: By formula

Calculate the input of the i-th node in the hidden layer, through the formula

Calculate the output of the i-th node in the hidden layer; M is the input sample dimension;

获取输出层第k个节点的输入，通过公式ψ(netk)计算输出层第k个节点的输出；q代表作用在第k个输出层节点上的输入维度(即为隐含层维度)；Step 5.3.4: By formula

Obtain the input of the kth node of the output layer, and calculate the output of the kth node of the output layer by the formula ψ(netk); q represents the input dimension (that is, the hidden layer dimension) acting on the kth output layer node;

计算P个训练样本的输出层误差；L为输出层样本维度；Step 5.3.5: By the formula

Calculate the output layer error of P training samples; L is the sample dimension of the output layer;

Step 5.3.6: Calculate the hidden layer error by referring to the method in step 5.3.5;

Step 5.3.7: Correct the output layer weight w _ki , the output layer threshold a _k , the hidden layer weight w _ij and the hidden layer threshold θ _i sequentially according to the error gradient descent method;

Step 5.3.8: Execute step 5.3.5 again to calculate the error of the output layer. If the error is less than ε, the training process of the BP neural network ends; otherwise, continue to execute step 5.3.2.

During the specific implementation, 30 sets of data are selected to form the training set. The training data of the training set are as follows:

Table 1 Historical values of grinding parameters.

Ball mill feeding amount Mill inlet water supply Pump pool supplementary water volume Cyclone feeding concentration Grinding size 11.80408021 4 7.869386805 36.25384938 31.47754722 18.96361676 4 12.64241118 65.93599079 50.5696447 23.30609519 4 15.5373968 90.23767278 62.14958719 25.9399415 4 17.29329433 110.1342072 69.17317734 27.53745004 4 18.35830003 126.4241118 73.43320011 28.50638795 4 19.00425863 139.7611576 76.01703453 29.0940785 4 19.39605233 150.6806072 77.58420933 29.45053083 4 19.63368722 159.6206964 78.534744889 29.6667301 4 19.77782007 166.9402224 79.11128028 29.79786159 4 19.86524106 172.9329434 79.460966424 29.87739686 4 19.91826457 177.8393683 79.67305828 29.92563743 4 19.95042496 181.a564093 79.80169983 29.95489682 4 19.96993122 185.1452844 79.87972486 29.97264354 4 19.98176236 187.8379875 79.92704944 29.98340747 4 19.98893831 190.0425863 79.95575325

29.98993612 4 19.99329075 191.8475592 79.97316299 29.99389595 4 19.99593063 193.325346 79.98372253 29.999629771 4 19.9975318 194.5352555 79.99012722 29.99775445 4 19.99850296 195.5258456 79.991401185 29.998638 4 19.999092 196.3368722 79.99636801 29.99917391 4 19.99944927 197.0008846 79.99779708 29.99949895 4 19.99966597 197.544532 79.99866386 29.9996961 4 19.9997974 197.9896329 79.99918959 29.99981567 4 19.99987712 198.3540506 79.99950846 29.9998882 4 19.99992547 198.6524106 79.99970187 29.99993219 4 19.99995479 198.8966871 79.99981917 29.99995887 4 19.99997258 199.0966838 79.99989032 29.99997505 4 19.99998337 199.2604273 79.99993348 29.99998487 4 19.99998991 199.3944891 79.99995965 30 4 19.99998991 199.4564891 79.99995988

Step 5.4: The real-time value of the grinding parameters is used as the test set of the trained BP neural network model, and the output is the predicted value of the grinding particle size. Step 5.4 is specifically:

Step 5.4.1: Take the real-time value of the grinding parameters taken from the industrial PC as the test set, including the real-time value of the ore feed to the ball mill, the real-time value of the water feed to the ball mill inlet, the real-time value of the supplementary water in the pump pool, and the real-time value of the rotation The real-time value of the ore concentration of the streamer;

Step 5.4.2: Normalize the test set data;

Step 5.4.3: Input the normalized test set into the BP neural network model trained in step 5.3 to obtain the grinding particle size prediction data.

Table 2 The real-time value of grinding parameters and grinding particle size and the predicted value of grinding particle size generated by simulation

Table 2 compares the predicted value of grinding particle size predicted by BP neural network with the real-time value. The comparison results show that the prediction effect basically meets the actual production requirements. And from the perspective of prediction effect, the larger the amount of data in this prediction method, the greater the prediction effect and stability.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the idea of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the present invention. within the scope of protection.

