CN117493803A - Sewage treatment process effluent quality prediction method based on recursive RBF neural network - Google Patents
Sewage treatment process effluent quality prediction method based on recursive RBF neural network Download PDFInfo
- Publication number
- CN117493803A CN117493803A CN202311451452.7A CN202311451452A CN117493803A CN 117493803 A CN117493803 A CN 117493803A CN 202311451452 A CN202311451452 A CN 202311451452A CN 117493803 A CN117493803 A CN 117493803A
- Authority
- CN
- China
- Prior art keywords
- pool
- tank
- treatment process
- neural network
- sewage treatment
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 96
- 230000008569 process Effects 0.000 title claims abstract description 61
- 239000010865 sewage Substances 0.000 title claims abstract description 56
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 46
- 238000013461 design Methods 0.000 claims abstract description 19
- 238000010606 normalization Methods 0.000 claims abstract description 14
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 10
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 56
- 210000002569 neuron Anatomy 0.000 claims description 50
- 238000004062 sedimentation Methods 0.000 claims description 45
- XKMRRTOUMJRJIA-UHFFFAOYSA-N ammonia nh3 Chemical compound N.N XKMRRTOUMJRJIA-UHFFFAOYSA-N 0.000 claims description 18
- 230000000306 recurrent effect Effects 0.000 claims description 17
- 239000010802 sludge Substances 0.000 claims description 15
- 238000005457 optimization Methods 0.000 claims description 13
- 238000010992 reflux Methods 0.000 claims description 13
- OKKJLVBELUTLKV-UHFFFAOYSA-N Methanol Chemical compound OC OKKJLVBELUTLKV-UHFFFAOYSA-N 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 9
- 229910052757 nitrogen Inorganic materials 0.000 claims description 9
- IJGRMHOSHXDMSA-UHFFFAOYSA-N nitrogen Substances N#N IJGRMHOSHXDMSA-UHFFFAOYSA-N 0.000 claims description 9
- 238000005516 engineering process Methods 0.000 claims description 8
- 230000035772 mutation Effects 0.000 claims description 8
- 230000009471 action Effects 0.000 claims description 7
- 230000000694 effects Effects 0.000 claims description 7
- OAICVXFJPJFONN-UHFFFAOYSA-N Phosphorus Chemical compound [P] OAICVXFJPJFONN-UHFFFAOYSA-N 0.000 claims description 6
- MMDJDBSEMBIJBB-UHFFFAOYSA-N [O-][N+]([O-])=O.[O-][N+]([O-])=O.[O-][N+]([O-])=O.[NH6+3] Chemical compound [O-][N+]([O-])=O.[O-][N+]([O-])=O.[O-][N+]([O-])=O.[NH6+3] MMDJDBSEMBIJBB-UHFFFAOYSA-N 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 6
- 210000003719 b-lymphocyte Anatomy 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 6
- 239000007788 liquid Substances 0.000 claims description 6
- NBIIXXVUZAFLBC-UHFFFAOYSA-K phosphate Chemical compound [O-]P([O-])([O-])=O NBIIXXVUZAFLBC-UHFFFAOYSA-K 0.000 claims description 6
- 229910052698 phosphorus Inorganic materials 0.000 claims description 6
- 239000011574 phosphorus Substances 0.000 claims description 6
- 238000007906 compression Methods 0.000 claims description 4
- 230000006835 compression Effects 0.000 claims description 4
- 238000013144 data compression Methods 0.000 claims description 4
- 230000003247 decreasing effect Effects 0.000 claims description 4
- 238000010187 selection method Methods 0.000 claims description 4
- VHUUQVKOLVNVRT-UHFFFAOYSA-N Ammonium hydroxide Chemical compound [NH4+].[OH-] VHUUQVKOLVNVRT-UHFFFAOYSA-N 0.000 claims description 3
- 101001104570 Homo sapiens Proteasome assembly chaperone 2 Proteins 0.000 claims description 3
- 101000625842 Homo sapiens Tubulin-specific chaperone E Proteins 0.000 claims description 3
- 102100024769 Tubulin-specific chaperone E Human genes 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 210000004027 cell Anatomy 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 239000004576 sand Substances 0.000 claims description 3
- 230000001502 supplementing effect Effects 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000004065 wastewater treatment Methods 0.000 claims description 3
- 230000001932 seasonal effect Effects 0.000 claims description 2
- 230000008859 change Effects 0.000 abstract description 12
- 230000007246 mechanism Effects 0.000 abstract description 5
- 230000008901 benefit Effects 0.000 abstract description 2
- 238000005094 computer simulation Methods 0.000 abstract description 2
- 230000002159 abnormal effect Effects 0.000 description 8
- 238000012545 processing Methods 0.000 description 6
- 238000011160 research Methods 0.000 description 5
- 230000002411 adverse Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 230000003631 expected effect Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000005842 biochemical reaction Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000013441 quality evaluation Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 239000002351 wastewater Substances 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A20/00—Water conservation; Efficient water supply; Efficient water use
- Y02A20/152—Water filtration
Abstract
The invention relates to a sewage treatment process effluent quality prediction method based on a recursion RBF neural network, wherein sewage is treated, data are collected in the treatment process, related variables are selected, and finally nonlinear normalization treatment is carried out on the data; the relevant variables are dynamically selected at each moment in time using a dynamic PLS algorithm based on a sliding window technique. Aiming at the problems that the operation working condition of the sewage treatment process is complex and an accurate mathematical mechanism model is difficult to establish, the invention utilizes the advantages of the RBF neural network in the aspect of nonlinear system modeling to establish the sewage treatment process model and predict the effluent quality of the sewage treatment process model. The invention combines the sewage treatment process with the RBF neural network in depth, designs the network input-output relationship which accords with the sewage treatment process, proposes to establish a dynamic model with the change of network scale, data dimension and key related variables, and improves the adaptability of the prediction model to working conditions.
