Dust measurement method and device used in industrial environment
Through sensor collection and the echo state network model is constructed, the problem of insufficient dust concentration prediction accuracy in industrial environments is solved, and more accurate dust concentration prediction and production safety guarantee are achieved.
Patent Information
- Application Number
- CN202410979817.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-11-12
AI Technical Summary
The existing dust concentration prediction methods have poor measurement accuracy at high and low concentrations, and cannot meet the precise monitoring needs of a large number of samples or multiple detection points in industrial environments.
The factory environment parameters and dust concentration data are collected through sensor sets, and the echo state network model is constructed, and the input data in the time dimension is predicted to improve the accuracy of dust concentration prediction.
It enhances the accuracy and decision-making ability of dust concentration prediction, avoids production safety risks caused by insensitive dust concentration prediction, and improves the generalization ability of the model on new data.
Smart Images

Figure CN118939907A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of dust detection, and in particular to a dust measurement method and device for use in an industrial environment. Background Art
[0002] Dust is a tiny solid particle generated during industrial production that can be suspended in the air. They are generated from a wide range of sources, including a series of industrial production processes such as cement, flour, metal material processing and ore blasting. Among them, metal dust (such as lead, copper, zinc, etc.) contains toxic substances, which pose a potential threat to human health and may cause serious diseases such as occupational pneumoconiosis. And the dust suspended in the industrial workshop will not only reduce the visual performance and work efficiency of the operators, but also affect the quality of the products. What's more serious is that it may also lead to unsafe behaviors and accidents. In addition, suspended dust may also damage precision instruments and cause direct economic losses to enterprises.
[0003] The Chinese patent with announcement number CN115146526B discloses a method for predicting dust concentration in a non-stationary process, and the method includes the following steps: S1: record the change value C(t) of dust concentration over time within a period of time; S2: decompose C(t) using the variational mode decomposition method to obtain the dust concentration deposition trend value over time, the dust concentration periodic change value with the mine working state, and the dust concentration fluctuation value with time with a stationary process; S3: respectively predict the dust concentration deposition trend value over time, the dust concentration periodic change value with the mine working state, and the dust concentration fluctuation value with time with a stationary process, and finally predict the dust concentration. However, the measurement accuracy of the above scheme at high and low concentrations is usually not ideal, and in the case of a large number of samples or multiple detection points, it cannot meet the needs of accurate monitoring. Therefore, it is very necessary to provide a dust measurement method and device for industrial environments to improve the prediction accuracy of dust concentration. Summary of the invention
[0004] In view of this, the present invention proposes a dust measurement method and device for industrial environments. The multi-dimensional data such as factory environmental parameters and dust concentration data are collected through a sensor set to enhance the data-based decision-making ability. At the same time, the input data in the time dimension is predicted and processed through the echo state network to improve the short-term memory of the echo state network and ensure the accuracy of dust concentration prediction in the factory.
[0005] The present invention provides a dust measurement method for an industrial environment, the method comprising:
[0006] Acquire a set of original sample data collected by a set of sensors disposed in the factory, wherein the set of original sample data includes factory environmental parameters, dust concentration data, and a first time period;
[0007] Constructing an initial echo state network model, dividing the original sample data set into a training data set and a test data set, and training the initial echo state network model with the training data set to obtain an echo state network model;
[0008] The environmental parameters, the dust concentration data and the first time period are input into the echo state network model to predict the dust concentration data in the factory in a second time period, wherein the first time period and the second time period have the same length and are continuously distributed.
[0009] On the basis of the above technical solution, preferably, the obtaining of the original sample data set collected by the sensor set arranged in the factory specifically includes:
[0010] According to the sensor set arranged in the factory, the temperature, humidity, light intensity and wind speed corresponding to the setting position of each sensor in the sensor set are respectively collected;
[0011] According to the light intensity corresponding to each sensor setting position and the concentration conversion function, the dust concentration data corresponding to the light intensity is calculated.
[0012] On the basis of the above technical solution, preferably, the expression of the concentration conversion function is:
[0013]
[0014]
[0015]
[0016] Wherein, I(θ) represents the scattered light intensity, I0 represents the initial incident light intensity, n represents the refractive index of the dust particles, θ represents the deflection angle between the incident light and the scattered light, d represents the diameter of the dust particles, λ represents the wavelength of the incident light, N represents the scattering intensity, ρ represents the particle concentration of the dust particles, and C represents the dust concentration data.
