Intelligent prediction method for bearing temperature of sewage treatment blower assisted by random event proxy model

Abstract

The invention provides a random event proxy model assisted sewage treatment blower bearing temperature intelligent prediction method, which solves the problem that the blower motor bearing temperature trend is difficult to characterize under the condition of adding lubricating oil, and further realizes accurate prediction of the blower motor bearing temperature. Firstly, a fuzzy neural network is used as a carrier of a data driving model, a bearing temperature prediction main model is constructed, model parameters are adjusted on line according to variable data such as outlet air pressure, a high-speed shaft vibration value and inlet air temperature, and blower bearing temperature prediction under the condition that lubricating oil is not added is achieved; secondly, a fuzzy rule is used for representing the influence of a lubricating oil adding event on the air blower bearing temperature, and an air blower operation random event agent model is designed; and finally, combining the main model and the agent model to predict the temperature of the blower bearing under the condition of adding the lubricating oil, thereby realizing accurate prediction of the temperature of the blower bearing under the influence of the added lubricating oil.

Description

Technical Field

[0001] The present invention designs a method for intelligently predicting the bearing temperature of a sewage treatment blower assisted by a random event proxy model, which realizes the intelligent prediction of the blower motor bearing temperature during the sewage treatment process. Since the bearing temperature of the sewage treatment blower is a key indicator for judging whether the blower is operating normally, it can directly reflect the working state and lubrication condition of the bearing. Therefore, predicting the blower bearing temperature is of great significance for understanding the operating status of the blower equipment. The prediction of the blower bearing temperature during the sewage treatment process is an important basis for early warning of equipment failures and provides important guidance for predictive maintenance of equipment. Background Art

[0002] With the rapid development of technologies such as industrialization, automation, and artificial intelligence, the level of automation in industries such as sewage treatment has gradually increased. With the large-scale use of mechanical equipment, higher requirements are placed on the stability and reliability of equipment. Excessive bearing temperature of sewage treatment blowers may lead to bearing damage, lubrication failure, and even blower shutdown. Therefore, it is of great significance to understand and monitor the operating status of equipment, discover equipment failures in a timely manner, and realize predictive maintenance of equipment to reduce the economic losses caused by unplanned downtime. In this context, data-driven models have the advantages of simple modeling and no need to clearly analyze the correlation between features. They have become an efficient intelligent prediction method. However, in the face of the complex operating environment of equipment, fixed-structure data-driven models are difficult to adapt to data fluctuations caused by changes in external conditions in a timely manner, which can easily lead to deviations in prediction results, increase unplanned downtime losses, increase maintenance costs, and may reduce the service life of equipment. Therefore, it is of great significance to study the prediction of blower bearing temperature, which can provide key support for predictive maintenance of equipment.

[0003] The present invention designs an intelligent prediction method for the bearing temperature of a sewage treatment blower based on the auxiliary prediction of a random event agent model. The method mainly analyzes the random events that are easily encountered during the operation of the equipment, designs a random event agent model based on fuzzy rule characterization, corrects the output of the neural network model, and realizes the prediction of the bearing temperature of the blower affected by the event. Summary of the invention

[0004] The present invention obtains an intelligent prediction method for the bearing temperature of a sewage treatment blower assisted by a random event proxy model. The method is designed with a fuzzy neural network (FNN) as the main framework of data-driven model prediction, and the FNN is trained online to achieve accurate prediction of the blower bearing temperature under normal equipment operation. In addition, by summarizing and analyzing the random events of adding bearing lubricating oil that are easily encountered during equipment operation, a random event proxy model based on fuzzy rule characterization is designed. The model applies knowledge and experience to summarize the functional trend of the impact of events on the prediction output. Finally, a model prediction value correction strategy is designed. The strategy introduces the degree of event impact as the proxy model weight to effectively control the correction degree to adjust the prediction result, so that the model maintains its accuracy when facing the impact of events, thereby improving the reliability of the bearing temperature prediction value.

