CN102778548B - Method for forecasting sludge volume index in sewage treatment process - Google Patents

Abstract

The invention relates to a method for forecasting sludge volume index in the sewage treatment process and belongs to the field of sewage treatment. The sewage treatment process is severe in production condition and serious in random interference and has the features of being strong in nonlinearity, large in time varying and severe in hysteresis. Sludge volume of different degrees exists in almost all of the municipal sewage plants and most of industrial sewage treatment plants in China every year. The main features of the sludge volume are that sludge settleability is worsened, the sludge volume index (SVI) represents the parameter of the sludge settleability, but the SVI which is criticality index is difficultly measured online. In order to solve the problem that the SVI in the sewage treatment process cannot be measured online, the forecasting method based on an integrated nerve network is adopted to achieve real-time forecasting of the SVI in the sewage treatment process, and good effect is obtained.

Description

A kind of sewage disposal process sludge bulking index forecasting method

Technical field

The present invention utilizes integrated neural network to realize the prediction of sludge bulking index SVI in sewage disposal process, and the concentration of SVI has directly determined the information of sludge settling in sewage disposal process, to the normal operation important of wastewater treatment; Forecasting Methodology is applied to sewage disposal system, both can have saved investment and operating cost, can monitor in time again wastewater treatment correlation parameter, impel sewage treatment plant's efficient stable operation; The monitoring of SVI, as the important step of wastewater treatment, is the important branch in advanced manufacturing technology field, has both belonged to control field, belongs to again water treatment field.

Background technology

Water resources problems has become the subject under discussion of countries in the world government first concern, and the United Nations's " World Water resource comprehensive assessment report " points out: water problems will seriously restrict 21 century global economy and social development, and may cause conflicting between country; Therefore, set up sewage treatment plant, protect to greatest extent water environment, realize freshwater resources sustainable utilization and benign cycle, become the strategic measure of Chinese government's water resources comprehensive utilization.

In " national environmental protection " 12 " planning " that State Council in 2011 prints and distributes, point out: it is that 17.7%, seven large water system state key is monitored section to be inferior to III class ratio be 45% that national state key monitoring surface water section in 2010 is inferior to V class ratio.Chinese Ministry of Environmental Protection points out in " China Environmental State Bulletin " for 2011: within 2010, national wastewater emission amount is 617.3 hundred million tons, increase by 4.7% than last year, and annual 343.3 billion cubic meters of totally disposing of sewage, water prevention and cure of pollution present situation is that part makes moderate progress, totally not yet containment, situation is still severe, and pressure continues to strengthen.Therefore, set up sewage treatment plant, protect to greatest extent water environment, realize freshwater resources sustainable utilization and benign cycle, become the strategic measure of Chinese government's water resources comprehensive utilization.But the operation conditions of sewage treatment plant but allows of no optimist: because facility operation rate of load condensate is low, Sewage Plant water inlet chemical oxygen demand (COD) concentration is low, the reasons such as incomplete are supervised, detected to water quality, in sewage disposal process, be difficult to ensure stability and the reliability of Sewage Plant operation.At present, the nearly all annual all sludge bulking of various degrees in municipal sewage plant and most of industrial sewage treatment plant of China.Sludge bulking not only makes sludge loss, and effluent quality exceeds standard, and even causes whole sewage disposal system collapse, endangers huge.Therefore, suppress sludge bulking occur, guarantee quality of sewage disposal up to standard be current problem demanding prompt solution.

The principal character of sludge bulking is that sludge settling property worsens, and SVI is the parameter that represents sludge settling property, is called again sludge bulking index, conventionally when sludge bulking occurs during higher than 150mL/g SVI.This key index of SVI is difficult to measure, and in sewage treatment plant's actual motion, obtains by artificial chemical examination, and it analyzes determination period generally needs multiple hours.The survey frequency of most of SVI of sewage treatment plant is 1-2 time weekly, is difficult to rely on the artificial laboratory values of SVI to obtain in time sludge bulking information.Simultaneously, owing to causing that the reason of sludge bulking is many-sided, and these factors influence each other, connect each other, restriction mutually, identify and predict for the sludge bulking generally occurring, this is the relevant knowledge that an engineering problem has related to again microorganism, needs the combination of multiple subject knowledges intersection.Therefore, study new method and realize the forecasting problem of SVI, become the important topic of sewage control engineering field research, and had important practical significance.