[0017] More preferably, the input layer of the echo state network model includes k neurons, the reserve pool includes N neurons, and the output layer includes L neurons;
[0018] The internal neuron update function expression of the reserve pool is:
[0019] x(l+1)=f(W inu(l+1)+Wx(l)+W back y(l))
[0020] Wherein, x(l+1) represents the state of the reservoir at the l+1th moment, f() represents the update function of the reservoir state in the echo state network model, and W in is the feedback matrix from the input layer to the storage pool, u(l+1) represents the input layer state at the l+1th moment, W represents the feedback matrix in the storage pool, and W back represents the bias term of the reserve pool, and y(l) represents the output layer state at the lth moment;
[0021] The output function expression of the output layer is:
[0022] y(l+1)=F(W out (u(l+1),x(l),y(l)))
[0023] Wherein, y(l+1) represents the output layer state at the l+1th moment, F() represents the output function of the echo state network model, x(l) represents the reserve pool state at the lth moment, and W out represents the feedback matrix of the output layer.
[0024] More preferably, the method further comprises:
[0025] Based on the position update function of each sensor in the sensor set, an optimized original sample data set is obtained, wherein the position update function includes a first position update function, a second position update function and a fixed measurement position function.
[0026] More preferably, the expression of the first position update function is:
[0027]
[0028] in, represents the position update function of the first sensor in the t+1th iteration, i represents the ith sensor, t represents the tth iteration, represents the position parameter of the jth dimension of the ith sensor in the tth iteration, α represents the first random variable, iter max represents the maximum number of iterations, R2 represents the dust concentration warning value, ST represents the dust concentration safety value, Q represents the second random variable that obeys the normal distribution, and L represents the standard parameter matrix.
[0029] More preferably, the expression of the second position update function is:
[0030]
[0031] in, represents the position update function of the second sensor in the t+1th iteration, represents the optimal position parameter of the current sensor in the t+1th iteration, represents the worst position parameter of the current sensor in the tth iteration, m represents the total number of set sensors, A + represents the standard pseudoinverse matrix.
[0032] More preferably, the expression of the fixed measurement position function is:
[0033]
[0034] in, represents the position update function in the t+1th iteration, represents the optimal position parameter of the current sensor in the tth iteration, β represents the step control parameter, K represents the random number corresponding to the step control parameter, and f i represents the position parameter of the i-th sensor, f w represents the optimal location parameter of the i-th sensor.
[0035] More preferably, the expression of the fixed measurement position function is:
[0036]
[0037] in, represents the position update function in the t+1th iteration, represents the optimal position parameter of the current sensor in the tth iteration, β represents the step control parameter, K represents the random number corresponding to the step control parameter, and f i represents the position parameter of the i-th sensor, f w represents the optimal location parameter of the i-th sensor.
[0038] In a second aspect of the present application, a dust measurement device for use in an industrial environment is provided, the dust measurement device comprising a data acquisition module, a model building module and a concentration prediction module, wherein:
[0039] The data acquisition module is used to acquire a set of original sample data collected by a set of sensors arranged in the factory, wherein the set of original sample data includes factory environmental parameters, dust concentration data and a first time period;
[0040] The model building module is used to build an initial echo state network model, divide the original sample data set into a training data set and a test data set, and train the initial echo state network model with the training data set to obtain an echo state network model;
[0041] The concentration prediction module is used to input the environmental parameters, the dust concentration data and the first time period into the echo state network model to predict the dust concentration data in the factory in a second time period, wherein the first time period and the second time period have the same length and are continuously distributed.
[0042] The dust measurement method and device provided by the present invention for use in an industrial environment have the following beneficial effects compared with the prior art:
[0043] (1) By collecting multi-dimensional data such as factory environmental parameters and dust concentration data through a collection of sensors, more accurate and comprehensive data support can be provided for the training of the echo state network model, thereby enhancing the data-based decision-making ability. At the same time, the echo state network is used to predict and process the input data in the time dimension to improve the short-term memory of the echo state network, so that dust concentration changes can be captured and predicted more accurately, ensuring the accuracy of dust concentration prediction in the factory and avoiding production safety risks caused by insensitive dust concentration prediction.
[0044] (2) By using a preset ratio to divide the original sample data set into a training set and a test set, it is beneficial to improve the pertinence and effectiveness of model training. At the same time, the initial echo state network model is iteratively trained based on the training set and the test set to continuously optimize the performance and accuracy of the model. A predetermined deviation value is set as the judgment criterion for model convergence to achieve adaptive convergence control of the echo state network model, which is beneficial to improving the generalization ability of the model on new data and enhancing the overall prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0046] Figure 1 A flow chart of a dust measurement method for an industrial environment provided by the present invention;
[0047] Figure 2 A structural principle diagram of the fiber Bragg grating dust concentration measurement system provided by the present invention;
[0048] Figure 3 A schematic diagram of the structure of the echo state network model provided by the present invention;
[0049] Figure 4 This is a schematic structural diagram of the dust measuring device provided by the present invention.