[0005] The present invention adopts the following technical solutions and implementation steps:

[0006] A method for intelligently predicting the bearing temperature of a sewage treatment blower assisted by a random event agent model comprises the following steps:

[0007] (1) Determine the input variables and output variables to be predicted required for the main model prediction: Intelligently predict the blower bearing temperature during sewage treatment, determine the outlet air pressure, high-speed shaft drive end bearing temperature, high-speed shaft vibration value, inlet air temperature, main motor current, and historical blower main motor bearing temperature as the relevant variables for predicting the blower main motor bearing temperature, take the blower main motor bearing temperature as the output variable, obtain the outlet air pressure data as the vector x1 = [x1(t-I+1),…,x1(th),…,x1(t)], and the high-speed shaft drive end bearing temperature data as the vector x2 = [x2(t-I+1 ),…,x2(th),…,x2(t)], the high-speed shaft vibration value data is vector x3=[x3(t-I+1),…,x3(th),…,x3(t)], the inlet air temperature data is vector x4=[x4(t-I+1),…,x4(th),…,x4(t)], the main motor current data is vector x5=[x5(t-I+1),…,x5(th),…,x5(t)], the historical blower main motor bearing temperature data is vector x6=[x6(t-I+1),…,x6(th),…,x6(t)], and the construction of all sample matrices is X=[x1,…,x p ,...,x o ] T ,x p =[x p (t-I+1),…,x p (th),…,x p(t)], h = 1, ..., I-1, p = 1, ..., o, where o and I are the number of relevant variables and the number of samples, respectively, and T represents the transpose of the matrix;

[0008] (2) Determine the fuzzy neural network model structure for blower shaft temperature prediction during normal operation: the network topology is divided into four layers, input layer, radial basis function layer, normalization layer, and output layer; the number of nodes in each layer of the network is denoted as: the number of neurons in the input layer is o, the number of neurons in the RBF layer is Ω, the number of neurons in the normalization layer is Ω, and the number of neurons in the output layer is 1; the center and standard deviation parameter matrix of the RBF layer neurons is expressed as C(t)=[c1(t),…,c q (t),...,c Ω (t)],c q (t) = [c 1q (t),…,c pq (t),...,c oq (t)],Θ(t)=[σ1(t),…,σ q (t),...,σ Ω (t)],σ q (t) = [σ 1q (t),…,σ pq (t),...,σ oq (t)], the standard deviation of the RBF layer is initialized to 1, and the center value of the RBF layer is initialized to a random value between 0 and 1; the output weight parameter matrix of the normalized layer neuron is expressed as w(t) = [w1(t),…,w q (t),...,w Ω (t)], the weights are initialized to random values ​​between 0 and 1; where p = 1, 2, ..., o, q = 1, 2, ..., Ω, c pq (t),σ pq (t), represents the center and standard deviation of the qth neuron in the RBF layer for the pth input in the input layer at time t, w q (t) represents the connection weight between the qth neuron in the normalized layer and the output, and the input sample vector at time t is determined to be X(t) = [x1(t),...,x p (t),...,x o (t)] T , the corresponding model output is the blower motor bearing temperature y(t), the network topology and calculation method are:

[0009] ① Input layer: The neurons in this layer are used to input the characteristic variables extracted from the blower into the prediction model. The output of the neurons in this layer is X(t) = [x1(t),...,x p (t),...,x o (t)] T ;

[0010] ②RBF layer: used to simulate the relationship between characteristic variables and axle temperature. The output u of the qth neuron in this layer is q (t) is

[0011]

[0012] ③ Normalization layer: This layer of neurons includes Ω, and the output of the qth neuron is φ q (t) is

[0013]

[0014] ④ Output layer: This layer has one neuron, and the output y(t) is calculated as

[0015]

[0016] Among them, w q (t) is the qth weight in the normalized layer, and y(t) represents the output of the output neuron at the tth moment;

[0017] (3) The fuzzy rules of the agent model for the event of adding lubricant to the bearing are designed as the operating condition vectors at different times, expressed as m(τ) = [m1(τ), ..., m i (τ),...,m k (τ)], i=1,...,k,k=2, m1(τ) represents the main motor current, m2(τ) ​​represents the main motor bearing temperature, τ is the time after the event occurs, represented by the difference between the current time and the time when the event occurs;

[0018] Rule 1: When the motor current is large, the motor shaft temperature is high, that is, m1 is greater than I up , m2 is greater than T up When up Indicates the conditional threshold of large current, T up The condition threshold for the high shaft temperature is 70% of the warning value of the bearing operating parameters specified by the factory. At this time, the temperature rise caused by the refueling event is expressed as

