CN115630101A - Hydrological parameter intelligent monitoring and water resource big data management system - Google Patents

Abstract

The invention discloses a hydrological parameter intelligent monitoring and water resource big data management system, which comprises a hydrological condition detection and control subsystem and a water resource big data management subsystem of the Internet of things, and realizes intelligent detection and adjustment of hydrological parameters and water resource big data management; aiming at the problem of low automation degree of traditional hydrological data monitoring and water resource management, the hydrological data are automatically acquired by setting a plurality of fixed monitoring nodes by using the technologies of Internet of things, wireless communication and intelligent control, and then the acquired data are sent to a cloud platform in real time through a wireless communication unit for processing, analyzing, storing and sharing, so that the timely early warning and automatic adjustment of the hydrological data are guaranteed, and the centralized monitoring and management of the distributed hydrological data are realized.

Description

Intelligent monitoring of hydrological parameters and big data management system of water resources

technical field

The invention relates to the technical field of hydrological data monitoring and management, in particular to a system for intelligent monitoring of hydrological parameters and water resources big data management.

Background technique

Hydrological monitoring can not only provide a strong data basis for research on flood control and disaster prevention, but also play a vital role in making decisions about the sustainable use of water resources. my country has a vast territory, with numerous rivers, lakes, reservoirs, channels, groundwater and drinking water resources distributed in a criss-cross pattern, forming a unique water resource system. However, in recent years, abnormal weather has often been encountered. When natural weather with heavy precipitation occurs, the water level and flow of water resources will skyrocket, thereby posing a great threat to river embankments and reservoir dams, and even some areas have The danger of dike and dam breaks will cause losses to people's life safety and property safety, so the need for hydrological monitoring is more urgent. Through the close combination of the Internet of Things, artificial intelligence, big data, cloud services and hydrological monitoring, the water resource intelligent monitoring and water resource information management system of the Internet of Things technology is applied to the remote monitoring of rivers, lakes, reservoirs, channels, groundwater and drinking water resources. On-line analysis of water level, flow and water quality, rainwater telemetry, remote video monitoring and other scenarios, timely grasp the water level and flow of water resources, reservoir water level, precipitation and other hydrological element data, and realize the storage, query, statistical analysis and various forms of hydrological monitoring data display.

Contents of the invention

The invention discloses a hydrological parameter intelligent monitoring and water resource big data management system. The invention aims at the problem of low automation of traditional hydrological data monitoring and water resource management. The monitoring nodes can automatically obtain hydrological data, and then send the collected data to the cloud platform in real time through the wireless communication unit for processing, analysis, storage and sharing, which provides guarantee for timely early warning and automatic adjustment of hydrological data, and realizes the distributed Centralized monitoring and management of hydrological data.

In order to solve the above problems, the present invention adopts the following technical solutions:

The intelligent monitoring of hydrological parameters and the big data management system of water resources are characterized in that the system includes a water regime detection and control subsystem and a water resources big data management subsystem of the Internet of Things to realize intelligent detection and adjustment of hydrological parameters and big data management of water resources.

The further technical improvement scheme of the present invention is:

The water regime detection and control subsystem includes Elman neural network-NARX neural network model, fuzzy recurrent neural network-NARX neural network controller, AANN self-associative neural network model, PI controller-NARX neural network controller, PI controller, parameters The detection module and the parameter prediction module, the output of multiple sets of upstream water area rainfall, water flow and water level sensors are used as the input of the parameter prediction module, the water level setting value, the output of the parameter prediction module, and the output of the AANN self-associative neural network model are respectively used as the Elman neural network-NARX neural network The corresponding input of network model, the difference of Elman neural network-NARX neural network model output and AANN self-associative neural network model output is as water level error, and water level error and error change rate are as the input of fuzzy recursive neural network-NARX neural network controller, each The water flow sensor group in the downstream water area is used as the input of the corresponding parameter detection module, the difference between the output of the fuzzy recurrent neural network-NARX neural network controller and the output of the corresponding PI controller and the output of the corresponding parameter detection module is used as the water flow error, and the water flow error is used as The input of the corresponding PI controller-NARX neural network controller, the output of the PI controller-NARX neural network controller is used as the control quantity of the corresponding pumping device; the water level sensor group of each downstream water area is used as the input of the corresponding parameter detection module, and the water level is set The difference between the fixed value and the output of the parameter detection module is used as the water level difference, the water level difference and the rate of change of the water level difference are used as the input of the corresponding PI controller, and the output of multiple parameter detection modules is used as the corresponding input of the AANN self-associative neural network model. The control subsystem realizes the detection and control of multi-area water level and water flow. The water regime detection and control subsystem is shown in Figure 1.

The further technical improvement scheme of the present invention is:

The parameter detection module is composed of multiple noise reduction self-encoder neural network-NARX neural network models, adaptive AP clusterers, multiple PSO wavelet adaptive neural network models and ESN neural network models. The measurement parameters of each are used as the input of the corresponding noise reduction autoencoder neural network-NARX neural network model, and the output of multiple noise reduction autoencoder neural network-NARX neural network models is used as the input of the adaptive AP clusterer, and the adaptive AP clustering The output values of different types of noise reduction autoencoder neural network-NARX neural network models are respectively used as the input of the corresponding PSO wavelet adaptive neural network model, and the outputs of multiple PSO wavelet adaptive neural network models are used as the ESN neural network The corresponding input of the model and the output of the ESN neural network model are used as the output of the parameter detection module; the parameter detection module is shown in Figure 2.

The further technical improvement scheme of the present invention is:

Parameter prediction module includes parameter detection module, TDL beat delay line A, metabolic GM (1,1) trend model, NARX neural network model A, NARX neural network model B, TDL beat delay line B, TDL beat delay line C , TDL beat delay line D and interval hesitant fuzzy number BAM neural network-ANFIS adaptive neuro-fuzzy reasoning model, the output of the parameter detection module is used as the input of TDL beat delay line A, and the output of TDL beat delay line A is used as The input of the metabolic GM (1, 1) trend model, the difference between the output of the TDL beat delay line A and the output of the metabolic GM (1, 1) trend model and the output of the metabolic GM (1, 1) trend model are respectively used as the NARX neural network model A And the input of NARX neural network model B, the output of NARX neural network model A and NARX neural network model B are respectively used as the input of TDL pressing delay line B and TDL pressing delay line C, TDL pressing delay line B, TDL pressing delay line Line C and the output of the BAM neural network-ANFIS adaptive neuro-fuzzy inference model of interval hesitant fuzzy numbers are respectively used as the corresponding input of the BAM neural network-ANFIS adaptive neuro-fuzzy inference model of interval hesitant fuzzy numbers, and the BAM neural network of interval hesitant fuzzy numbers The four parameters output by the network-ANFIS adaptive neuro-fuzzy inference model are a, b, c and d, respectively, a and b form the interval number [a, b] as the minimum value of the detected parameter, and c and d form the interval number [c, d] is the maximum value of the detected parameter, the interval number [a, b] and the interval number [c, d] constitute ([a, b], [c, d]) as the interval hesitation of the detected parameter Fuzzy number, interval hesitation fuzzy number BAM neural network-ANFIS adaptive neuro-fuzzy inference model outputs the interval hesitation fuzzy number of the detected parameter. The parameter prediction module is shown in Figure 2.

