Wind speed and wind pressure IoT intelligent big data system

Abstract

The present invention discloses an intelligent big data system of wind speed and wind pressure Internet of Things. The system consists of two parts: an environmental parameter acquisition and control platform and a wind speed prediction subsystem. The environmental parameter acquisition and control platform realizes detection, adjustment and monitoring of environmental parameters; the wind speed prediction subsystem realizes accurate prediction of wind speed and wind pressure in the detected environment. The present invention effectively solves the problem that the existing environmental parameter detection does not have the influence of nonlinearity and large hysteresis of wind speed and wind pressure parameter changes and large and complex environmental areas on wind speed and wind pressure detection, and does not accurately detect and predict environmental wind speed and wind pressure, thereby greatly affecting the regulation and production management of environmental wind speed and wind pressure.

Description

Technical Field

[0001] The present invention relates to the technical field of wind speed and wind pressure information parameter detection, prediction and control, and in particular to a wind speed and wind pressure Internet of Things intelligent big data system. Background Art

[0002] Wind speed and wind pressure measurement and regulation systems are widely used in various fields, such as agriculture, electricity, steel, petrochemical, mining, energy conservation and industry. Wind speed measuring devices are mainly divided into hot ball type, impeller type and differential pressure type according to the principle. The advantages of the hot ball type are high sensitivity, large range, and adaptability to environmental measurement. The disadvantage is that the platinum wire connected to the hot ball in the probe is relatively fragile, which can easily cause the probe to be damaged during use and cannot be repaired; the impeller type mainly relies on the wind to drive the impeller to rotate and generate electromagnetic signals for measurement. The advantage of this method is that the instrument is relatively durable and is often used for long-term measurement. The three-cup anemometer used in meteorological observation is also based on the same principle. The disadvantage is that the test accuracy is low and the sensitivity is slightly poor. It is not suitable for testing small wind speed changes; the differential pressure type gas flow rate and flow online detection equipment can effectively avoid clogging of the detection equipment. It is not only suitable for clean air flow wind speed measurement, but also can be used for material, dust gas and high temperature, high humidity gas flow rate detection and control. The measurement data is stable, the response is fast, and the maintenance is convenient. The data can be transmitted to the factory central control room in real time, providing technical support for the realization of intelligent production. Based on the characteristics of low wind speed measurement accuracy and weak anti-interference, this patent adopts a big data processing method to intelligently process and adjust the big data output by wind speed and wind pressure sensors, and invents an intelligent big data system for wind speed and wind pressure Internet of Things to improve the accuracy of wind speed and wind pressure detection and the reliability of adjustment. Summary of the invention

[0003] The present invention provides an intelligent big data system of wind speed and wind pressure Internet of Things. The present invention effectively solves the problems that the existing environmental parameter detection does not have the nonlinearity and large hysteresis according to the changes of wind speed and wind pressure parameters and the influence of large and complex environmental areas on wind speed and wind pressure detection, and does not accurately detect and predict environmental wind speed and wind pressure, thereby greatly affecting the regulation and production management of environmental wind speed and wind pressure.

[0004] The present invention is achieved through the following technical solutions:

[0005] The wind speed and wind pressure IoT intelligent big data system consists of two parts: an environmental parameter acquisition and control platform and a wind speed prediction subsystem. The environmental parameter acquisition and control platform realizes the detection, adjustment and monitoring of environmental parameters; the wind speed prediction subsystem realizes the accurate prediction of the wind speed and wind pressure of the detected environment.

[0006] The further technical improvement scheme of the present invention is:

[0007] The environmental parameter acquisition and control platform consists of detection nodes, control nodes, gateway nodes, on-site monitoring terminals, cloud platforms and mobile phone APPs. The detection nodes collect environmental parameters and upload them to the cloud platform through the gateway nodes, and use the data provided by the cloud platform to the mobile phone APP. The mobile phone APP can monitor the detected environmental parameters in real time and adjust the external devices of the control nodes through the cloud platform. The detection nodes and control nodes are responsible for collecting the detected environmental parameter information and adjusting the environmental equipment. The detection nodes, control nodes, on-site monitoring terminals, cloud platforms and mobile phone APPs are communicated two-way through the gateway nodes to realize the collection of detected environmental parameters and control of environmental equipment. The structure of the environmental parameter acquisition and control platform is as follows: Figure 1 shown.

[0008] The further technical improvement scheme of the present invention is:

[0009] The wind speed prediction subsystem consists of an LSTM neural network model, an Adaline neural network model, a variational mode decomposition model, a K-means clustering classifier, a CNN convolution-NARX neural network model and a fuzzy wavelet neural network model of a Vague set, a wind pressure parameter detection module and a TS fuzzy neural network model of a Vague set; the time series wind speed values ​​of the detected environment perceived by multiple wind speed sensors are respectively used as the input of the LSTM neural network model and the Adaline neural network model, the difference between the outputs of the LSTM neural network model and the Adaline neural network model is used as the wind speed fluctuation value of the detected environment, the time series wind speed fluctuation value is used as the input of the variational mode decomposition model, the variational mode decomposition model outputs multiple modal function IMF components, the energy entropy of multiple IMF components is used as the input of the K-means clustering classifier, the energy entropy of multiple types of IMF components output by the K-means clustering classifier is respectively used as the input of multiple corresponding CNN convolution-NARX neural network models, the Adaline neural network model and multiple CNN convolutions The output of the product-NARX neural network model is used as the corresponding input of the fuzzy wavelet neural network model of the Vague set. The three parameters of the output of the fuzzy wavelet neural network model of the Vague set are x, t and 1-f, x is the real value of the detected wind speed, t is the credibility, 1-ft is the uncertainty, f is the unreliability, and the value of the Vague set of the detected wind speed composed of x, t and 1-f is [x, (t, 1-f)]. The output of multiple wind pressure sensors is used as the input of the wind pressure parameter detection module. The wind pressure parameter detection module and the Vague The outputs of the fuzzy wavelet neural network model of the ue set are used as the corresponding inputs of the beat delay line TDL, and the two beat delay lines TDL are used as the corresponding inputs of the TS fuzzy neural network model of the Vague set. The three parameters output by the TS fuzzy neural network model of the Vague set are y, s and 1-z, y is the predicted value of the detected wind speed, s is the credibility, 1-zs is the uncertainty, z is the unreliability, y, s and 1-z constitute the predicted value of the Vague set of the detected ambient wind speed [y, (s, 1-z)]. The functional structure of the wind speed prediction subsystem is as follows: Figure 2 shown.