[0019]

[0020] Among them, l 11 ,l 12 ,l 13 , is the function parameter of rule 1, obtained by fitting the data of bearing temperature rise under the refueling event;

[0021] Rule 2: When the motor current is normal and the motor shaft temperature is normal, that is, m1 is less than I n or, m2 is less than T nor , I norIndicates the normal current condition threshold, T nor The condition threshold for normal shaft temperature is 50% of the warning value of the bearing operating parameters specified by the factory, and the temperature rise at this time is expressed as

[0022]

[0023] In the formula, l 21 ,l 22 ,l 23 ,l 24 , is the function parameter of rule 2, D represents the temperature rise effect when the influence of the refueling event is stable, and each parameter is obtained by fitting the bearing temperature rise value data under the operating conditions corresponding to the refueling event;

[0024] After obtaining the condition input, the membership function is used for fuzzy reasoning, and the trapezoidal membership function is used to score the membership of the premise condition, as shown in the formula

[0025]

[0026] In the formula, m i is the value of the i-th condition, a and b are the two inflection points of the trapezoidal membership function, a1 and a2 are the values ​​of I nor ,T nor ,b1,b2 are I up ,T up ,θ 1i represents the membership of the i-th condition of the first rule; among them, the membership of rule one is

[0027] θ1(τ)=min(θ 11 (τ),θ 12 (τ)) (7) The rule membership of rule 2 is

[0028] θ2(τ)=min((1-θ 11 (τ)),(1-θ 12 (τ))) (8)

[0029] In formulas (7) and (8), min(.) means taking the value of the smallest element in the brackets;

[0030] The total output of the proxy model is

[0031] N(τ)=f1(τ)β1(τ)+f2(τ)β2(τ)(9)

[0032] Where N(τ) is the total output of the proxy model, β1(τ) and β2(τ) represent the normalized membership of the rules.

[0033]

[0034] (4) Training the main fuzzy neural network model

[0035] The model parameters are updated using the gradient descent algorithm to predict the motor bearing temperature when no refueling event occurs.

[0036] ①Define the loss function as

[0037]

[0038] Among them, e1(t) represents the difference between the actual temperature value at time t and the predicted output of the main model, y(t) is the expected output of the neural network model at time t, is the actual output;

[0039] ② Initialize the number of training rounds Lp1 = 1, and set the total number of training rounds MLp1 = 50;

[0040] ③ Set the initial training step number to Lp2 = 1, input samples for training, use dynamic sliding window to dynamically update model parameters, set the window size to P = 18, take the training samples of window size to update model parameters, and the sliding window matrix is ​​represented as X1 = [x1,...,x p ,...,x o ] T ,x p =[x p (t ζ -P+1),…,x p (t ζ -1),x p (t ζ )],t ζ Represents the last moment in the model sliding window;

[0041] ④ Input the sample at time t, calculate the model output loss L(t) according to equations (1), (2), (3), (12), and (13), and use the gradient descent method to adjust the model parameters at time t+1. The update rules for the connection weights of the output layer and the mean and standard deviation of the RBF layer are as follows:

[0042]

[0043] Where λ is the learning rate. When the number of training rounds is 50, a value between 0.001 and 0.1 is selected, which has little effect on the experimental results. C(t)=[c1(t),...,c q (t),...,c Ω (t)],c q (t) = [c 1q (t),...,c pq (t),...,c oq(t)],Θ(t)=[σ1(t),...,σ q (t),...,σ Ω (t)],σ q (t)=[σ 1q (t),...,σ pq (t),...,σ oq (t)],w(t)=[w1(t),...,w q (t),...,w Ω (t)],ΔC(t)=[Δc1(t),...,Δc q (t),...,Δc Ω (t)],Δc q (t) = [Δc 1q (t),...,Δc pq (t),...,Δc oq (t)],ΔΘ(t)=[Δσ1(t),...,Δσ q (t),...,Δσ Ω (t)],Δσ q (t) = [Δσ 1q (t),...,Δσ pq (t),...,Δσ oq (t)],Δw(t)=[Δw1(t),...,Δw q (t),...,Δw Ω (t)], where the partial derivative is calculated as follows

