CN116151121A - A Neural Network-Based NH4-N Soft-Sensing Method for Effluent Water - Google Patents

Abstract

The invention provides a neural network-based effluent NH4-N soft measurement method, which comprises the following steps: initializing a network structure and network parameters of the constructed echo state network; constructing a sub-reserve pool according to a singular value decomposition method based on the initialized echo state network; optimizing the sub-pool according to the condition number and the differential evolution algorithm; updating the optimized weight, input weight and state matrix of the sub-reserve pool, judging whether the iteration times are smaller than a preset iteration threshold, if so, jumping to the step of constructing the sub-reserve pool according to a singular value decomposition method based on the initialized echo state network; if yes, calculating an output weight; testing the echo state network according to the output weight and the test sample to obtain a determined effluent NH4-N detection model; and inputting the data to be detected into a determined effluent NH4-N detection model to obtain a detection result. The invention provides a high-efficiency and rapid solution for measuring key water quality parameters in the sewage treatment process.

Description

A soft-sensing method for effluent NH4-N based on neural network

Technical Field

The invention relates to the technical field of water treatment, and in particular to a soft measurement method of effluent NH4-N based on neural network.

Background Art

With the rapid development of urbanization and industrialization in today's society, my country's water environment has been severely damaged. Sewage discharge not only seriously affects the daily life of residents, but also destroys the ecological balance of nature. In order to reduce the discharge of sewage and realize the recycling of water, sewage treatment plants have been established all over the country. In the sewage treatment process, NH4-N concentration is an important parameter to measure the performance of sewage treatment process (WWTP). However, since the sewage treatment process is a complex system with high nonlinearity, large hysteresis, large time variation, multivariate coupling and other characteristics, and the maintenance cost is high, its prediction is still an unresolved problem. Therefore, how to predict the concentration of effluent NH4-N at low cost and high efficiency is very necessary for the assessment of effluent water quality and the stable operation of sewage treatment plants.

Summary of the invention

In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a soft measurement method for effluent NH4-N based on neural network. The present invention combines the differential evolution algorithm with condition number analysis to construct a new echo state network, which can predict the ammonia nitrogen concentration in the sewage treatment process.

To achieve the above object, the present invention provides the following solutions:

A soft measurement method for effluent NH4-N based on neural network, comprising:

Constructing an echo state network, and initializing the network structure and network parameters of the echo state network; the input variables of the echo state network include inlet water temperature, total suspended solids, dissolved oxygen concentration, pH value, and outlet water redox potential; the output variables of the echo state network include ammonia nitrogen concentration;

Based on the initialized echo state network, a sub-reservoir is constructed according to the singular value decomposition method;

Optimizing the sub-reservoir according to a condition number and a differential evolution algorithm;

The weights, input weights and state matrix of the optimized sub-reserve pool are updated, and it is determined whether the number of iterations is less than a preset iteration threshold. If so, jump to the step of "constructing a sub-reserve pool based on the initialized echo state network according to the singular value decomposition method"; if so, calculate the output weights;

The echo state network is tested according to the output weights and the test samples to obtain a determined effluent NH4-N detection model;

The data to be tested is input into the determined effluent NH4-N detection model to obtain the test results.

Preferably, initializing the network structure and network parameters of the echo state network includes:

Determine the initial structure of the echo state network as 5-N-1; wherein N represents the number of nodes in the sub-reservoir; and the size of N is gradually increased;

其中，u_k表示第k组输入样本，t_k表示第k组实际输出值，

表示输入样本维度为n，L为样本总数；The sigmoid function is used as the network activation function G(·), the initial iteration number i＝1, the maximum iteration number i _max ≤30, and the training sample

Among them, _uk represents the kth group of input samples, _tk represents the kth group of actual output values,

Indicates that the input sample dimension is n, and L is the total number of samples;

Randomly initialize the network input weights W ⁱⁿ and the internal weights W of the reserve pool between (0,1).