[0010] The further technical improvement scheme of the present invention is:

[0011] The wind pressure parameter detection module is composed of a FOA-wavelet neural network model with a time-delay unit, a GM (1, 1) grey prediction model, a K-means clustering classifier, a CNN convolution-NARX neural network model and a fuzzy wavelet neural network model of a Vague set; the time series parameter values ​​of the detected environment perceived by the wind pressure sensor are respectively used as the input of the corresponding FOA-wavelet neural network model with a time-delay unit and the GM (1, 1) grey prediction model, the difference between the output of the FOA-wavelet neural network model with a time-delay unit and the GM (1, 1) grey prediction model is used as the parameter fluctuation value of the detected environment, and multiple time series parameter fluctuation values ​​and multiple GM (1, 1) grey prediction model outputs are respectively used as the corresponding K-means The input of the clustering classifier, the multiple types of time series parameter fluctuation values ​​output by the K-means clustering classifier, and the output of the GM (1, 1) grey prediction model are respectively used as the input of the corresponding CNN convolution-NARX neural network model. The output of the CNN convolution-NARX neural network model is used as the corresponding input of the Vague set fuzzy wavelet neural network model. The three parameters output by the Vague set fuzzy wavelet neural network model are x, t and 1-f, respectively. x is the real value of the detected parameter, t is the credibility, f is the unreliability, and 1-ft is the uncertainty. The value of the Vague set of the detected parameter composed of x, t and 1-f is [x, (t, 1-f)]. The output of the Vague set fuzzy wavelet neural network model is used as the output of the wind pressure parameter detection module. The function and structure of the wind pressure parameter detection module are as follows: Figure 3 shown.

[0012] Compared with the prior art, the present invention has the following obvious advantages:

[0013] 1. In view of the uncertainty and randomness of the problems of wind speed and wind pressure sensor precision error, interference and measurement anomaly in the process of measuring wind speed and wind pressure parameters, the present invention converts multiple wind speed sensor output values ​​and multiple wind pressure sensor output values ​​into the TS fuzzy neural network model form of Vague set and the fuzzy wavelet neural network model form of Vague set respectively through a wind speed prediction subsystem and a wind pressure parameter detection module, which effectively handles the fuzziness, dynamics and uncertainty of the measurement parameters of multiple wind speed and wind pressure sensors, and improves the objectivity and credibility of the detection parameters of the wind speed and wind pressure sensors.

[0014] Second, the LSTM neural network model used in the present invention is a recurrent neural network with 4 interaction layers in a repetitive network. It can not only extract information from the sequence data of the input wind speed sensor output parameters like a standard recurrent neural network, but also retain the information of the long-term correlation of the input wind speed sensor output from the previous remote step. In addition, since the sampling interval of the wind speed sensor output parameters is relatively small, the wind speed sensor output parameters have long-term spatial and temporal correlations, and the LSTM neural network model has enough long-term memory to process the spatiotemporal relationship between the wind speed sensor output parameters, thereby improving the accuracy and robustness of processing the wind speed sensor output parameters.

[0015] 3. The variational mode decomposition model of the present invention can decompose the time series wind speed fluctuation values ​​into a series of intrinsic mode functions IMF, and continuously iteratively update the center frequency and frequency band bandwidth of each component, separate the adaptive frequency components of the original time series wind speed fluctuation values, and extract the characteristic frequency components containing the time series wind speed fluctuation values. The variational mode decomposition model can effectively overcome the modal aliasing problem to achieve denoising of the time series wind speed fluctuation values. The dense peak features of the evolution curve of the time series wind speed fluctuation values ​​after denoising disappear and gradually become smooth. The variational mode decomposition model improves the accuracy and robustness of processing the time series wind speed fluctuation values.

[0016] 4. According to the characteristics of wind speed parameter sample differences, the present invention constructs a K-means clustering classifier to classify the sample parameters of multiple IMF component energy entropy of the detected wind speed parameter fluctuation values, and designs multiple CNN convolution-NARX neural network models to predict the detected wind speed parameter fluctuation values. In the process of predicting the detected wind speed parameter fluctuation values, the K-means clustering classifier extracts the multiple IMF component energy entropies of the detected parameter fluctuation values ​​with similar causes from the time-space characteristic data, so as to establish a more targeted and corresponding CNN convolution-NARX neural network model that can predict the detected wind speed parameter fluctuation values ​​according to the different characteristics of the multiple IMF component energy entropies of the detected wind speed parameter fluctuation values ​​to improve the prediction accuracy and robustness.