Description
Technical Field
The invention relates to the technical field of data acquisition and processing, in particular to a wastewater quality prediction method in a wastewater treatment process based on a recurrent RBF neural network.
Background
The volume of raw data in the sewage treatment process is very large, and the construction of a prediction model is very challenging. The main manifestations are: 1) The data contains a large amount of redundant information, so that not only is the useful information difficult to obtain, but also the generalization capability of the prediction model is reduced;
2) The data contains a large amount of abnormal data, and the abnormal data has great adverse effect on the model and even can cause failure in model construction; 3) The data dimensions are not uniform, the importance of the data can be estimated in error due to the non-uniform dimensions, and modeling difficulty is increased.
The sewage treatment process is a complex biochemical reaction process, and the whole process is closely connected and has continuity. The water quality of the effluent is not only changed by the current working condition, but also influenced by the historical working condition and the historical water quality of the effluent. Because the sewage treatment process has a large hysteresis characteristic, the influence of the historical effluent quality and the historical working condition on the current effluent quality cannot be ignored.
Disclosure of Invention
The invention provides a method for predicting the effluent quality in the sewage treatment process based on a recursive RBF neural network, which aims at reducing the data dimension, reducing the data scale, reducing the influence of abnormal data, fully playing the role of each variable, reducing the influence of the working condition mutation on the effluent quality and the hysteresis characteristic in the sewage treatment process.
The technical scheme of the invention is as follows:
a sewage treatment process effluent quality prediction method based on a recursion RBF neural network includes the steps that sewage is treated, data are collected in the treatment process, relevant variables are selected, and finally nonlinear normalization treatment is carried out on the data;
the related variables are dynamically selected at each moment by adopting a dynamic PLS algorithm based on a sliding window technology, and the design selection method comprises the following steps:
wherein X (t) is the acquired data at the time t, X (t) is the modeling data set at the time t, and L is the length of the sliding window, and the value of the sliding window is set according to the working condition state; when t is more than or equal to L, PLS algorithm analysis is carried out on the data from t-L to t, and when t is less than L, PLS algorithm analysis is carried out on all sample data, and key relevant variables of the water quality prediction index of the water outlet in the time period are selected; and constructing a recursive RBF neural network by taking the selected key variable as input and the corresponding effluent quality prediction index as output.
Preferably, the adopted data compression technology is to perform one-step logarithmic normalization processing after linear compression, and the logarithmic normalization processing rule formula is as follows:
wherein x is min And x max Respectively the minimum and the maximum of the normalized data set, and a is an integer between 2 and 10.
Preferably, the partial least squares method is used to screen and decompose the acquired data.
Preferably, sewage is treated by a sewage treatment system, data acquisition is carried out in the treatment process, and the number of data acquisition points is 10 and contains 57 acquisition amounts;
the sewage treatment system comprises a primary sedimentation tank, an A tank, a B tank and a secondary sedimentation tank which are sequentially communicated, wherein a water inlet is formed in the primary sedimentation tank, a water outlet is formed in the secondary sedimentation tank, the bottom of the primary sedimentation tank and the bottom of the secondary sedimentation tank are communicated with a sludge treatment device, the bottom of the secondary sedimentation tank is communicated with the bottom of the A tank, the A tank comprises an anaerobic zone of the A tank and an anoxic zone of the A tank, the B tank is an aerobic zone of the B tank and is divided into three parts, and the anaerobic zone of the A tank is communicated with the bottom of the last part of the aerobic zone of the B tank through an outer return pipe.