[0044]

[0045] ⑤ Determine whether the maximum number of training rounds has been reached. If Lp1 = 50, the training ends and the trained model is used for the tth round. ζ +1 time; otherwise, determine whether the maximum number of training steps has been reached. If Lp2=P, then Lp1=Lp1+1, and return to ③ to continue training; otherwise, the number of training steps increases, Lp2=Lp2+1, t=t+1, and return to ④ to continue training;

[0046] (5) Use the proxy model to correct the output of the main prediction model

[0047] To achieve the prediction of the motor bearing temperature when adding lubricating oil to the blower bearing temperature, if lubricating oil is added at time t1;

[0048] ① Through the steps in Section (4), the uncorrected main model training is realized, where t ζ =t-1, t is the current moment, the trained model predicts the fan bearing temperature at the current moment under normal working conditions through equations (1), (2), and (3);

[0049] ② Through the steps in Section (3), according to the corresponding conditions of the fan operation time, use formula (9) to calculate the temperature difference caused by the refueling event at the sampling time τ after the event occurs, where τ = t-t1;

[0050] ③ In order to improve the adaptability of the rules, the proxy model output correction weight v is set, and v is initialized to a value between 1 and 2. The total output of the blower motor bearing temperature prediction model at time t under the influence of the refueling event is

[0051]

[0052] ④ The proxy model corrects the weight update and defines the loss function as

[0053]

[0054] The parameter v is updated once every moment after the event occurs. The sample and rule conditions at time t are input. The model output loss L2(t) is calculated according to equations (1), (2), (3), (9), and (18). The weight parameter at time t+1 is adjusted using the gradient descent method. The learning rate is twice the learning rate of the main model. The update rule is:

[0055]

[0056] ⑤Use instruments such as ultrasonic or infrared thermal imaging to observe the time T1 required for a uniform oil film to form inside the bearing. When τ>T1, stop the proxy model output and use the trained main model output formula (3) as the temperature prediction value; otherwise, t=t+1, return to step ①, and use formula (18) as the bearing temperature prediction output.

[0057] The creativity of the present invention is mainly reflected in:

[0058] The present invention aims at the fact that in the process of predicting the bearing temperature of the sewage treatment blower motor, the distribution characteristics of the input data may fluctuate due to the occurrence of random events. It is difficult for data-driven models such as fuzzy neural networks with fixed structures to capture and adapt to such dynamic changes. The prediction results lack the knowledge guidance of changes in operating conditions, resulting in a decrease in prediction performance. An intelligent prediction model for the bearing temperature of the sewage treatment blower assisted by a random event proxy model is established. The model uses fuzzy rules based on knowledge and experience to perform knowledge compensation on the prediction results, effectively correcting the prediction deviation of the fuzzy neural network within a period of time after the bearing is added with lubricating oil, and improving the robustness and accuracy of the bearing temperature prediction value. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1This is a result diagram of the blower motor shaft temperature prediction of the present invention, and the enlarged part shows the impact range of a refueling event;

[0060] Figure 2 It is an error diagram of the blower motor shaft temperature prediction of the present invention, and the enlarged part shows the influence interval of a refueling event;

[0061] Figure 3 This is a result diagram of the blower motor shaft temperature prediction of the present invention, and the enlarged part shows the influence interval of two refueling events occurring at different times;

[0062] Figure 4 It is an error diagram of the blower motor shaft temperature prediction of the present invention, and the enlarged part shows the influence interval of two refueling events occurring at different times; DETAILED DESCRIPTION

[0063] The present invention designs an intelligent prediction method for the bearing temperature of a sewage treatment blower motor based on an event proxy model. Six variables, namely, outlet air pressure, high-speed shaft drive end bearing temperature, high-speed shaft vibration value, inlet air temperature, main motor current, and historical blower main motor bearing temperature, are selected as input variables of the main model for predicting the blower main motor bearing temperature. The main motor current and historical blower main motor bearing temperature are used as input variables of the proxy model. The temperature difference of the bearing temperature caused by the refueling event relative to normal operation is used as the output variable of the proxy model.