[0017] 5. In the CNN convolution-NARX neural network model adopted by the present invention, the CNN convolutional neural network is a deep feedforward neural network, and its typical structure is composed of an input layer, a convolutional layer, a pooling layer and a fully connected layer. The CNN convolutional neural network performs operations such as convolution and pooling on the input data, and extracts local features of the data by establishing multiple filters to obtain robust features with translation and rotation invariance. The input of the NARX neural network model includes the output of the CNN convolutional neural network for a period of time and the historical feedback of the NARX neural network model output. This part of the feedback input can be considered to include the historical information parameters of the CNN convolutional neural network output for a period of time. The NARX neural network model is a dynamic neural network model that can effectively predict the nonlinear and non-stationary time series of the CNN convolutional neural network output, and can improve the prediction accuracy of the CNN convolutional neural network output time series when the non-stationarity of the time series is reduced; the NARX neural network model introduces a dynamic recursive network of the model by introducing a delay module and output feedback. It introduces the delayed feedback of the input and output vectors into the network training to form a new input vector, and has good nonlinear mapping ability. The input of the network model includes not only the original input data, but also the trained output data. The generalization ability of the network is improved, making it more accurate and robust than traditional neural networks in the time series prediction of nonlinear detected parameters.

[0018] 6. The three parameters output by the ANFIS fuzzy neural network model of the Vague set proposed in the present invention are x, t and 1-f, respectively, where x is the real value of the predicted value of the detected wind speed parameter, t is the credibility, 1-ft is the uncertainty, and f is the unreliability. The values ​​of the Vague set of the predicted value of the detected wind speed parameter constituted by x, t and 1-f are [x, (t, 1-f)]. It has the function of approximating any linear and nonlinear function with arbitrary precision, and has a fast convergence speed, a small number of samples required, a fast model operation speed, reliable results, and good effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is the environmental parameter collection and control platform of the present invention;

[0020] Figure 2 The wind speed prediction subsystem of the present invention;

[0021] Figure 3 The wind pressure parameter detection module of the present invention;

[0022] Figure 4 A detection node of the present invention;

[0023] Figure 5 A control node of the present invention;

[0024] Figure 6 A gateway node of the present invention;

[0025] Figure 7 It is the on-site monitoring end software of the present invention. DETAILED DESCRIPTION

[0026] Combined with Figure 1-7 , the technical solution of the present invention is further described:

[0027] 1. System overall functional design

[0028] The wind speed and wind pressure IoT intelligent big data system of the present invention can detect and predict environmental factor parameters. The system consists of two parts: an environmental parameter acquisition and control platform and a wind speed prediction subsystem. The environmental parameter acquisition and control platform includes environmental parameter detection nodes, control nodes, gateway nodes, field monitoring terminals, cloud platforms and mobile APPs. The detection nodes and control nodes are self-organized to form a LoRa monitoring network to realize LoRa communication between the detection nodes, control nodes and gateway nodes; the detection nodes send the collected environmental parameters to the field monitoring terminal and the cloud platform through the gateway nodes, and the gateway nodes and the cloud platform realize two-way transmission of environmental parameters and related control information between the field monitoring terminal and the mobile APP. The mobile APP is designed using the open source framework APP provided by Gizwits. You only need to integrate the APP SDK provided by Gizwits in the mobile APP to connect to the Gizwits platform and realize remote detection and regulation functions based on the mobile APP. The structure of the environmental parameter acquisition and control platform is shown in Figure 1 .

[0029] 2. Design of Detection Node

[0030] A large number of detection nodes 1 based on the LoRa wireless sensor network are used as environmental parameter sensing terminals. The detection nodes realize information interaction between gateway nodes through the self-organizing LoRa wireless sensor network. The detection nodes include sensors for collecting environmental wind speed, wind pressure, humidity, temperature and light intensity parameters and corresponding signal conditioning circuits, STM32 microprocessors and LoRa communication modules SX127X; the software of the detection nodes mainly realizes LoRa communication and the collection and preprocessing of environmental parameters. 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. See the detection node structure for details. Figure 4 .

[0031] 3. Design of Control Node

[0032] The control node realizes information interaction between gateway nodes through the self-organizing LoRa network. The control node includes 4 digital-to-analog conversion circuits corresponding to the control external devices, an STM32 microprocessor, 4 external device controllers and a LoRa communication module SX127X; the 4 external device controllers are a temperature controller, a humidity controller, a wind speed controller and a light controller. The control node structure is shown in Figure 5 .

[0033] 4. Gateway Node Design

[0034] The gateway node includes SX127X module, NB-IoT module, STM32 microcontroller and RS232 interface. The gateway node includes SX127X module to realize the self-organizing network of communication between detection node and control node, NB-IoT module to realize the two-way data interaction between gateway and cloud platform, RS232 interface to connect to the field monitoring terminal, and realize the information interaction between gateway and field monitoring terminal. See the structure of gateway node Figure 6 .

[0035] 5. On-site monitoring software

[0036] The field monitoring terminal is an industrial control computer. The field monitoring terminal 3 mainly realizes the collection, prediction and adjustment of environmental parameters and realizes information interaction with the gateway node. The main functions of the field monitoring terminal are communication parameter setting, data analysis and data management, and environmental big data processing. The management software uses Microsoft Visual++6.0 as the development tool and calls the system's Mscomm communication control to design the communication program. The field monitoring terminal software functions are shown in Figure 7 The wind speed prediction subsystem consists of LSTM neural network model, Adaline neural network model, variational mode decomposition model, K-means clustering classifier, CNN convolution-NARX neural network model and Vague set fuzzy wavelet neural network model, wind pressure parameter detection module and Vague set TS fuzzy neural network model; the design process of the wind speed prediction subsystem is as follows:

