CN115557600A - A biochemical reaction artificial neural network intelligent aeration device and its control method - Google Patents

Abstract

The invention discloses a biochemical reaction artificial neural network intelligent aeration device and a control method thereof. The invention deploys a neural network model in a controller, and uses the neural network model to collect data according to the influent water quality monitoring component and the effluent water quality monitoring component. The data can be used to adjust the network parameters, so that the present invention can optimize the calculation process according to the water quality, water quantity and external environment, and accurately control the aeration amount.

Description

A biochemical reaction artificial neural network intelligent aeration device and its control method

technical field

The invention relates to the technical field of sewage treatment, in particular to an artificial neural network intelligent aeration device for biochemical reactions and a control method thereof.

Background technique

The SBR water treatment process is the sequence batch activated sludge method, also known as the intermittent activated sludge method sewage treatment process. The SBR process is based on the process of activated sludge biological treatment of wastewater that degrades organic matter, ammonia nitrogen and other pollutants in sewage with suspended microorganisms under aerobic-anoxic-anaerobic conditions. In order to provide aerobic conditions, aeration is required in the SBR biochemical reaction tank, which is often provided in the form of an aerator; in the anoxic-anaerobic stage, the aeration is stopped.

The amount of aeration directly determines the growth activities of aerobic microorganisms, and plays an important role in the biochemical reaction stages such as organic matter degradation and nitrification; if the amount of aeration is insufficient, it will lead to insufficient dissolved oxygen in the water, and aerobic bacteria cannot survive normally, so that Sludge activity is inhibited, nitrification is insufficient, and the chemical oxygen demand and ammonia nitrogen of the effluent exceed the standard, which affects the water quality of the effluent; if the aeration is excessive, that is, the aeration has exceeded the actual oxygen required by the activated sludge to treat sewage, resulting in activated sewage. Sludge self-oxidation will cause adverse consequences such as sludge aging, deflocculation and sludge bulking. Excessive aeration not only has a negative impact on the process, but also increases unnecessary energy consumption.

The existing aeration technology cannot be adjusted and optimized according to the water quality, water quantity and external environment, making the aeration volume inaccurate.

Contents of the invention

Aiming at the above-mentioned deficiencies in the prior art, the invention provides a biochemical reaction artificial neural network intelligent aeration device and its control method to solve the problem that the existing aeration technology cannot be adjusted and optimized according to the water quality, water quantity and external environment, so that The problem of inaccurate aeration volume.

In order to achieve the purpose of the above invention, the technical solution adopted in the present invention is: an artificial neural network intelligent aeration device for biochemical reactions, including: a controller, an aerator, a regulating tank, an SBR biochemical tank, an MBR membrane separation tank, and a water outlet tank , Influent water quality monitoring components and effluent water quality monitoring components;

A lifting pump is set in the regulating tank to pump the sewage into the regulating tank into the SBR biochemical pool; the SBR biochemical pool is used for biochemical reaction of the sewage, and the sewage after the biochemical reaction is poured into the MBR membrane for separation pool; the MBR membrane separation pool is used to separate the sludge from the sewage after the biochemical reaction to obtain filtered water; the outlet pool is used to hold the filtered water; the influent water quality monitoring component is inserted into the adjustment pool, Used to collect influent data and send it to the controller; the effluent water quality monitoring component is inserted into the SBR biochemical tank to collect water quality data during the aeration process and send it to the controller; the controller is used to collect the water quality data according to the influent data and Water quality data, to control the aeration rate of the aerator; the aerator is used to input oxygen to the SBR biochemical tank.

Further, the influent water quality monitoring component includes: flowmeter, COD detector, TP detector, TN detector, DO detector, temperature detector and pH detector; the effluent water quality monitoring component includes: COD detector , TP detector, TN detector, DO detector, sludge concentration detector, pH detector and ORP detector.

A method for controlling an artificial neural network intelligent aeration device for biochemical reactions, comprising the following steps:

S1. According to the data collected by the influent water quality monitoring component and the effluent water quality monitoring component, the neural network model is trained to obtain a preliminary trained neural network model;

S2. Deploy the initially trained neural network model in the controller;

S3, collect the data of the influent water quality monitoring component and the effluent water quality monitoring component in real time, and input it into the controller, and adjust the network parameters of the neural network model of preliminary training;

S4. Calculate the aeration volume of the aerator according to the adjusted neural network model.