[0064] The experimental data comes from the sampling records of equipment operation variables of a sewage treatment plant, including the operation event records of each device. The sampling interval is one hour. After the experimental samples are screened, 2200 sets of data are left. The data are arranged in time sequence and gradually placed in the sliding window. The number of sliding windows is initialized to P = 18;

[0065] The present invention adopts the following technical solutions and implementation steps:

[0066] A method for intelligently predicting the bearing temperature of a sewage treatment blower assisted by a random event agent model comprises the following steps:

[0067] (1) Determine the input variables and output variables to be predicted required for the main model prediction: Intelligently predict the blower bearing temperature during sewage treatment, determine the outlet air pressure, high-speed shaft drive end bearing temperature, high-speed shaft vibration value, inlet air temperature, main motor current, and historical blower main motor bearing temperature as the relevant variables for predicting the blower main motor bearing temperature, take the blower main motor bearing temperature as the output variable, obtain the outlet air pressure data as the vector x1 = [x1(t-I+1),…,x1(th),…,x1(t)], and the high-speed shaft drive end bearing temperature data as the vector x2 = [x2(t-I+1 ),…,x2(th),…,x2(t)], the high-speed shaft vibration value data is vector x3=[x3(t-I+1),…,x3(th),…,x3(t)], the inlet air temperature data is vector x4=[x4(t-I+1),…,x4(th),…,x4(t)], the main motor current data is vector x5=[x5(t-I+1),…,x5(th),…,x5(t)], the historical blower main motor bearing temperature data is vector x6=[x6(t-I+1),…,x6(th),…,x6(t)], and the training sample matrix is ​​constructed as X=[x1,…,x p ,...,x o ] T ,x p =[x p (t-I+1),…,x p (th),…,x p (t)], h = 0, ..., I-1, p = 1, ..., o, where o = 6 and I = 2200 are the number of relevant variables and the number of samples, respectively, and T represents the transpose of the matrix;

[0068] (2) Determine the fuzzy neural network model structure for blower shaft temperature prediction during normal operation: the network topology is divided into four layers, input layer, radial basis function layer, normalization layer, and output layer; the number of nodes in each layer of the network is initialized to 6 neurons in the input layer, 7 neurons in the RBF layer, 7 neurons in the normalization layer, and 1 neuron in the output layer; the RBF layer neuron center and standard deviation parameter matrix is ​​expressed as C(t) = [c1(t),…,c q (t),...,c Ω (t)],c q (t) = [c 1q (t),…,c pq (t),...,c oq (t)],Θ(t)=[σ1(t),…,σ q (t),...,σ Ω (t)],σ q (t)=[σ1q (t),…,σ pq (t),...,σ oq (t)], the standard deviation of the RBF layer is initialized to 1, and the center value of the RBF layer is initialized to a random value between 0 and 1; the output weight parameter matrix of the normalized layer neuron is expressed as w(t) = [w1(t),…,w q (t),...,w Ω (t)], the weights are initialized to random values ​​between 0 and 1; where p = 1, 2, …, o, o = 6, q = 1, 2, …, Ω, Ω = 7, c pq (t),σ pq (t), represents the center and standard deviation of the qth neuron in the RBF layer for the pth input in the input layer at time t, w q (t) represents the connection weight between the qth neuron in the normalized layer and the output, and the input sample vector at time t is determined to be X(t) = [x1(t),...,x p (t),...,x o (t)] T , the corresponding model output is the blower motor bearing temperature y(t), the network topology and calculation method are

[0069] ① Input layer: The neurons in this layer are used to input the characteristic variables extracted from the blower into the prediction model. The output of the neurons in this layer is X(t) = [x1(t),...,x p (t),...,x o (t)] T ;

[0070] ②RBF layer: used to simulate the relationship between characteristic variables and axle temperature. The output u of the qth neuron in this layer is q (t) is

[0071]

[0072] ③ Normalization layer: This layer of neurons includes Ω, and the output of the qth neuron is φ q (t) is

[0073]

[0074] ④ Output layer: This layer has one neuron, and the output y(t) is calculated as

[0075]

[0076] Among them, w q (t) is the qth weight in the normalized layer, and y(t) represents the output of the output neuron at the tth moment;

[0077] (3) The fuzzy rules of the refueling event agent model are designed as the operating condition vectors at different times, expressed as m(τ) = [m1(τ), ..., m i (τ),...,m k (τ)], i=1,...,k,k=2, τ is the time after the event occurs, m1(τ) represents the main motor current, and m2(τ) ​​represents the main motor bearing temperature;

[0078] Rule 1: When the motor current is large, the motor shaft temperature is high, that is, m1 is greater than I up ,m2 is greater than T up When up Indicates the conditional threshold of large current, T up Indicates the condition threshold of higher shaft temperature. The temperature rise at this time is expressed as

[0079]

[0080] Rule 2: When the motor current is normal and the motor shaft temperature is normal, that is, m1 is less than I nor ,m2 is less than T nor ,I nor Indicates the normal current condition threshold, T nor The condition threshold of normal shaft temperature is expressed as

[0081]

[0082] After obtaining the condition input, the membership function is used for fuzzy reasoning, and the trapezoidal membership function is used to score the membership of the premise condition, as shown in the formula

[0083]

[0084] In the formula, m i is the value of the ith condition, a1=I nor =30,a2=T nor =45,b1=I up =45,b2=T up =65,θ 1i represents the membership of the i-th condition of the first rule; among them, the membership of rule one is

[0085] θ1(τ)=min(θ 11 (τ),θ 12 (τ)) (29)

[0086] The rule membership of rule 2 is

[0087] θ2(τ)=min((1-θ 11 (τ)),(1-θ 12(τ))) (30)

[0088] In formulas (29) and (30), min(.) means taking the value of the smallest element in the brackets;

[0089] The total output of the proxy model is

[0090] N(τ+1)=f1β1(τ)+f2β2(τ) (31)

[0091] Where N(τ) is the total output of the proxy model, β1(τ) and β2(τ) represent the normalized membership of the rules.

[0092]

[0093] (4) Training the main fuzzy neural network model

[0094] The model parameters are updated using the gradient descent algorithm to predict the motor bearing temperature when no refueling event occurs.

[0095] ①Define the loss function as

[0096]

[0097] Among them, e1(t) represents the difference between the actual temperature value at time t and the predicted output of the main model, y(t) is the expected output of the neural network model at time t, is the actual output;

[0098] ② Initialize the number of training rounds Lp1 = 1, and set the total number of training rounds MLp1 = 50;

[0099] ③ Set the initial training step number to Lp2 = 1, input samples for training, use dynamic sliding window to dynamically update model parameters, set the window size to P = 18, take the training samples of window size to update model parameters, and the sliding window matrix is ​​represented as X1 = [x1,...,x p ,...,x o ] T ,x p =[x p (t ζ -P+1),…,x p (t ζ -1),x p (t ζ )],t ζ Represents the last moment in the model sliding window;

[0100] ④ Input the sample at time t, calculate the model output loss L(t) according to equations (23), (24), (25), (34), (35), and use the gradient descent method to adjust the model parameters at time t+1. The update rules for the connection weights of the output layer and the mean and standard deviation of the RBF layer are:

[0101]

[0102] Where λ = 0.005, C(t) = [c1(t), ..., c q (t),...,c Ω (t)],c q (t) = [c 1q (t),...,c pq (t),...,c oq (t)],Θ(t)=[σ1(t),...,σ q (t),...,σ Ω (t)],σ q (t) = [σ 1q (t),...,σ pq (t),...,σ oq (t)],w(t)=[w1(t),...,w q (t),...,w Ω (t)],ΔC(t)=[Δc1(t),...,Δc q (t),...,Δc Ω (t)],Δc q (t) = [Δc 1q (t),...,Δc pq (t),...,Δc oq (t)],ΔΘ(t)=[Δσ1(t),...,Δσ q (t),...,Δσ Ω (t)],Δσ q (t) = [Δσ 1q (t),...,Δσ pq (t),...,Δσ oq (t)],Δw(t)=[Δw1(t),...,Δw q (t),...,Δw Ω (t)], where the partial derivative is calculated as follows

[0103]

[0104] ⑤ Determine whether the maximum number of training rounds has been reached. If Lp1 = 50, the training ends and the trained model is used for the tth round. ζ+1 time; otherwise, determine whether the maximum number of training steps has been reached. If Lp2=P, then Lp1=Lp1+1, and return to ③ to continue training; otherwise, the number of training steps increases, Lp2=Lp2+1, t=t+1, and return to ④ to continue training;