The further technical improvement scheme of the present invention is:

Noise reduction self-encoder neural network-NARX neural network model, BAM neural network-ANFIS adaptive neuro-fuzzy inference model, BAM neural network-NARX neural network model, Elman neural network-NARX neural network model, fuzzy recurrent neural network-NARX neural network Controller, PI controller-NARX neural network controller is characterized by series connection of noise reduction self-encoder neural network and NARX neural network model, series connection of BAM neural network and ANFIS adaptive neuro-fuzzy reasoning model, series connection of BAM neural network and NARX neural network model , Elman neural network and NARX neural network model connected in series, fuzzy recurrent neural network connected in series with NARX neural network controller, PI controller connected in series with NARX neural network controller.

The further technical improvement scheme of the present invention is:

The water resources big data management subsystem of the Internet of Things includes a hydrological parameter measurement terminal, a hydrological gateway, an on-site monitoring terminal, a hydrological parameter control terminal, a hydrological parameter cloud platform, and a hydrological monitoring mobile APP. The hydrological parameter measurement terminal is responsible for collecting the hydrological parameters of the detected water area Information, there is a hydrological detection and control subsystem in the on-site monitoring terminal. Through the hydrological gateway, the two-way communication between the hydrological parameter measurement terminal, the hydrological parameter control terminal, the on-site monitoring terminal, the hydrological parameter cloud platform and the hydrological monitoring mobile phone APP is realized, and the hydrological parameter is realized. Intelligent regulation. The water resource big data management subsystem of the Internet of Things is shown in Figure 3.

The further technical improvement scheme of the present invention is:

The hydrological parameter measurement terminal includes water level, water flow, rainfall, PH value, water temperature and dissolved oxygen sensor group and corresponding signal conditioning circuit, STM32 microprocessor, GPS module, camera and GPRS wireless transmission module for collecting hydrological parameters; hydrological parameter measurement See Figure 4 for the terminal.

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

1. In the process of hydrological parameter measurement, the present invention aims at the uncertainty and randomness of the hydrological parameter measurement sensor accuracy error, interference and measurement anomaly. The present invention converts the output value of the hydrological parameter measurement sensor into interval hesitation through the parameter prediction module The fuzzy number BAM neural network-ANFIS adaptive neuro-fuzzy reasoning model form, they effectively deal with the fuzziness, dynamics and uncertainty of hydrological parameter measurement, and improve the objectivity and credibility of the hydrological parameter sensor detection parameters.

Two, the BAM neural network of the present invention's BAM neural network-ANFIS fuzzy neural network model is a kind of associative memory neural network model, and the BAM neural network output is input as the ANFIS fuzzy neural network model, and the BAM neural network can realize two-way different associative models, through two-way The associative memory matrix stores the hydrological parameter input sample data of the BAM neural network summarized in advance. When the BAM neural network has new hydrological parameter information input, the BAM neural network recalls the corresponding output results in parallel as ANFIS Fuzzy neural network model input, the two-way associative memory network of BAM neural network is a two-layer nonlinear feedback neural network, which has the functions of associative memory, distributed storage and self-learning. When adding an input signal to a layer of BAM neural network When , another layer of BAM neural network can get the output signal of hydrological parameters.

Three, the self-adaptive wavelet neural network model of PSO of the present invention, the transfer function of hidden layer uses wavelet function in this model, by adaptively adjusting the parameter of wavelet function, can more effectively input and extract signal time-frequency characteristic, have stable structure, algorithm Simple, strong global search ability, fast convergence speed, strong generalization ability, etc., the adaptive wavelet neural network of PSO can effectively predict the input signal, has the characteristics of small noise influence, good stability, fast convergence speed, and identification High precision. It avoids the blindness of BP network structure design, the linear distribution of network weight coefficients and the convexity of the learning objective function, so that the network training process fundamentally avoids problems such as local optimization. Strong function learning ability, can approximate any nonlinear function with high precision.

Four, the present invention adopts the self-adaptive wavelet neural network of PSO, has avoided requiring activation function to be differentiable in the gradient descent method, and the process calculation to function derivation, and the iterative formula is simple during each particle search, thus calculation speed is lower than gradient again The method is much faster. By adjusting the parameters in the iterative formula, it can also jump out of the local extremum and perform global optimization, which simply and effectively improves the training speed of the network. The adaptive wavelet neural network model of PSO algorithm has smaller error, faster convergence speed and stronger generalization ability. The wavelet function is used as the excitation function of the hidden layer, and the PSO algorithm is used to adjust the weight, stretching parameters, and translation parameters. The adaptive wavelet neural network model of the PSO algorithm has the advantages of simple algorithm, stable structure, and fast calculation convergence speed. , Strong global optimization ability, high recognition accuracy, and strong generalization ability.

5. The present invention adopts the metabolic GM (1, 1) trend model to predict the input parameters with a long time span. Using the metabolic GM(1, 1) trend model can predict the future time value according to the input parameter value, use the above method to predict the future time value of each input parameter, and add the future value of multiple input parameters to the metabolic GM(1, 1) In the original sequence of the trend model, the data at the beginning of the input sequence is correspondingly removed for modeling, and then the future value of the input parameter is predicted. By analogy, the future values of the output parameters of the metabolic GM(1,1) trend model are predicted. This method is called the equal-dimensional gray number supplementary model, which can realize long-term prediction and more accurately grasp the change trend of input parameter values.

Description of drawings

Fig. 1 is the water regime detection and control subsystem of the present invention;

Fig. 2 is a parameter detection module and a parameter prediction module of the present invention;

Fig. 3 is the water resources big data management subsystem of the Internet of Things of the present invention;

Fig. 4 is the hydrological parameter measurement terminal of the present invention;

Fig. 5 is the hydrological parameter control end of the present invention;

Fig. 6 is the hydrological gateway of the present invention;

Fig. 7 is the software structure of the field control terminal of the present invention.