[0050] Explanation of the accompanying drawings: 1. Dust measuring device; 11. Data acquisition module; 12. Model building module; 13. Model building module. DETAILED DESCRIPTION
[0051] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0052] Before introducing the embodiments of the present invention, some terms and their abbreviations involved in the embodiments of the present invention are first defined and explained.
[0053] The principle that fiber Bragg gratings can be used to measure dust concentration is mainly based on the scattering effect of fiber Bragg gratings. When there are dust particles in the air, these particles will scatter the light signal, thereby changing the propagation characteristics of the light signal in the fiber Bragg grating, and then the dust concentration can be indirectly measured by detecting this change. Specifically, fiber Bragg gratings are usually composed of periodic changes in the refractive index in the optical fiber. This periodic structure can cause the light signal to diffract and scatter in the optical fiber. When the light signal passes through the optical fiber, if there are dust particles, these particles will scatter the light signal, causing the propagation characteristics of the light signal in the optical fiber to change, such as light intensity, phase, etc. By detecting this change in the light signal, the concentration of dust particles in the air can be inferred.
[0054] The present application embodiment discloses a dust measurement method for an industrial environment, such as Figure 1 As shown, the method comprises steps S1-S3.
[0055] Step S1, obtaining an original sample data set collected by a sensor set arranged in a factory, wherein the original sample data set includes factory environmental parameters, dust concentration data and a first time period.
[0056] In this embodiment, a method is provided as follows: Figure 2 The fiber Bragg grating dust concentration measurement system shown in the figure includes a main controller, an air compressor, a dust collector, a variable frequency fan, an anemometer, a temperature sensor, a humidity sensor, a standard sieve, a fiber Bragg grating collimator, a dust box, an air outlet, a transmission optical fiber, a fiber Bragg grating regulator and a computer.
[0057] The fiber grating dust concentration measurement system is mainly composed of three modules, namely, a dust generation module, a data acquisition module, and a data processing module. The dust generation module includes a dust absorber, a compressed air pump, a variable frequency fan, an anemometer, a standard sieve, a sensor, and a dust box. Among them, the dust absorber is used to collect dust in the air; the dust absorber can be provided with an air inlet and connected to the compressed air pump. The compressed air pump is connected to the dust absorber for pressurizing the dust absorber. The dust absorber is connected to the dust box through an air duct provided with a variable frequency fan; the dust box is used to collect and store dust. An anemometer 105 can also be provided on the air duct to calibrate the air flow rate of the variable frequency fan. The sensor can be provided on the dust box or inside the dust box. The sensor can include a temperature sensor for measuring the temperature data of the dust; and a humidity sensor for measuring the humidity data of the dust. At the front end of the dust box, a standard sieve is also provided on the side close to the dust absorber, which can be used to screen out impurities in the dust. An air outlet for discharging dust may also be provided at the rear end of the dust box.
[0058] The data acquisition module may include: at least one fiber grating collimator, which is used to collect optical power data when there is dust or no dust in the dust box; wherein at least one fiber grating collimator is connected to the fiber grating regulator through a transmission optical fiber; the fiber grating regulator is connected to the data processing module, which is used to send the optical power data to the data processing module; wherein the optical power data is used to represent the dust environment parameters. The data processing module is used to perform data processing on the dust environment parameters using a dust prediction model based on an echo state network to obtain the prediction results of the dust concentration detection data of the target dust environment. For example, the data processing module can be a computing terminal. For example, the computing terminal can be an intelligent terminal device such as a desktop computer or a laptop computer.
[0059] The data processing module is used to construct an input data sequence from time data, optical power data, temperature data, humidity data, and wind speed data, and input the input data sequence into a dust prediction model based on an echo state network to obtain dust concentration detection data output by the model.
[0060] Specifically, the dust detection system is started through the main controller, the dust collector collects dust in the air, and the pressure in the dust collector is increased by controlling the air compressor to discharge the dust; the variable frequency fan is controlled to simulate the actual air flow rate, and the anemometer is controlled to calibrate the air flow rate so that the air flow rate is within the predetermined parameters. The impurities that do not belong to the dust in the dust are screened out through the standard sieve, and the temperature and humidity sensors are controlled to work to directly measure the temperature and humidity in the dust. The dust in the dust box can be discharged through the air outlet, and the fiber grating collimator directly measures the optical power when there is dust or no dust, and transmits the data to the fiber grating demodulator through the transmission fiber. Finally, the dust concentration is predicted in the computing terminal using the trained dust prediction model.