[0037] 1. LSTM neural network model design

[0038] The time series wind speed values ​​of the detected environment sensed by multiple wind speed sensors are used as the input of the LSTM neural network model and the Adaline neural network model respectively. The difference between the output of the LSTM neural network model and the Adaline neural network model is used as the wind speed fluctuation value of the detected environment. The LSTM neural network model introduces the mechanism of memory cell and hidden layer state to control the information transmission between hidden layers. There are three gates (Gates) calculation structures in the memory cell of an LSTM neural network, namely the input gate, forget gate and output gate. Among them, the input gate can control the addition or filtering of new wind speed information; the forget gate can forget the wind speed detection information that needs to be discarded and retain the useful information in the past; the output gate can make the memory cell only output the wind speed detection information related to the current time step. These three gate structures perform operations such as matrix multiplication and nonlinear summation in the memory cell, so that the memory will not decay in continuous iterations. The Long Short-Term Memory (LSTM) structural unit consists of a cell, an input gate, an output gate, and a forget gate. The LSTM neural network model is a short-term memory model that can last for a long time and is suitable for predicting changes in the wind speed input of a time series. The LSTM neural network model effectively prevents the gradient from disappearing during RNN training. The Long Short-Term Memory (LSTM) network is a special RNN. The LSTM neural network model can learn long-term wind speed detection dependency information while avoiding the gradient disappearance problem. LSTM adds a structure called a memory cell to the neural nodes of the hidden layer of the RNN in the internal structure of the neuron to memorize the dynamic change information of the wind speed detected in the past, and adds three gate (Input, Forget, Output) structures to control the use of historical information on the wind speed detected. Suppose the input is the time series value of the wind speed detection input (x1, x2,…, x T ), the hidden layer state is (h1, h2, …, h T ), then at time t we have:

[0039] i t =sigmoid(W hi h t-1 +W xi X t ) (1)

[0040] f t =sigmoid(W hf h t-1 +W hf X t ) (2)

[0041] c t =f t ⊙c t-1 +i t ⊙tanh(W hc h t-1 +W xc X t ) (3)

[0042] o t =sigmoid(W ho h t-1 +W hx X t +W co c t ) (4)

[0043] h t =o t ⊙tanh(c t ) (5)

[0044] where i t 、f t , o t represents the input gate, forget gate and output gate, c t Represents cell unit, Wh represents the weight of recursive connection, Wx represents the weight from input layer to hidden layer, sigmoid and tanh are two activation functions, and the output of LSTM neural network model is the nonlinear value of the detected ambient wind speed.

[0045] 2. Adaline neural network model design

[0046] The time series wind speed values ​​of the detected environment sensed by multiple wind speed sensors are respectively used as the input of the LSTM neural network model and the Adaline neural network model. The difference between the outputs of the LSTM neural network model and the Adaline neural network model is used as the wind speed fluctuation value of the detected environment. The adaptive linear unit (Adaptive Linear Element) of the Adaline neural network model is one of the early neural network models. The input signal of the model can be written in the form of a vector: X(K) = [x0(K), x1(K), ...x n (K)] T , each set of input signals corresponds to a set of weight vectors, which are expressed as: W(K) = [k0(K), k1(K), ... k(K)], when x0(K) is equal to negative 1, it is the bias value of the Adaline neural network model that determines the excitation or inhibition state of the neuron. The network output can be defined according to the input vector and weight vector of the Adaline neural network model as:

[0047]

[0048] In the Adaline neural network model, there is a special input, namely the ideal response output d(K), which is sent to the Adaline neural network model, and then compared with the network output y(K), and the difference is sent to the learning algorithm mechanism to adjust the weight vector until the optimal weight vector is obtained. y(K) and d(K) tend to be consistent. The adjustment process of the weight vector is the learning process of the network. The learning algorithm is the core part of the learning process. The weight optimization search algorithm of the Adaline neural network model adopts the LMS algorithm least squares method. The Adaline neural network model outputs the linear value of the wind speed in the detected environment.

[0049] 3. Variational mode decomposition model design

[0050] The difference between the output of the LSTM neural network model and the Adaline neural network is used as the wind speed fluctuation value of the detected environment, and the time series wind speed fluctuation value is used as the input of the variational mode decomposition model. The variational mode decomposition model outputs multiple modal function IMF components, and the energy entropy of multiple IMF components is used as the input of the K-means clustering classifier; the variational mode decomposition model is an adaptive, non-recursive signal time-frequency analysis method, which can decompose the time series wind speed fluctuation value signal into several wind speed fluctuation value sub-signals, namely, the IMF component u k , and minimize the bandwidth and u of all IMF components k The amplitude modulation and frequency modulation function can be expressed as:

[0051] u k (t) =A k cos[φ k (t) ] (7)

[0052] Where φ k (t) is a non-decreasing function, A k (t) For the envelope, construct a constrained variational problem to solve u k , quadratic penalty terms and Lagrange multipliers are introduced to solve the variational problem, making it an unconstrained problem. The variational mode decomposition model can decompose the time series wind speed fluctuation value signal to be decomposed into several IMF components. The energy entropy value can measure the regularity of the time series wind speed fluctuation, indicating the energy characteristics of the time series wind speed fluctuation signal in different frequency bands. When the time series wind speed fluctuation value mutates, the energy will also change. The energy of the mth IMF component is defined as:

[0053]

[0054] Where x m(i) is the mth component after the time series wind speed fluctuation signal sample is decomposed, n is the number of sampling points, and the energy entropy of the mth IMF component is:

[0055]

[0056] 4. K-means clustering classifier design

[0057] Multiple IMF component energy entropies are used as inputs of the K-means clustering classifier, and multiple types of IMF component energy entropies output by the K-means clustering classifier are used as inputs of multiple corresponding CNN convolution-NARX neural network models; the core idea of ​​the K-means clustering algorithm is to divide n data objects into K classes, and minimize the sum of squares of all data objects in each class to the cluster center point of the class, but its clustering time is relatively long. In order to achieve fast clustering of data, retain the efficiency of the K-means clustering classifier and expand the application scope of the K-means clustering classifier to discrete data, the calculation process of the K-means clustering classifier is as follows:

[0058] (1) From the entire sample X, let I = 1, and randomly select K data objects in X as the initial clustering center m j (I) , where j = 1, 2,…,K.