Detailed ways

In conjunction with accompanying drawing 1-7, the technical solution of the present invention is further described:

1. Water regime detection and control subsystem steps:

1. Construct a parameter detection module, which consists of multiple noise reduction autoencoder neural network-NARX neural network models, adaptive AP clusterers, multiple PSO wavelet adaptive neural network models and ESN neural network models; multiple The measurement parameters output by the group parameter sensor for a period of time are respectively used as the input of the corresponding noise reduction autoencoder neural network-NARX neural network model, and the outputs of multiple noise reduction autoencoder neural network-NARX neural network models are used as the adaptive AP clusterer. Input, the adaptive AP clusterer outputs different types of noise reduction self-encoder neural network-NARX neural network model output values are respectively used as the input of the corresponding PSO wavelet adaptive neural network model, multiple PSO wavelet adaptive neural network The output of the model is used as the corresponding input of the ESN neural network model, and the output of the ESN neural network model is used as the output of the parameter detection module;

(1), Noise Reduction Autoencoder Neural Network-NARX Neural Network Model Design

Noise Reduction Autoencoder Neural Network-NARX Neural Network Model is the Noise Reduction Autoencoder Neural Network output as the input of the NARX Neural Network model, the Noise Reduction Autoencoder Neural Network is a dimensionality reduction method by training a multilayer neural network with a small central layer , to transform high-dimensional data into low-dimensional data. Denoising autoencoder neural network (DAE) is a typical three-layer neural network, which has an encoding process between the hidden layer and the input layer, and a decoding process between the output layer and the hidden layer. The noise reduction self-encoder neural network obtains the encoded representation (encoder) through the encoding operation of the input data, and obtains the reconstructed input data (decoder) through the output decoding operation of the hidden layer. The data in the hidden layer is the dimensionality reduction data. . Then a reconstruction error function is defined to measure the learning effect of the denoising autoencoder neural network. Constraints can be added based on the error function to generate various types of noise reduction autoencoder neural networks, encoders and decoders, and loss functions are as follows:

Encoder: h=δ(Wx+b) (1)

decoder:

Loss function:

为输出向量，δ为激励函数，一般使用Sigmoid函数或tanh函数。降噪自编码神经网络分为编码过程和解码过程，输入层到隐藏层为编码过程，隐藏层到输出层为解码过程。降噪自编码神经网络的目标是利用误差函数使输入和输出尽量相近，通过反向传播最小化误差函数，得到自编码网络最优的权值和偏置，为建立深度自编码网络模型做准备。根据自编码网络编码和解码原理，利用含有噪声测量数据得到编码数据和解码数据，通过解码数据和原始数据构造误差函数，通过反向传播最小化误差函数，得到最优的网络权值和偏置。通过加入噪声来破坏测量数据，然后将损坏的数据作为输入层输入到神经网络中。降噪自编码神经网络的重构结果应与测量参数原始数据近似，通过这种方法，可以消除测量参数扰动，并获得稳定的结构。原始测量参数通过加入噪声得到干扰输入，然后输入到编码器中得到特征表达，再通过解码器映射到输出层。NARX神经网络模型的当时输出不仅取决于过去的NARX神经网络模型输出y(t-n)，还取决于NARX神经网络模型当前的降噪自编码神经网络输出向量X(t)以及降噪自编码神经网络输出向量的延迟阶数等。NARX神经网络模型的降噪自编码神经网络输出通过时延层传递给隐层，隐层对降噪自编码神经网络输出的信号进行处理后传递到输出层，输出层将隐层输出信号做线性加权获得最终的神经网络输出信号，时延层将NARX神经网络模型反馈的信号和输入层的降噪自编码神经网络输出的信号进行延时，然后输送到隐层。NARX神经网络模型具有非线性映射能力、良好的鲁棒性和自适应性等特点，适宜对参数传感器输出预测值的多个高频波动部分进一步处理。NARX神经网络模型(Nonlinear Auto-Regression with External input neuralnetwork)是一种动态的前馈神经网络，NARX神经网络是一个有着外部输入的非线性自回归网络，它有一个多步时延的动态特性，并通过反馈连接封闭网络的若干层，NARX神经网络模型是非线性动态系统中应用最广泛的一种回归动态神经网络，其性能普遍优于全回归神经网络。NARX神经网络模型主要由输入层、隐层、输出层及输入和输出延时层构成，在应用前一般要事先确定输入和输出的延时阶数、隐层神经元个数，NARX神经网络模型的当时输出不仅取决于过去的NARX神经网络模型输出y(t-n)，还取决于NARX神经网络模型当前的粒子群优化自适应小波神经网络输出向量X(t)以及粒子群优化自适应小波神经网络输出向量的延迟阶数等。NARX神经网络模型的粒子群优化自适应小波神经网络输出通过时延层传递给隐层，隐层对粒子群优化自适应小波神经网络输出的信号进行处理后传递到输出层，输出层将隐层输出信号做线性加权获得最终的神经网络输出信号，时延层将NARX神经网络模型反馈的信号和输入层的粒子群优化自适应小波神经网络输出的信号进行延时，然后输送到隐层。NARX神经网络模型具有非线性映射能力、良好的鲁棒性和自适应性等特点。x(t)表示NARX神经网络模型的输入，即降噪自编码神经网络输出；m表示外部输入的延迟阶数；y(t)是神经网络的输出，即下一时段的降噪自编码神经网络输出的预测值；n是输出延迟阶数；s为隐含层神经元的个数；由此可以得到第j个隐含单元的输出为：The training process of AE is similar to that of the BP neural network. W and W' are weight matrixes, b and b' are biases, h is the output value of the hidden layer, x is the input vector,

is the output vector, and δ is the activation function, generally using the Sigmoid function or tanh function. The noise reduction self-encoder neural network is divided into an encoding process and a decoding process. The encoding process is from the input layer to the hidden layer, and the decoding process is from the hidden layer to the output layer. The goal of the noise reduction self-encoder neural network is to use the error function to make the input and output as close as possible, minimize the error function through backpropagation, and obtain the optimal weight and bias of the self-encoder network to prepare for the establishment of a deep self-encoder network model . According to the principle of self-encoding network encoding and decoding, the encoded data and decoded data are obtained by using the noise-containing measurement data, and the error function is constructed through the decoded data and the original data, and the error function is minimized through backpropagation to obtain the optimal network weight and bias. . The measured data is corrupted by adding noise, and the corrupted data is fed into the neural network as an input layer. The reconstruction result of the noise-reduced autoencoder neural network should be similar to the original data of the measured parameters. By this method, the disturbance of the measured parameters can be eliminated and a stable structure can be obtained. The original measurement parameters are obtained by adding noise to get the interference input, and then input into the encoder to obtain the feature expression, and then mapped to the output layer through the decoder. The current output of the NARX neural network model depends not only on the past NARX neural network model output y(tn), but also on the current noise reduction autoencoder neural network output vector X(t) of the NARX neural network model and the noise reduction autoencoder neural network The delay order of the output vector, etc. The output of the noise-reducing self-encoding neural network of the NARX neural network model is transmitted to the hidden layer through the delay layer. The hidden layer processes the signal output by the noise-reducing self-encoding neural network and then passes it to the output layer. The final neural network output signal is obtained by weighting, and the delay layer delays the signal fed back by the NARX neural network model and the signal output by the noise reduction self-encoding neural network of the input layer, and then sends it to the hidden layer. The NARX neural network model has the characteristics of nonlinear mapping ability, good robustness and adaptability, and is suitable for further processing of multiple high-frequency fluctuations in the output prediction values of parameter sensors. The NARX neural network model (Nonlinear Auto-Regression with External input neural network) is a dynamic feedforward neural network. The NARX neural network is a nonlinear auto-regressive network with external input. It has a dynamic characteristic of multi-step delay. And connect several layers of the closed network through feedback. The NARX neural network model is the most widely used regression dynamic neural network in nonlinear dynamic systems, and its performance is generally better than that of the full regression neural network. The NARX neural network model is mainly composed of an input layer, a hidden layer, an output layer, and an input and output delay layer. Before application, it is generally necessary to determine the delay order of the input and output, the number of neurons in the hidden layer, and the NARX neural network model. The current output depends not only on the past NARX neural network model output y(tn), but also on the current particle swarm optimization adaptive wavelet neural network output vector X(t) of the NARX neural network model and the particle swarm optimization adaptive wavelet neural network The delay order of the output vector, etc. The particle swarm optimization adaptive wavelet neural network output of the NARX neural network model is transmitted to the hidden layer through the delay layer. The hidden layer processes the signal output by the particle swarm optimization adaptive wavelet neural network and then passes it to the output layer. The output signal is linearly weighted to obtain the final neural network output signal. The delay layer delays the signal fed back by the NARX neural network model and the output signal of the particle swarm optimization adaptive wavelet neural network in the input layer, and then sends it to the hidden layer. The NARX neural network model has the characteristics of nonlinear mapping ability, good robustness and adaptability. x(t) represents the input of the NARX neural network model, that is, the output of the noise reduction self-encoder neural network; m represents the delay order of the external input; y(t) is the output of the neural network, that is, the noise reduction self-encoder neural The predicted value of the network output; n is the output delay order; s is the number of neurons in the hidden layer; thus, the output of the jth hidden unit can be obtained as:

In the above formula, w _ji is the connection weight between the i-th input and the j-th hidden neuron, and b _j is the bias value of the j-th hidden neuron.

(2), adaptive AP clusterer design

Adaptive AP clusterer completes clustering through "message passing" between data objects, mainly based on the similarity between data points (using negative Euclidean distance as the standard), using attractiveness responsibility and attribution The two kinds of messages of degree availability are cyclically updated and iterated, and finally the optimal clustering result is found. Assuming a data set X={x ₁ , x ₂ ,…, x _n } composed of N data points, the similarity between any two data points is:

Among them, the value on the main diagonal of simi(i, k) is replaced by the bias parameter value p, the larger p indicates that the probability of the point being selected as a representative point is greater. Therefore, the final number of clusters will change as p changes, and p is generally set as the median of simi(i, k) without prior knowledge. Define R(i, k) as the degree of attraction of candidate representative point k to each data point i, and A(i, k) as the degree to which data point i supports k as a representative point. The larger R(i, k)+A(i, k), the more likely the representative point k is the data center (exemplar). The specific adaptive AP clusterer algorithm flow is as follows:

A. The initial attractiveness R(i, k) and belongingness R(i, k) are zero matrices isomorphic to the similarity matrix simi(i, k).

B. Set p=-50, lamda=0.5, update R(i, k) and A(i, k) continuously until the constraint condition is met, and record the number of clusters as K1.

C. Let p=p-10, update R(i, k) and A(i, k) continuously until the constraint condition is met, and obtain a series of cluster numbers K2, 3, ... l (according to experience lmax=10) .

D. In steps B and C, if it is detected that the algorithm is oscillating and cannot converge, then lamda (value range 0.5-0.9) will eliminate the oscillating with a step size of 0.1 until the algorithm converges.

E. Use the silhouette coefficient index to evaluate the clustering quality and number of clusters obtained in steps B and C. The larger the index, the better the clustering quality, and the corresponding number of clusters K is the optimal number of clusters.

F. The adaptive AP clusterer mainly improves the accuracy and rapidity of the algorithm by adaptively adjusting the bias parameters and damping factors of the original AP clusterer. This algorithm uses the silhouette coefficient as the evaluation index of clustering effectiveness and clustering quality, uses the degree of oscillation as the index to judge whether the algorithm converges after oscillation, adaptively adjusts and obtains the optimal combination of bias parameters and damping factors, and finally obtains the optimal Clustering results.

(3), PSO wavelet adaptive neural network model design

Adaptive wavelet neural network of PSO The adaptive wavelet neural network is based on the wavelet theory, using a nonlinear wavelet base instead of the commonly used nonlinear Sigmoid function and combining with artificial neural network to propose a feedforward network, the hidden layer The transfer function of the method uses a wavelet function, and by adaptively adjusting the parameters of the wavelet function, the time-frequency characteristics of the input parameters can be extracted more effectively. It uses the wavelet function as the activation function of the neurons, and the expansion and translation factors of the wavelet, as well as the connection weight, are adaptively adjusted during the optimization of the error energy function. Assuming that the input signal of the wavelet neural network can be expressed as a one-dimensional vector x _i (i=1, 2, ..., n) of an input parameter, the output signal is expressed as y _k (k = 1, 2, ..., m), and the wavelet The formula for calculating the value of the output layer of the neural network is:

为小波基函数，b_j为小波基函数的平移因子，a_j小波基函数的伸缩因子，ω_jk为隐含层j节点和输出层k节点间的连接权值。本专利中的自适应小波神经网络的权值和阈值的修正算法采用梯度修正法来更新网络权值和小波基函数参数，从而使小波神经网络输出不断逼近期望输出。采用PSO的自适应小波神经网络，避免了梯度下降法中要求激活函数可微，以及对函数求导的过程计算，并且各个粒子搜索时迭代公式简单，因而计算速度又比梯度下降法快得多。而且通过对迭代公式中参数的调整，还能很好地跳出局部极值。初始化一群随机粒子，然后通过迭代找到最优解。在每一次迭代中，粒子通过跟踪二个“极值”来更新自己。第一个就是粒子本身所找到的最优解pbest，这个解称为个体极值；另一个是整个种群目前找到的最优解，这个解称为全局极值gbest。用PSO的自适应小波神经网络，就是首先将自适应小波神经网络的各种参数列为粒子的位置向量X，将均方误差能量函数式设为用于优化的目标函数，通过粒子群优算法的基本公式进行迭代，寻求最优解。PSO的自适应小波神经网络训练算法如下：In the formula, ω _ij is 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 is the expansion factor of the wavelet basis function, ω _jk is the connection weight between the hidden layer j node and the output layer k node. The weight and threshold correction algorithm of the self-adaptive wavelet neural network in this patent uses a gradient correction method to update the network weight and wavelet basis function parameters, so that the output of the wavelet neural network is constantly approaching the desired output. The adaptive wavelet neural network of PSO avoids the differentiable activation function in the gradient descent method and the process calculation of the function derivation, and the iteration formula is simple when each particle is searched, so the calculation speed is much faster than the gradient descent method. . And by adjusting the parameters in the iterative formula, it can also jump out of the local extremum well. Initialize a group of random particles, and then find the optimal solution through iteration. In each iteration, the particle updates itself by tracking two "extrema". The first is the optimal solution pbest found by the particle itself, which is called the individual extremum; the other is the optimal solution currently found by the entire population, and this solution is called the global extremum gbest. Using the adaptive wavelet neural network of PSO is to first list the various parameters of the adaptive wavelet neural network as the position vector X of the particle, set the mean square error energy function as the objective function for optimization, and use the particle swarm optimization algorithm The basic formula is iterated to find the optimal solution. The adaptive wavelet neural network training algorithm of PSO is as follows:

A. Initialize the network structure, determine the number of neurons in the hidden layer of the network, and determine the dimension D of the target search space.