Preferably, the first collecting point is arranged in front of the water inlet of the primary sedimentation tank, the second collecting point is arranged in the anaerobic zone of the tank A, the third collecting point is arranged in the anoxic zone of the tank A, the fourth to sixth collecting points are sequentially arranged in the aerobic zone of the tank B, the seventh collecting point is arranged in the secondary sedimentation tank, the eighth collecting point is arranged in the outer return pipe, the ninth collecting point is arranged in the sludge treatment device, and the tenth collecting point is arranged at the sludge outlet at the bottom of the secondary sedimentation tank;
the first collecting point collects the inflow water flow, the inflow water pH value, the inflow water COD and the inflow water ammonia nitrogen NH 3 N and the total plant treated water quantity;
the second collecting point collects the anaerobic ORP of the pool A, the anoxic front ORP of the pool A and the anoxic last nitrate nitrogen NO of the pool A 3 N, A pool anoxic end MLSS, A pool anoxic end liquid level, A pool aerobic pre-DO, A pool aerobic mid-DO and A pool good pre-ammonia nitrogen NH 3 N, primary ammonia nitrogen at the end of the pool A, the temperature of the pool A and the air flow rate of the pool A are 1;
the third collecting point collects the second aerobic DO of the A pool, the second good last orthophosphate of the A pool and the first good last ammonia nitrogen NH of the A pool 3 N, A pool air flow 2, pool A air flow 3, pool A internal reflux pump 2 frequency;
the fourth collecting point collects the anaerobic ORP of the B pool, the anoxic front ORP of the B pool and the anoxic last nitrate nitrogen NO of the B pool 3 N, B pool anoxic end MLSS, B pool anoxic end liquid level, B pool-aerobic pre-DO, B pool-pre-well ammonia nitrogen NH 3 N, a dosing pump flowmeter 1, a dosing pump 1 control command, secondary A pool ammonia nitrogen, secondary A oxygen-eliminating to anoxic rotary pump 1 frequency, secondary B oxygen-eliminating to anoxic rotary pump 2 frequency, B pool air flow 1, B pool internal reflux pump 3 frequency;
the fifth collecting point collects DO in the first aerobic tank of the B tank, DO in the second aerobic tank of the B tank, orthophosphate in the second good tank of the B tank, a dosing pump flowmeter 2, a dosing pump 2 control command, the temperature of the B tank, the air flow rate of the B tank 2 and the frequency of a reflux pump 4 in the B tank;
collecting PAC2 series biological cell flowmeter, secondary series B cell end test ammonia nitrogen, methanol addition amount and B cell air flow 3 at a sixth collection point;
a seventh collecting point collects positive phosphorus OP at the water outlet end of the secondary sedimentation tank, total phosphorus OP at the water outlet end of the secondary sedimentation tank, water outlet PH of the secondary sedimentation tank and total nitrogen TN at the water outlet end of the secondary sedimentation tank;
an eighth acquisition point acquires the external reflux flowmeter;
a ninth collection point collects the sludge discharge amount of the residual sludge;
and the tenth acquisition point acquires the backwash water quantity of the sand filter.
Preferably, the recursive RBF neural network design comprises the steps of:
the historical effluent quality index is used as a recursion input for predicting the current effluent quality index, and a memory factor is designed in a recursion link for adjusting the effect of the historical effluent quality;
the average working condition is represented by the average distance between working conditions, the mutation degree of the working condition at a certain moment is quantified by the distance between the working condition at the certain moment and the average working condition, and the memory factor is set according to the negative correlation of the mutation degree, wherein the specific expression formula is as follows:
wherein t is the current time, N is the hysteresis period, the adjustment is carried out according to the seasonal variation, generally 2-3 weeks are taken, k is the adjustment parameter, and can be obtained through learning,is the average working condition of the time period from N-t to t, x i Is the working condition of the ith moment in the time period from N-t to t.
Preferably, the recursive RBF neural network parameter optimization method comprises the following steps: network optimization is performed using the instantaneous error function as a cost function,
preferably, in order to solve the problem that the substitution of the cost function causes insufficient sample learning, thereby reducing the network performance, the parameter optimization is adjusted according to the full gradient information, and the full gradient calculation formula is as follows:
wherein θ n (t) is the nth parameter to be optimized in the network, y (t) is the network output at time t, y (t-i) is the ith time in the period of N-tNetwork output of carving, phi k (t) is the output of the kth neuron of the designed neural network,
wherein x (t) = [ x ] 1 (t),x 2 (t),…,x d (t),...,x D (t)] T Is the input vector of the network, x (t) ε R D ,d＝1,2,...,D,μ k (t)＝[μ k,1 (t),μ k,2 (t),...,μ k,D (t)]Is the center vector, μ, of the kth hidden neuron k (t)∈R D ,k＝1,2,...,K,||x(t)-μ k (t) || represents x (t) and μ k Euclidean distance, sigma, between (t) k (t) is the width of action of the kth hidden neuron.
Preferably, the structure adjustment of the recurrent RBF neural network includes the steps of:
the integrity definition expression at the time t of the neuron is as follows:
wherein phi (t) = [ phi ] 1 (t),φ 2 (t),…,φ k (t)]Is the neuron output, kappa (t) is the kernel matrix of the RBF neural network, τ 2 (x(t))∈(0,1)，τ 2 The larger (x (t)) means the better the integrity of R (μ (t)). Wherein the method comprises the steps of
Wherein,
preferably, the orthogonality definition expression at time t of the neuron is:
wherein mu k (t)＝[μ k,1 (t),μ k,2 (t),...,μ k,D (t)]Is the center vector, μ, of the kth hidden neuron k (t)∈R D ,k＝1,2,...,K,||μ i (t)-μ j (t) || represents the Euclidean distance between the centers of the ith and jth neurons, σ i (t) is the width of action of the ith hidden neuron.
In the network construction stage, increasing and decreasing the neurons according to the integrality and the orthogonality of the neurons, and supplementing the integrality of the network by adding new neurons when the integrality of the neurons is insufficient; when a problem of poor orthogonality occurs between neurons, redundant information of the network is reduced by deleting the neurons.
The invention has the beneficial effects that:
(1) Aiming at the problems that the operation working condition of the sewage treatment process is complex and an accurate mathematical mechanism model is difficult to establish, the invention utilizes the advantages of the RBF neural network in the aspect of nonlinear system modeling to establish the sewage treatment process model and predict the effluent quality of the sewage treatment process model. The invention combines the sewage treatment process with the RBF neural network in depth, designs the network input-output relationship which accords with the sewage treatment process, and firstly proposes to establish a dynamic model with the change of network scale, data dimension and key related variables, and improves the adaptability of the prediction model to working conditions. This is the main innovation of the research of the present invention.