The present invention proposes a kind of sludge bulking index SVI Forecasting Methodology, by building integrated neural network model, select one group both to maintain close ties with SVI, the variable of easily measuring is again as the input of integrated neural network, predict by integrated neural network for the immeasurable variable relevant to SVI, finally realize the prediction to SVI, thereby guarantee to find in time sludge bulking, reduce sludge bulking and occur, ensure the normal operation of sewage treatment plant.

Summary of the invention

The present invention has obtained the Forecasting Methodology of sludge bulking index SVI in a kind of sewage disposal process based on integrated neural network; The method is by analyzing sewage disposal process, in numerous measurable variables, select the input of one group of variable that has close ties with SVI and easily measure as integrated neural network, predict by integrated neural network for the immeasurable variable relevant to SVI, finally realize the prediction to SVI;

The present invention has adopted following technical scheme and performing step:

1. a Forecasting Methodology of sludge bulking index SVI,

(1) be designed for the integrated neural network initial topology structure that SVI predicts; Network is divided into two parts: Part I comprises input layer, hidden layer, output layer, and Part II also comprises input layer, hidden layer, output layer; Part I be input as auxiliary variable, be output as biochemical oxygen demand BOD; Part II be input as auxiliary variable and biochemical oxygen demand BOD, be output as sludge bulking index SVI;

Initialization neural network:

Part I neural network is the connected mode of N-M1-1, and input layer is N, and hidden layer neuron is M1, and output layer neuron is 1; The weights of Part I neural network are carried out to random assignment; The input table of neural network is shown X ₁=[x ₁, x ₂..., x _n] ^t, [x ₁, x ₂..., x _n] ^tfor [x ₁, x ₂..., x _n] transposition, the desired output of Part I neural network is expressed as y _d1; If total P training sample, p training sample is X ₁(p)=[x ₁(p), x ₂(p) ..., x _n(p)] ^t, during with p training sample neural network training, the output of Part I neural network can be described as:

Wherein, M1 is the hidden layer neuron number of Part I neural network, X ₁(p)=[x ₁(p), x ₂(p) ..., x _n(p)] ^tinput vector, j hidden layer neuron of Part I neural network and the neuronic connection weights of output layer; the output of j hidden layer neuron of Part I neural network,

Wherein, f (x)=(1+e ^-x) ^-1, the connection weights of i input layer of Part I neural network and j hidden layer neuron, x _ibe the output of i input layer of Part I neural network, the output of Part I neural network input layer equals its input value;

Part II neural network is the connected mode of N+1-M2-1, and input layer is N+1, and hidden layer neuron is M2, and output layer neuron is 1; The weights of Part II neural network are carried out to random assignment; The input table of Part II neural network is shown X ₂=[x ₁, x ₂..., x _n, y ₁] ^t, [x ₁, x ₂..., x _n, y ₁] ^tfor [x ₁, x ₂..., x _n, y ₁] transposition, the desired output of Part II neural network is expressed as y _d2; If total P training sample, p training sample is X ₂(p)=[x ₁(p), x ₂(p) ..., x _n(p), y ₁(p)] ^t, during with p training sample neural network training, the output of Part II neural network can be described as:

Wherein, M2 is the hidden layer neuron number of Part II neural network, X ₂(p)=[x ₁(p), x ₂(p) ..., x _n(p), y ₁(p)] ^tthe input vector of Part II neural network, the output y of Part I neural network ₁(p) also as the input of Part II neural network, k hidden layer neuron of Part II neural network and the neuronic connection weights of output layer; the output of k hidden layer neuron of Part II neural network,

Wherein, function f is identical with the expression-form in formula (2), the connection weights of k input layer of Part II neural network and l hidden layer neuron, x _kthe output of k input layer of Part II neural network, x _n+1(p)=y ₁(p), the output of the input layer of Part II neural network equals its input value;

Definition error function is:

$E\left(p\right)=\frac{1}{P}\underset{p=1}{\overset{P}{\mathrm{\Σ}}}{(y_2\left(p\right)-y_{d2}\left(p\right))}^T(y_2\left(p\right)-y_{d2}\left(p\right)),---\left(5\right)$