[0061] As the core component of the dust detection system, the master controller can be used to coordinate and manage the various parts of the system to achieve the overall functions and goals of the system.
[0062] The dust collector is a device used to collect dust particles from the air. The main function of the air compressor is to compress the gas from a low pressure state to a high pressure state. In this system, it increases the pressure in the dust collector and discharges the dust from the dust collector. The variable frequency fan is a fan device that uses variable frequency speed regulation technology. Its main feature is that it can adjust the air volume and air pressure output of the fan by changing the speed of the motor to simulate the wind speed in the real environment. By adjusting the frequency and speed of the fan, the speed and direction of the air flow can be controlled. The air compressor and the variable frequency fan are used to discharge the dust in the dust absorber, and the wind speed of the variable frequency fan can be adjusted according to actual requirements to simulate the real environment. The anemometer is a device used to measure the air flow rate (i.e. wind speed). The standard sieve refers to a device used for particle sorting and screening, which is used to screen out non-dust impurities in the air to ensure the accuracy and reliability of subsequent measurements. The temperature and humidity sensor is used to measure the temperature and humidity in the environment. The dust box is a container for collecting and storing dust and impurities in the environment. The air outlet is used to discharge impure air in the dust box to avoid dust accumulation in the dust box. Fiber Bragg grating collimator is an optical device used to control and adjust the output light beam of the optical fiber. Its main function is to convert the output light beam of the optical fiber into a parallel light beam, thereby realizing the collimation and focusing of the light beam. Transmission fiber is a device that uses optical fiber materials to transmit optical signals. Fiber Bragg grating demodulator is an optical device used in optical fiber communication systems. Its main function is to demodulate the received optical signal and extract the transmitted information or data from it. The computing terminal processes, calculates, analyzes and stores the input data. The data processing module is used to obtain the dust concentration based on the optical power.
[0063] Specifically, the principle that fiber Bragg grating can be used to measure dust concentration is mainly based on the scattering effect of fiber Bragg grating. When there are dust particles in the air, these particles will scatter the light signal, thereby changing the propagation characteristics of the light signal in the fiber Bragg grating, and then the dust concentration can be indirectly measured by detecting this change. Specifically, fiber Bragg grating is usually composed of periodic refractive index changes in the optical fiber. This periodic structure can cause the light signal to diffract and scatter in the optical fiber. When the light signal passes through the optical fiber, if there are dust particles, these particles will scatter with the light signal, causing the propagation characteristics of the light signal in the optical fiber to change, such as light intensity, phase, etc. By detecting this change in the light signal, the concentration of dust particles in the air can be inferred. In general, fiber Bragg grating uses the scattering effect of light signals and dust particles to measure dust concentration.
[0064] In this embodiment, step S1 also includes steps S11 to S12.
[0065] Step S11 , based on the sensor set installed in the factory, respectively collect the temperature, humidity, light intensity and wind speed corresponding to each sensor installation position in the sensor set.
[0066] In this step, based on the factory's production process, equipment layout, and raw material / product storage conditions, analyze the areas where dust may be generated and the pollution characteristics, use computational fluid dynamics (CFD) simulation or other predictive models to predict the dust distribution inside the factory, and based on the above dust distribution analysis, determine key areas and potential high-concentration points as key locations for sensor deployment. For large factories, a regional sensor deployment strategy can be adopted to improve the overall monitoring coverage. At the same time, according to the monitoring objectives and environmental conditions, select appropriate sensor types, such as light scattering, photoelectric, inductive, etc., and consider the sensor's measurement range, response speed, stability and other indicators, and select sensors that meet the actual needs of the factory to improve the accuracy and reliability of overall monitoring.
[0067] Step S12, calculating dust concentration data corresponding to the light intensity according to the light intensity corresponding to each sensor setting position and the concentration conversion function.
[0068] In this step, the basic principle of detecting dust concentration by optical method is that a beam of linear polarized light is irradiated on the particle, and the dust particle is assumed to be an ideal spherical particle and regarded as a point. According to the law of light transmission, the intensity of scattered light is:
[0069]
[0070] Generally speaking, since the intensity of scattered light can be superimposed when the particle scattering is incoherent, the intensity of scattered light can be further expressed as:
[0071]
[0072] If the dust particle size distribution has little effect on the dust concentration, the impact on the measurement result can be ignored. Without considering the particle size distribution, the relationship between particle concentration and mass concentration is shown in the following formula:
[0073]
[0074] The expression of the concentration conversion function can be obtained as follows:
[0075]
[0076] Wherein, I(θ) represents the scattered light intensity, I0 represents the initial incident light intensity, n represents the refractive index of the dust particles, θ represents the deflection angle between the incident light and the scattered light, d represents the diameter of the dust particles, λ represents the wavelength of the incident light, N represents the scattering intensity, ρ represents the particle concentration of the dust particles, and C represents the dust concentration data.
