abercrombie pariss,Pipe network water quality pred

kingu08pgg9

Water pipe network and the secondary pollution forecast

Abstract This article describes how to use neural network prediction of water quality pipeline and pipe network for the secondary pollution of water,[link widoczny dla zalogowanych], puts forward corresponding countermeasures to improve the water pipe network and ensure that residents some reference to health.
Keywords neural networks; tap water; Forecast

1 Introduction
China is a serious drought and water shortage in the country, is the world's 13 most water-poor individuals are one of the countries. Since China's accession to the WTO, along with industrial development, the water appears environmental degradation, declining water quality and other conditions have severely affected production and living of residents. Pipe network water quality of life and the lives of important people. Pipe network water quality is to ensure the health status of the key. And so precious drinking water, through the underground pipeline, Unfortunately, in this process, drinking water and pipeline as Therefore, strengthening the network of water quality prediction and forecasting for the prevention of secondary pollution of water pipe network and the timely processing of secondary pollution occurs is significant.
This article is a water pipe network in the first-hand on the basis of measured data pipe network water quality prediction. The layout of the aqueduct are very different,[link widoczny dla zalogowanych], pipeline itself can be regarded as a system, and we discussed the water quality time series measured data is subject to linear or nonlinear relationship is not easy to say it, it is difficult to establish a specific mathematical models to predict future water quality data. This article will use the BP neural network model to predict the water quality of tap water, and predicted values were compared with the actual value of the analysis. For the second pollution of the water pipe network,[link widoczny dla zalogowanych], and proposes countermeasures.
2 BP neural network's basic profile
1) BP network
BP network usually has one or more hidden layers, hidden layer neurons are used in the S-type transformation function, the output layer neurons pure linear transformation functions. Figure 1 depicts a BP with one hidden layer network.

Figure 1 BP network model structure
2) BP network learning process
is currently in the practical application of neural networks, most of the neural network model is the use of BP network and its variations, BP network is the core of the feedforward network section, and reflects the essence of artificial neural network part. BP neural network generally consists of an input layer, multiple hidden layer and one output layers, the implementation of all connections between the layers. Hidden layer neurons are used in the S-type transformation function, the output layer of neurons using pure linear transfer function. BP network learning process consists of four parts:
(1) smooth transmission mode: the input mode from the input layer to output layer by the spread of the middle layer.
(2) Back Propagation: hope for the network and the network output error of the difference between the actual output signal from the output layer to pass through the middle layer correction to the input layer, layer by layer connection weights value.
(3) memory training:
(4) Learning Convergence: Network convergence that networks tend to tend to the minimum global error
3 predict water quality parameters of the BP pipeline network
BP network can be used for pattern recognition, which uses a specific output vector will be linked with the input vector. The model we build is based on a large number of measured data, the database every day on the PH value, sulfate, nitrate nitrogen, ammonia nitrogen, total hardness, permanent hardness, chloride, total alkalinity, oxygen consumption volume of the nine water quality parameters measured records. We chose commonly used 3-layer BP neural network to predict future water quality parameters of a nine day.
To make the network predicted and measured values can be achieved almost infinite we established a network of training to get our set of neural network within the error range each forecasting model weights and threshold. Then read out by accessing the database WaterQualityRecords been assigned to the measured data and input variables Variable; the latest experimental data to assign TrueValue day to test whether the predictive value of our request basis. Here is the m read language.
connA = database ('WaterQualityRecords','','');
% connect to the database
CursorA = exec (connA, 'SELECT * FROM WaterPlant');% execute SQL statements and open cursors
CursorA = fetch (cursorA);
% Read data into MATLAB cell array
WaterDataBase = cursorA.Data;
% Read data to the WaterDataBase
For i = 1:600
% 600 days prior to the data assigned to Variable
Variable = WaterDataBase (i , Smile

End ;
TrueValue = WaterDataBase (601, Smile

;
% 601 days of data assigned to TrueValue
with the database can be connected to the network after the self-learning process to train the weights of each network and the threshold value. Using the random array to produce the network's initial weights and threshold. Neural network trained to avoid cross-validation method we used: about the measured data in our database is divided into training set, test set and test set. First, the training set used to train the network, according to BP algorithm to adjust the network structure and parameters; then tested the trained network test set to further optimize the network structure and parameters,[link widoczny dla zalogowanych], and ultimately determine the best training a network; the final test set of unknown samples with test and test the network accuracy. The following is the process of learning and prediction m language:
Matrix = zeros (30, width);
NeuralI = 100;
InData = zeros (NeuralI, 6);
GoalData = zeros (6,[link widoczny dla zalogowanych],6);
Neuralo = 6;
[R, Q] = size (InData);
[S2, Q] = size (GoalData);
S1 = 6;
[W1 b1] = rands (S1, R);
[W2 b2] = rands (S2, S1);
Max_epoch = 400;% maximum training steps
Err_goal = 0.01;% training target
Lr = 0.01;% learning rate
NNTWARN OFF
For j = 1: width
P = PingMeiWaterBase (j, 1: end);
Simdata = P (end-NeuralI +1-31 + x: end-31 + x )';
For i = 1: Neuralo
InData (:, i) = P (end-NeuralI-Neuralo-1 + i: end-Neuralo-2 + i )';
GoalData (:, i) = P (end-2 * Neuralo + i: end-Neuralo-1 + i )';
End
For i = 1: max_epoch
A1 = tansig (w1 * Indata, b1);
A2 = purelin (w2 * A1, b2);
Error = GoalData-A2;
D2 = deltalin (A2, error);
D1 = deltatan (A1, D2);
[Dw1, db1] = learnbp (Indata, D1, lr);
[Dw2, db2] = learnbp (A1, D2, lr);
W1 = w1 + dw1;
W2 = w2 + dw2;

	Forum SIMSON JAWA ROMET (!!) Strona Główna -> Romet Ogar	Wszystkie czasy w strefie EET (Europa)
Strona 1 z 1