(2) The invention designs a relatively complete data compression and fusion technology, pre-processes sewage treatment data, solves the problems of abnormal data, non-uniform dimension, large data volume and the like, and provides data support for subsequent model design, which is a second innovation of the invention.
(3) The method starts from each link established by the prediction model, firstly, the data is preprocessed, the influence of data dimension, abnormal data and the like on the prediction model is eliminated, and a unique structure and parameter adjustment method is designed according to the characteristics of RBF neural networks in the prediction model, so that the performance of the prediction model is further improved. The integrity of the predictive model design is a further innovation of the present invention.
Drawings
FIG. 1 is a scheme and technique of the present invention;
FIG. 2 is a graph of the expected effect of nonlinear normalization of the present invention;
FIG. 3 is a block diagram of a recurrent RBF neural network of the present invention;
fig. 4 is a structural diagram of the apparatus of the present invention.
In fig. 4, (1) -first, second, third, fourth, fifth, sixth, seventh, eighth, ninth, and tenth acquisition points, (2) -second, third, fourth, fifth, sixth, seventh, eighth, and ninth acquisition points, respectively.
Detailed Description
(1) For the problem of difficult data processing
On the basis of analysis of a variable mechanism in the correlation process of each index of the water quality of the water and actual factory data acquisition, the method adopts a partial least square method to screen and decompose the acquired data, respectively extracts data with stronger correlation with each index of the water quality of the water, and deletes the data which has no or weak correlation with each index of the water quality of the water, so as to avoid adverse effects of multiple coupling of variables in the modeling process. In order to reduce the influence of abnormal data and reasonably exert the predictive value of each related variable, the original data is processed by a nonlinear normalization method.
(2) Aiming at the problem that a mechanism model is difficult to construct
On the basis of fully considering the characteristics of the sewage treatment operation mechanism, the invention designs the recursion RBF neural network with the memory factors, which reflects the relation between the effluent quality and the working condition change and the historical effluent quality, plays the role of historical data and also prevents the problem of slow response of the prediction model to the working condition mutation.
(3) Difficulty in determining structure and parameters of prediction model
In order to obtain a proper network structure, the invention designs a self-organizing network structure adjusting method based on the integrity and the orthogonality of network neurons on the basis of the research of RBF neural networks, and the neurons are increased or decreased according to the sewage treatment running state data in the model construction process to obtain the proper network structure. In order to obtain optimal network parameters, a full-gradient-based parameter optimization algorithm is designed to rapidly acquire sample information and accelerate parameter optimization of a model.
Study protocol adopted
Overall research method
The invention designs a model for predicting the effluent quality in the sewage treatment process based on a recurrent RBF neural network with a memory factor according to the indication that the national sewage environment quality evaluation and monitoring technology is developed towards the intelligent, automatic and networking direction around the problem that the effluent quality in the sewage treatment process is difficult to measure rapidly and accurately. A specific implementation method is shown in fig. 1. The sewage is treated by three stages, wherein the first stage treatment is treated by a primary sedimentation tank, the second stage treatment is treated by anaerobic-aerobic reaction, and the third stage treatment is treated by secondary sedimentation by a secondary sedimentation tank.
Technical route
In order to achieve the research object of the present invention, the technical route adopted is summarized as follows:
selection of key related variables
In order to eliminate the adverse effect of the original data and acquire the data constructed by the model, the preprocessing of the original data mainly involves the selection of key related variables and the nonlinear normalization operation of the data.
1) The selection of key related variables is a dynamic process because of the sewage treatment process, and the operation conditions of different sewage treatment systems have great differences. Along with the change of the working condition of the sewage treatment process, the change of each index of the water quality of the effluent is different, and the correlation degree of the parameters of each link to each index of the water quality of the effluent is also different. In order to obtain variables related to various indexes of the effluent quality as much as possible, a plurality of sensors and monitoring points are distributed at various stages of the sewage treatment process, and 57 process variables listed in table 1 are obtained. In order to cope with the working condition change in the sewage treatment process, the invention aims to design a data selection method based on a sliding window technology, and the number and the variable types of related variables are dynamically changed at different moments. The sewage is treated by a sewage treatment system, data acquisition is carried out in the treatment process, and the number of data acquisition points is 10 and 9 (see figure 4), and the data acquisition points contain 57 acquisition amounts;
the sewage treatment system comprises a primary sedimentation tank, an A tank, a B tank and a secondary sedimentation tank which are sequentially communicated, wherein a water inlet is formed in the primary sedimentation tank, a water outlet is formed in the secondary sedimentation tank, the bottom of the primary sedimentation tank and the bottom of the secondary sedimentation tank are communicated with a sludge treatment device, the bottom of the secondary sedimentation tank is communicated with the bottom of the A tank, the A tank comprises an anaerobic zone of the A tank and an anoxic zone of the A tank, the B tank is an aerobic zone of the B tank and is divided into three parts, and the anaerobic zone of the A tank is communicated with the bottom of the last part of the aerobic zone of the B tank through an outer return pipe.