Further, the neural network model in the step S1 includes: an input layer, a hidden layer and an output layer;

The relationship between the input layer and the output layer is:

为第k个输入量对应的实际输出，C_jk为第j个隐藏层神经元第k个输入量对应的径向基中心，σ_j为第j个隐藏层神经元的径向基宽度；D_jk为第j个隐藏层神经元第k个输入量的权重，L为隐藏层神经元的数量，w_j为第j个隐藏层神经元的权重，N为隐藏层神经元的数量。Among them, Y is the output of the output layer, X _k is the kth input of the input layer, K is the number of inputs,

is the actual output corresponding to the kth input quantity, C _jk is the radial basis center corresponding to the kth input quantity of the jth hidden layer neuron, σ _j is the radial basis width of the jth hidden layer neuron; D _jk is the weight of the kth input of the jth hidden layer neuron, L is the number of hidden layer neurons, _wj is the weight of the jth hidden layer neuron, and N is the number of hidden layer neurons.

Further, said obtaining the radial basis center includes the following steps:

A1. Calculate the distance between all input quantities to obtain a set of sample distances;

A2. Select a plurality of different sample distances from the sample distance set as the initial center;

A3. Calculate the distance from multiple input quantities to each initial center;

A4. Determine whether there is a distance smaller than the distance threshold. If yes, add the input amount to the corresponding initial center to obtain multiple similar sets. If not, use the distance as the new initial center, and jump to step A3 until all Input quantities are distributed to similar sets;

A5. Obtain the radial basis center according to all input quantities in the similarity set.

The beneficial effect of the above further scheme is: first select a plurality of different sample distances from the sample distance set as the initial center, obtain multiple initial centers, find the input quantity close to the initial center, and classify it into a similar set, The distances of the input quantities in the similar set are all lower than the distance threshold. When there is an input quantity that is far away from the existing initial center, that is, when the distance threshold is not met, the distance is used as the new initial center, so that each The input quantities can be divided into respective similar sets, and the radial basis center is found adaptively.

Further, the formula for calculating the radial basis center in the step A5 is:

Among them, C _jk is the radial basis center corresponding to the k-th input of the j-th hidden layer neuron in a similar set, P is the number of inputs in the similar set, and X _k is the k-th input of the input layer.

Further, the weight update formula of the jth hidden layer neuron is:

w _j ⁱ⁺¹ ＝w _j ⁱ -△w _j ⁱ

Among them, w _j ⁱ⁺¹ is the weight of the jth hidden layer neuron in the i+1th iteration, w _j ⁱ is the weight of the jth hidden layer neuron in the ith iteration, and △w _j ⁱ is the weight of the jth hidden layer neuron in the ith iteration. The weight variation of the jth hidden layer neuron in the i iteration.

Further, the formula of the weight variation Δw _j ⁱ is:

为第i-1次迭代的梯度均值估计量，θ_i-1为第i-1次迭代的梯度统计离散度估计量，λ_i-1为第i-1次迭代的衰减因子，λ_i-2为第i-2次迭代的衰减因子，

为第i-2次迭代的梯度均值估计量，θ_i-2为第i-2次迭代的梯度统计离散度估计量，Yⁱ为第i次迭代的输出层的输出，ε为常数。Wherein, n is the moving step size,

is the gradient mean estimator of the i-1th iteration, θi _-1 is the gradient statistical dispersion estimator of the i-1th iteration, λi _-1 is the attenuation factor of the i-1th iteration, _{λi- 2} is the attenuation factor of the i-2th iteration,

is the gradient mean estimator of the i-2 iteration, θ _i-2 is the gradient statistical dispersion estimator of the i-2 iteration, Y ⁱ is the output of the output layer of the i-th iteration, and ε is a constant.

In summary, the beneficial effects of the present invention are: the present invention deploys the neural network model in the controller, and adjusts the network parameters according to the data collected by the influent water quality monitoring component and the effluent water quality monitoring component through the neural network model, so that the present invention It can optimize the calculation process according to water quality, water quantity and external environment, and precisely control the aeration amount.

Description of drawings

Fig. 1 is the schematic diagram of the artificial neural network intelligent aeration device of a kind of biochemical reaction;

Fig. 2 is a flowchart of a control method of an artificial neural network intelligent aeration device for biochemical reactions.

detailed description

The specific embodiments of the present invention are described below so that those skilled in the art can understand the present invention, but it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes Within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are included in the protection list.

As shown in Figure 1, an artificial neural network intelligent aeration device for biochemical reactions includes: a controller, an aerator, a regulating tank, an SBR biochemical tank, an MBR membrane separation tank, an outlet tank, an inlet water quality monitoring component, and an outlet water Water quality monitoring components;

A lifting pump is set in the regulating tank to pump the sewage into the regulating tank into the SBR biochemical pool; the SBR biochemical pool is used for biochemical reaction of the sewage, and the sewage after the biochemical reaction is poured into the MBR membrane for separation pool; the MBR membrane separation pool is used to separate the sludge from the sewage after the biochemical reaction to obtain filtered water; the outlet pool is used to hold the filtered water; the influent water quality monitoring component is inserted into the adjustment pool, Used to collect influent data and send it to the controller; the effluent water quality monitoring component is inserted into the SBR biochemical tank to collect water quality data during the aeration process and send it to the controller; the controller is used to collect the water quality data according to the influent data and Water quality data, to control the aeration rate of the aerator; the aerator is used to input oxygen to the SBR biochemical tank.

The influent water quality monitoring component includes: a flow meter, a COD detector, a TP detector, a TN detector, a DO detector, a temperature detector, a pH detector and a sludge concentration detector; the outlet water quality monitoring component includes: COD detector, TP detector, TN detector, DO detector, sludge concentration detector, pH detector and ORP detector.

As shown in Figure 2, a control method of an artificial neural network intelligent aeration device for a biochemical reaction comprises the following steps:

S1. According to the data collected by the influent water quality monitoring component and the effluent water quality monitoring component, the neural network model is trained to obtain a preliminary trained neural network model;

S2. Deploy the initially trained neural network model in the controller;

S3, collect the data of the influent water quality monitoring component and the effluent water quality monitoring component in real time, and input it into the controller, and adjust the network parameters of the neural network model of preliminary training;

S4. Calculate the aeration volume of the aerator according to the adjusted neural network model.

Through the data collected before and after the biochemical reaction, the reference aeration volume for the next operation cycle is obtained through machine learning. As the operating time of the system increases, the stored aeration volume process data increases, and the more aeration volume data that can be referenced in the future , the aeration volume will become more and more accurate, and the effluent water quality will be more stable, achieving the goals of intelligence, high efficiency and energy saving.

In step S1, through the data collected by the influent water quality monitoring component and the data collected by the effluent water quality monitoring component, the corresponding relationship between the influent and effluent data after aeration by the aerator is initially found, and the neural network model for preliminary training is obtained. Apply the initially trained neural network model to the biochemical process of the biochemical pool. In the actual operation process, the network parameters of the neural network model are continuously fine-tuned again through the data of the water after aeration, which is equivalent to the actual operation process. In training the neural network model, the network parameters are changed to adapt to the external environment.

The data collected by the influent water quality monitoring component is the input data of the neural network model, and the data collected by the effluent water quality monitoring component is the actual data after passing through the aerator. According to the input data and actual data, the network parameters of the neural network model can be adjusted. .

The neural network model in the step S1 includes: an input layer, a hidden layer and an output layer;

The relationship between the input layer and the output layer is:

为第k个输入量对应的实际输出，C_jk为第j个隐藏层神经元第k个输入量对应的径向基中心，σ_j为第j个隐藏层神经元的径向基宽度；D_jk为第j个隐藏层神经元第k个输入量的权重，L为隐藏层神经元的数量，w_j为第j个隐藏层神经元的权重，N为隐藏层神经元的数量。Among them, Y is the output of the output layer, X _k is the kth input of the input layer, K is the number of inputs,

is the actual output corresponding to the kth input quantity, C _jk is the radial basis center corresponding to the kth input quantity of the jth hidden layer neuron, σ _j is the radial basis width of the jth hidden layer neuron; D _jk is the weight of the kth input of the jth hidden layer neuron, L is the number of hidden layer neurons, _wj is the weight of the jth hidden layer neuron, and N is the number of hidden layer neurons.

The data of the kth input quantity X _k of the input layer comes from the data collected by the influent water quality monitoring component, and the actual output corresponding to the kth input quantity comes from the data collected by the effluent water quality monitoring component.

The acquisition of the radial basis center includes the following steps:

A1. Calculate the distance between all input quantities to obtain a set of sample distances;

A2. Select a plurality of different sample distances from the sample distance set as the initial center;

A3. Calculate the distance from multiple input quantities to each initial center;

A4. Determine whether there is a distance smaller than the distance threshold. If yes, add the input amount to the corresponding initial center to obtain multiple similar sets. If not, use the distance as the new initial center, and jump to step A3 until all Input quantities are distributed to similar sets;

A5. Obtain the radial basis center according to all input quantities in the similarity set.

In this embodiment, the types of input quantities include: reaction water temperature, sludge concentration, influent flow rate, pH value of the regulating tank, COD value, NH3-N value, ORP value and TP value.

The formula for calculating the radial basis center in the step A5 is:

Among them, C _jk is the radial basis center corresponding to the k-th input of the j-th hidden layer neuron in a similar set, P is the number of inputs in the similar set, and X _k is the k-th input of the input layer.

The weight update formula of the jth hidden layer neuron is:

w _j ⁱ⁺¹ ＝w _j ⁱ -Δw _j ⁱ

Among them, w _j ⁱ⁺¹ is the weight of the jth hidden layer neuron in the i+1th iteration, w _j ⁱ is the weight of the jth hidden layer neuron in the ith iteration, Δw _j ⁱ is the i The weight variation of the jth hidden layer neuron in the iteration.

The formula of the weight variation Δw _j ⁱ is:

为第i-1次迭代的梯度均值估计量，θ_i-1为第i-1次迭代的梯度统计离散度估计量，λ_i-1为第i-1次迭代的衰减因子，λ_i-2为第i-2次迭代的衰减因子，

为第i-2次迭代的梯度均值估计量，θ_i-2为第i-2次迭代的梯度统计离散度估计量，Yⁱ为第i次迭代的输出层的输出，ε为常数。Wherein, n is the moving step size,

is the gradient mean estimator of the i-1th iteration, θi _-1 is the gradient statistical dispersion estimator of the i-1th iteration, λi _-1 is the attenuation factor of the i-1th iteration, _{λi- 2} is the attenuation factor of the i-2th iteration,

is the gradient mean estimator of the i-2 iteration, θ _i-2 is the gradient statistical dispersion estimator of the i-2 iteration, Y ⁱ is the output of the output layer of the i-th iteration, and ε is a constant.

In this embodiment, the relationship formula between the input layer and the output layer, the process of obtaining the radial basis center, the formula of the radial basis center, the weight update formula of the neuron in the hidden layer and the formula of the weight variation are applicable to the neuron of the present invention. All training processes of the network model.

Claims (8)

1. An artificial neural network intelligent aeration device for biochemical reactions, characterized in that it includes: a controller, an aerator, a regulating tank, an SBR biochemical tank, an MBR membrane separation tank, a water outlet tank, an inlet water quality monitoring component and an outlet water Water quality monitoring components; A lifting pump is set in the regulating tank to pump the sewage into the regulating tank into the SBR biochemical pool; the SBR biochemical pool is used for biochemical reaction of the sewage, and the sewage after the biochemical reaction is poured into the MBR membrane for separation pool; the MBR membrane separation pool is used to separate the sludge from the sewage after the biochemical reaction to obtain filtered water; the outlet pool is used to hold the filtered water; the influent water quality monitoring component is inserted into the adjustment pool, Used to collect influent data and send it to the controller; the effluent water quality monitoring component is inserted into the SBR biochemical tank to collect water quality data during the aeration process and send it to the controller; the controller is used to collect the water quality data according to the influent data and Water quality data, to control the aeration rate of the aerator; the aerator is used to input oxygen to the SBR biochemical tank. 2. The artificial neural network intelligent aeration device for biochemical reactions according to claim 1, wherein the influent water quality monitoring components include: flowmeter, COD detector, TP detector, TN detector, DO detector instrument, temperature detector and pH detector; the effluent water quality monitoring components include: COD detector, TP detector, TN detector, DO detector, sludge concentration detector, pH detector and ORP detector. 3. the control method of the artificial neural network intelligent aeration device of biochemical reaction according to claim 1, is characterized in that, comprises the following steps: S1. According to the data collected by the influent water quality monitoring component and the effluent water quality monitoring component, the neural network model is trained to obtain a preliminary trained neural network model; S2. Deploy the initially trained neural network model in the controller; S3, collect the data of the influent water quality monitoring component and the effluent water quality monitoring component in real time, and input it into the controller, and adjust the network parameters of the neural network model of preliminary training; S4. Calculate the aeration volume of the aerator according to the adjusted neural network model. 4. the control method of the artificial neural network intelligent aeration device of biochemical reaction according to claim 3, is characterized in that, in described step S1, neural network model comprises: input layer, hidden layer and output layer; The relationship between the input layer and the output layer is:

Among them, Y is the output of the output layer, X _k is the kth input of the input layer, K is the number of inputs,

is the actual output corresponding to the kth input quantity, C _jk is the radial basis center corresponding to the kth input quantity of the jth hidden layer neuron, σ _j is the radial basis width of the jth hidden layer neuron; D _jk is the weight of the kth input of the jth hidden layer neuron, L is the number of hidden layer neurons, _wj is the weight of the jth hidden layer neuron, and N is the number of hidden layer neurons. 5. the control method of the artificial neural network intelligent aeration device of biochemical reaction according to claim 4, is characterized in that, described acquisition radial basis center comprises the following steps: A1. Calculate the distance between all input quantities to obtain a set of sample distances; A2. Select a plurality of different sample distances from the sample distance set as the initial center; A3. Calculate the distance from multiple input quantities to each initial center; A4. Determine whether there is a distance smaller than the distance threshold. If yes, add the input amount to the corresponding initial center to obtain multiple similar sets. If not, use the distance as the new initial center, and jump to step A3 until all Input quantities are distributed to similar sets; A5. Obtain the radial basis center according to all input quantities in the similarity set. 6. the control method of the artificial neural network intelligent aeration device of biochemical reaction according to claim 5, is characterized in that, the calculation formula of radial basis center in described step A5 is:

Among them, C _jk is the radial basis center corresponding to the k-th input of the j-th hidden layer neuron in a similar set, P is the number of inputs in the similar set, and X _k is the k-th input of the input layer. 7. the control method of the artificial neural network intelligent aeration device of biochemical reaction according to claim 4, is characterized in that, the weight updating formula of described j hidden layer neuron is: w _j ⁱ⁺¹ ＝w _j ⁱ -Δw _j ⁱ Among them, w _j ⁱ⁺¹ is the weight of the jth hidden layer neuron in the i+1th iteration, w _j ⁱ is the weight of the jth hidden layer neuron in the ith iteration, Δw _j ⁱ is the i The weight variation of the jth hidden layer neuron in the iteration. 8. the control method of the artificial neural network intelligent aeration device of biochemical reaction according to claim 7, is characterized in that, the formula of described weight variation Δw _j ⁱ is:

Wherein, n is the moving step size,

is the gradient mean estimator of the i-1th iteration, θi _-1 is the gradient statistical dispersion estimator of the i-1th iteration, λi _-1 is the attenuation factor of the i-1th iteration, _{λi- 2} is the attenuation factor of the i-2th iteration,

is the gradient mean estimator of the i-2 iteration, θ _i-2 is the gradient statistical dispersion estimator of the i-2 iteration, Y ⁱ is the output of the output layer of the i-th iteration, and ε is a constant.

Images