Preferably, the constructing of the sub-reservoir based on the initialized echo state network according to the singular value decomposition method comprises:

以及两个正交矩阵U_i和V_i；其中，

为随机生成的在(0,1)之间数值，n_k为子储备池节点个数；Randomly generate a diagonal matrix

And two orthogonal matrices U _i and V _i ; where,

is a randomly generated value between (0,1), and n _k is the number of nodes in the sub-reservation pool;

The sub-reservoir ΔW _i =U _i S _i V _i is constructed using the diagonal matrix and the two orthogonal matrices.

Preferably, the optimizing the sub-reserve pool according to the condition number and the differential evolution algorithm comprises:

Construct a fitness function fitness(ΔW _i )=κ(H _i )-κ(H _i-1 ); where κ(H _i ) represents the condition number of the state matrix corresponding to the reserve pool after the new sub-reserve pool is added;

其中n∈{1,2…,NP}，并设置最大迭代次数G；Initialize the initial population NP = 50, mutation operator F = 0.7, crossover operator CR = 0.5, and initialize the population

Where n∈{1,2…,NP}, and set the maximum number of iterations G;

得到变异个体μⁿ；其中，r₁,r₂,r₃∈[1,NP]为三个互不相同的整数；According to the formula

Get the mutant individual μ ⁿ ; where r ₁ , r ₂ , r ₃ ∈[1,NP] are three different integers;

完成交叉操作，对于每一个独立个体

得到交叉个体

其中，j∈{1,2…,10}，rand∈(0,1)是随机生成的；pass

Complete the crossover operation, for each independent individual

Get crossover individuals

Among them, j∈{1,2…,10}, rand∈(0,1) is randomly generated;

When the algorithm reaches the maximum number of iterations, the algorithm process is terminated, the optimal individual is returned, and a new sub-reserve pool is constructed using the optimal individual.

Preferably, the weights, input weights and state matrix of the optimized sub-reserve pool are updated, and it is determined whether the number of iterations is less than a preset iteration threshold. If so, the process jumps to the step of "constructing a sub-reserve pool based on the initialized echo state network according to a singular value decomposition method"; if so, the output weights are calculated, including:

Update the reserve pool weight to

Update the input weights to

Update the state matrix to H = [H ₁ ,…,H _i ];

If i≥i _max is satisfied, the next step is executed, otherwise the process returns to the step "constructing a sub-reservoir based on the initialized echo state network according to the singular value decomposition method";

The output weight W ^out is calculated by the formula W ^out =H ^Γ T, wherein Γ represents pseudo-inverse calculation, and T is the target output data.

According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

The present invention provides a soft measurement method for effluent NH4-N based on a neural network, comprising: constructing an echo state network, and initializing the network structure and network parameters of the echo state network; the input variables of the echo state network include inlet water temperature, total suspended solids, dissolved oxygen concentration, pH value and effluent redox potential; the output variables of the echo state network include ammonia nitrogen concentration; based on the initialized echo state network, constructing a sub-reserve pool according to a singular value decomposition method; optimizing the sub-reserve pool according to a condition number and a differential evolution algorithm; updating the weights, input weights and state matrix of the optimized sub-reserve pool, and judging whether the number of iterations is less than a preset iteration threshold, if so, jumping to the step of "constructing a sub-reserve pool according to a singular value decomposition method based on the initialized echo state network"; if so, calculating the output weight; testing the echo state network according to the output weight and a test sample to obtain a determined effluent NH4-N detection model; inputting the data to be detected into the determined effluent NH4-N detection model to obtain a detection result. The present invention provides an efficient and rapid solution for measuring key water quality parameters in the sewage treatment process.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

FIG1 is a flow chart of a method provided by an embodiment of the present invention;

FIG2 is a topological structure diagram of a neural network provided by an embodiment of the present invention;

FIG3 is a distribution diagram of absolute values of output weights provided by an embodiment of the present invention;

FIG4 is a diagram showing the prediction results of effluent NH4-N concentration provided by an embodiment of the present invention;

FIG5 is a diagram showing an error in prediction of effluent NH4-N concentration according to an embodiment of the present invention.