The first collecting point is arranged in front of a water inlet of the primary sedimentation tank, the second collecting point is arranged in an anaerobic zone of the tank A, the third collecting point is arranged in an anoxic zone of the tank A, the fourth to sixth collecting points are sequentially arranged in three parts of an aerobic zone of the tank B, the seventh collecting point is arranged in the secondary sedimentation tank, the eighth collecting point is arranged in an outer return pipe, the ninth collecting point is arranged in a sludge treatment device, and the tenth collecting point is arranged at a sludge outlet at the bottom of the secondary sedimentation tank;
the first collecting point collects the inflow water flow, the inflow water pH value, the inflow water COD and the inflow water ammonia nitrogen NH 3 N and the total plant treated water quantity;
the second collecting point collects the anaerobic ORP of the pool A, the anoxic front ORP of the pool A and the anoxic last nitrate nitrogen NO of the pool A 3 N, A pool anoxic end MLSS, A pool anoxic end liquid level, A pool aerobic pre-DO, A pool aerobic mid-DO and A pool good pre-ammonia nitrogen NH 3 N, primary ammonia nitrogen at the end of the pool A, the temperature of the pool A and the air flow rate of the pool A are 1;
the third collecting point collects the second aerobic DO of the A pool, the second good last orthophosphate of the A pool and the first good last ammonia nitrogen NH of the A pool 3 N, A pool air flow 2, pool A air flow 3, pool A internal reflux pump 2 frequency;
the fourth collecting point collects the anaerobic ORP of the B pool, the anoxic front ORP of the B pool and the anoxic last nitrate nitrogen NO of the B pool 3 N, B pool anoxicLast MLSS, anoxic last liquid level of B pool, aerobic pre-DO of B pool, good pre-ammonia nitrogen NH of B pool 3 N, a dosing pump flowmeter 1, a dosing pump 1 control command, secondary A pool ammonia nitrogen, secondary A oxygen-eliminating to anoxic rotary pump 1 frequency, secondary B oxygen-eliminating to anoxic rotary pump 2 frequency, B pool air flow 1, B pool internal reflux pump 3 frequency; the fifth collecting point collects DO in the first aerobic tank of the B tank, DO in the second aerobic tank of the B tank, orthophosphate in the second good tank of the B tank, a dosing pump flowmeter 2, a dosing pump 2 control command, the temperature of the B tank, the air flow rate of the B tank 2 and the frequency of a reflux pump 4 in the B tank;
collecting PAC2 series biological cell flowmeter, secondary series B cell end test ammonia nitrogen, methanol addition amount and B cell air flow 3 at a sixth collection point;
a seventh collecting point collects positive phosphorus OP at the water outlet end of the secondary sedimentation tank, total phosphorus OP at the water outlet end of the secondary sedimentation tank, water outlet PH of the secondary sedimentation tank and total nitrogen TN at the water outlet end of the secondary sedimentation tank;
an eighth acquisition point acquires the external reflux flowmeter;
a ninth collection point collects the sludge discharge amount of the residual sludge;
and the tenth acquisition point acquires the backwash water quantity of the sand filter.
TABLE 1 Sewage treatment Process variable
In order to select the key related variables, the invention adopts a dynamic PLS algorithm based on a sliding window technology to dynamically select the key related variables at each moment, and the design selection method comprises the following steps:
wherein X (t) is the acquired data at time t, X (t) is the modeling data set at time t, and L is the length of the sliding window, and the value of the sliding window is set according to the working condition state. In short, PLS algorithm analysis is carried out on the data from t-L to t, and key relevant variables of the water quality prediction index of the water outlet in the time period are selected. And constructing the RBF neural network by taking the selected key variable as input and the corresponding effluent quality prediction index as output.
2) Nonlinear normalization of data
In view of the complexity of the sewage treatment process, in order to reduce the influence of abnormal data in the normalization process, nonlinear logarithmic compression is adopted in the normalization process of the data, so that even if the original data with larger deviation abnormal data is compressed and then uniformly distributed in a smaller area, the expected effect is shown in fig. 2. The adopted data compression technology is that after the common linear compression, one-step logarithmic processing is carried out, and the logarithmic normalization processing rule formula is as follows:
wherein x is min And x max The minimum and maximum values of the normalized dataset, respectively.
Recursive RBF neural network design with memory factors
1) Recursive RBF neural network structure design
Because the sewage treatment process has continuity, the invention designs a recursive RBF neural network for constructing a model. The recursive structure is shown in fig. 3, and the historical effluent quality index is used as a recursive input for predicting the current effluent quality index. In consideration of the fact that the influence of the historical effluent quality on the current effluent quality in the sewage treatment process is different due to the change of working conditions, a memory factor is designed in a recursion link for adjusting the effect of the historical effluent quality, and the problem that a prediction model is insensitive or oversensitive to the sudden change of the working conditions is solved.
1) Memory factor design method
The memory factor is mainly used for adjusting the prediction effect of the historical effluent quality on the current effluent quality. The water quality of the effluent generally changes along with the change of the working condition, but the change of the water quality of the effluent has hysteresis phenomenon compared with the change of the working condition. In order to solve the problem that the water quality of the discharged water is insensitive to the working condition, the memory factor is changed according to the change of the working condition. The memory factor design scheme to be adopted is that average working conditions are represented by average distances among working conditions, the mutation degree of the working conditions at a certain moment is quantified by the distances between the working conditions and the average working conditions, the memory factor is set according to the negative correlation of the mutation degree, and a specific building formula is as follows:
where N is the hysteresis period and k is the adjustment parameter.