P is training sample sum, y _d2and y (p) ₂(p) be respectively desired output and the actual output of p moment integrated neural network Part II, the object of training integrated neural network is to make the error function of formula (5) definition reach expectation value;

(2) sample data is proofreaied and correct;

If A data sample { X ₁(1), X ₁(2) ..., X ₁(A) }, average is χ, and the deviation of each sample is D (a)=X ₁(a)-χ, a=1,2 ..., A, calculates standard deviation:

$\mathrm{\σ}=\sqrt{\underset{a=1}{\overset{A}{\mathrm{\Σ}}}\frac{{(X_1\left(a\right)-\mathrm{\χ})}^2}{A-1}},---\left(6\right)$

If some sample X ₁(a) deviation meets:

|D(a)|≥3σ,，a=1,2,…,A,(7)

Think sample X ₁(a) be abnormal data, should give rejecting, obtain the data after proofreading and correct, these data are as training sample and the test sample book of neural network;

Its feature is further comprising the steps of:

(3) with the data training integrated neural network after proofreading and correct, be specially:

1. a given integrated neural network, the hidden layer neuron of Part I neural network is M1, and the hidden layer neuron of Part II neural network is M2, and M1 and M2 are the positive integer that is less than 200, initialization neural network weight with the initial weight of neural network is 0 to 1 random number;

2. adjust the weights v of integrated neural network Part I neural network according to formula (8) ¹;

Wherein, be moore-Penrose contrary, y _d1for the desired output of Part I neural network;

3. adjust the weights v of integrated neural network Part II neural network according to formula (9) ²;

Wherein, be moore-Penrose contrary, y _d2for the desired output of Part II neural network;

4. the value of error function (5) reaches anticipation error E _dwithin≤0.01 o'clock, stop calculating; Otherwise turn to step 2. to continue training;

(4) test sample book is detected: the input using test sample book data as the integrated neural network training, the output of integrated neural network Part II neural network is predicting the outcome of SVI;

Creativeness of the present invention is mainly reflected in:

(1) the present invention is directed in current sewage disposal process sludge bulking index SVI long measuring period, the problem that can not detect online, feature that can Nonlinear Function Approximation according to neural network, adopt integrated neural network to predict SVI, compare and other neural metwork training methods, because integrated neural network utilizes the contrary method training weights of Moore-Penrose, have the advantages that real-time is good and precision is high;

(2) the present invention is by having adopted integrated neural network to realize the mapping between auxiliary variable discharge Qin, dissolved oxygen concentration DO, acidity-basicity ph, chemical oxygen demand COD, total nitrogen TN and SVI; Measure and the BOD value relevant to SVI for being difficult to, first integrated neural network can predict BOD, and then realize the prediction of SVI, improved the precision of prediction of SVI, has the advantages that precision is high; Also saved the complex process of development sensor and reduced operating cost simultaneously;

To note especially: the present invention just for convenience, employing be that SVI is predicted, this invention is also applicable to other indexes of sludge bulking---SDI etc. equally, all should belong to scope of the present invention as long as adopt principle of the present invention to predict.

Brief description of the drawings

Fig. 1 is integrated neural network topological structure of the present invention;

Fig. 2 is the present invention figure that predicts the outcome, and wherein solid black lines is SVI measured value, and red dotted line is to adopt the contrary method of Moore-Penrose to obtain SVI predicted value, and blue dot is scribed ss the SVI predicted value that adopts Gradient Descent method to obtain;

Fig. 3 is the present invention's Error Graph that predicts the outcome, and red dotted line is the error that predicts the outcome that adopts the contrary method of Moore-Penrose, and blue dot is scribed ss the error that predicts the outcome that adopts Gradient Descent method;

Table 1-8 is experimental data of the present invention, and table 1-5 is for detecting sample Q _in, DO, pH, COD, TN value, table 6 is BOD predicted value, table 7 is SVI measured value, table 8 is SVI predicted value of the present invention.