[0077] According to the principle of optical dust detection, the measurement results reflect the concentration of dust particles under optical power. Therefore, the smaller the optical power, the fewer dust particles are absorbed, and the more realistic the test results are. The measurement results of dust concentration using optical power are more reliable than those using point light sources. However, the non-parallelism of fiber Bragg gratings will affect the test results, and the absorption intensity of the passing dust particles is inconsistent, resulting in errors in the test results. Therefore, it is necessary to improve parallelism (that is, select highly consistent sensors for measurement) to improve the reliability and authenticity of dust concentration test results.
[0078] Step S2, constructing an initial echo state network model, dividing the original sample data set into a training data set and a test data set, and training the initial echo state network model with the training data set to obtain an echo state network model.
[0079] In this embodiment, if Figure 3 As shown, the input layer of the echo state network model contains k neurons, the reserve pool contains N neurons, and the output layer contains L neurons;
[0080] The core structure of the echo state network is a randomly generated reserve pool, similar to the hidden layer of a general neural network. However, its size is much larger than that of a traditional neural network hidden layer, usually around dozens or hundreds. Its internal neuron connections are sparse, and the sparsity SD is generally 2% to 5%.
[0081] At time l, since the echo state network needs to input multiple variables (optical power, temperature, humidity, dust concentration, etc.), the sequence of input units is u(l) = [u1(l), u2(l), ... u k (l)], the state sequence of the reservoir is x(l) = [x1(l), x2(l), ... x N (l)], the output unit sequence is y(l) = [y1(l), y2(l), ...y L (l)].
[0082] The internal neuron update function expression of the reserve pool is:
[0083] x(l+1)=f(W in u(l+1)+Wx(l)+W back y(l))
[0084] Where x(l+1) represents the state of the reservoir at the l+1th moment, f() represents the update function of the reservoir state in the echo state network model, and W in is the feedback matrix from the input layer to the storage pool, u(l+1) represents the input layer state at the l+1th moment, W represents the feedback matrix in the storage pool, and W back represents the bias term of the reserve pool, and y(l) represents the output layer state at the lth moment.
[0085] W is not required back , that is, let W back ≤0, the update equation of the reserve pool state is obtained as:
[0086] x(l+1)=f(W in u(l+1)+Wx(l))
[0087] The output function expression of the output layer is:
[0088] y(l+1)=F(W out (u(l+1),x(l),y(l)))
[0089] Among them, y(l+1) represents the output layer state at the l+1th moment, F() represents the output function of the echo state network model, x(l) represents the state of the reservoir at the lth moment, and W out represents the feedback matrix of the output layer.
[0090] The echo state network model is highly efficient because it only needs to learn and train the feedback matrix W of the output layer. out , the feedback matrix W from the input layer to the reservoir inAs well as the feedback matrix W in the reserve pool, once determined, it remains unchanged. At the same time, the echo state network has the function of instantaneously memorizing the input signal u(n), that is, the echo state.
[0091] The echo state network has the condition of echo state: the spectral radius of the feedback matrix W in the reserve pool is less than 1.
[0092] That is, the maximum spectral radius of the eigenvalue that satisfies W is: |λ| max <1.
[0093] For the randomly generated feedback matrix in the reserve pool, find the maximum value of the feedback matrix update:
[0094]
[0095] Among them, W new represents the new internal feedback matrix.
[0096] In one example, an optimized original sample data set is obtained based on a position update function of each sensor in the sensor set, wherein the position update function includes a first position update function, a second position update function, and a fixed measurement position function.
[0097] The expression of the first position update function is:
[0098]
[0099] in, represents the position update function of the first sensor in the t+1th iteration, i represents the ith sensor, t represents the tth iteration, represents the position parameter of the jth dimension of the ith sensor in the tth iteration, α represents the first random variable, α∈0,1, iter max Indicates the maximum number of iterations, R2 indicates the dust concentration warning value, ST indicates the dust concentration safety value, and satisfies R2∈[0,1] and ST∈[0.5,1], Q indicates the second random variable that obeys the normal distribution, and L indicates the standard parameter matrix, which is a 1×d matrix, where each element of the L matrix is 1. R2<ST indicates that the surrounding dust has not been measured, and this range can be widely measured. R2≥ST indicates that the surrounding dust has been measured, and it is not necessary to measure in this range, and it can be measured elsewhere.