[0059] (2) Let d(i, j) represent K cluster centers m j (I) For each object x in the large data sample X i The distance between is:

[0060]

[0061] Use formula (10) to find the smallest Euclidean distance d among all (i, j) values ​​of d(i, j), and find the smallest Euclidean distance d at the cluster center m. j (I) The same cluster S j Store object x in i . Let m j (I+1) Represents the new cluster center point, and its calculation formula is:

[0062]

[0063] N in formula (11) j Represents the number of data objects in the jth class.

[0064] (4) Set judgment criteria to determine whether the criteria are met. If so, proceed to the next step. If not, go to step (2).

[0065] (5) Output the clustering results of the big data. Usually, a judgment criterion is used to determine whether to terminate the loop. That is, when the division results obtained in the Ith iteration and the I-1th iteration are the same, the division is considered reasonable and the iteration is terminated.

[0066] 5. CNN convolution-NARX neural network model design

[0067] The energy entropy of multiple types of IMF components output by the K-means clustering classifier are used as inputs of multiple corresponding CNN convolution-NARX neural network models respectively. The outputs of the Adaline neural network and multiple CNN convolution-NARX neural network models are used as the corresponding inputs of the fuzzy wavelet neural network model of the Vague set; the CNN convolution-NARX neural network model uses the output of the CNN convolution neural network as the input of the NARX neural network model. The CNN convolution neural network model can automatically mine and extract sensitive spatial features that characterize the time series wind speed fluctuation values ​​from a large number of IMF component energy entropy values ​​of time series wind speed fluctuation values. The CNN convolution neural network model structure mainly includes four parts: ① Input layer (Input). The input layer is the input of the CNN convolution neural network model, and generally the IMF component energy entropy of the time series wind speed fluctuation value is directly input. ② Convolution layer (Conv). Due to the large dimension of the input layer data, it is difficult for the CNN convolutional neural network model to directly and comprehensively perceive the IMF component energy entropy input information of all time series wind speed fluctuation values. It is necessary to divide the input data into several parts for local perception, and then obtain global information through weight sharing, while reducing the complexity of the CNN convolutional neural network model structure. This process is the main function of the convolution layer. The specific process is to use a convolution kernel of a specific size to traverse and convolve the IMF component energy entropy input signal of the time series wind speed fluctuation value with a fixed step size, so as to realize the mining and extraction of sensitive features of the IMF component energy entropy input signal of the time series wind speed fluctuation value. ③Pool layer (Pool, also known as downsampling layer). Since the data sample dimension obtained after the convolution operation is still large, it is necessary to compress the data volume and extract key information to avoid long model training time and overfitting. Therefore, a pooling layer is connected after the convolution layer to reduce the dimension. Considering the peak characteristics of the defect characteristics, the maximum pooling method is used for downsampling. ④, fully connected layer. After all the convolution and pooling operations, the IMF component energy entropy feature extraction data of the time series wind speed fluctuation value enters the fully connected layer. Each neural layer in this layer is fully connected with all the neurons in the previous layer, and the local feature information of the IMF component energy entropy value of the time series wind speed fluctuation value extracted by the convolution layer and the pooling layer is integrated.At the same time, in order to avoid overfitting, the dropout technology is added to this layer. The output value of the last fully connected layer will be passed to the output layer. The pooling results of the last layer are connected together in a head-to-tail manner to form an output layer and serve as the input of the NARX neural network model. The NARX neural network model is a dynamic recursive neural network with output feedback connection. In terms of topological connection relationship, it is equivalent to a BP neural network with input delay plus a delayed feedback connection from output to input. Its structure consists of an input layer, a delay layer, a hidden layer and an output layer. The input layer nodes are used for signal input, the delay layer nodes are used for the time delay of input signals and output feedback signals, the hidden layer nodes use activation functions to perform nonlinear operations on delayed signals, and the output layer nodes are used to linearly weight the hidden layer output to obtain the final network output. The output h of the i-th hidden layer node of the NARX neural network model. i for:

[0068]

[0069] The output of the jth output layer node of the NARX neural network is o j for:

[0070]