Structure and parameter optimization method of recursive RBF neural network
1) Parameter optimization of recursive RBF neural network
The data volume of the sewage treatment process is large, and the data scale is still large even though the data pretreatment is carried out. The large data volume is challenging to model training, and in order to reduce parameter optimization difficulty, the invention uses instantaneous error function in the parameter optimization process of the recurrent RBF neural network
Instead of cumulative error function
As a learned cost function. Wherein e i Is the error at time i. The replacement essence of the cost function is to reduce the repeated learning process of the model on the sample, which can cause insufficient learning on the sample, thereby reducing the network performance. In order to solve the problem, the parameter optimization of the invention is to be adjusted according to the full gradient information, and the full gradient calculation formula is as follows:
wherein θ n And (t) is the nth parameter to be optimized in the network, and y (t) is the network output at the moment t.
2) Structure adjustment of recursive RBF neural network
The structure adjustment of the network adopts the early research results of the invention responsible person, namely, the real-time adjustment is carried out according to the integrity and the orthogonality of the network neurons, and the adjustment method has proved to be scientific and effective.
The integrity definition expression at time t of the neuron is:
wherein x (t) = [ x ] 1 (t),x 2 (t),…,x d (t),...,x D (t)] T Is the input vector of the network, x (t) ε R D ,d＝1,2,...,D,φ(t)＝[φ 1 (t),φ 2 (t),…,φ k (t)]Is the neuron output and κ (t) is the kernel matrix of the RBF neural network. As defined, τ 2 (x(t))∈(0,1)，τ 2 A larger (x (t)) means a better integrity of ψ (μ (t)). Wherein the method comprises the steps of
Wherein,
wherein mu k (t)＝[μ k,1 (t),μ k,2 (t),...,μ k,D (t)]Is the center vector, μ, of the kth hidden neuron k (t)∈R D ,k＝1,2,...,K,||μ i (t)-μ j (t) || represents the Euclidean distance between the centers of the ith and jth neurons, σ i (t) is the width of action of the ith hidden neuron;
the orthogonality definition expression at time t of the neuron is:
wherein mu k (t)＝[μ k,1 (t),μ k,2 (t),...,μ k,D (t)]Is the center vector, μ, of the kth hidden neuron k (t)∈R D ,k＝1,2,...,K,||x(t)-μ k (t) || represents x (t) and μ k Euclidean distance, sigma, between (t) k (t) is the width of action of the kth hidden neuron.
In the network construction stage, increasing and decreasing the neurons according to the integrality and the orthogonality of the neurons, and supplementing the integrality of the network by adding new neurons when the integrality of the neurons is insufficient; when a problem of poor orthogonality occurs between neurons, redundant information of the network is reduced by deleting the neurons.
The present invention is not limited to the above embodiments, but is capable of modification and variation in detail, and other modifications and variations can be made by those skilled in the art without departing from the scope of the present invention.
Claims (10)
1. A sewage treatment process effluent quality prediction method based on a recursion RBF neural network is characterized in that sewage is treated, data are collected in the treatment process, related variables are selected, and finally nonlinear normalization treatment is carried out on the data;
the related variables are dynamically selected at each moment by adopting a dynamic PLS algorithm based on a sliding window technology, and the design selection method comprises the following steps:
wherein X (t) is the acquired data at the time t, X (t) is the modeling data set at the time t, and L is the length of the sliding window, and the value of the sliding window is set according to the working condition state; when t is more than or equal to L, PLS algorithm analysis is carried out on the data from t-L to t, and when t is less than L, PLS algorithm analysis is carried out on all sample data, and key relevant variables of the water quality prediction index of the water outlet in the time period are selected; and constructing a recursive RBF neural network by taking the selected key variable as input and the corresponding effluent quality prediction index as output.
2. The method for predicting the effluent quality in the sewage treatment process based on the recurrent RBF neural network as set forth in claim 1, wherein the data compression technique is a linear compression, and then a logarithmic normalization process is performed, and a logarithmic normalization process rule formula is:
wherein x is min And x max Respectively the minimum and the maximum of the normalized data set, and a is an integer between 2 and 10.
3. The method for predicting the effluent quality of a sewage treatment process based on a recurrent RBF neural network as recited in claim 1, wherein the collected data is screened and decomposed by a partial least squares method.
4. The method for predicting the effluent quality of a sewage treatment process based on a recurrent RBF neural network as claimed in claim 3, wherein the sewage is treated by a sewage treatment system, data acquisition is carried out in the treatment process, and the number of data acquisition points is 10 and the number of data acquisition points comprises 57 acquisition amounts;
the sewage treatment system comprises a primary sedimentation tank, an A tank, a B tank and a secondary sedimentation tank which are sequentially communicated, wherein a water inlet is formed in the primary sedimentation tank, a water outlet is formed in the secondary sedimentation tank, the bottom of the primary sedimentation tank and the bottom of the secondary sedimentation tank are communicated with a sludge treatment device, the bottom of the secondary sedimentation tank is communicated with the bottom of the A tank, the A tank comprises an anaerobic zone of the A tank and an anoxic zone of the A tank, the B tank is an aerobic zone of the B tank and is divided into three parts, and the anaerobic zone of the A tank is communicated with the bottom of the last part of the aerobic zone of the B tank through an outer return pipe.