[0100] The expression of the second position update function is:
[0101]
[0102] in, represents the position update function of the second sensor in the t+1th iteration, represents the optimal position parameter of the current sensor in the t+1th iteration, represents the worst position parameter of the current sensor in the tth iteration, m represents the total number of set sensors, A + represents the standard pseudo-inverse matrix, A is a 1×d matrix where each element is randomly assigned a value of 1 or -1, and A + =A T (AA T ) -1 ,when , it means that the current position has been measured and needs to be measured at another position.
[0103] The expression of the fixed measurement position function is:
[0104]
[0105] in, represents the position update function in the t+1th iteration, represents the optimal position parameter of the current sensor in the tth iteration, β represents the step control parameter, K represents the random number corresponding to the step control parameter, K∈[-1,1], f i represents the position parameter of the i-th sensor, f w represents the optimal position parameter of the i-th sensor. i >f w When f i ≤f w , it means the current position has not been measured.
[0106] In this embodiment, step S2 also includes steps S21 to S24.
[0107] Step S21, dividing the optimized original sample data set into a training data set and a test data set according to a preset ratio.
[0108] In this step, the original data is divided into a training set and a test set in chronological order, ensuring that the data in the test data set is later than the data in the training data set in time. The ratio of the training data set to the test data set is 7:3. At the same time, a sliding window method can be used to take the data of the first N time steps as input to predict the concentration value of the next time step. This makes the prediction ability of the echo state network model at future time points closer to the actual deployment requirements, and uses the time dependence contained in the data to improve the prediction accuracy of the echo state network model.
[0109] In an example, assuming that the original data sequence is {M(1), M(2), ..., M(T)}, the input and output samples can be constructed using a sliding time window. Then the output sample can be expressed as [M(t-J+1), M(t-J+2), ..., M(t)], and the output sample can be expressed as M(t+1), where J represents the length of the time window. At the same time, different window lengths J are selected, and the prediction performance of the model under various window lengths is evaluated, and the mean absolute error (MAE) is calculated as the evaluation index: MAE = (1 / S) × Σ|y true -y pred |, where S is the number of test samples, y true is the true value, y pred To predict the value, choose the optimal window length J* that minimizes MAE.
[0110] The original data set D is randomly divided into K mutually exclusive subsets {D1, D2, ..., D H}, for each h∈[1,H], use D\D H (other subsets except Dk) as training sets, D H As the test set, train the model parameters and calculate the performance index P on the test set h, The final performance indicator P is the average of H evaluation results: P = (1 / H) × ΣP h .
[0111] Step S22, based on the training data set and the test data set, iteratively train the initial echo state network model to obtain a concentration recognition result.
[0112] In this embodiment, the hyperparameters of the initial echo state network model are determined, such as the input dimension (corresponding to the time window length N), the number of hidden layer neurons, the connection matrix, etc. The appropriate initial values of the hyperparameters are selected based on experience or grid search. For example: input dimension N = 10; the number of hidden layer neurons M = 100; the connection matrix W is a sparse random matrix. An efficient gradient descent algorithm such as the Adam optimizer can be used to update the parameters, and an early stopping mechanism can be introduced to terminate the training process when the performance of the validation set is no longer improved.
[0113] In an example, the ESN parameters are first initialized, the input dimension N corresponds to the time window length, the number of neurons in the hidden layer (reserve pool) M, and the input weight matrix W in The size of is M×(N+1), the size of the reserve pool weight matrix W is M×M, and the feedback weight matrix W back The size is M×1; the input vector u(t)=[x(t),1], where x(t) is the input at time step t, and the update formula x(l+1)=f(W inu(l+1)+Wx(l)+W back y(l)), where f() is the activation function, usually the tanh function; the output weight matrix W out The size of W is 1×(N+1+M), and ridge regression can be used to solve W out , at this time W out =(X T ×X+λ×I) -1 ×X T ×Y, where X is the state sequence of the reservoir, Y is the target output sequence, and λ is the regularization parameter.
[0114] Step S23, determining whether the difference between the concentration recognition result and the dust concentration data is less than a preset deviation value.
[0115] Step S24: If the difference between the concentration recognition result and the dust concentration data is less than the preset deviation value, the model training operation is terminated, and the initial echo state network model is used as the echo state network model.
[0116] Step S3, inputting the environmental parameters, dust concentration data and the first time period into the echo state network model to predict the dust concentration data in the factory in the second time period, wherein the first time period and the second time period have the same length and are continuously distributed.
[0117] By using time, temperature, humidity and wind speed as independent variables for dust concentration prediction, the model input is time, temperature, humidity and wind speed, and their corresponding time series are averaged once every four hours. The output based on the weather conditions and air quality data of the previous four hours is the prediction of the average dust concentration in the next four hours, and the prediction accuracy of the echo state network model is evaluated.