Embodiment

The present invention chooses the auxiliary variable Q of prediction SVI _in, DO, pH, COD, TN, BOD, wherein Q _indischarge, DO is dissolved oxygen concentration in aeration tank, pH is the potential of hydrogen of water quality in aeration tank, COD is the chemical oxygen demand (COD) of effluent quality, TN is the total nitrogen concentration of effluent quality, BOD is the biochemical oxygen demand of effluent quality, but BOD is difficult to measure, and need to predict it by integrated neural network Part I neural network; Q _inunit be m ³/ day, pH does not have unit, and other unit is mg/litre;

Experimental data is from certain sewage treatment plant water analysis daily sheet in 2008.Experiment sample is remaining 160 groups of data after data pre-service, 160 groups of whole data samples are divided into two parts: wherein 100 groups of data are used as training sample, and all the other 60 groups of data are as test sample book, and test sample book data are as shown in table 1-8; Fig. 1 is SVI neural network prediction model, and the input of integrated neural network Part I is respectively Q _in, DO, pH, COD, TN, be output as BOD, adopt the connected mode of 5-10-1, i.e. 5 input neurons, 10 hidden layer neuron, 1 output neuron; The input of Part II is respectively Q _in, DO, pH, COD, TN, BOD, be output as SVI, adopt the connected mode of 6-20-1, i.e. 6 input neurons, 20 hidden layer neuron, 1 output neuron;

Integrated neural network algorithm concrete steps are as follows:

(1) initialization neural network: Part I neural network is determined the connected mode of 5-10-1, to the weights of Part I neural network with carry out random assignment, its value is 0 to 1 random number, and input is respectively Q _in, DO, pH, COD, TN value, be output as the value of BOD; Part II neural network is determined the connected mode of 6-20-1, to the weights of Part II neural network with carry out random assignment, its value is 0 to 1 random number, and input is respectively Q _in, DO, pH, COD, TN, BOD value, be output as the value of SVI;

(2) sample data is proofreaied and correct, get respectively Q _in, 160 groups of DO, pH, COD, TN, BOD, SVI data of having proofreaied and correct, wherein 100 groups for training;

(3) with the training sample data neural network training after proofreading and correct, in training process, first, integrated neural network Part I neural network weight is adjusted; Secondly, integrated neural network Part II neural network weight is adjusted; Thereby realize the training of neural network;

Be specially:

1. train given initialization integrated neural network, error E is expected in design _d=0.01;

2. adjust the weights of Part I neural network according to formula (8)

3. adjust the weights of Part II neural network according to formula (9)

4. the calculated value of error function (5) reaches anticipation error E _din time, stops calculating; Otherwise turn to step 2. to continue training;

(4) test data is detected: the input using test sample book data as the neural network training, data are in table 1-5, and integrated neural network Part I is output as predicting the outcome of BOD, and data are in table 6; SVI is in table 7 for the actual output of system, and the output of integrated neural network Part II is predicting the outcome of SVI, and data are in table 8; Fig. 2 is that SVI predicts the outcome, and Fig. 3 is SVI predicated error;

Fig. 2 is SVI prediction case figure, X-axis: input sample point, and Y-axis: sludge bulking index SVI (mg/L), solid line is real system output valve, dotted line is the output valve of integrated neural network Part II; Fig. 3 is predicated error, and result proves the validity of the method.

Training data:

Table 1. auxiliary variable Q _ininput value (m ³/ day)

44101	39024	32229	35023	36924	38572	41115	36107	29156	39246
										38568	38655	34193	36332	32484	37724	36446	35636	34746	34893
37102	41598	38058	40716	40868	36358	40879	44150	45779	41230
										39891	32257	40498	40221	46669	34669	41824	51520	39421	36131
33251	35789	40106	45191	43308	37615	42596	41948	34647	36967
										34879	34365	34291	34886	38731	39308	44198	39003	34487	35198

The input value (mg/L) of table 2. auxiliary variable DO

1.5	3	2	2.5	1.5	3	3	2	2.5	2
										0.7	1.5	2	3.5	0.9	1	1	1.2	1	1.2
2	1.2	1	3.5	1.5	2	1.2	1	3	0.35
										2	3.5	1.5	2	1.75	1.2	1.2	2	1	1
1	1.5	0.6	2	1.4	1.2	3	1.5	1	1
										1	2	2	3	1.2	3	3	1.2	0.7	0.8