5. The method for predicting the effluent quality in the sewage treatment process based on the recurrent RBF neural network according to claim 4, wherein the first collecting point is arranged in front of a water inlet of the primary sedimentation tank, the second collecting point is arranged in an anaerobic zone of the tank A, the third collecting point is arranged in an anoxic zone of the tank A, the fourth to sixth collecting points are sequentially arranged in three parts of an aerobic zone of the tank B, the seventh collecting point is arranged in the secondary sedimentation tank, the eighth collecting point is arranged in an outer reflux pipe, the ninth collecting point is arranged in the sludge treatment device, and the tenth collecting point is arranged at a sludge outlet at the bottom of the secondary sedimentation tank;
the first collecting point collects the inflow water flow, the inflow water pH value, the inflow water COD and the inflow water ammonia nitrogen NH 3 N and the total plant treated water quantity;
the second collecting point collects the anaerobic ORP of the pool A, the anoxic front ORP of the pool A and the anoxic last nitrate nitrogen NO of the pool A 3 N, A pool anoxic end MLSS, A pool anoxic end liquid level, A pool aerobic pre-DO, A pool aerobic mid-DO and A pool good pre-ammonia nitrogen NH 3 N, primary ammonia nitrogen at the end of the pool A, the temperature of the pool A and the air flow rate of the pool A are 1;
the third collecting point collects the second aerobic DO of the A pool, the second good last orthophosphate of the A pool and the first good last ammonia nitrogen NH of the A pool 3 N, A pool air flow 2, pool A air flow 3, pool A internal reflux pump 2 frequency;
the fourth collecting point collects the anaerobic ORP of the B pool, the anoxic front ORP of the B pool and the anoxic last nitrate nitrogen NO of the B pool 3 N, B pool anoxic end MLSS, B pool anoxic end liquid level, B pool-aerobic pre-DO, B pool-pre-well ammonia nitrogen NH 3 N, dosing pump flowmeter 1, dosing pump 1 control command, secondary A pool ammonia nitrogen, secondary A oxygen-eliminating to anoxic rotary pump 1 frequency, secondary B oxygen-eliminatingOxygen to anoxic rotary pump 2 frequency, B pool air flow 1, B pool internal reflux pump 3 frequency;
the fifth collecting point collects DO in the first aerobic tank of the B tank, DO in the second aerobic tank of the B tank, orthophosphate in the second good tank of the B tank, a dosing pump flowmeter 2, a dosing pump 2 control command, the temperature of the B tank, the air flow rate of the B tank 2 and the frequency of a reflux pump 4 in the B tank;
collecting PAC2 series biological cell flowmeter, secondary series B cell end test ammonia nitrogen, methanol addition amount and B cell air flow 3 at a sixth collection point;
a seventh acquisition point acquires positive phosphorus OP at the water outlet end of the secondary sedimentation tank, total phosphorus OP at the water outlet end of the secondary sedimentation tank, the water outlet pH value of the secondary sedimentation tank and total nitrogen TN at the water outlet end of the secondary sedimentation tank;
an eighth acquisition point acquires the external reflux flowmeter;
a ninth collection point collects the sludge discharge amount of the residual sludge;
and the tenth acquisition point acquires the backwash water quantity of the sand filter.
6. The method for predicting the effluent quality of a wastewater treatment process based on a recurrent RBF neural network as recited in claim 1, wherein the recurrent RBF neural network design comprises the steps of:
the historical effluent quality index is used as a recursion input for predicting the current effluent quality index, and a memory factor is designed in a recursion link for adjusting the effect of the historical effluent quality;
the average working condition is represented by the average distance between working conditions, the mutation degree of the working condition at a certain moment is quantified by the distance between the working condition at the certain moment and the average working condition, and the memory factor is set according to the negative correlation of the mutation degree, wherein the specific expression formula is as follows:
wherein t is the current time, N is the hysteresis period, the adjustment is carried out according to the seasonal variation, generally 2-3 weeks are taken, k is the adjustment parameter, and can be obtained through learning,is the average working condition of the time period from N-t to t, x i Is the working condition of the ith moment in the time period from N-t to t.
7. The method for predicting the effluent quality of a sewage treatment process based on a recurrent RBF neural network as recited in claim 6, wherein the parameter optimization method of the recurrent RBF neural network is as follows: network optimization is performed using the instantaneous error function as a cost function,
8. the method for predicting the effluent quality of a sewage treatment process based on a recurrent RBF neural network as claimed in claim 7, wherein the parameter optimization is adjusted according to full gradient information, and the full gradient calculation formula is as follows:
wherein θ n (t) is the N-th parameter to be optimized in the network, y (t) is the network output at time t, y (t-i) is the network output at the i-th time in the period of N-t, phi k (t) is the output of the kth neuron of the designed neural network,
wherein x (t) = [ x ] 1 (t),x 2 (t),…,x d (t),...,x D (t)]T is the input vector of the network, x (T) ∈R D ,d＝1,2,...,D,μ k (t)＝[μ k,1 (t),μ k,2 (t),...,μ k,D (t)]Is the center vector, μ, of the kth hidden neuron k (t)∈R D ,k＝1,2,...,K,||x(t)-μ k (t) || represents x (t) and μ k (t)The Euclidean distance between sigma k (t) is the width of action of the kth hidden neuron.