[0118] In this embodiment, the prediction results are displayed in the form of charts, such as line graphs, bar graphs, etc., to intuitively reflect the changing trend of dust concentration. At the same time, the predicted values are compared with the actual monitored values to intuitively demonstrate the accuracy of the model prediction. According to the process characteristics and safety requirements, multi-level dust concentration warning thresholds are set. When the predicted value exceeds the threshold, warning information is promptly sent to relevant personnel via email, text messages, etc., where the warning information should include the prediction results, analysis of the reasons for exceeding the standard, and prevention and control measures to be taken.
[0119] In an example, taking a manufacturing factory as an example, according to its production process and safety management requirements, the following three-level dust concentration warning thresholds can be set.
[0120] Level 1 warning threshold - threshold setting: dust concentration> 0.5mg / m 3; Trigger condition: It is predicted that the dust concentration will exceed the threshold within the next hour; Warning measures: Send a warning SMS to the workshop supervisor to remind him to check the equipment operation status, start the automatic monitoring system, and monitor the changes in dust concentration in real time.
[0121] Second level warning threshold - threshold setting: dust concentration>1.0mg / m 3 ; Trigger condition: It is predicted that the dust concentration will exceed the threshold within the next 30 minutes; Early warning measures: Immediately send an email warning to the factory director and environmental protection person in charge, automatically start the emergency dust removal equipment, strengthen dust control in the workshop, issue a workshop suspension notice, and require workers to temporarily evacuate.
[0122] Level 3 warning threshold - threshold setting: dust concentration>2.0mg / m 3 ; Trigger condition: It is predicted that the dust concentration will exceed the threshold within the next 10 minutes; Early warning measures: Immediately report to the local environmental protection department, activate the emergency plan, stop production completely, evacuate all workers in the workshop, mobilize the emergency response team, and take emergency dust control measures.
[0123] At the same time, the prediction results can also be integrated into the factory's production control system to provide decision support for managers. According to the prediction results, short-term production plan adjustment plans can be formulated, such as adjusting process parameters, increasing the operation of dust removal equipment, etc. For medium- and long-term production planning, it can also rely on the prediction results to deploy prevention and control measures in advance, such as optimizing equipment layout, improving process flow, etc., and linking the prediction results with the factory's safety, environmental protection and other indicators to provide a basis for comprehensive production management decisions. Establish a mechanism for regular retraining of the model, collect the latest environmental and dust data at any time, and retrain the ESN model. According to the actual application effect, the hyperparameters of the ESN model, such as the size of the hidden layer, the regularization coefficient, etc., can be adjusted to further improve the prediction accuracy. Using the idea of transfer learning, learn from the data and models of other factories, enhance the generalization ability of ESN, and establish a model management and version control mechanism to ensure that the latest optimized model is used in the production environment.
[0124] By collecting multi-dimensional data such as factory environmental parameters and dust concentration data through a collection of sensors, more accurate and comprehensive data support can be provided for the training of the echo state network model, enhancing the data-based decision-making ability. At the same time, the echo state network is used to predict and process input data in the time dimension to improve the short-term memory of the echo state network. This enables more accurate capture and prediction of dust concentration changes, ensuring the accuracy of dust concentration prediction within the factory and avoiding production safety risks caused by insensitive dust concentration prediction.
[0125] Based on the above method, the present application embodiment discloses a dust measurement device for industrial environment, referring to Figure 4The dust measurement device 1 includes a data acquisition module 11, a model building module 12 and a concentration prediction module 13, wherein:
[0126] The data acquisition module 11 is used to acquire an original sample data set collected by a sensor set arranged in the factory, wherein the original sample data set includes factory environmental parameters, dust concentration data and a first time period;
[0127] The model building module 12 is used to build an initial echo state network model, divide the original sample data set into a training data set and a test data set, and train the initial echo state network model with the training data set to obtain the echo state network model;
[0128] The concentration prediction module 13 is used to input environmental parameters, dust concentration data and a first time period into the echo state network model to predict the dust concentration data in the factory in a second time period, wherein the first time period and the second time period have the same length and are continuously distributed.
[0129] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A dust measurement method for industrial environments, characterized in that: The method comprises: Acquire a set of original sample data collected by a set of sensors disposed in the factory, wherein the set of original sample data includes factory environmental parameters, dust concentration data, and a first time period; Constructing an initial echo state network model, dividing the original sample data set into a training data set and a test data set, and training the initial echo state network model with the training data set to obtain an echo state network model; The environmental parameters, the dust concentration data and the first time period are input into the echo state network model to predict the dust concentration data in the factory in a second time period, wherein the first time period and the second time period have the same length and are continuously distributed.