Claims (6)

1. An online real-time prediction system for grinding particle size based on the Internet of Things, characterized in that it comprises: The data acquisition unit is used to simulate and generate the real-time values of the grinding parameters and the real-time values of the grinding particle size, and perform data processing on the real-time values of the grinding parameters to obtain standard voltage data; The wireless data transmission unit is used to send the standard voltage data to the data management unit; The data management unit is used to receive the standard voltage data, restore the standard voltage data to the real-time value of the grinding parameter and store it; The MES data reading unit is used to read the real-time value of the grinding parameter stored in the data management unit, and store the historical value of the grinding parameter; The grinding particle size prediction unit is used to train the BP neural network model with the historical value of the grinding parameter as the input of the BP neural network, and input the real-time value of the grinding parameter into the trained BP neural network model for grinding particle size prediction. predict; A BP neural network model evaluation unit is used to compare the real-time value of the predicted grinding particle size with the real-time value of the simulated grinding particle size to evaluate the BP neural network model; The data acquisition unit includes: The real-time data generation module is used to establish a grinding virtual object model to simulate the industrial site data collection situation, and generate a set of real-time values of grinding parameters at intervals of 5S; The data processing module is used to standardize the real-time values of the generated grinding parameters and convert them into standard voltage data of [0,5V]; Grinding particle size prediction unit includes: Initialization module, for setting up the initial model based on BP neural network including input layer, hidden layer and output layer; The parameter setting module is used for setting the parameters of the initial model; The learning module uses the historical value of the grinding parameter as the training set of the initial model of the BP neural network to train the initial model of the BP neural network to obtain the trained BP neural network model; The grinding particle size prediction module uses the real-time value of the grinding parameters as the test set of the trained BP neural network model, and the output is the predicted value of the grinding particle size. 2. The online real-time prediction system of grinding particle size based on the Internet of Things as claimed in claim 1, wherein the data management unit comprises: The data display module is used to convert the received standard voltage data into a standard voltage signal label to display; The data conversion module is used to restore the received standard voltage quantity to the real-time value of the grinding parameter; The data storage module is used to store the real-time values of the grinding parameters. 3. A method for online real-time prediction of grinding particle size based on the Internet of Things, is characterized in that, comprises the steps: Step 1: Simulate and generate real-time values of grinding parameters and grinding particle size, and perform data processing on the real-time values of grinding parameters to obtain standard voltage data; Step 2: Send the standard voltage data to the data management unit; Step 3: Receive the standard voltage data, restore the standard voltage data to the real-time value of the grinding parameters and store it; Step 4: read the real-time value of the grinding parameter stored in the data management unit, and store the historical value of the grinding parameter; Step 5: use the historical value of the grinding parameter as the input of the BP neural network to train the BP neural network model, and input the real-time value of the grinding parameter into the trained BP neural network model to predict the grinding particle size; Step 6: compare the real-time value of the predicted grinding particle size with the real-time value of the simulated grinding particle size to evaluate the BP neural network model; Said step 5 includes: Step 5.1: Establish an initial model based on BP neural network including input layer, hidden layer and output layer; Step 5.2: Set parameters for the initial model; Step 5.3: use the historical value of the grinding parameter as the training set of the initial model of the BP neural network to train the initial model of the BP neural network, constantly adjust the weights and thresholds of each layer, and obtain the trained BP neural network model; Step 5.4: Use the real-time value of the grinding parameter as the test set of the trained BP neural network model, and the output is the predicted value of the grinding particle size; Said step 1 includes: Step 1.1: Establish a grinding virtual object model to simulate industrial field data collection, and generate a set of real-time values of grinding parameters at intervals of 5 seconds; Step 1.2: The data processing module standardizes the real-time values of the generated grinding parameters and converts them into standard voltage data of [0,5V]. 4. the online real-time prediction method of grinding particle size based on Internet of Things as claimed in claim 3, is characterized in that, described step 3 comprises: Step 3.1: Convert the received standard voltage data into a standard voltage signal label and display it; Step 3.2: Restore the received standard voltage quantity to the real-time value of the grinding parameter; Step 3.3: Store real-time values of grinding parameters. 5. the online real-time prediction method of grinding particle size based on Internet of Things as claimed in claim 3, is characterized in that, described step 5.1 comprises: Step 5.1.1: Set the number of nodes in the input layer to 4, and the nodes correspond to 4 grinding parameters, including: ball mill feed volume, ball mill inlet water feed volume, pump pool supplementary water volume, and cyclone feed concentration; Step 5.1.2: Set the number of hidden layers to 2, and the number of hidden layer neurons to 30 and 20 respectively. 6. The online real-time prediction method of grinding particle size based on the Internet of Things as claimed in claim 3, wherein said step 5.2 comprises: Step 5.2.1: Set the hidden layer weight w _ij , the hidden layer threshold θ _i , the output layer weight w _ki and the output layer threshold a _k , where j represents the jth input sample, and i represents the ith hidden layer layer node, k represents the kth output layer node; Step 5.2.2: Set the minimum expected target error to 1e-5, and the learning rate to 0.05; Step 5.2.3: The activation function of the hidden layer and the output layer of the BP neural network is selected as an S-type logarithmic function, namely

Among them, n is 4 s-dimensional input column vectors.

Images