DETAILED DESCRIPTION

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

The terms "first", "second", "third" and "fourth" in the specification and claims of the present application and the drawings are used to distinguish different objects rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a series of steps, processes, methods, etc. are not limited to the listed steps, but may optionally include steps that are not listed, or may optionally include other step elements inherent to these processes, methods, products or devices.

The purpose of the present invention is to provide a soft measurement method for effluent NH4-N based on neural network, which can provide an efficient and fast solution for the measurement of key water quality parameters in the sewage treatment process.

In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

FIG1 is a flow chart of a method provided by an embodiment of the present invention. As shown in FIG1 , the present invention provides a soft measurement method for effluent NH4-N based on a neural network, comprising:

Step 100: construct an echo state network, and initialize the network structure and network parameters of the echo state network; the input variables of the echo state network include inlet water temperature, total suspended solids, dissolved oxygen concentration, pH value and outlet water redox potential; the output variables of the echo state network include ammonia nitrogen concentration;

Step 200: Based on the initialized echo state network, construct a sub-reservoir according to a singular value decomposition method;

Step 300: Optimizing the sub-reservoir according to the condition number and differential evolution algorithm;

Step 400: Update the weights, input weights and state matrix of the optimized sub-reservoir, and determine whether the number of iterations is less than a preset iteration threshold. If so, jump to the step of "constructing a sub-reservoir based on the initialized echo state network according to the singular value decomposition method"; if so, calculate the output weights;

Step 500: testing the echo state network according to the output weights and the test samples to obtain a determined outlet water NH4-N detection model;

Step 600: input the data to be detected into the determined effluent NH4-N detection model to obtain the detection result.

Preferably, initializing the network structure and network parameters of the echo state network includes:

Determine the initial structure of the echo state network as 5-N-1; wherein N represents the number of nodes in the sub-reservoir; and the size of N is gradually increased;

其中，u_k表示第k组输入样本，t_k表示第k组实际输出值，

表示输入样本维度为n，L为样本总数；The sigmoid function is used as the network activation function G(·), the initial iteration number i＝1, the maximum iteration number i _max ≤30, and the training sample

Among them, _uk represents the kth group of input samples, _tk represents the kth group of actual output values,

Indicates that the input sample dimension is n, and L is the total number of samples;

Randomly initialize the network input weights W ⁱⁿ and the internal weights W of the reserve pool between (0,1).

Specifically, the effluent NH4-N soft measurement method based on the CNEESN neural network in this embodiment is a continuously growing network, and the main operation process is as follows: first, a small-scale reserve pool is generated and the corresponding weights are initialized, and then the singular values randomly generated by the singular value decomposition (SVD) method are optimized according to the condition number analysis and differential evolution algorithm, and a new sub-reserve pool is constructed using the optimized singular values, and then the new sub-reserve pool is added to the network, as shown in Figure 2. This embodiment uses condition number analysis and differential evolution algorithms to make the condition number of the network's reserve pool as small as possible after each sub-reserve pool is added to the network, so as to train to obtain a better output weight, and then predict the ammonia nitrogen concentration in the sewage treatment process. This embodiment includes the following steps:

Step 1: Initialize network structure and parameters

Step 1.1: Initialize the network structure

Taking the inlet water temperature, total suspended solids, dissolved oxygen concentration, pH value and outlet redox potential as input variables and ammonia nitrogen concentration as output variable, the initial structure of the echo state network is determined to be 5-N-1, where N represents the number of sub-reservoir nodes. The number of reserve pool nodes N of a typical echo state network is 50≤N≤1000, but the model proposed by this method is a growth model, and the size of N gradually increases. The initial N in this network is 10, that is, the network contains 5 input nodes, 10 reserve pool nodes, and 1 output node.

Step 1.2: Initialize network parameters

u_k表示第k组输入样本，t_k表示第k组实际输出值，

表示输入样本维度为n，L为样本总数；随机初始化网络输入权值Wⁱⁿ和储备池内部权值W在(0,1)之间。The sigmoid function is used as the network activation function G(·), the initial iteration number i＝1, the maximum iteration number i _max ≤30, and the training sample

u _k represents the kth group of input samples, t _k represents the kth group of actual output values,

Indicates that the input sample dimension is n, L is the total number of samples; the network input weight W ⁱⁿ and the internal weight W of the reserve pool are randomly initialized between (0,1).