B. Determine the number of particles m, initialize the position vector and velocity vector of the particles, bring the position vector and velocity vector of the particles into the iterative formula of the algorithm to update, use the error energy function as the objective function to perform optimization calculations, and record each particle so far The searched optimal position pbest and the optimal position gbest found so far by the entire particle swarm.

C. Search the entire particle swarm to the optimal position gbest so far, map it to network weights and thresholds for this learning, and use the error energy function as the fitness of the particles for calculation.

D. If the value of the error energy function is within the allowable error range of the actual problem, the iteration is complete; otherwise, switch back to the algorithm and continue the iteration.

(4), ESN neural network model design

The ESN neural network model (Echo state network, ESN) is a new type of dynamic neural network, which has all the advantages of the dynamic neural network. Better adapt to nonlinear system identification. The "reserve pool" is to transform the middle connection part of the traditional dynamic neural network into a randomly connected "reserve pool". The whole learning process is actually the process of learning how to connect the "reserve pool". The "reserve pool" is actually a randomly generated large-scale recursive structure in which neurons are sparsely connected to each other, and SD is usually used to represent the percentage of interconnected neurons in the total neurons N. The state equation of the ESN neural network model is:

为神经网络的输出偏差或可以代表噪声；f＝f[f₁，f₂，…，f_n]为“储备池”内部神经元的n个激活函数；f_i为双曲正切函数；f_out为ESN神经网络模型的ε个输出函数。In the formula, W is the weight matrix of the reserve pool, W _in is the input weight matrix; W _back is the feedback weight matrix; x(n) represents the internal state of the neural network; W _out is the kernel reserve pool of the ESN neural network model, neural The connection weight matrix between the input of the network and the output of the neural network;

is the output deviation of the neural network or can represent noise; f=f[f ₁ , f ₂ ,..., f _n ] is the n activation functions of neurons inside the "reserve pool"; f _i is the hyperbolic tangent function; f _out are the ε output functions of the ESN neural network model.

2. Build parameter prediction module design

Parameter prediction module includes parameter detection module, TDL beat delay line A, metabolic GM (1,1) trend model, NARX neural network model A, NARX neural network model B, TDL beat delay line B, TDL beat delay line C , TDL beat delay line D and BAM neural network-ANFIS adaptive neuro-fuzzy reasoning model of interval hesitant fuzzy numbers; the output of the parameter detection module is used as the input of TDL beat delay line A, and the output of TDL beat delay line A is used as The input of the metabolic GM (1, 1) trend model, the difference between the output of the TDL beat delay line A and the output of the metabolic GM (1, 1) trend model and the output of the metabolic GM (1, 1) trend model are respectively used as the NARX neural network model A And the input of NARX neural network model B, the output of NARX neural network model A and NARX neural network model B are respectively used as the input of TDL pressing delay line B and TDL pressing delay line C, TDL pressing delay line B, TDL pressing delay line Line C and the output of the BAM neural network-ANFIS adaptive neuro-fuzzy inference model of interval hesitant fuzzy numbers are respectively used as the corresponding input of the BAM neural network-ANFIS adaptive neuro-fuzzy inference model of interval hesitant fuzzy numbers, and the BAM neural network of interval hesitant fuzzy numbers The four parameters output by the network-ANFIS adaptive neuro-fuzzy inference model are a, b, c and d, respectively, a and b form the interval number [a, b] as the minimum value of the detected parameter, and c and d form the interval number [c, d] is the maximum value of the detected parameter, the interval number [a, b] and the interval number [c, d] constitute ([a, b], [c, d]) as the interval hesitation of the detected parameter Fuzzy number, interval hesitation fuzzy number BAM neural network-ANFIS adaptive neuro-fuzzy inference model outputs the interval hesitation fuzzy number of the detected parameter.

(1) Metabolic GM (1, 1) trend model design

The GM(1,1) gray prediction method has more advantages than the traditional statistical prediction method. It does not need to determine whether the predictor variables obey the normal distribution, does not need large sample statistics, and does not need to output according to the ESN neural network model. The prediction model can be changed at any time due to changes. Through the cumulative generation technology, a unified differential equation model is established, and the cumulative ESN neural network model outputs the original value to restore the prediction result. The differential equation model has higher prediction accuracy. The essence of establishing a GM (1, 1) trend model is to accumulate and generate the input raw data once, so that the generated sequence shows a certain law, and by establishing a differential equation model, a fitting curve is obtained to predict the trend of the output of the ESN neural network model. Make predictions. The present invention adopts the metabolic GM (1, 1) trend model to predict the time span of the output trend of the ESN neural network model, and the metabolic GM (1, 1) trend model can predict the future time output trend parameter value according to the ESN neural network model value, and use After the value of each output trend parameter predicted by the above method, the trend value of the output parameter at the predicted future time is added to the original sequence of output parameters of the ESN neural network model, and a data model at the beginning of the sequence is removed accordingly, and then the prediction is performed. A future output parameter trend value. By analogy, the trend value of the output parameter at the future time is predicted. This method is called the metabolic model, which can realize the trend prediction of the output parameters for a longer period of time, and can grasp the change trend of the output parameter values more accurately.

(2) Design of BAM neural network-ANFIS adaptive neuro-fuzzy reasoning model for interval hesitant fuzzy numbers

The BAM neural network-ANFIS adaptive neuro-fuzzy inference model of interval hesitant fuzzy numbers is the output of the BAM neural network as the input of the ANFIS adaptive neuro-fuzzy inference model. In the topology of the BAM neural network model, the initial mode of the network input is x(t), After being weighted by the weight matrix W ₁ , it reaches the output terminal y, after the nonlinear transformation of the transfer characteristic f _y of the output node and weighted by the W ₂ matrix, it returns to the input terminal x, and then passes through the non-linear transformation of the transfer characteristic f _x of the output node at the x terminal. The linear transformation becomes the output of the input terminal x, and this operation process is repeated, and the state transition equation of the BAM neural network model is shown in formula (8).

The neural network and fuzzy control are organically combined in the ANFIS adaptive neural reasoning model, which can not only play the advantages of both, but also make up for their respective shortcomings. The fuzzy membership function and fuzzy rules in the ANFIS adaptive neural network fuzzy system are obtained by learning a large amount of known data. The biggest feature of the ANFIS adaptive neural inference model is the modeling method based on data, not based on experience or is intuitively given arbitrarily. This is especially important for systems whose properties are not fully understood or whose properties are very complex. The main operation steps of the ANFIS adaptive neural reasoning model are as follows:

Layer 1: Fuzzify the input data, and the corresponding output of each node can be expressed as:

The formula n is the number of membership functions for each input, and the membership functions adopt Gaussian membership functions.

Layer 2: Realize rule operations, output the applicability of rules, and use multiplication for rule operations in the ANFIS adaptive neural inference model.

Layer 3: Normalize the applicability of each rule:

Layer 4: The transfer function of each node is a linear function, representing a local linear model, and the output of each adaptive node i is:

Layer 5: The single node of this layer is a fixed node, and the output of the ANFIS adaptive neural reasoning model is:

The conditional parameters that determine the shape of the membership function and the conclusion parameters of the inference rules in the ANFIS adaptive neural inference model can be trained through the learning process. The parameters are adjusted using the linear least squares estimation algorithm combined with the gradient descent algorithm. In each iteration of the ANFIS adaptive neural reasoning model, the input signal is transmitted along the forward direction of the network until the fourth layer. At this time, the conditional parameters are fixed, and the least square estimation algorithm is used to adjust the network parameters; the signal continues to be forwarded along the network until the output layer ( i.e. layer 5). The ANFIS adaptive neural inference model backpropagates the obtained error signal along the network, and uses the gradient method to update the conditional parameters. By adjusting the given condition parameters in the ANFIS adaptive neural inference model in this way, the global optimal point of the conclusion parameters can be obtained, which can not only reduce the dimension of the search space in the gradient method, but also improve the ANFIS adaptive neural inference model. The convergence rate of the parameter.

3. Water regime detection and control subsystem design

The water regime detection and control subsystem includes Elman neural network-NARX neural network model, fuzzy recurrent neural network-NARX neural network controller, AANN self-associative neural network model, PI controller-NARX neural network controller, PI controller, parameters Detection module and parameter prediction module;

(1), Elman neural network-NARX neural network model design

The Elman neural network output of the Elman neural network-NARX neural network model is used as the input of the NARX neural network model. The Elman neural network can be regarded as a forward neural network with local memory units and local feedback connections. In addition to the hidden layer, there are A special association layer. This layer receives feedback signals from the hidden layer, and each hidden layer node has a corresponding association layer node connection. The association layer takes the hidden layer state at the previous moment together with the network input at the current moment as the input of the hidden layer, which is equivalent to state feedback. The transfer function of the hidden layer is generally a Sigmoid function, the output layer is a linear function, and the association layer is also a linear function. In order to effectively solve the problem of approximation accuracy in the adjustment of water level parameters, the role of the association layer is enhanced. The current output of the NARX neural network model not only depends on the past NARX neural network model output y(t-n), but also depends on the current Elman neural network output vector X(t) of the NARX neural network model and the delay order of the Elman neural network output vector wait. The Elman neural network output of the NARX neural network model is transmitted to the hidden layer through the delay layer. The hidden layer processes the signal output by the Elman neural network and then passes it to the output layer. The output layer linearly weights the output signal of the hidden layer to obtain the final neural network. The output signal, the delay layer delays the signal fed back by the NARX neural network model and the signal output by the Elman neural network of the input layer, and then sends it to the hidden layer.

(2), fuzzy recurrent neural network-NARX neural network controller design

分别代表第k层的第i个节点的输入和输出，则网络内部的信号传递过程和各层之间的输入输出关系可以描述如下。第I层：输入层，该层的各输入节点直接与输入变量相连接，网络的输入和输出表示为：The fuzzy recurrent neural network-NARX neural network controller's fuzzy recurrent neural network output is used as the input of the NARX neural network controller. The fuzzy recurrent neural network consists of 4 layers: input layer, member function layer, rule layer and output layer. The network contains n input nodes, where each input node corresponds to m conditional nodes, m represents the number of rules, nm rule nodes, and 1 output node. The first layer introduces the input into the network; the second layer fuzzifies the input, and the membership function adopted is a Gaussian function; the third layer corresponds to fuzzy reasoning; the fourth layer corresponds to the defuzzification operation. use

represent the input and output of the i-th node in the k-th layer respectively, then the signal transmission process inside the network and the input-output relationship between each layer can be described as follows. Layer I: the input layer, each input node of this layer is directly connected to the input variable, and the input and output of the network are expressed as:

和

为网络输入层第i个节点的输入和输出，N表示迭代的次数。第II层：成员函数层，该层的节点将输入变量进行模糊化，每一个节点代表一个隶属函数，采用高斯基函数作为隶属函数，网络的输入和输出表示为：In the formula

and

is the input and output of the i-th node of the network input layer, and N represents the number of iterations. Layer II: membership function layer, the nodes in this layer fuzzify the input variables, each node represents a membership function, and the Gaussian function is used as the membership function, and the input and output of the network are expressed as:

In the formula, m _ij and σ _ij represent the mean center and width value of the jth Gaussian function of the i-th language variable in the second layer, respectively, and m is the number of all language variables corresponding to the input node. Layer III: The fuzzy reasoning layer, that is, the rule layer, adds dynamic feedback to make the network have better learning efficiency. The internal variable h _k is introduced in the feedback link, and the sigmoid function is selected as the activation function of the internal variable in the feedback link. The input and output of the network are expressed as:

是第三层的输出量，m表示完全连接时的规则数。反馈环节主要是计算内部变量的值和内部变量相应隶属函数的激活强度。该激活强度与第3层的规则节点匹配度相关。反馈环节引入的内部变量，包含两种类型的节点：承接节点和反馈节点，承接节点使用加权求和来计算内部变量实现去模糊化的功能；内部变量表示的隐藏规则的模糊推理的结果。反馈节点采用sigmoid函数作为模糊隶属度函数，实现内部变量的模糊化。第IV层：去模糊化层，即输出层。该层节点对输入量进行求和操作。网络的输入和输出表示为：In the formula, ω _jk is the connection weight of the recursive part. The neurons in this layer represent the antecedent part of the fuzzy logic rules. The nodes in this layer perform ∏ operation on the output of the second layer and the feedback of the third layer.

is the output of the third layer, and m represents the number of rules when fully connected. The feedback link is mainly to calculate the value of the internal variable and the activation intensity of the corresponding membership function of the internal variable. This activation strength is related to the level 3 regular node matching degree. The internal variables introduced in the feedback link include two types of nodes: the receiving node and the feedback node. The receiving node uses the weighted sum to calculate the internal variables to realize the function of defuzzification; the internal variable represents the result of the fuzzy reasoning of the hidden rules. The feedback node uses the sigmoid function as the fuzzy membership function to realize the fuzzification of internal variables. Layer IV: Defuzzification layer, namely the output layer. This layer node performs a summation operation on the input quantities. The input and output of the network are expressed as:

In the formula, λ _j is the connection weight of the output layer. The recurrent neural network has the performance of approaching a highly nonlinear dynamic system. The training error and test error of the recurrent neural network added with internal variables are respectively significantly reduced. The fuzzy recurrent neural network of this patent The weight of the neural network is trained by using the gradient descent algorithm with cross-validation. By introducing internal variables in the feedback link, the output of the rule layer is weighted and summed, and then the defuzzification output is used as the feedback amount, and the feedback amount and the membership The output of the degree function layer is used as the input of the next moment of the rule layer. The output of the fuzzy recurrent neural network contains the history information of the activation intensity and output of the rule layer, which enhances the ability of the fuzzy recurrent neural network to adapt to the nonlinear dynamic system. The current output of the NARX neural network model not only depends on the past NARX neural network model output y(tn), but also depends on the current fuzzy recurrent neural network output vector X(t) of the NARX neural network model and the delay of the fuzzy recurrent neural network output vector order etc. The output of the fuzzy recurrent neural network of the NARX neural network model is transmitted to the hidden layer through the delay layer. The hidden layer processes the output signal of the fuzzy recurrent neural network and then passes it to the output layer. The neural network outputs the signal, and the delay layer delays the signal fed back by the NARX neural network model and the signal output by the fuzzy recurrent neural network of the input layer, and then sends it to the hidden layer.

(3), AANN self-associative neural network model design

The AANN auto-associative neural network model is a feedforward auto-associative neural network (Auto-associative neural network, AANN) with a special structure. The AANN auto-associative neural network model structure includes an input layer, a certain number of hidden layers and a output layer. Firstly, through the parameter input layer, mapping layer and bottleneck layer, the compression of the output data information of multiple water level parameter detection modules is realized, and the output of multiple water level parameter detection modules is extracted from the high-dimensional parameter space output by multiple water level parameter detection modules. The most representative low-dimensional subspace of , and effectively filter out the noise and measurement errors in the output data of multiple water level parameter detection modules, and then realize multiple water level parameter detection modules through the bottleneck layer, demapping layer and output layer The decompression of the output and output restores the previously compressed information to each input value, thereby realizing the reconstruction of the output and output data of each multiple water level parameter detection modules. In order to achieve the purpose of compressing the output information of multiple water level parameter detection modules, the number of nodes in the bottleneck layer of the AANN self-associative neural network model is significantly smaller than that of the input layer. The simple single mapping among them, except that the excitation function of the output layer of AANN self-associative neural network adopts the linear function, the other layers all adopt the non-linear excitation function. In essence, the first layer of the hidden layer of the AANN self-associative neural network model is called the mapping layer, and the node transfer function of the mapping layer may be an S-type function or other similar nonlinear functions; the second layer of the hidden layer It is called the bottleneck layer. The dimension of the bottleneck layer is the smallest in the network. Its transfer function may be linear or nonlinear. The bottleneck layer avoids the easy-to-implement one-to-one mapping relationship between the output and the input. It Make the network encode and compress the output signals of multiple water level parameter detection modules, and decode and decompress the output data of multiple water level parameter detection modules after the bottleneck layer to generate estimated values of the output signals of multiple water level parameter detection modules; the hidden layer The third or last layer is called the demapping layer. The node transfer function of the demapping layer is usually a nonlinear S-type function. The AANN self-associative neural network is trained with the error back propagation algorithm.

(4), PI controller-NARX neural network controller design

PI Controller - The PI controller output of the NARX neural network controller is used as the input of the NARX neural network controller, and the current output of the NARX neural network model depends not only on the past NARX neural network model output y(t-n), but also depends on the NARX neural network Model the current PI controller output vector X(t) and the delay order of the PI controller output vector, etc. The PI controller output of the NARX neural network model is transmitted to the hidden layer through the delay layer. The hidden layer processes the signal output by the PI controller and then passes it to the output layer. The output layer linearly weights the output signal of the hidden layer to obtain the final neural network. The output signal, the delay layer delays the signal fed back by the NARX neural network model and the signal output by the PI controller of the input layer, and then sends it to the hidden layer.

2. Water resources big data management subsystem of the Internet of Things

The water resources big data management subsystem of the Internet of Things includes a hydrological parameter measurement terminal, an on-site monitoring terminal, a hydrological parameter control terminal, a hydrological gateway, a hydrological parameter cloud platform, and a hydrological monitoring mobile phone APP. The hydrological parameter measurement terminal is responsible for collecting the hydrological parameters of the detected water area Information, there is a hydrological detection and control subsystem in the on-site monitoring terminal. Through the hydrological gateway, the two-way communication between the hydrological parameter measurement terminal, the hydrological parameter control terminal, the on-site monitoring terminal, the hydrological parameter cloud platform and the hydrological monitoring mobile phone APP is realized, and the hydrological parameter is realized. Intelligent regulation. The hydrological monitoring mobile APP accesses the hydrological parameter cloud platform through the 5G network to realize remote monitoring of hydrological parameters, and the water regime detection and control subsystem is implemented in entitlement 1.

1. Design of the overall function of the system

The water resources big data management subsystem of the Internet of Things of the present invention realizes the detection and adjustment of hydrological parameters, and a plurality of hydrological parameter measurement terminals of the system are constructed into a wireless monitoring network in a self-organized manner to realize parameter measurement terminals, hydrological parameter Two-way wireless communication between the control terminal, the hydrological gateway, the on-site monitoring terminal, the hydrological parameter cloud platform and the hydrological monitoring mobile APP; the water level parameter measurement terminal sends the detected hydrological parameters to the on-site monitoring terminal and the hydrological parameter cloud platform through the hydrological gateway. The water regime detection and control subsystem processes and intelligently adjusts the hydrological parameters; on the hydrological monitoring mobile phone APP, the real-time monitoring of the hydrological parameters is realized by accessing the hydrological parameter cloud platform. The water resource big data management subsystem of the Internet of Things is shown in Figure 3.

2. Design of hydrological parameter measurement terminal

A large number of hydrological parameter measurement terminals based on wireless sensor networks are used as hydrological parameter sensing terminals. The hydrological parameter measurement terminal and the hydrological parameter control terminal realize the information interaction between the on-site monitoring terminal, the hydrological gateway and the hydrological parameter cloud platform through the self-organizing wireless network. The hydrological parameter measurement terminal includes the water level, water flow rate, rainfall, PH value, water temperature, dissolved oxygen sensor group and other sensor groups that affect the hydrological parameters and the corresponding signal conditioning circuit, STM32 microprocessor, GPS module, camera and GPRS wireless transmission module; relying on high-precision GPS positioning technology, establish river geographic graphics, monitor the coordinate information of lead fish in real time, automatically control the rotation angle of camera and searchlight, monitor the upstream and downstream of the section, and self-tracking video surveillance can track in real time lead fish position or manually track lead fish position. The software of the hydrological parameter measurement terminal mainly realizes the wireless communication and the collection and preprocessing of hydrological parameters. The software adopts C language programming, which has a high degree of compatibility, which greatly improves the work efficiency of software design and development, and enhances the reliability, readability and portability of program codes. The terminal structure of hydrological parameter measurement is shown in Figure 4.

3. Design of hydrological parameter control terminal

The hydrological parameter control terminal includes STM32 single-chip microcomputer, GPRS wireless transmission module and pumping device. The water regime detection and control subsystem is designed in the on-site monitoring terminal to realize the detection and adjustment of hydrological parameters. The software is designed in C language, with a high degree of compatibility. The working efficiency of software design and development is improved, and the reliability, readability and portability of program codes are enhanced. The structure of the hydrological parameter control terminal is shown in Figure 5.

3. Hydrological gateway design

The hydrological gateway includes a GPRS wireless transmission module, NB-IoT module, STM32 microcontroller and RS232 interface. The hydrological gateway includes a GPRS wireless transmission module to realize the self-organizing communication network between the hydrological parameter measurement terminal and the hydrological control terminal. The NB-IoT module realizes The two-way data interaction between the gateway and the hydrological parameter cloud platform, the hydrological parameter control terminal, the on-site control terminal and the hydrological monitoring mobile APP, the RS232 interface is connected to the on-site monitoring terminal to realize the information interaction between the gateway and the on-site monitoring terminal. The hydrological gateway is shown in Figure 6.

4. On-site monitoring software

The on-site monitoring terminal is an industrial control computer. The on-site monitoring terminal mainly realizes the collection of hydrological parameters and the intelligent adjustment of hydrological parameters, and realizes the information between the parameter measurement terminal and the hydrological parameter control terminal, as well as the hydrological parameter cloud platform and the hydrological monitoring mobile APP. Interaction, the main functions of the on-site monitoring terminal are communication parameter setting, water resources data analysis and data management, water regime detection and control subsystem. The management software selects Microsoft Visual++6.0 as the development tool, and calls the system's Mscomm communication control to design the communication program. The software functions of the on-site monitoring terminal are shown in Figure 7.

The technical means disclosed in the solutions of the present invention are not limited to the technical means disclosed in the above embodiments, but also include technical solutions composed of any combination of the above technical features. It should be pointed out that for those skilled in the art, some 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 (7)

1. The hydrological parameter intelligent monitoring and water resource big data management system is characterized by comprising a hydrological condition detection and control subsystem and a water resource big data management subsystem of the Internet of things, and the hydrological parameter intelligent detection, adjustment and water resource big data management are realized;

the regimen detection and control subsystem comprises an Elman neural network-NARX neural network model, a fuzzy recurrent neural network-NARX neural network controller, an AANN auto-associative neural network model, a PI controller-NARX neural network controller, a PI controller, a parameter detection module and a parameter prediction module.

2. The intelligent hydrological parameter monitoring and water resource big data management system according to claim 1, wherein: the rainfall, the water flow and the water level sensor output of a plurality of groups of upstream water areas are used as the input of a parameter prediction module, the water level set value, the parameter prediction module output and the AANN self-association neural network model output are respectively used as the corresponding input of an Elman neural network-NARX neural network model, the difference between the output of the Elman neural network-NARX neural network model and the output of the AANN self-association neural network model is used as a water level error, the water level error and the error change rate are used as the input of a fuzzy recurrent neural network-NARX neural network controller, each downstream water area water flow sensor group is used as the input of a corresponding parameter detection module, the difference between the output of the fuzzy recurrent neural network-NARX neural network controller and the output of a corresponding PI controller and the output of a corresponding parameter detection module is used as a water flow error, the water flow error is used as the input of a corresponding PI controller-NARX neural network controller, and the PI controller-NARX neural network controller outputs as the control quantity of a corresponding water pumping device; the water level sensor group of each downstream water area is used as the input of a corresponding parameter detection module, the difference between a water level set value and the output of the parameter detection module is used as the water level difference, the water level difference and the change rate of the water level difference are used as the input of a corresponding PI controller, the output of a plurality of parameter detection modules is used as the corresponding input of an AANN auto-associative neural network model, and the water situation detection and control subsystem realizes the detection and control of the water level and the water flow of multiple areas.

3. The intelligent hydrologic parameter monitoring and water resource big data management system of claim 1, characterized in that: the parameter detection module is composed of a noise reduction self-coding neural network-NARX neural network model, a self-adaptive AP (access point) clustering device, a PS0 wavelet self-adaptive neural network model and an ESN neural network model, measurement parameters output by a parameter sensor for a period of time are respectively used as the input of the corresponding noise reduction self-coding neural network-NARX neural network model, the output of the noise reduction self-coding neural network-NARX neural network model is used as the input of the self-adaptive AP clustering device, the output values of different types of noise reduction self-coding neural network-NARX neural network models are respectively used as the input of the corresponding PS0 wavelet self-adaptive neural network model, the output of the PSO wavelet self-adaptive neural network model is used as the corresponding input of the ESN neural network model, and the output of the ESN neural network model is used as the output of the parameter detection module.

4. The intelligent hydrologic parameter monitoring and water resource big data management system of claim 1, characterized in that: the parameter prediction module comprises a parameter detection module, a TDL beat-to-beat delay line A, a metabolism GM (1, 1) trend model, a NARX neural network model A, a NARX neural network model B, a TDL beat-to-beat delay line C, a TDL beat-to-beat delay line D and a BAM neural network-ANFIS adaptive neural fuzzy inference model of interval hesitation fuzzy numbers, wherein the output of the parameter detection module is used as the input of the TDL beat-to-beat delay line A, the output of the TDL beat-to-beat delay line A is used as the input of a metabolism GM (1, 1) trend model, the difference between the output of the TDL beat-to-beat delay line A and the output of the metabolism GM (1, 1) trend model are respectively used as the input of the NARX neural network model A and the NARX neural network model B, the output of the NARX neural network model A and the output of the NARX neural network model B are respectively used as the input of a TDL beat-to-beat delay line B and a TDL beat-to-beat delay line C, the output of the BAM neural network-ANFIS adaptive neural fuzzy inference model of the TDL beat-to-beat delay line B, the TDL beat-to-beat delay line C and the interval hesitation fuzzy number is respectively used as the corresponding input of the BAM neural network-ANFIS adaptive neural fuzzy inference model of the interval hesitation fuzzy number, 4 parameters output by the BAM neural network-ANFIS adaptive neural fuzzy inference model of the interval hesitation fuzzy number are respectively a, B, C and D, and the BAM neural network-ANFIS adaptive neural fuzzy inference model of the interval hesitation fuzzy number outputs the interval hesitation fuzzy number of the detected parameters.

5. The intelligent hydrologic parameter monitoring and water resource big data management system of claim 4, characterized in that: the a and b form interval number [ a, b ] as the minimum value of the detected parameter, the c and d form interval number [ c, d ] as the maximum value of the detected parameter, and the interval number [ a, b ] and the interval number [ c, d ] form ([ a, b ], [ c, d ]) as the interval hesitation fuzzy number of the detected parameter.

6. The intelligent hydrologic parameter monitoring and water resource big data management system of claim 1, characterized in that: the big data management subsystem of water resource of thing networking is responsible for gathering the hydrology parameter information who is detected the waters for hydrology parameter measurement end, has the regimen to detect and control subsystem in the on-the-spot monitoring end, realizes hydrology parameter measurement end, hydrology parameter control end, on-the-spot monitoring end, hydrology parameter cloud platform and hydrology monitoring cell-phone APP's two-way communication through the hydrology gateway, realizes the intelligent regulation of hydrology parameter.

7. The intelligent hydrological parameter monitoring and water resource big data management system according to claim 6, wherein: the hydrological parameter measuring end comprises a water level, water flow, rainfall, a PH value, a water temperature and dissolved oxygen sensor group for collecting hydrological parameters, a corresponding signal conditioning circuit, an STM32 microprocessor, a GPS module, a camera and a GPRS wireless transmission module.

Images