2. The method according to claim 1, characterized in that The obtaining of the original sample data set collected by the sensor set arranged in the factory specifically includes: According to the sensor set arranged in the factory, the temperature, humidity, light intensity and wind speed corresponding to the setting position of each sensor in the sensor set are respectively collected; According to the light intensity corresponding to each sensor setting position and the concentration conversion function, the dust concentration data corresponding to the light intensity is calculated.
3. The method according to claim 2, characterized in that The expression of the concentration conversion function is: Wherein, I(θ) represents the scattered light intensity, I0 represents the initial incident light intensity, n represents the refractive index of the dust particles, θ represents the deflection angle between the incident light and the scattered light, d represents the diameter of the dust particles, λ represents the wavelength of the incident light, N represents the scattering intensity, ρ represents the particle concentration of the dust particles, and C represents the dust concentration data.
4. The method according to claim 1, characterized in that The input layer of the echo state network model includes k neurons, the reserve pool includes N neurons, and the output layer includes L neurons; The internal neuron update function expression of the reserve pool is: x(l+1)=f(W in u(l+1)+Wx(l)+W back y(l)) Wherein, x(l+1) represents the state of the reservoir at the l+1th moment, f() represents the update function of the reservoir state in the echo state network model, and W in is the feedback matrix from the input layer to the storage pool, u(l+1) represents the input layer state at the l+1th moment, W represents the feedback matrix in the storage pool, and W back represents the bias term of the reserve pool, and y(l) represents the output layer state at the lth moment; The output function expression of the output layer is: y(l+1)=F(W out (u(l+1),x(l),y(l))) Wherein, y(l+1) represents the output layer state at the l+1th moment, F() represents the output function of the echo state network model, x(l) represents the reserve pool state at the lth moment, and W out represents the feedback matrix of the output layer.
5. The method according to claim 1, characterized in that The method further comprises: Based on the position update function of each sensor in the sensor set, an optimized original sample data set is obtained, wherein the position update function includes a first position update function, a second position update function and a fixed measurement position function.
6. The method according to claim 5, characterized in that The expression of the first position update function is: in, represents the position update function of the first sensor in the t+1th iteration, i represents the ith sensor, t represents the tth iteration, represents the position parameter of the jth dimension of the ith sensor in the tth iteration, α represents the first random variable, iter max represents the maximum number of iterations, R2 represents the dust concentration warning value, ST represents the dust concentration safety value, Q represents the second random variable that obeys the normal distribution, and L represents the standard parameter matrix.
7. The method according to claim 5, characterized in that The expression of the second position update function is: in, represents the position update function of the second sensor in the t+1th iteration, represents the optimal position parameter of the current sensor in the t+1th iteration, represents the worst position parameter of the current sensor in the tth iteration, m represents the total number of set sensors, A + represents the standard pseudoinverse matrix.
8. The method according to claim 5, characterized in that The expression of the fixed measurement position function is: in, represents the position update function in the t+1th iteration, represents the optimal position parameter of the current sensor in the tth iteration, β represents the step control parameter, K represents the random number corresponding to the step control parameter, and f i represents the position parameter of the i-th sensor, f w represents the optimal location parameter of the i-th sensor.
9. The method according to claim 5, characterized in that The dividing the original sample data set into a training data set and a test data set, and training the initial echo state network model with the training data set to obtain the echo state network model, specifically includes: Divide the optimized original sample data set into a training data set and a test data set according to a preset ratio; Based on the training data set and the test data set, iteratively training the initial echo state network model to obtain a concentration recognition result; Determine whether the difference between the concentration identification result and the dust concentration data is less than a preset deviation value; If the difference between the concentration recognition result and the dust concentration data is less than the preset deviation value, the model training operation is terminated, and the initial echo state network model is used as the echo state network model.
10. A dust measuring device for use in an industrial environment, characterized in that: The dust measurement device (1) comprises a data acquisition module (11), a model building module (12) and a concentration prediction module (13), wherein: The data acquisition module (11) is used to acquire an original sample data set collected by a sensor set arranged in the factory, wherein the original sample data set includes factory environmental parameters, dust concentration data and a first time period; The model building module (12) is used to build an initial echo state network model, divide the original sample data set into a training data set and a test data set, and train the initial echo state network model with the training data set to obtain an echo state network model; The concentration prediction module (13) is used to input the environmental parameters, the dust concentration data and the first time period into the echo state network model to predict the dust concentration data in the factory in a second time period, wherein the first time period and the second time period have the same length and are continuously distributed.