9. The method for predicting the quality of effluent from a wastewater treatment process based on a recurrent RBF neural network as claimed in claim 8, wherein the structural adjustment of the recurrent RBF neural network comprises the steps of:
the integrity definition expression at the time t of the neuron is as follows:
wherein phi (t) = [ phi ] 1 (t),φ 2 (t),…,φ k (t)]Is the neuron output, kappa (t) is the kernel matrix of the RBF neural network, τ 2 (x(t))∈(0,1)，τ 2 A larger (x (t)) means better integrity of ψ (μ (t)), where
Wherein,
wherein mu k (t)＝[μ k,1 (t),μ k,2 (t),...,μ k,D (t)]Is the center vector, μ, of the kth hidden neuron k (t)∈R D ,k＝1,2,...,K,||μ i (t)-μ j (t) || represents the Euclidean distance between the centers of the ith and jth neurons, σ i (t) is the width of action of the ith hidden neuron.
10. The method for predicting the effluent quality of a sewage treatment process based on a recurrent RBF neural network as claimed in claim 9, wherein the orthogonality defining expression at the time t of the neurons is:
wherein mu k (t)＝[μ k,1 (t),μ k,2 (t),...,μ k,D (t)]Is the center vector, μ, of the kth hidden neuron k (t)∈R D ,k＝1,2,...,K,||μ i (t)-μ j (t) || represents the Euclidean distance between the centers of the ith and jth neurons, σ i (t) is the width of action of the ith hidden neuron;
in the network construction stage, increasing and decreasing the neurons according to the integrality and the orthogonality of the neurons, and supplementing the integrality of the network by adding new neurons when the integrality of the neurons is insufficient; when a problem of poor orthogonality occurs between neurons, redundant information of the network is reduced by deleting the neurons.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311451452.7A CN117493803A (en) | 2023-11-02 | 2023-11-02 | Sewage treatment process effluent quality prediction method based on recursive RBF neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311451452.7A CN117493803A (en) | 2023-11-02 | 2023-11-02 | Sewage treatment process effluent quality prediction method based on recursive RBF neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117493803A true CN117493803A (en) | 2024-02-02 |
Family
ID=89684219
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311451452.7A Pending CN117493803A (en) | 2023-11-02 | 2023-11-02 | Sewage treatment process effluent quality prediction method based on recursive RBF neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117493803A (en) |
-
2023
- 2023-11-02 CN CN202311451452.7A patent/CN117493803A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107358021B (en) | DO prediction model establishment method based on BP neural network optimization | |
| US10570024B2 (en) | Method for effluent total nitrogen-based on a recurrent self-organizing RBF neural network | |
| CN105510546B (en) | A kind of biochemical oxygen demand (BOD) BOD intelligent detecting methods based on self-organizing Recurrent RBF Neural Networks | |
| CN111553468B (en) | Method for accurately predicting water quality of effluent of sewage treatment plant | |
| CN111291937A (en) | Method for predicting quality of treated sewage based on combination of support vector classification and GRU neural network | |
| CN102854296B (en) | Sewage-disposal soft measurement method on basis of integrated neural network | |
| CN111354423A (en) | Effluent ammonia nitrogen concentration prediction method of self-organizing recursive fuzzy neural network based on multivariate time series analysis | |
| CN108562709A (en) | A kind of sewage disposal system water quality monitoring method for early warning based on convolution self-encoding encoder extreme learning machine | |
| CN110889085A (en) | Intelligent wastewater monitoring method and system based on complex network multiple online regression | |
| CN107247888B (en) | Method for soft measurement of total phosphorus TP (thermal transfer profile) in sewage treatment effluent based on storage pool network | |
| CN112989704B (en) | IRFM-CMNN effluent BOD concentration prediction method based on DE algorithm | |
| CN101556458B (en) | Automatic control algorithm for feeding vitriol in tap water by coagulation | |
| CN114275912B (en) | Aeration system dissolved oxygen control method based on adaptive neural network model | |
| CN111693667A (en) | Water quality detection system and method based on gated recursive array | |
| CN111125907B (en) | Sewage treatment ammonia nitrogen soft measurement method based on hybrid intelligent model | |
| CN106706491B (en) | Intelligent detection method for membrane bioreactor-MBR water permeability | |
| Wang et al. | A full-view management method based on artificial neural networks for energy and material-savings in wastewater treatment plants | |
| CN114781249A (en) | High-density clarification tank dosage prediction and control method based on multidimensional scoring model | |
| CN201330211Y (en) | Working parameter self-optimizing simulation system for sewage treatment plant | |
| CN117493803A (en) | Sewage treatment process effluent quality prediction method based on recursive RBF neural network | |
| CN114149076B (en) | Intelligent debugging system of anaerobic ammonia oxidation sewage treatment system | |
| CN113222324B (en) | Sewage quality monitoring method based on PLS-PSO-RBF neural network model | |
| Zhang et al. | Neighborhood component analysis for modeling papermaking wastewater treatment processes | |
| Zeng et al. | A Convolutional Neural Network-based Prediction Mechanism for Sewage Treatment | |
| CN113838542B (en) | Intelligent prediction method and system for chemical oxygen demand |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |