An intelligent decision-making method for sludge bulking based on type-two fuzzy width learning

Abstract

The present invention proposes an intelligent decision-making method for sludge bulking based on type-II fuzzy width learning, which is aimed at the problems of frequent sludge bulking events and huge harm in sewage treatment plants. The present invention uses a width learning network to establish a sludge bulking decision model, and uses type-II fuzzy neurons to replace the characteristic layer neurons of the width learning network to achieve accurate sludge bulking decision-making and reduce the incidence of sludge bulking. Experimental results show that this method can provide accurate decision-making opinions, reduce the harm caused by sludge bulking, ensure the safety of the sewage treatment process, and promote the efficient and stable operation of sewage treatment plants.

Description

Technical Field

[0001] Based on the operating characteristics of the sewage treatment process, the present invention uses a type-two fuzzy width learning network to construct a sludge bulking intelligent decision-making model, which provides decision-making opinions for inhibiting sludge bulking. Type-two fuzzy neurons are used to replace the characteristic layer neurons of the width learning network to achieve accurate sludge bulking decisions and reduce the incidence of sludge bulking. Applying intelligent decision-making methods to sewage treatment systems has an important impact on the safe, stable and efficient operation of sewage treatment. It is an important branch of the field of advanced manufacturing technology, belonging to both the field of water treatment and the field of intelligent control. Therefore, sludge bulking intelligent decision-making is of great significance in sewage treatment systems. Background Art

[0002] The activated sludge sewage treatment technology mixes activated sludge with wastewater and aerates it to decompose and separate organic pollutants in the wastewater, achieving the purification effect. At present, more than 90% of sewage treatment plants in my country use the activated sludge process. However, due to the insufficient automation level of sewage treatment equipment and imperfect monitoring network in my country's sewage treatment plants, sludge bulking occurs frequently. Sludge bulking will reduce the sludge settling rate, which will lead to difficulties in mud-water separation, affect the effluent water quality, and seriously endanger the normal and stable operation of the sewage treatment plant. Therefore, in-depth analysis of the sludge bulking phenomenon and research on sludge bulking decision-making methods are of great research significance for ensuring the stable operation of the sewage treatment process and improving treatment efficiency.

[0003] At present, a large number of studies on sludge bulking decision-making have been carried out, but the results are not optimistic. On the one hand, due to the complexity of the sludge bulking mechanism, the sludge bulking decision-making method based on the mechanism model cannot cover the comprehensive operation mechanism and is difficult to meet the requirements of system operation. On the other hand, although some decision-making methods based on expert systems have achieved certain results, the sewage treatment process is a dynamic, complex and time-varying system, which makes it difficult for traditional sludge bulking decision-making models to adapt to dynamically changing working conditions and cannot achieve accurate decision-making results. Therefore, considering the serious hazards of sludge bulking, designing a real-time and accurate sludge bulking decision-making method has important theoretical significance and application value for realizing safe, stable and efficient operation of the sewage treatment process, inhibiting sludge bulking, and improving effluent water quality.

[0004] The present invention designs a sludge bulking intelligent decision-making method based on type-II fuzzy width learning. By constructing a decision-making model based on type-II fuzzy width learning network, type-II fuzzy neurons are used to replace the characteristic layer neurons of the width learning network to achieve accurate sludge bulking decision-making and reduce the incidence of sludge bulking. The present invention not only solves the problem of suppressing sludge bulking in sewage treatment plants, but also improves the effluent water quality and improves the stability and safety of sewage treatment plant operation. Summary of the invention

[0005] The present invention obtains an intelligent decision-making method for sludge bulking based on type-2 fuzzy width learning. The method constructs a decision model based on a type-2 fuzzy width learning network, wherein type-2 fuzzy neurons are used to replace the feature layer neurons of the width learning network, which solves the problem of poor accuracy of traditional decision-making methods and achieves accurate decision-making on sludge bulking.

[0006] The present invention adopts the following technical solutions and implementation steps:

[0007] 1. A sludge bulking intelligent decision-making method based on type II fuzzy width learning, characterized in that a decision model based on a width learning network is established to provide decision support for inhibiting sludge bulking, wherein type II fuzzy neurons are used to replace the characteristic layer neurons of the width learning network to achieve accurate sludge bulking decision-making and reduce the incidence of sludge bulking, comprising the following steps:

[0008] (1) Operation data collection: Sludge bulking is an abnormal phenomenon in which the sludge structure is extremely loose, the volume increases, it floats, and it is difficult to settle and separate. It is easy to cause the sludge settling performance to decline and the sludge to be lost, resulting in the effluent water quality exceeding the standard, and even causing the sewage treatment efficiency to decline and the operation system to collapse. Therefore, it is necessary to provide decision support for inhibiting sludge bulking. Taking the activated sludge sewage treatment system as the research object, the operation data are collected through the instruments installed on site, including: dissolved oxygen, effluent total nitrogen, effluent total phosphorus, effluent sludge load, aeration tank temperature, effluent pH, so as to realize real-time data collection; the data collected by the instrument is transmitted to the programmable logic controller through the communication protocol, and the programmable logic controller transmits the operation data to the host computer through the communication protocol. The data in the host computer is transmitted to the data processing server through the local area network;

[0009] (2) Operation data preprocessing: Taking operation data as the research object, process variables related to sludge bulking are selected as input variables of the intelligent decision-making model: dissolved oxygen, effluent total nitrogen, effluent total phosphorus, effluent sludge load, aeration tank temperature, and effluent pH; the obtained input variables are normalized to [0,1]; operation suggestions are used as output variables of the sludge bulking intelligent decision-making model; all sample data are divided into two groups, one group contains P training samples and the other group contains M test samples, and P>M is generally required;

[0010] (3) Design of a type-two fuzzy width learning model for sludge bulking decision-making: A type-two fuzzy width learning network is used to establish a sludge bulking decision-making model. The type-two fuzzy width learning network structure has four layers, including an input layer, an interval type-two fuzzy neural network layer, an enhancement layer, and an output layer. Its structure is a 6-IJ-10 connection mode. The number of neurons in the input layer of the model for sludge bulking decision-making is determined to be 6, the number of subnetworks in the interval type-two fuzzy neural network layer is I, and I is any positive integer between [1, 10]. The number of neuron groups in the enhancement layer is J, and J is any positive integer between [2, 20]. The number of neurons in the output layer is 10. P training samples are used to train the type-two fuzzy width learning network. The input of the sludge bulking decision-making model is x(t) = [x1(t), x2(t), …, x P (t)],x n (t) = [x 1n (t),x 2n (t),x 3n (t),x 4n (t),x 5n (t),x 6n (t)] is the nth sample at the tth iteration after normalization, n = 1, 2, ..., P, x 1n (t) is the dissolved oxygen concentration in the nth sample at the tth iteration after normalization, x 2n (t) is the total nitrogen concentration of the effluent in the nth sample at the tth iteration after normalization, x 3n (t) is the total phosphorus concentration in the effluent of the nth sample at the tth iteration after normalization, x 4n (t) is the effluent sludge load concentration in the nth sample at the tth iteration after normalization, x 5n (t) is the aeration tank temperature value in the nth sample at the tth iteration after normalization, x 6n (t) is the pH value of the effluent in the nth sample at the tth iteration after normalization. The output of the sludge bulking decision model is the operation recommendation Y d (t) = [Y d1 (t),Y d2 (t),…,Y dP (t)],Y dn (t) = [Y 1dn (t),Y 2dn (t),…,Y 10dn (t)] is the recommended operation for the nth sample at the tth iteration, Y 1dn (t) is operation suggestion 1, which is to increase the aeration volume of the biochemical pool and control the dissolved oxygen concentration to 2-4 mg / L. 2dn (t) is operation suggestion 2, that is, adding urea to the biochemical pool to control the ratio of biological oxygen demand and nitrogen concentration to be greater than 100 mg / L:3 mg / L, Y 3dn(t) Operation suggestion 3 is to add sodium phosphate to the biochemical pool to control the ratio of biological oxygen demand and phosphorus concentration to be greater than 100 mg / L:1 mg / L. 4dn (t) is operation suggestion 4, that is, to adjust the operating parameters and increase the sludge return ratio to 25%-50%, Y 5dn (t) is operation suggestion 5, that is, to introduce boiler steam to raise the sewage temperature to above 22°C, Y 6dn (t) Operation suggestion 6 is to add industrial sulfuric acid or calcium hydroxide to adjust the pH in the aeration tank to 6.5-8.5. 7dn (t) is operation suggestion 7, which is to increase the sludge load to above 0.2-0.3 kgBOD5 / kgMLSS.d, Y 8dn (t) is Operation Recommendation 8, which is to shorten the sludge age in the aeration tank to 7-15 days. 9dn (t) is operation suggestion 9, which is to increase the nitration solution reflux ratio to more than 300%, Y 10dn (t) is operation suggestion 10, which is to increase the concentration of fungicide in the aeration tank to 12-14 mg / L. The layers in the decision model based on the type-II fuzzy width learning network are expressed as follows:

[0011] Type II fuzzy width learning network input layer: This layer consists of 6 neurons, and the output of each neuron is as follows:

[0012] u s (t) = x s (t),s=1,2,...,6, (1)

[0013] Among them, u s (t) is the output value of the sth neuron in the input layer of the type-II fuzzy width learning network at the tth iteration;

[0014] Type-II fuzzy width learning network interval type-II fuzzy neural network layer: The interval type-II fuzzy neural network layer is composed of I subnetworks, and each interval type-II fuzzy neural network subnetwork structure has five layers, including input layer, membership function layer, activation layer, consequent layer and output layer; its structure is a 6-LQVV connection mode, which determines that the number of input layer neurons of each interval type-II fuzzy neural network is 6, the number of membership function layer neurons is L, and L is any positive integer between [2,20]; the number of activation layer neurons is Q, and Q is any positive integer between [2,20]; the number of consequent layer neurons is V, and the number of output layer neurons is V, and V is any positive integer between [2,10]; the output of each layer can be expressed as:

[0015] Interval type-2 fuzzy neural network subnetwork input layer: This layer consists of 6 neurons, and the output of each neuron is as follows:

[0016] rs (t) = u s (t), (2)

[0017] Among them, r s (t) is the output value of the sth neuron in the input layer of the interval type-2 fuzzy neural network subnetwork at the tth iteration;

[0018] Interval type-2 fuzzy neural network subnetwork membership function layer: This layer consists of L neurons, and the output of this layer can be expressed as:

[0019]

[0020]

[0021] Among them, μ ls (t) is the output value of the lth neuron in the membership function layer of the interval type-II fuzzy neural network subnetwork at the tth iteration, ξ ls (t) is the uncertainty center of the sth input and lth membership function layer neuron at the tth iteration, ξ ls (t) is the lower bound of the uncertainty center of the s-th input and the l-th membership function layer neuron at the t-th iteration, and takes a random value in the interval (0,1); is the upper bound of the uncertainty center of the sth input and the lth membership function layer neuron at the tth iteration, randomly taking values ​​in the interval (0,1], and the lower bound of the uncertainty center is less than the upper bound; σ ls (t) is the fixed standard deviation of the sth input and the lth membership function layer neuron at the tth iteration, and takes a random value in the interval (0,1);

[0022] Interval type-2 fuzzy neural network subnetwork activation layer: This layer consists of Q neurons, and the output of this layer can be expressed as:

[0023]

[0024] Among them, f q (t) is the activation intensity of the qth neuron in the activation layer of the interval type-2 fuzzy neural network subnetwork at the tth iteration, f q (t) and are the lower and upper bounds of the activation strength of the qth neuron in the activation layer at the tth iteration;

[0025] Interval type-2 fuzzy neural network subnetwork subsequent layer: This layer consists of V neurons, and the output of this layer can be expressed as:

[0026]

[0027]

[0028] in, o v (t) and are the lower and upper bounds of the output of the vth neuron in the consequent layer of the interval type-2 fuzzy neural network subnetwork at the tth iteration, w vq (n) and are the lower and upper bounds of the weight between the vth neuron in the posterior layer and the qth neuron in the activation layer at the tth iteration, both of which are randomly selected in the interval [-1,1], and the lower bound of the weight is smaller than the upper bound;

[0029] Interval type-2 fuzzy neural network subnetwork output layer: This layer consists of V neurons, and the output of this layer can be expressed as:

[0030]

[0031] Among them, v (t) is the output value of the vth neuron in the output layer of the interval type-2 fuzzy neural network subnetwork at the tth iteration;

[0032] Type II fuzzy width learning network enhancement layer: This layer consists of J groups of neurons, and the output of this layer can be expressed as:

[0033] H(t)=[h1(t),h2(t),...,h J (t)], (9)

[0034] h j (t) = ζ j (O(t)w hj (t)+β hj (t)),j=1,2,...,J, (10)

[0035] Where H(t) is the output value of the enhanced layer of the type-II fuzzy width learning network at the tth iteration, h j (t) is the output value of the jth group of neurons in the enhancement layer at the tth iteration, ζ j () is the activation function of the jth group of neurons in the enhancement layer, O(t) = [o1(t), o2(t), …, o I (t)] is the output value of the interval type-2 fuzzy neural network layer at the tth iteration, o i (t)=[o1(t),o2(t),…,o V (t)] is the output value of the ith subnetwork of the interval type-II fuzzy neural network layer at the tth iteration, i = 1, 2, ..., I, w hj (t) and β hj(t) is the weight and bias value between the jth group of neurons in the interval type-II fuzzy neural network layer and the enhancement layer at the tth iteration;

[0036] Output layer of type-2 fuzzy width learning network: This layer consists of 10 neurons, and the output of this layer can be expressed as:

[0037]

[0038] Where Y(t) is the output value of the output layer of the type-II fuzzy width learning network at the tth iteration, w i (t) is the weight between the output layer of the type-II fuzzy width learning network and the i-th sub-network of the interval type-II fuzzy neural network layer at the t-th iteration, w j (t) is the weight between the jth group of neurons in the output layer and enhancement layer of the type-II fuzzy width learning network at the tth iteration;

[0039] (4) Training the decision model based on the type-two fuzzy width learning network, specifically:

[0040] ① Assume that the current iteration number t = 1, and the maximum iteration number of the decision model based on the type-two fuzzy width learning network is K = 30;

[0041] ② Given the initial interval type-2 fuzzy neural network layer subnetwork number of type-2 fuzzy width learning network is I, the number of enhancement layer neuron groups is J, the number of membership function layer neurons of each interval type-2 fuzzy neural network layer subnetwork is L, the number of activation layer neurons is Q, the number of consequent layer neurons is V, and the number of output layer neurons is V;

[0042] ③ According to formulas (1)-(10), the output Y(t) of the type-two fuzzy width learning network is calculated. The accuracy of the output of the type-two fuzzy width learning network is expressed as:

[0043] A(t)=TF(t) / P, (12)

[0044] Among them, A(t) is the output accuracy of the fuzzy width learning network at the tth iteration, and TF(t) is the number of samples whose output of the fuzzy width learning network is the same as the actual value at the tth iteration. The pseudo-inverse algorithm is used to adjust the parameters of the type-2 fuzzy width learning network:

[0045] W(t)=D(t) + Y d (t), (13)

[0046]

[0047] Where W(t)=[W i (t), W j(t)] is the weight between the output layer of the type-II fuzzy width learning network and the interval type-II fuzzy neural network layer and the enhancement layer at the tth iteration, W i (t) is the weight between the output layer of the type-II fuzzy width learning network and the interval type-II fuzzy neural network layer at the tth iteration, W j (t) is the weight between the output layer and the enhancement layer of the type-II fuzzy width learning network at the tth iteration; D(t) = [O(t, H(t)], λ is a positive constant, randomly selected in the interval (0,1], and E is the unit matrix;

[0048] ④ Let t = t + 1. If t ≤ K or A(t) > 0.05, return to step ③. If t > K and A(t) ≤ 0.05, stop the calculation and exit the loop to complete the training.

[0049] (5) Intelligent Decision-making on Sludge Bulking

[0050] Using the trained type-two fuzzy width learning network sludge bulking decision model, the dissolved oxygen, effluent total nitrogen, effluent total phosphorus, effluent sludge load, aeration tank temperature, and effluent pH of M test samples are used as the input variables of the model to obtain the sludge bulking category output value of the model, which is the operation suggestion corresponding to the category of the sample to be tested, including: Operation suggestion 1: Increase the aeration volume of the biochemical pool and control the dissolved oxygen concentration to 2-4 mg / L; Operation suggestion 2: Add urea to the biochemical pool to control the ratio of biological oxygen demand and nitrogen concentration to be greater than 100 mg / L:3 mg / L; Operation suggestion 3: Add sodium phosphate to the biochemical pool to control the ratio of biological oxygen demand and phosphorus concentration to be greater than 100 mg / L:1mg / L, Operation suggestion 4: Adjust the operating parameters, increase the sludge return ratio to 25%-50%, Operation suggestion 5: Introduce boiler steam to increase the sewage temperature to above 22°C, Operation suggestion 6: Add industrial sulfuric acid or calcium hydroxide to adjust the pH, and control the pH in the aeration tank to 6.5-8.5, Operation suggestion 7: Increase the sludge load to above 0.2-0.3kgBOD5 / kgMLSS.d, Operation suggestion 8: Shorten the sludge age of the aeration tank to 7-15d, Operation suggestion 9: Increase the nitrification liquid return ratio to more than 300%, Operation suggestion 10: Increase the fungicide concentration in the aeration tank to 12-14mg / L to provide decision support for the production process.

[0051] The creativity of the present invention is mainly reflected in:

[0052] Aiming at the problem that sludge swelling is difficult to suppress in current sewage treatment, the present invention proposes a sludge swelling intelligent decision-making method based on type-II fuzzy width learning, which uses type-II fuzzy neurons to replace the feature layer neurons of the width learning network to improve the robustness of the network, solves the problem of suppressing sludge swelling in sewage treatment plants, and meets the demand for accurate suppression of sludge swelling in sewage treatment plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a topological diagram of the type II fuzzy width learning network structure of the present invention;

[0054] Figure 2 This is a sludge bulking decision training effect diagram of the present invention, wherein the red cross represents the actual sludge bulking type, and the blue circle represents the output value based on the type-two fuzzy width learning network decision model;

[0055] Figure 3 This is a sludge bulking decision test effect diagram of the present invention, wherein the red cross is the actual sludge bulking type, and the blue circle is the output value based on the type-two fuzzy width learning network decision model. DETAILED DESCRIPTION

[0056] The experimental data comes from the 2020 data of a sewage treatment plant: the actual detection data of dissolved oxygen, effluent total nitrogen, effluent total phosphorus, effluent sludge load, aeration tank temperature, and effluent pH are taken as process sample data. After eliminating abnormal experimental samples, 100 groups of available data remain, of which 60 groups are used as training samples and the remaining 40 groups are used as test samples.

[0057] The present invention adopts the following technical solutions and implementation steps:

[0058] 1. A sludge bulking intelligent decision-making method based on type II fuzzy width learning, characterized in that a decision model based on a width learning network is established to provide decision support for inhibiting sludge bulking, wherein type II fuzzy neurons are used to replace the characteristic layer neurons of the width learning network to achieve accurate sludge bulking decision-making and reduce the incidence of sludge bulking, comprising the following steps:

[0059] (1) Operation data collection: Sludge bulking is an abnormal phenomenon in which the sludge structure is extremely loose, the volume increases, it floats, and it is difficult to settle and separate. It is easy to cause the sludge settling performance to decline and the sludge to be lost, resulting in the effluent water quality exceeding the standard, and even causing the sewage treatment efficiency to decline and the operation system to collapse. Therefore, it is necessary to provide decision support for inhibiting sludge bulking. Taking the activated sludge sewage treatment system as the research object, the operation data are collected through the instruments installed on site, including: dissolved oxygen, effluent total nitrogen, effluent total phosphorus, effluent sludge load, aeration tank temperature, effluent pH, so as to realize real-time data collection; the data collected by the instrument is transmitted to the programmable logic controller through the communication protocol, and the programmable logic controller transmits the operation data to the host computer through the communication protocol. The data in the host computer is transmitted to the data processing server through the local area network;

[0060] (2) Operation data preprocessing: Taking operation data as the research object, process variables related to sludge bulking are selected as input variables of the intelligent decision-making model: dissolved oxygen, effluent total nitrogen, effluent total phosphorus, effluent sludge load, aeration tank temperature, and effluent pH; the obtained input variables are normalized to [0,1]; operation suggestions are used as output variables of the sludge bulking intelligent decision-making model; all sample data are divided into two groups, one group contains 60 training samples and the other group contains 40 test samples;

[0061] (3) Design of a type-two fuzzy width learning model for sludge bulking decision-making: A type-two fuzzy width learning network is used to establish a sludge bulking decision-making model. The type-two fuzzy width learning network structure has four layers, including an input layer, an interval type-two fuzzy neural network layer, an enhancement layer, and an output layer. Its structure is a 6-IJ-10 connection mode. The number of neurons in the input layer of the model for sludge bulking decision-making is determined to be 6, the number of subnetworks in the interval type-two fuzzy neural network layer is I, and I is any positive integer between [1, 10]. The number of neuron groups in the enhancement layer is J, and J is any positive integer between [2, 20]. The number of neurons in the output layer is 10. 60 training samples are used to train the type-two fuzzy width learning network. The input of the sludge bulking decision-making model is x(t) = [x1(t), x2(t), …, x 60 (t)],x n (t) = [x 1n (t),x 2n (t),x 3n (t),x 4n (t),x 5n (t),x 6n (t)] is the nth sample at the tth iteration after normalization, n = 1, 2, ..., 60, x 1n (t) is the dissolved oxygen concentration in the nth sample at the tth iteration after normalization, x 2n (t) is the total nitrogen concentration of the effluent in the nth sample at the tth iteration after normalization, x 3n (t) is the total phosphorus concentration in the effluent of the nth sample at the tth iteration after normalization, x 4n (t) is the effluent sludge load concentration in the nth sample at the tth iteration after normalization, x 5n (t) is the aeration tank temperature value in the nth sample at the tth iteration after normalization, x 6n (t) is the pH value of the effluent in the nth sample at the tth iteration after normalization. The output of the sludge bulking decision model is the operation recommendation Y d (t) = [Y d1 (t),Y d2 (t),…,Y d60 (t)],Y dn (t) = [Y 1dn(t),Y 2dn (t),…,Y 10dn (t)] is the recommended operation for the nth sample at the tth iteration, Y 1dn (t) is operation suggestion 1, which is to increase the aeration volume of the biochemical pool and control the dissolved oxygen concentration to 2-4 mg / L. 2dn (t) is operation suggestion 2, that is, adding urea to the biochemical pool to control the ratio of biological oxygen demand and nitrogen concentration to be greater than 100 mg / L:3 mg / L, Y 3dn (t) Operation suggestion 3 is to add sodium phosphate to the biochemical pool to control the ratio of biological oxygen demand and phosphorus concentration to be greater than 100 mg / L:1 mg / L. 4dn (t) is operation suggestion 4, that is, to adjust the operating parameters and increase the sludge return ratio to 25%-50%, Y 5dn (t) is operation suggestion 5, that is, to introduce boiler steam to raise the sewage temperature to above 22°C, Y 6dn (t) Operation suggestion 6 is to add industrial sulfuric acid or calcium hydroxide to adjust the pH in the aeration tank to 6.5-8.5. 7dn (t) is operation suggestion 7, which is to increase the sludge load to above 0.2-0.3 kgBOD5 / kgMLSS.d, Y 8dn (t) is Operation Recommendation 8, which is to shorten the sludge age in the aeration tank to 7-15 days. 9dn (t) is operation suggestion 9, which is to increase the nitration solution reflux ratio to more than 300%, Y 10dn (t) is operation suggestion 10, which is to increase the concentration of fungicide in the aeration tank to 12-14 mg / L. The layers in the decision model based on the type-II fuzzy width learning network are expressed as follows:

[0062] Type II fuzzy width learning network input layer: This layer consists of 6 neurons, and the output of each neuron is as follows:

[0063] u s (t) = x s (t),s=1,2,...,6, (1)

[0064] Among them, u s (t) is the output value of the sth neuron in the input layer of the type-II fuzzy width learning network at the tth iteration;

[0065] Type-II fuzzy width learning network interval type-II fuzzy neural network layer: The interval type-II fuzzy neural network layer is composed of I subnetworks, and each interval type-II fuzzy neural network subnetwork structure has five layers, including input layer, membership function layer, activation layer, consequent layer and output layer; its structure is a 6-LQVV connection mode, which determines that the number of input layer neurons of each interval type-II fuzzy neural network is 6, the number of membership function layer neurons is L, and L is any positive integer between [2,20]; the number of activation layer neurons is Q, and Q is any positive integer between [2,20]; the number of consequent layer neurons is V, and the number of output layer neurons is V, and V is any positive integer between [2,10]; the output of each layer can be expressed as:

[0066] Interval type-2 fuzzy neural network subnetwork input layer: This layer consists of 6 neurons, and the output of each neuron is as follows:

[0067] r s (t) = u s (t), (2)

[0068] Among them, r s (t) is the output value of the sth neuron in the input layer of the interval type-2 fuzzy neural network subnetwork at the tth iteration;

[0069] Interval type-2 fuzzy neural network subnetwork membership function layer: This layer consists of L neurons, and the output of this layer can be expressed as:

[0070]

[0071]

[0072] Among them, μ ls (t) is the output value of the lth neuron in the membership function layer of the interval type-II fuzzy neural network subnetwork at the tth iteration, ξ ls (t) is the uncertainty center of the sth input and lth membership function layer neuron at the tth iteration, ξ ls (t) is the lower bound of the uncertainty center of the s-th input and the l-th membership function layer neuron at the t-th iteration, and takes a random value in the interval (0,1); is the upper bound of the uncertainty center of the sth input and the lth membership function layer neuron at the tth iteration, randomly taking values ​​in the interval (0,1], and the lower bound of the uncertainty center is less than the upper bound; σ ls (t) is the fixed standard deviation of the sth input and the lth membership function layer neuron at the tth iteration, and takes a random value in the interval (0,1);

[0073] Interval type-2 fuzzy neural network subnetwork activation layer: This layer consists of Q neurons, and the output of this layer can be expressed as:

[0074]

[0075] Among them, f q (t) is the activation intensity of the qth neuron in the activation layer of the interval type-2 fuzzy neural network subnetwork at the tth iteration, f q (t) and are the lower and upper bounds of the activation strength of the qth neuron in the activation layer at the tth iteration;

[0076] Interval type-2 fuzzy neural network subnetwork subsequent layer: This layer consists of V neurons, and the output of this layer can be expressed as:

[0077]

[0078]

[0079] in, o v (t) and are the lower and upper bounds of the output of the vth neuron in the consequent layer of the interval type-2 fuzzy neural network subnetwork at the tth iteration, w vq (n) and are the lower and upper bounds of the weight between the vth neuron in the posterior layer and the qth neuron in the activation layer at the tth iteration, both of which are randomly selected in the interval [-1,1], and the lower bound of the weight is smaller than the upper bound;

[0080] Interval type-2 fuzzy neural network subnetwork output layer: This layer consists of V neurons, and the output of this layer can be expressed as:

[0081]

[0082] Among them, v (t) is the output value of the vth neuron in the output layer of the interval type-2 fuzzy neural network subnetwork at the tth iteration;

[0083] Type II fuzzy width learning network enhancement layer: This layer consists of J groups of neurons, and the output of this layer can be expressed as:

[0084] H(t)=[h1(t),h2(t),...,h J (t)], (9)

[0085] h j (t) = ζ j (O(t)w hj (t)+β hj(t)),j=1,2,...,J, (10)

[0086] Where H(t) is the output value of the enhanced layer of the type-II fuzzy width learning network at the tth iteration, h j (t) is the output value of the jth group of neurons in the enhancement layer at the tth iteration, ζ j () is the activation function of the jth group of neurons in the enhancement layer, O(t) = [o1(t), o2(t), …, o I (t)] is the output value of the interval type-2 fuzzy neural network layer at the tth iteration, o i (t)=[o1(t),o2(t),…,o V (t)] is the output value of the ith subnetwork of the interval type-II fuzzy neural network layer at the tth iteration, i = 1, 2, ..., I, w hj (t) and β hj (t) is the weight and bias value between the jth group of neurons in the interval type-II fuzzy neural network layer and the enhancement layer at the tth iteration;

[0087] Output layer of type-2 fuzzy width learning network: This layer consists of 10 neurons, and the output of this layer can be expressed as:

[0088]

[0089] Where Y(t) is the output value of the output layer of the type-II fuzzy width learning network at the tth iteration, w i (t) is the weight between the output layer of the type-II fuzzy width learning network and the i-th sub-network of the interval type-II fuzzy neural network layer at the t-th iteration, w j (t) is the weight between the jth group of neurons in the output layer and enhancement layer of the type-II fuzzy width learning network at the tth iteration;

[0090] (4) Training the decision model based on the type-two fuzzy width learning network, specifically:

[0091] ① Assume that the current iteration number t = 1, and the maximum iteration number of the decision model based on the type-two fuzzy width learning network is K = 30;

[0092] ② Given the initial interval type-2 fuzzy neural network layer subnetwork number of type-2 fuzzy width learning network is I, the number of enhancement layer neuron groups is J, the number of membership function layer neurons of each interval type-2 fuzzy neural network layer subnetwork is L, the number of activation layer neurons is Q, the number of consequent layer neurons is V, and the number of output layer neurons is V;

[0093] ③ According to formulas (1)-(10), the output Y(t) of the type-two fuzzy width learning network is calculated. The accuracy of the output of the type-two fuzzy width learning network is expressed as:

[0094] A(t)=TF(t) / P, (12)

[0095] Among them, A(t) is the output accuracy of the fuzzy width learning network at the tth iteration, and TF(t) is the number of samples whose output of the fuzzy width learning network is the same as the actual value at the tth iteration. The pseudo-inverse algorithm is used to adjust the parameters of the type-2 fuzzy width learning network:

[0096] W(t)=D(t) + Y d (t), (13)

[0097]

[0098] Where W(t)=[W i (t), W j (t)] is the weight between the output layer of the type-II fuzzy width learning network and the interval type-II fuzzy neural network layer and the enhancement layer at the tth iteration, W i (t) is the weight between the output layer of the type-II fuzzy width learning network and the interval type-II fuzzy neural network layer at the tth iteration, W j (t) is the weight between the output layer and the enhancement layer of the type-II fuzzy width learning network at the tth iteration; D(t) = [O(t, H(t)], λ is a positive constant, randomly selected in the interval (0,1], and E is the unit matrix;

[0099] ④ Let t = t + 1. If t ≤ K or A(t) > 0.05, return to step ③. If t > K and A(t) ≤ 0.05, stop the calculation and exit the loop to complete the training.

[0100] (5) Intelligent Decision-making on Sludge Bulking

[0101] The trained type-two fuzzy width learning network sludge bulking decision model is used to obtain the optimal network parameters and structure of the model, such as Figure 1 As shown; the training results of sludge bulking intelligent decision-making method are as follows Figure 2 As shown, X-axis: number of training samples, Y-axis: output value of sludge bulking type training, red cross is the actual sludge bulking type, blue circle is the output value based on the type-two fuzzy width learning network decision model;

[0102] Using the trained sludge bulking decision model, the dissolved oxygen, effluent total nitrogen, effluent total phosphorus, effluent sludge load, aeration tank temperature, and effluent pH of 40 test samples are used as the input variables of the model to obtain the sludge bulking category output value of the model, which is the operation suggestion corresponding to the category to which the sample to be tested belongs, including: Operation suggestion 1: Increase the aeration volume of the biochemical pool and control the dissolved oxygen concentration to 2-4 mg / L, Operation suggestion 2: Add urea to the biochemical pool to control the ratio of biological oxygen demand and nitrogen concentration to be greater than 100 mg / L:3 mg / L, Operation suggestion 3: Add sodium phosphate to the biochemical pool to control the ratio of biological oxygen demand and phosphorus concentration to be greater than 100 mg / L :1mg / L, operation suggestion 4: adjust the operating parameters, increase the sludge return ratio to 25%-50%, operation suggestion 5: introduce boiler steam to increase the sewage temperature to above 22℃, operation suggestion 6: add industrial sulfuric acid or calcium hydroxide to adjust the pH, control the pH in the aeration tank to 6.5-8.5, operation suggestion 7: increase the sludge load to above 0.2-0.3kgBOD5 / kgMLSS.d, operation suggestion 8: shorten the sludge age in the aeration tank to 7-15d, operation suggestion 9: increase the nitrification liquid return ratio to more than 300%, operation suggestion 10: increase the fungicide concentration in the aeration tank to 12-14mg / L; the test results of the sludge expansion intelligent decision-making method are as follows Figure 3 As shown, X-axis: number of test samples, Y-axis: output value of sludge bulking type test, red cross is the actual sludge bulking type, blue circle is the output value of the decision model based on type-2 fuzzy width learning network; the experimental results show the effectiveness of the decision-making method based on type-2 fuzzy width learning network.

Claims

1. A sludge bulking intelligent decision-making method based on type-II fuzzy width learning, characterized in that: A decision model based on a width learning network is established to provide decision support for inhibiting sludge bulking. Type II fuzzy neurons are used to replace the feature layer neurons of the width learning network to achieve accurate sludge bulking decision-making and reduce the incidence of sludge bulking, including the following steps: (1) Operation data collection: Sludge bulking is an abnormal phenomenon in which the sludge structure is extremely loose, the volume increases, it floats, and it is difficult to settle and separate. Since it can easily cause the sludge settling performance to decline and the sludge to be lost, resulting in the effluent water quality exceeding the standard, and even causing the sewage treatment efficiency to decline and the operation system to crash, it is necessary to provide decision support for inhibiting sludge bulking. Taking the activated sludge sewage treatment system as the research object, the operation data are collected through the instruments installed on site, including: dissolved oxygen, effluent total nitrogen, effluent total phosphorus, effluent sludge load, aeration tank temperature, effluent pH, so as to realize real-time data collection. The data collected by the instrument is transmitted to the programmable logic controller through the communication protocol, and the programmable logic controller transmits the operation data to the host computer through the communication protocol. The data in the host computer is transmitted to the data processing server through the local area network. (2) Operation data preprocessing: Taking operation data as the research object, process variables related to sludge bulking are selected as input variables of the intelligent decision-making model: dissolved oxygen, effluent total nitrogen, effluent total phosphorus, effluent sludge load, aeration tank temperature, and effluent pH; the obtained input variables are normalized to [0,1]; operation suggestions are used as output variables of the sludge bulking intelligent decision-making model; all sample data are divided into two groups, one group contains P training samples and the other group contains M test samples, requiring P>M; (3) Design of a type-two fuzzy width learning model for sludge bulking decision-making: A type-two fuzzy width learning network is used to establish a sludge bulking decision-making model. The type-two fuzzy width learning network structure has four layers, including an input layer, an interval type-two fuzzy neural network layer, an enhancement layer, and an output layer. Its structure is a 6-IJ-10 connection mode. The number of neurons in the input layer of the model for sludge bulking decision-making is determined to be 6, the number of subnetworks in the interval type-two fuzzy neural network layer is I, and I is any positive integer between [1, 10]. The number of neuron groups in the enhancement layer is J, and J is any positive integer between [2, 20]. The number of neurons in the output layer is 10. P training samples are used to train the type-two fuzzy width learning network. The input of the sludge bulking decision-making model is x(t) = [x1(t), x2(t), …, x P (t)],x n (t) = [x 1n (t),x 2n (t),x 3n (t),x 4n (t),x 5n (t),x 6n (t)] is the nth sample at the tth iteration after normalization, n = 1, 2, ..., P, x 1n (t) is the dissolved oxygen concentration in the nth sample at the tth iteration after normalization, x 2n (t) is the total nitrogen concentration of the effluent in the nth sample at the tth iteration after normalization, x 3n (t) is the total phosphorus concentration in the effluent of the nth sample at the tth iteration after normalization, x 4n (t) is the effluent sludge load concentration in the nth sample at the tth iteration after normalization, x 5n (t) is the aeration tank temperature value in the nth sample at the tth iteration after normalization, x 6n (t) is the pH value of the effluent in the nth sample at the tth iteration after normalization. The output of the sludge bulking decision model is the operation recommendation Y d (t) = [Y d1 (t),Y d2 (t),…,Y dP (t)],Y dn (t) = [Y 1dn (t),Y 2dn (t),…,Y 10dn (t)] is the recommended operation for the nth sample at the tth iteration, Y 1dn (t) is operation suggestion 1, which is to increase the aeration volume of the biochemical pool and control the dissolved oxygen concentration to 2-3 mg / L. 2dn (t) is operation suggestion 2, that is, adding urea to the biochemical pool to control the ratio of biological oxygen demand to nitrogen concentration to be greater than 100:

3. 3dn (t) is operation suggestion 3, that is, adding sodium phosphate to the biochemical pool to control the ratio of biological oxygen demand to phosphorus concentration to be greater than 100:

1. 4dn (t) is operation suggestion 4, that is, to adjust the operating parameters and increase the sludge return ratio to 25%-50%, Y 5dn (t) is operation suggestion 5, that is, to introduce boiler steam to raise the sewage temperature to above 22°C, Y 6dn (t) Operation suggestion 6 is to add industrial sulfuric acid or calcium hydroxide to adjust the pH in the aeration tank to 6.5-8.

5. 7dn (t) is Operation Recommendation 7, which is to increase the residual sludge discharge to 13000m 3 / d or above, Y 8dn (t) is Operation Recommendation 8, which is to reduce the sludge concentration in the aeration tank to 1200 mg / L. 9dn (t) is operation suggestion 9, which is to increase the nitration solution reflux ratio to more than 300%, Y 10dn (t) is operation suggestion 10, which is to increase the concentration of fungicide in the aeration tank to 12-14 mg / L. The layers in the decision model based on the type-II fuzzy width learning network are expressed as follows: Type II fuzzy width learning network input layer: This layer consists of 6 neurons, and the output of each neuron is as follows: u s (t)=x s (t),s=1,2,...,6, (1) Among them, u s (t) is the output value of the sth neuron in the input layer of the type-II fuzzy width learning network at the tth iteration; Type-II fuzzy width learning network interval type-II fuzzy neural network layer: The interval type-II fuzzy neural network layer is composed of I subnetworks, and each interval type-II fuzzy neural network subnetwork structure has five layers, including input layer, membership function layer, activation layer, consequent layer and output layer; its structure is a 6-LQVV connection mode, which determines that the number of input layer neurons of each interval type-II fuzzy neural network is 6, the number of membership function layer neurons is L, and L is any positive integer between [2,20]; the number of activation layer neurons is Q, and Q is any positive integer between [2,20]; the number of consequent layer neurons is V, and the number of output layer neurons is V, and V is any positive integer between [2,10]; the output of each layer is expressed as: Interval type-2 fuzzy neural network subnetwork input layer: This layer consists of 6 neurons, and the output of each neuron is as follows: r s (t)=u s (t), (2) Among them, r s (t) is the output value of the sth neuron in the input layer of the interval type-2 fuzzy neural network subnetwork at the tth iteration; Interval type-2 fuzzy neural network subnetwork membership function layer: This layer consists of L neurons, and the output of this layer is expressed as: Among them, μ ls (t) is the output value of the lth neuron in the membership function layer of the interval type-II fuzzy neural network subnetwork at the tth iteration, ξ ls (t) is the uncertainty center of the sth input and lth membership function layer neuron at the tth iteration, ξ ls (t) is the lower bound of the uncertainty center of the s-th input and the l-th membership function layer neuron at the t-th iteration, and takes a random value in the interval (0,1); ξ ls (t) is the upper bound of the uncertainty center of the s-th input and the l-th membership function layer neuron at the t-th iteration, and takes a random value in the interval (0,1], and the lower bound of the uncertainty center is smaller than the upper bound; σ ls (t) is the fixed standard deviation of the sth input and the lth membership function layer neuron at the tth iteration, and takes a random value in the interval (0,1); Interval type-2 fuzzy neural network subnetwork activation layer: This layer consists of Q neurons, and the output of this layer is expressed as: Among them, f q (t) is the activation intensity of the qth neuron in the activation layer of the interval type-2 fuzzy neural network subnetwork at the tth iteration, f q (t) and are the lower and upper bounds of the activation strength of the qth neuron in the activation layer at the tth iteration; Interval type-2 fuzzy neural network subnetwork subsequent layer: This layer consists of V neurons, and the output of this layer is expressed as: in, o v (t) and are the lower and upper bounds of the output of the vth neuron in the consequent layer of the interval type-2 fuzzy neural network subnetwork at the tth iteration, w vq (n) and are the lower and upper bounds of the weight between the vth neuron in the posterior layer and the qth neuron in the activation layer at the tth iteration, both of which are randomly selected in the interval [-1,1], and the lower bound of the weight is smaller than the upper bound; Interval type-2 fuzzy neural network subnetwork output layer: This layer consists of V neurons, and the output of this layer is expressed as: Among them, v (t) is the output value of the vth neuron in the output layer of the interval type-2 fuzzy neural network subnetwork at the tth iteration; Type II fuzzy width learning network enhancement layer: This layer consists of J groups of neurons, and the output of this layer is expressed as: H(t)=[h1(t),h2(t),...,h J (t)],(9)h j (t)=ζ j (O(t)w hj (t)+β hj (t)),j=1,2,...,J, (10) Where H(t) is the output value of the enhanced layer of the type-II fuzzy width learning network at the tth iteration, h j (t) is the output value of the jth group of neurons in the enhancement layer at the tth iteration, ζ j () is the activation function of the jth group of neurons in the enhancement layer, O(t) = [o1(t), o2(t), …, o I (t)] is the output value of the interval type-2 fuzzy neural network layer at the tth iteration, o i (t)=[o1(t),o2(t),…,o V (t)] is the output value of the ith subnetwork of the interval type-II fuzzy neural network layer at the tth iteration, i = 1, 2, ..., I, w hj (t) and β hj (t) is the weight and bias value between the jth group of neurons in the interval type-II fuzzy neural network layer and the enhancement layer at the tth iteration; Output layer of type-2 fuzzy width learning network: This layer consists of 10 neurons, and the output of this layer is expressed as: Where Y(t) is the output value of the output layer of the type-II fuzzy width learning network at the tth iteration, w i (t) is the weight between the output layer of the type-II fuzzy width learning network and the i-th sub-network of the interval type-II fuzzy neural network layer at the t-th iteration, w j (t) is the weight between the jth group of neurons in the output layer and enhancement layer of the type-II fuzzy width learning network at the tth iteration; (4) Training the decision model based on the type-two fuzzy width learning network, specifically: ① Assume that the current iteration number t = 1, and the maximum iteration number of the decision model based on the type-two fuzzy width learning network is K = 30; ② Given the initial interval type-2 fuzzy neural network layer subnetwork number of type-2 fuzzy width learning network is I, the number of enhancement layer neuron groups is J, the number of membership function layer neurons of each interval type-2 fuzzy neural network layer subnetwork is L, the number of activation layer neurons is Q, the number of consequent layer neurons is V, and the number of output layer neurons is V; ③ According to formulas (1)-(10), the output Y(t) of the type-two fuzzy width learning network is calculated. The accuracy of the output of the type-two fuzzy width learning network is expressed as: A(t)=TF(t) / P, (12) Among them, A(t) is the output accuracy of the fuzzy width learning network at the tth iteration, TF(t) is the number of samples whose output of the fuzzy width learning network is the same as the actual value at the tth iteration; the pseudo-inverse algorithm is used to adjust the parameters of the type-2 fuzzy width learning network: W(t)=D(t) + Y d (t), (13) Where W(t)=[W i (t), W j (t)] is the weight between the output layer of the type-II fuzzy width learning network and the interval type-II fuzzy neural network layer and the enhancement layer at the tth iteration, W i (t) is the weight between the output layer of the type-II fuzzy width learning network and the interval type-II fuzzy neural network layer at the tth iteration, W j (t) is the weight between the output layer and the enhancement layer of the type-II fuzzy width learning network at the tth iteration; D(t) = [O(t, H(t)], λ is a positive constant, randomly selected in the interval (0,1], and E is the unit matrix; ④ Let t = t + 1. If t ≤ K or A(t) > 0.05, return to step ③. If t > K and A(t) ≤ 0.05, stop the calculation and exit the loop to complete the training. (5) Intelligent Decision-making on Sludge Bulking Using the trained type-two fuzzy width learning network sludge bulking decision model, the dissolved oxygen, effluent total nitrogen, effluent total phosphorus, effluent sludge load, aeration tank temperature, and effluent pH of M test samples are used as the input variables of the model to obtain the sludge bulking category output value of the model, which is the operation suggestion corresponding to the category of the sample to be tested, including: Operation suggestion 1: Increase the aeration volume of the biochemical pool and control the dissolved oxygen concentration to 2-4 mg / L; Operation suggestion 2: Add urea to the biochemical pool to control the ratio of biological oxygen demand and nitrogen concentration to be greater than 100 mg / L:3 mg / L; Operation suggestion 3: Add sodium phosphate to the biochemical pool to control the ratio of biological oxygen demand and phosphorus concentration to be greater than 10 0mg / L:1mg / L, Operation suggestion 4: Adjust the operating parameters, increase the sludge return ratio to 25%-50%, Operation suggestion 5: Introduce boiler steam to increase the sewage temperature to above 22°C, Operation suggestion 6: Add industrial sulfuric acid or calcium hydroxide to adjust the pH, and control the pH in the aeration tank to 6.5-8.5, Operation suggestion 7: Increase the sludge load to 0.2-0.3kgBOD5 / kgMLSS.d, Operation suggestion 8: Shorten the sludge age of the aeration tank to 7-15d, Operation suggestion 9: Increase the nitrification liquid return ratio to more than 300%, Operation suggestion 10: Increase the fungicide concentration in the aeration tank to 12-14mg / L to provide decision support for the production process.

Images