[0071] 6. Design of Fuzzy Wavelet Neural Network Model for Vague Sets

[0072] The outputs of the Adaline neural network model and multiple CNN convolution-NARX neural network models are used as the corresponding inputs of the fuzzy wavelet neural network model of the Vague set. The three parameters of the output of the fuzzy wavelet neural network model of the Vague set are x, t and 1-f, respectively. x is the real value of the detected wind speed, t is the credibility, 1-ft is the uncertainty, and f is the unreliability. The value of the Vague set of the detected wind speed composed of x, t and 1-f is [x, (t, 1-f)]; this patent applies fuzzy neural network for fuzzy reasoning, and combines the characteristics of multi-resolution analysis of wavelets, uses wavelet functions as the excitation function of neural network neurons, and constructs a fuzzy wavelet network (Fuzzy Wavelet Network, FWNN) of the Vague set. This fuzzy wavelet neural network based on FWNN has good intelligence, robustness, stability and rapidity of index tracking. The fuzzy wavelet neural network of the Vague set includes two parts: fuzzy neural network (FNN) and wavelet neural network (WNN). The fuzzy neural network of the Vague set contains 4 basic layers: the first layer is the input layer, each input vector corresponds to a neuron; each neuron in the second layer represents a language variable value; each neuron in the third layer represents a fuzzy rule; and the fourth layer is the normalization layer. At the same time, the input of the fuzzy neural network is used as the input of the wavelet neural network, and each fuzzy rule corresponds to a wavelet network. The wavelet basis function is a wavelet basis group obtained by translating the wavelet function. Therefore, the wavelet neural network generated by different scale functions can capture the characteristics of different time domains and frequency domains, and different fuzzy reasoning selects the corresponding wavelet network. Wavelets have the characteristics of multi-resolution analysis. If the wavelet function is used as the excitation function of the neural network neuron, since the scaling and translation of each neuron can be adjusted, the selection of low-scale parameters can learn smooth functions, and the increase of scale can learn local singular functions with higher accuracy, which is higher than the ANN with the same number of neurons and parameters. The fuzzy wavelet network of this patent is realized by five basic layers: input, fuzzification, reasoning, wavelet network layer and defuzzification layer. The number of neural network nodes in each layer is n, n×M, M, M and 3 respectively. Once the number of input n and rules M is determined, the structure of the FWNN model is determined. The input of the Vague set fuzzy wavelet neural network is X = [x1, x2, ... x n ], T i is the number of wavelets corresponding to the i-th rule; w ik is the weight coefficient; is the wavelet function, The output value of the linear combination of the local model wavelet network corresponding to rule i is:

[0073]

[0074] The first layer is the input layer: each node in this layer is directly connected to each component xj of the input vector, and the input value X = [x1, x2, ...x n ] is passed to the next layer; the second layer calculates the membership function value corresponding to each input variable; the third layer calculates the applicability of each rule; the fourth layer is the wavelet network layer output, which is mainly used for output compensation; the fifth layer is the control signal output layer, also known as the anti-fuzzification layer, in which defuzzification calculation is performed. The three parameters output by the fuzzy wavelet neural network model of the Vague set are x, t and 1-f, x is the real value of the detected wind speed, t is the credibility, 1-ft is the uncertainty, f is the unreliability, and the value of the Vague set of the detected wind speed composed of x, t and 1-f is [x, (t, 1-f)];

[0075] 7. Design of wind pressure parameter detection module

[0076] The outputs of multiple wind pressure sensors are used as the input of the wind pressure parameter detection module, and the outputs of the wind pressure parameter detection module and the fuzzy wavelet neural network model of the Vague set are used as the corresponding inputs of the TS fuzzy neural network model of the Vague set; the wind pressure parameter detection module is composed of a FOA-wavelet neural network model with a time lag unit, a GM (1, 1) grey prediction model, a K-means clustering classifier, a CNN convolution-NARX neural network model and a fuzzy wavelet neural network model of the Vague set; the time series parameter values ​​of the detected environment perceived by the wind pressure sensor are used as the inputs of the corresponding FOA-wavelet neural network model with a time lag unit and the GM (1, 1) grey prediction model, and the difference between the outputs of the FOA-wavelet neural network model with a time lag unit and the GM (1, 1) grey prediction model is used as the parameter fluctuation value of the detected environment, and the time series The column parameter fluctuation value and the GM (1, 1) grey prediction model output are respectively used as the input of the corresponding K-means clustering classifier, the multiple types of time series parameter fluctuation values ​​output by the two K-means clustering classifiers and the GM (1, 1) grey prediction model output are respectively used as the input of the corresponding CNN convolution-NARX neural network model, and the output of the CNN convolution-NARX neural network model is used as the corresponding input of the fuzzy wavelet neural network model of the Vague set. The three parameters of the fuzzy wavelet neural network model of the Vague set are x, t and 1-f, x is the real value of the detected parameter, t is the credibility, f is the unreliability, 1-ft is the uncertainty, and the value of the Vague set of the detected parameter constituted by x, t and 1-f is [x, (t, 1-f)], and the output of the fuzzy wavelet neural network model of the Vague set is used as the output of the wind pressure parameter detection module.

[0077] , Design of FOA-wavelet neural network model with time-delay unit

[0078] The time series parameter values ​​of the wind pressure sensor sensing the detected environment are respectively used as the input of the corresponding FOA-wavelet neural network model with time lag unit and the GM (1, 1) grey prediction model. The difference between the output of the FOA-wavelet neural network model with time lag unit and the GM (1, 1) grey prediction model is used as the parameter fluctuation value of the detected environment. The time series parameter fluctuation value and the output of the GM (1, 1) grey prediction model are respectively used as the input of the corresponding K-means clustering classifier; the output of the FOA-wavelet neural network model with time lag unit is the beat delay line as the input of the fruit fly optimization algorithm-wavelet neural network model. The fruit fly optimization algorithm simulates the foraging behavior of fruit flies in the biological world and can be used to find the global optimal solution. First, the fruit fly population is initialized, the fruit fly population sizepop, the maximum number of iterations maxgen and the fruit fly position (X, Y) are initialized, and the initial position of the fruit fly is assigned by random initialization. The parameters optimized by this patent are the weight coefficient, wavelet scaling factor and translation factor of the wavelet neural network model; the fruit fly searches by smell, judges the distance between it and the food according to its position, and calculates the position concentration judgment value; the obtained concentration judgment value is substituted into the concentration judgment function, that is, the objective function in this problem, and the odor concentration corresponding to the fruit flies at different positions is obtained, and the fruit fly with the best odor concentration is selected; the position of the fruit fly with the best concentration is set as the best position after this iteration, and other fruit flies move closer to the position of the fruit fly; repeat the above steps, and if the best concentration is better than the best concentration after the last iteration, update the best position until the maximum number of iterations is reached. The advantages of the fruit fly algorithm are fast running speed, simple and easy to understand, and good convergence performance and global search ability. The wavelet neural network uses the wavelet function as the excitation function of the neuron and combines it with a feedforward network proposed by the artificial neural network. The scaling, translation factor and connection weight of the wavelet in the wavelet neural network model are adaptively adjusted in the process of optimizing the error energy function. Assume that the input signal of the wavelet neural network model can be represented as a one-dimensional vector x output by a parameter sensor i (i=1,2,…,n), the output signal is represented by y k (k=1,2,…,m), the calculation formula of the output layer prediction value of the wavelet neural network model is:

[0079]

[0080] In the formula, ω ij The connection weight between the input layer i node and the hidden layer j node, is the wavelet basis function, b j is the translation factor of the wavelet basis function, a j The scaling factor of the wavelet basis function, ω jkis the connection weight between the hidden layer j node and the output layer k node. The weight and threshold correction algorithm of the wavelet neural network model in this patent adopts the gradient correction method to update the network weight and wavelet basis function parameters, so that the output of the wavelet neural network model continuously approaches the expected output.

[0081] (2) Design of GM (1, 1) grey prediction model

[0082] The grey prediction method has more advantages than the traditional statistical prediction method. It does not need to determine whether the predicted variable obeys the normal distribution, does not require a large sample statistic, and does not need to change the prediction model at any time according to the change of the input variable. Through the cumulative generation technology, a unified differential equation model is established. The original value of the accumulated sensor output is restored to obtain the prediction result. The differential equation model has a higher prediction accuracy. The essence of establishing the GM (1, 1) grey prediction model is to perform a cumulative generation of the original data output by the sensor so that the generated series presents a certain regularity. By establishing the differential equation model, the fitting curve is obtained to predict the sensor output value.

[0083] (3) K-means clustering classifier design

[0084] The time series parameter fluctuation value and the GM (1, 1) grey prediction model output are respectively used as the input of the corresponding K-means clustering classifier, and the multiple types of time series parameter fluctuation values ​​and the GM (1, 1) grey prediction model output of the two K-means clustering classifiers are respectively used as the input of the corresponding CNN convolution-NARX neural network model. The design process of the K-means clustering classifier refers to the design process of the K-means clustering classifier in design step 4 of this patent.

[0085] (4) CNN convolution-NARX neural network model design

[0086] The multiple types of time series parameter fluctuation values ​​output by the two K-means clustering classifiers and the GM (1, 1) grey prediction model output are respectively used as the input of the corresponding CNN convolution-NARX neural network model, and the output of the CNN convolution-NARX neural network model is used as the corresponding input of the fuzzy wavelet neural network model of the Vague set. The design process of the CNN convolution-NARX neural network model refers to the design of the CNN convolution-NARX neural network model in step 5 of this patent.

[0087] (5) Design of fuzzy wavelet neural network model based on Vague set

[0088] The output of the CNN convolution-NARX neural network model is used as the corresponding input of the fuzzy wavelet neural network model of the Vague set. The three parameters of the fuzzy wavelet neural network model of the Vague set are x, t and 1-f, respectively. x is the real value of the detected parameter, t is the credibility, f is the unreliability, and 1-ft is the uncertainty. The value of the Vague set of the detected parameter constituted by x, t and 1-f is [x, (t, 1-f)]. The output of the fuzzy wavelet neural network model of the Vague set is used as the output of the wind pressure parameter detection module. The design process of the fuzzy wavelet neural network model of the Vague set refers to the design of the fuzzy wavelet neural network model of the Vague set in step 6 of this patent.

[0089] 8. Design of TS fuzzy neural network model based on Vague set

[0090] The outputs of the wind pressure parameter detection module and the Vague set fuzzy wavelet neural network model are used as the corresponding beat delay line TDL inputs, and the two beat delay lines TDL are used as the corresponding inputs of the Vague set TS fuzzy neural network model. The three parameters output by the Vague set TS fuzzy neural network model are y, s and 1-z, respectively. y is the predicted value of the detected wind speed, s is the credibility, 1-zs is the uncertainty, and z is the unreliability. The Vague set predicted value of the detected ambient wind speed composed of y, s and 1-z is [y, (s, 1-z)]. The Vague set TS fuzzy neural network model includes a precondition network and a postcondition network. (1) The precondition network, the first layer of the precondition network is the input layer, and the number of nodes in this layer is n. The second layer is the fuzzification layer, which fuzzifies the input data, and each neuron executes the corresponding membership function The third layer is the fuzzy rule layer. The number of nodes in the fourth layer is m, and this layer implements normalized calculation. (2) Consequence network. The first layer is the input layer, where the input value x0 of the 0th node is 1, and its function is to provide the constant term of the fuzzy rule consequence. The second layer has m nodes, and its function is to calculate the consequence of each rule:

[0091]

[0092] Layer 3 computes system output:

[0093]

[0094] Adjust the central value c of the membership function of the second layer through the network learning algorithm j and width b j and the connection weight p of the subsequent network jk For simplicity, the parameter p jkThe consequent of each rule is fixed, and the subsequent of each rule becomes the connection weight of one layer in the simplified structure. This simplified structure has the same structure as the TS fuzzy neural network classifier of the conventional model, and the calculation results of the conventional model can be applied.