The input value of table 3. auxiliary variable pH

7.8	7.7	7.6	7.9	8	7.8	7.8	7.7	7.7	7.8
										8.2	7.9	8	7.9	7.5	7.9	7.7	8	7.7	8
7.8	8.2	7.8	8.1	8.1	7.7	7.6	8.1	7.8	7.6
										7.6	7.5	8.1	8.1	7.8	7.8	7.8	7.3	7.9	7.9
7.6	7.4	7.8	8	7.9	7.8	7.7	7.7	7.5	7.6
										7.5	7.6	7.9	7.7	7.5	7.8	7.7	7.8	7.9	7.7

The input value (mg/L) of table 4. auxiliary variable COD

66.3	69.2	72.7	77.1	57.6	66.1	67.5	53.9	61.8	66.1
										66	74.4	70.2	68.1	75.3	71.6	62.5	65.8	74.6	77.8
76.5	71.2	71.9	66.1	69.6	68.1	69.7	74.1	62.3	55.3
										69.9	42.2	50.8	54.8	64.9	56.4	53.1	56.5	65.7	49.4
53.6	46.5	51.1	32.8	30	43.4	26.7	47.6	44.3	56.4

50.5

50

50.5

57.5

63.3

65.4

36.4

55.6

55.6

64.6

The input value (mg/L) of table 5. auxiliary variable TN

4.5	6.5	4.5	4.2	4.5	4	6.5	7.5	4	6.5
										5.5	4	5.5	5.5	4	6	6	4.5	7	8
3	4.5	4.5	4.5	6	4.5	7	5.5	7.5	3
										3.5	7.5	3.5	4.6	4.5	3.5	2.5	4	3.3	7.5
6.5	6.5	7	3	2.5	3	7.5	5	6	7
										7.5	6	6.5	8.5	3	4	2.5	4.5	6	3

The predicted value (mg/L) of table 6. auxiliary variable BOD

81	94.8	81	89.6	95.6	96.5	84.6	90.6	84.2	89
										71.1	79.7	82.4	82.6	80.6	80.6	76	85.6	82.8	89
86.2	82.3	82.9	88.7	86.3	72.2	88.3	77.4	92.3	87.5
										89.4	95	92.7	84.9	86.7	88	88.4	90.1	97.9	90.9
90.2	79.5	93.2	96.1	98	91.8	96.5	85.2	91.5	90.7
										94.9	92.8	96	95.5	90.8	88.4	98.5	94.8	92.2	90.9

The measured value (mg/L) of table 7.SVI

100	95	95.5	99.6	98	89.9	95	92	94.1	121.3
										100	99.7	99.7	99.2	113.3	110.7	90.9	93	92.7	90.8
91.1	89.8	94.1	92.7	93.5	93.6	109.6	99.6	99.5	99.8
										99.6	89.7	99.8	99.8	102.1	87.6	94.1	89.7	99.7	99.6
91.8	91.5	93.6	103.2	90.4	95.9	99.6	93.4	93.9	91.8
										88.3	89.1	93.3	95.6	96.7	98.9	88.5	99.8	95.7	98.5

The predicted value (mg/L) of table 8.SVI

100	95	95.5	99.7	97.9	89.1	94.7	92.1	94.9	121.3
										99.2	100.2	100.4	99.2	113.3	110.7	90.9	92.9	92.7	90.7
91.1	89.7	93.9	93.7	93.7	94.1	109.9	99.2	99.7	99.8
										99.8	89.7	100.4	99.8	102.2	87.6	94.1	89.6	99.6	99.6
91.9	92.2	93.6	102.9	90.1	95.4	100.0	93.4	93.9	91.8
										88.3	89.1	93.3	95.3	96.7	98.9	88.5	100.2	95.0	98.6

Claims (1)

1. a Forecasting Methodology of sludge bulking index SVI, its feature comprises the following steps:

(1) be designed for the integrated neural network that SVI predicts, integrated neural network is divided into two parts: Part I neural network comprises input layer, hidden layer, output layer, and Part II neural network also comprises input layer, hidden layer, output layer; Part I neural network be input as auxiliary variable, be output as biochemical oxygen demand BOD; Part II neural network be input as auxiliary variable and biochemical oxygen demand BOD, be output as sludge bulking index SVI;

Initialization neural network:

Part I neural network is the connected mode of N-M1-1, and input layer is N, and hidden layer neuron is M1, and output layer neuron is 1; The weights of Part I neural network are carried out to random assignment; The input table of neural network is shown X ₁=[x ₁, x ₂..., x _n] ^t, [x ₁, x ₂..., x _n] ^tfor [x ₁, x ₂..., x _n] transposition, the desired output of Part I neural network is expressed as y _d1; If total P training sample, p training sample is X ₁(p)=[x ₁(p), x ₂(p) ..., x _n(p)] ^t, during with p training sample neural network training, the output of Part I neural network can be described as:

Wherein, M1 is the hidden layer neuron number of Part I neural network, X ₁(p)=[x ₁(p), x ₂(p) ..., x _n(p)] ^tinput vector, j hidden layer neuron of Part I neural network and the neuronic connection weights of output layer; the output of j hidden layer neuron of Part I neural network,

Wherein, f (x)=(1+e ^-x) ^-1, the connection weights of i input layer of Part I neural network and j hidden layer neuron, x _ibe the output of i input layer of Part I neural network, the output of input layer equals its input value;

Part II neural network is the connected mode of N+1-M2-1, and input layer is N+1, and hidden layer neuron is M2, and output layer neuron is 1; The weights of Part II neural network are carried out to random assignment; The input table of Part II neural network is shown X ₂=[x ₁, x ₂..., x _n, y ₁] ^t, [x ₁, x ₂..., x _n, y ₁] ^tfor [x ₁, x ₂..., x _n, y ₁] transposition, the desired output of Part II neural network is expressed as y _d2; If total P training sample, p training sample is X ₂(p)=[x ₁(p), x ₂(p) ..., x _n(p), y ₁(p)] ^t, during with p training sample neural network training, the output of Part II neural network can be described as:

Wherein, M2 is the hidden layer neuron number of Part II neural network, X ₂(p)=[x ₁(p), x ₂(p) ..., x _n(p), y ₁(p)] ^tbe input vector, the output of Part I neural network is as the input of Part II neural network, k hidden layer neuron of Part II neural network and the neuronic connection weights of output layer; the output of k hidden layer neuron of Part II neural network,

Wherein, function f is identical with the expression-form in formula (2), Part I neural network, the connection weights of k input layer of Part II neural network and l hidden layer neuron, x _kthe output of k input layer of Part II neural network, x _n+1(p)=y ₁(p), the output of input layer equals its input value;

Definition error function is:

P is training sample sum, y _d2and y (p) ₂(p) be respectively desired output and the actual output of integrated neural network Part II while being input as p training sample, the object of training integrated neural network is to make the error function of formula (5) definition reach expectation value;

(2) sample data is proofreaied and correct;

If A data sample { X ₁(1), X ₁(2) ..., X ₁(A) }, average is χ, and the deviation of each sample is D (a)=X ₁(a)-χ, a=1,2 ..., A, calculates standard deviation:

If some sample X ₁(a) deviation meets:

|D(a)|≥3σ，a＝1,2,…,A, (7)

Think sample X ₁(a) be abnormal data, should give rejecting, obtain the data after proofreading and correct, these data are as training sample and the test sample book of neural network;

Its feature is further comprising the steps of:

(3) with the data training integrated neural network after proofreading and correct, be specially:

1. a given integrated neural network, the hidden layer neuron of Part I neural network is M1, and the hidden layer neuron of Part II neural network is M2, and M1 and M2 are the positive integer that is less than 200, initialization neural network weight with the initial weight of neural network is 0 to 1 random number;

2. adjust the weights v of integrated neural network Part I neural network according to formula (8) ¹;

Wherein, be moore-Penrose contrary, y _d1for the desired output of Part I neural network;

3. adjust the weights v of integrated neural network Part II neural network according to formula (9) ²;

Wherein, be moore-Penrose contrary, y _d2for the desired output of Part II neural network;

4. the value of error function (5) reaches anticipation error E _dwithin≤0.01 o'clock, stop calculating; Otherwise turn to step 2. to continue training;

(4) test sample book is detected: the input using test sample book data as the integrated neural network training, the output of integrated neural network Part II is predicting the outcome of SVI.