Preferably, the constructing of the sub-reservoir based on the initialized echo state network according to the singular value decomposition method comprises:

以及两个正交矩阵U_i和V_i；其中，

为随机生成的在(0,1)之间数值，n_k为子储备池节点个数；Randomly generate a diagonal matrix

And two orthogonal matrices U _i and V _i ; where,

is a randomly generated value between (0,1), and n _k is the number of nodes in the sub-reservation pool;

The sub-reservoir ΔW _i =U _i S _i V _i is constructed using the diagonal matrix and the two orthogonal matrices.

以及两个正交矩阵U_i和V_i,利用三个矩阵可以构造子储备池ΔW_i＝U_iS_iV_i。通过这种方法可以确保ΔW_i和S_i有相同的奇异值，从而保证ESP特性。Optionally, step 2 of this embodiment is to construct a sub-reserve pool according to singular value decomposition. This embodiment randomly generates a diagonal matrix

And two orthogonal matrices U _i and V _i , using three matrices, the sub-reservoir ΔW _i = U _i S _i V _i can be constructed. This method can ensure that ΔW _i and S _i have the same singular value, thus ensuring the ESP characteristics.

Preferably, the optimizing the sub-reserve pool according to the condition number and the differential evolution algorithm comprises:

Construct a fitness function fitness(ΔW _i )=κ(H _i )-κ(H _i-1 ); where κ(H _i ) represents the condition number of the state matrix corresponding to the reserve pool after the new sub-reserve pool is added;

其中n∈{1,2…,NP}，并设置最大迭代次数G；Initialize the initial population NP = 50, mutation operator F = 0.7, crossover operator CR = 0.5, and initialize the population

Where n∈{1,2…,NP}, and set the maximum number of iterations G;

得到变异个体μⁿ；其中，r₁,r₂,r₃∈[1,NP]为三个互不相同的整数；According to the formula

Get the mutant individual μ ⁿ ; where r ₁ , r ₂ , r ₃ ∈[1,NP] are three different integers;

完成交叉操作，对于每一个独立个体

得到交叉个体

其中，j∈{1,2…,10}，rand∈(0,1)是随机生成的；pass

Complete the crossover operation, for each independent individual

Get crossover individuals

Among them, j∈{1,2…,10}, rand∈(0,1) is randomly generated;

When the algorithm reaches the maximum number of iterations, the algorithm process is terminated, the optimal individual is returned, and a new sub-reserve pool is constructed using the optimal individual.

Specifically, step 3 in this embodiment is to optimize the sub-reserve pool according to the condition number and the differential evolution algorithm, and the specific steps are as follows:

Step 3.1: Design fitness function According to the theoretical analysis of matrix condition number, we know that the larger the condition number of a matrix, the easier it is to produce ill-conditioned solutions when performing matrix operations. Therefore, this embodiment designs a fitness function to measure the performance of the newly generated sub-reserve pool. The specific calculation method is as follows:

fitness(ΔW _i )=κ(H _i )-κ(H _i-1 ) (1)

Where κ(H _i ) represents the condition number of the state matrix corresponding to the reserve pool after the new sub-reservoir is added.

Step 3.2: Use differential evolution algorithm to optimize the singular values generated by SVD

其中n∈{1,2…,NP}，并设置最大迭代次数G。Initialize the initial population NP = 50, mutation operator F = 0.7, crossover operator CR = 0.5. Initialize the population

Where n∈{1,2…,NP}, and the maximum number of iterations G is set.

The mutant individual μ ⁿ is obtained by formula (2):

Where r ₁ , r ₂ , r ₃ ∈[1,NP] are three different integers.

The crossover operation is completed by formula (3), and for each independent individual σ _j n, the crossover individual γ _j n is obtained

Among them, j∈{1,2…,10}, rand∈(0,1) is randomly generated.

The selection operation is completed by formula (4)

fitness(·) represents the fitness function calculation in formula (1).

After the algorithm reaches the maximum number of iterations, the algorithm process is terminated and the optimal individual is returned. This individual is used to construct a new sub-reserve pool.

Preferably, the weights, input weights and state matrix of the optimized sub-reserve pool are updated, and it is determined whether the number of iterations is less than a preset iteration threshold. If so, the process jumps to the step of "constructing a sub-reserve pool based on the initialized echo state network according to a singular value decomposition method"; if so, the output weights are calculated, including:

Update the reserve pool weight to

Update the input weights to

Update the state matrix to H = [H ₁ ,…,H _i ];

If i≥i _max is satisfied, the next step is executed, otherwise the process returns to the step "constructing a sub-reservoir based on the initialized echo state network according to the singular value decomposition method";

The output weight W ^out is calculated by the formula W ^out =H ^Γ T, wherein Γ represents pseudo-inverse calculation, and T is the target output data.

输入权值更新为

状态矩阵更新为H＝[H₁,…,H_i]。Furthermore, step 4 in this embodiment is to update the structure and parameters of the network. In this embodiment, the weight of the reserve pool is updated to

The input weights are updated to

The state matrix is updated to H = [H ₁ ,…,H _i ].

If i≥i _max is satisfied, execute step 5, otherwise return to step 2.

Furthermore, as shown in FIG3 , step 5 in this embodiment: calculating the output weight. The output weight can be calculated by formula (5):

W ^out =H ^Γ T (5)

Where Γ represents pseudo-inverse calculation and T is the target output data.

In addition, step 6 of this embodiment is to test the network, using the output weight W ^out obtained in the above steps, input the test sample, and test the network. This embodiment applies the tested network to the prediction of effluent NH4-N concentration, and the results and errors are shown in Figures 4 and 5 respectively.

The data samples in this embodiment are as follows, where Tables 1-12 are experimental data of the present invention. Tables 1-5 are training input samples: inlet water temperature, dissolved oxygen at the end of aerobic stage, total suspended solids at the end of aerobic stage, pH value of effluent, and redox potential of effluent, Table 6 is the concentration of ammonia nitrogen in the effluent of the training sample, Tables 7-11 are test input samples: inlet water temperature, dissolved oxygen at the end of aerobic stage, total suspended solids at the end of aerobic stage, pH value of effluent, and redox potential of effluent, and Table 12 is the concentration of ammonia nitrogen in the effluent of the test sample.

Table 1 Auxiliary variables Inlet water temperature (℃)

Table 2 Auxiliary variables dissolved oxygen (mg/L)

Table 3 Auxiliary variables Total suspended solids (mg/L)

Table 4 Auxiliary variables pH value

Table 5 Auxiliary variables Redox potential

Table 6 Measured effluent NH4-N concentration (mg/L)

The test samples are as follows:

Table 7 Auxiliary variables Inlet water temperature (℃)

26.6664 25.5925 26.0751 26.8655 24.9307 24.9436 25.2516 25.8255 24.9177 25.4691 25.6463 23.6239 26.7961 23.3835 25.5664 25.6231 23.6806 24.1833 25.5388 25.7410 25.9991 25.5576 24.9465 24.9725 24.7418 27.2087 25.8663 26.7136 24.9061 25.6696 24.6813 23.2770 23.8631 24.9667 26.8065 24.4801 24.8874 25.4850 22.9625 25.2472 25.9962 27.1094 25.6289 25.4081 24.2291 25.4720 27.2028 25.3994 25.5649 24.6698 24.9018 24.5476 25.3617 23.7378 24.3022 24.9840 22.8098 25.0100 25.2979 25.0303 27.0784 24.2721 24.4198 24.9826 25.6667 23.0559 23.7307 25.4778 25.3893 25.5126 25.6725 25.4067 25.0534 23.1565 25.0881 24.9119 24.9667 24.9480 24.8686 26.9098 25.9305 23.2841 25.3486 25.2993 24.5188 25.4371 24.9480 27.2933 25.9845 25.4618 25.2212 27.0562 23.1027 24.8614 25.0852 24.5591 25.4153 25.6260 26.9349 25.3501 25.3486 24.3796 25.2936 23.6253 24.5404 24.2047 26.7917 24.6051 25.6158 24.6368 22.9115 24.9047 25.2559 26.5046 27.1331 25.9641 24.9999 26.0429 23.6295 24.6698 24.7908 24.7490 26.0283 23.0630 25.4952 25.1589 23.2032 23.5598 25.6522 23.3310 25.5402 23.1551 23.8745 24.6152 26.7858 25.1633 25.9436 23.6295 25.1532 25.8255 24.8052 25.2950 25.1778 23.9902 27.3334 27.1880 23.4745 26.9556 25.3399 23.4048 25.9539 26.8153 25.6740 25.4458 26.0400 25.1315 24.8225 24.9494 23.4318 25.5053 26.6723 26.8212 23.0956 25.4981 25.2299 23.5769 23.6096 23.1381 23.7006 25.5068 23.5114 25.6405 25.1488 23.8717 26.9763 27.2147 26.9526 25.1040 23.6422 25.1285 25.1300 23.8477 23.4190 23.0191 24.9595 24.1218 23.6338 25.2849 23.6295 26.7652

Table 8 Auxiliary variables Dissolved oxygen (mg/L)

0.0562 0.0366 0.0480 0.0492 0.0497 0.0355 0.0298 0.0680 0.0501 1.4971 0.0307 0.9334 0.0395 0.3574 0.0322 0.0480 1.1828 0.4821 0.0792 0.0507 0.0692 0.0401 4.8300 0.0371 0.0606 0.0518 0.0811 0.0602 0.0542 0.0325 0.0450 0.3821 0.8550 0.0382 0.0559 0.3764 5.7695 1.5715 1.0503 0.0367 0.1061 0.0507 0.0311 0.2692 0.4867 0.0310 0.0581 0.0504 0.0509 0.0501 0.0457 0.3747 0.0375 0.2519 0.4357 0.0396 0.2746 0.0319 0.0336 3.3770 0.0644 0.4701 5.4844 0.0391 0.0305 0.5861 0.2103 0.0344 0.5381 0.3484 0.0391 2.2688 0.0882 0.2611 0.0321 0.0716 0.0659 0.0359 0.0427 0.0656 0.0531 0.3268 0.0311 0.0296 5.3460 0.1882 0.0288 0.0394 0.0860 0.1166 0.0336 0.0735 0.3412 0.0423 0.0567 0.1819 0.0303 0.0308 0.0654 0.0301 0.5169 0.4193 0.0911 0.8742 0.1841 0.4882 0.0972 0.1170 0.0318 0.0765 0.3444 0.0666 0.0340 0.0582 0.0585 0.0609 0.0339 0.0683 0.9117 5.2513 0.0416 0.0291 0.0689 0.2962 0.0333 0.0450 0.2693 0.2476 0.0603 0.3696 0.0374 0.3343 0.5440 0.4043 0.0587 0.0342 0.0633 0.2651 0.0380 0.0793 0.0307 0.0460 0.0352 0.8274 0.0383 0.0382 0.3765 0.0484 0.0438 0.9111 0.0425 0.0411 0.0398 2.0565 0.0398 0.0739 0.0663 0.0301 0.3594 0.4145 0.1327 0.0987 0.2734 1.1829 0.0315 0.4537 0.2459 0.3024 0.5110 1.2652 0.3495 0.0317 0.0395 0.4844 0.0470 0.0514 0.0417 0.0336 1.2551 0.0351 0.0843 0.2665 0.7375 0.7197 0.0962 0.7756 0.2465 0.1265 0.8817 0.0793

Table 9 Auxiliary variables Total suspended solids (mg/L)

Table 10 Auxiliary variable pH value

Table 11 Auxiliary variables Redox potential

Table 12 Measured effluent NH4-N concentration (mg/L)

The beneficial effects of the present invention are as follows:

The invention combines condition number analysis with a differential evolution algorithm to optimize the ESN network structure, thereby avoiding the problem of ill-conditioned solutions, improving the robustness of the network, and increasing its anti-interference performance.

The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims (5)

1. A soft measurement method for effluent NH4-N based on neural network, characterized by comprising: Constructing an echo state network, and initializing the network structure and network parameters of the echo state network; the input variables of the echo state network include inlet water temperature, total suspended solids, dissolved oxygen concentration, pH value, and outlet water redox potential; the output variables of the echo state network include ammonia nitrogen concentration; Based on the initialized echo state network, a sub-reservoir is constructed according to the singular value decomposition method; Optimizing the sub-reservoir according to a condition number and a differential evolution algorithm; The weights, input weights and state matrix of the optimized sub-reserve pool are updated, and it is determined whether the number of iterations is less than a preset iteration threshold. If so, jump to the step of "constructing a sub-reserve pool based on the initialized echo state network according to the singular value decomposition method"; if so, calculate the output weights; The echo state network is tested according to the output weights and the test samples to obtain a determined effluent NH4-N detection model; The data to be tested is input into the determined effluent NH4-N detection model to obtain the test results. 2. The method for soft-sensing NH4-N in water outlet based on neural network according to claim 1, characterized in that the network structure and network parameters of the echo state network are initialized, comprising: Determine the initial structure of the echo state network as 5-N-1; wherein N represents the number of nodes in the sub-reservoir; and the size of N is gradually increased; The sigmoid function is used as the network activation function G(·), the initial iteration number i＝1, the maximum iteration number i _max ≤30, and the training sample

Among them, _uk represents the kth group of input samples, _tk represents the kth group of actual output values,

Indicates that the input sample dimension is n, and L is the total number of samples; Randomly initialize the network input weights W ⁱⁿ and the internal weights W of the reserve pool between (0,1). 3. The effluent NH4-N soft-sensing method based on neural network according to claim 2 is characterized in that the sub-reserve pool is constructed based on the initialized echo state network according to the singular value decomposition method, including: Randomly generate a diagonal matrix

And two orthogonal matrices U _i and V _i ; where,

is a randomly generated value between (0,1), and n _k is the number of nodes in the sub-reservation pool; The sub-reservoir ΔW _i =U _i S _i V _i is constructed using the diagonal matrix and the two orthogonal matrices. 4. The method for soft sensing of effluent NH4-N based on neural network according to claim 3, characterized in that the sub-reserve pool is optimized according to the condition number and differential evolution algorithm, comprising: Construct a fitness function fitness(ΔW _i )=κ(H _i )-κ(H _i-1 ); where κ(H _i ) represents the condition number of the state matrix corresponding to the reserve pool after the new sub-reserve pool is added; Initialize the initial population NP = 50, mutation operator F = 0.7, crossover operator CR = 0.5, and initialize the population

Where n∈{1,2…,NP}, and set the maximum number of iterations G; According to the formula

Get the mutant individual μ ⁿ ; where r ₁ , r ₂ , r ₃ ∈[1,NP] are three different integers; pass

Complete the crossover operation, for each independent individual

Get crossover individuals

Among them, j∈{1,2…,10}, rand∈(0,1) is randomly generated; When the algorithm reaches the maximum number of iterations, the algorithm process is terminated, the optimal individual is returned, and a new sub-reserve pool is constructed using the optimal individual. 5. The effluent NH4-N soft sensor method based on neural network according to claim 4 is characterized in that the weights, input weights and state matrix of the optimized sub-reserve pool are updated, and it is judged whether the number of iterations is less than a preset iteration threshold. If so, jump to the step of "constructing a sub-reserve pool based on the initialized echo state network according to the singular value decomposition method"; if so, calculate the output weight, including: Update the reserve pool weight to

Update the input weights to

Update the state matrix to H = [H ₁ , ..., H _i ]; If i≥i _max is satisfied, the next step is executed, otherwise the process returns to the step "constructing a sub-reservoir based on the initialized echo state network according to the singular value decomposition method"; The output weight W ^out is calculated by the formula W ^out =H ^Γ T, wherein Γ represents pseudo-inverse calculation, and T is the target output data.

Images