[0095] 6. Design example of environmental parameter acquisition and control platform

[0096] According to the actual situation of the environmental big data detection system, the system arranges the plane layout and installation diagram of the detection nodes, gateway nodes and on-site monitoring terminals of the environmental parameter collection and control platform. The sensors of the detection nodes are evenly arranged in various positions of the detected and controlled environment according to the needs of detection. The environmental parameters are collected and controlled through the system.

[0097] The technical means disclosed in the scheme of the present invention are not limited to the technical means disclosed in the above-mentioned implementation mode, but also include technical schemes composed of any combination of the above-mentioned technical features. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications are also regarded as the protection scope of the present invention.

Claims

1. Wind speed and wind pressure IoT intelligent big data system, characterized by: The system consists of two parts: an environmental parameter acquisition and control platform and a wind speed prediction subsystem. The environmental parameter acquisition and control platform realizes the detection, adjustment and monitoring of environmental parameters; the wind speed prediction subsystem realizes the accurate prediction of the wind speed and wind pressure of the detected environment; The wind speed prediction subsystem is composed of an LSTM neural network model, an Adaline neural network model, a variational mode decomposition model, a K-means clustering classifier, a CNN convolution-NARX neural network model, a Vague set fuzzy wavelet neural network model, a wind pressure parameter detection module, a beat delay line TDL and a Vague set TS fuzzy neural network model; The time series wind speed values ​​of the detected environment perceived by multiple wind speed sensors are respectively used as the input of the LSTM neural network model and the Adaline neural network model. The difference between the outputs of the LSTM neural network model and the Adaline neural network model is used as the wind speed fluctuation value of the detected environment. The time series wind speed fluctuation value is used as the input of the variational mode decomposition model. The variational mode decomposition model outputs multiple modal function IMF components. The energy entropies of multiple IMF components are used as the input of the K-means clustering classifier. The energy entropies of multiple types of IMF components output by the K-means clustering classifier are respectively used as the input of multiple corresponding CNN convolution-NARX neural network models. The outputs of the Adaline neural network model and multiple CNN convolution-NARX neural network models are used as the corresponding inputs of the fuzzy wavelet neural network model of the Vague set. The three parameters output by the fuzzy wavelet neural network model of the Vague set are x, t and 1-f, respectively. x is the real value of the detected wind speed, t is the credibility, 1-f is the sum of the credibility and uncertainty, and 1-ft is the uncertainty. f is the unreliability, x, t and 1-f constitute the value of the Vague set of the detected wind speed is [x, (t, 1-f)], the outputs of multiple wind pressure sensors are used as the input of the wind pressure parameter detection module, the outputs of the wind pressure parameter detection module and the fuzzy wavelet neural network model of the Vague set are respectively used as the corresponding beat delay line TDL inputs, the two beat delay lines TDL are respectively used as the corresponding inputs of the TS fuzzy neural network model of the Vague set, the three parameters output by the TS fuzzy neural network model of the Vague set are y, s and 1-z, y is the predicted value of the detected wind speed, s is the reliability, 1-z is the sum of the reliability and uncertainty, 1-zs is the uncertainty, z is the unreliability, y, s and 1-z constitute the predicted value of the Vague set of the detected environmental wind speed is [y, (s, 1-z)]; The wind pressure parameter detection module is composed of a FOA-wavelet neural network model with a time-delay unit, a GM (1, 1) grey prediction model, a K-means clustering classifier, a CNN convolution-NARX neural network model and a fuzzy wavelet neural network model of a Vague set; The time series parameter values ​​of the detected environment perceived by the wind pressure sensor are respectively used as the input of the corresponding FOA-wavelet neural network model with time lag unit and the GM(1,1) grey prediction model. The difference between the outputs of the FOA-wavelet neural network model with time lag unit and the GM(1,1) grey prediction model is used as the parameter fluctuation value of the detected environment. Multiple time series parameter fluctuation values ​​and multiple GM(1,1) grey prediction model outputs are respectively used as the input of the corresponding K-means clustering classifier. Multiple types of time series parameter fluctuation values ​​and GM(1,1) grey prediction model outputs of the two K-means clustering classifiers are The output of the grey prediction model is used as the input of the corresponding CNN convolution-NARX neural network model. The output of the CNN convolution-NARX neural network model is used as the corresponding input of the fuzzy wavelet neural network model of the vague set. The three parameters of the output of the fuzzy wavelet neural network model of the vague set are x, t and 1-f, where x is the real value of the detected parameter, t is the credibility, f is the unreliability, 1-f is the sum of the credibility and uncertainty, 1-ft is the uncertainty, and the value of the vague set of the detected parameter composed of x, t and 1-f is [x, (t, 1-f)]. The output of the fuzzy wavelet neural network model of the vague set is used as the output of the wind pressure parameter detection module.

2. The wind speed and wind pressure IoT intelligent big data system according to claim 1 is characterized by: The environmental parameter acquisition and control platform is mainly composed of detection nodes, control nodes, gateway nodes, on-site monitoring terminals, cloud platforms and mobile phone APP.

3. The wind speed and wind pressure IoT intelligent big data system according to claim 2 is characterized by: The detection node collects environmental parameters and uploads them to the cloud platform via the gateway node, and uses the data provided by the cloud platform to the mobile phone APP. The mobile phone APP can monitor the detected environmental parameters in real time and adjust the external devices of the control nodes provided by the cloud platform. The detection node and the control node are responsible for collecting the detected environmental parameter information and adjusting the environmental equipment. The gateway node is used to realize two-way communication among the detection node, the control node, the on-site monitoring terminal, the cloud platform and the mobile phone APP, thereby realizing the collection of the detected environmental parameters and the control of the environmental equipment.

Images