[0105] (5) Use the proxy model to correct the output of the main prediction model

[0106] To achieve the prediction of the motor bearing temperature when adding lubricating oil to the blower bearing temperature, if lubricating oil is added at time t1;

[0107] ① Through the steps in Section (4), the uncorrected main model training is realized, where t ζ =t-1, t is the current moment, the trained model predicts the fan bearing temperature at the current moment under normal working conditions through equations (23), (24), and (25);

[0108] ② Through the steps in Section (3), according to the corresponding conditions of the fan operation time, use formula (31) to calculate the temperature difference caused by the refueling event at time τ after the event occurs, τ = t-t1;

[0109] ③ Set the proxy model output correction weight v, initialize v = 1.4, and the total output of the blower motor bearing temperature prediction model at the current time t under the influence of the refueling event is

[0110]

[0111] ④ The proxy model corrects the weight update and defines the loss function as

[0112]

[0113] The parameter v is updated once every moment after the event occurs. The sample and rule conditions at time t are input. The model output loss L2(t) is calculated according to equations (23), (24), (25), (31), and (41). The weight parameter at time t+1 is adjusted using the gradient descent method. The update rule is:

[0114]

[0115] ⑤ When τ>T1, T1=72, the proxy model output is stopped, and the trained main model output formula (25) is used as the temperature prediction value; otherwise t=t+1, return to step ①, and use formula (40) as the bearing temperature prediction output.

Claims

1. A method for intelligent prediction of bearing temperature of sewage treatment blower assisted by random event agent model, characterized in that The following steps are involved: (1) Determine the input variables and output variables to be predicted required for the main model prediction: Intelligently predict the blower bearing temperature during sewage treatment, determine the outlet air pressure, high-speed shaft drive end bearing temperature, high-speed shaft vibration value, inlet air temperature, main motor current, and historical blower main motor bearing temperature as the relevant variables for predicting the blower main motor bearing temperature, take the blower main motor bearing temperature as the output variable, obtain the outlet air pressure data as the vector x1 = [x1(t-I+1),…,x1(th),…,x1(t)], and the high-speed shaft drive end bearing temperature data as the vector x2 = [x2(t-I+1 ),…,x2(th),…,x2(t)], the high-speed shaft vibration value data is vector x3=[x3(t-I+1),…,x3(th),…,x3(t)], the inlet air temperature data is vector x4=[x4(t-I+1),…,x4(th),…,x4(t)], the main motor current data is vector x5=[x5(t-I+1),…,x5(th),…,x5(t)], the historical blower main motor bearing temperature data is vector x6=[x6(t-I+1),…,x6(th),…,x6(t)], and the construction of all sample matrices is X=[x1,…,x p ,...,x o ] T ,x p =[x p (t-I+1),…,x p (th),…,x p (t)], h = 1, ..., I-1, p = 1, ..., o, where o and I are the number of relevant variables and the number of samples, respectively, and T represents the transpose of the matrix; (2) Determine the fuzzy neural network model structure for blower shaft temperature prediction during normal operation: the network topology is divided into four layers, input layer, radial basis function layer, normalization layer, and output layer; the number of nodes in each layer of the network is denoted as: the number of neurons in the input layer is o, the number of neurons in the RBF layer is Ω, the number of neurons in the normalization layer is Ω, and the number of neurons in the output layer is 1; the center and standard deviation parameter matrix of the RBF layer neurons is expressed as C(t)=[c1(t),…,c q (t),...,c Ω (t)],c q (t) = [c 1q (t),…,c pq (t),...,c oq (t)],Θ(t)=[σ1(t),…,σ q (t),...,σ Ω (t)],σ q (t)=[σ 1q (t),…,σ pq (t),...,σ oq (t)], the standard deviation of the RBF layer is initialized to 1, and the center value of the RBF layer is initialized to a random value between 0 and 1; The output weight parameter matrix of the normalized layer neurons is expressed as w(t)=[w1(t),…,w q (t),...,w Ω (t)], the weights are initialized to random values ​​between 0 and 1; where p = 1, 2, ..., o, q = 1, 2, ..., Ω, c pq (t),σ pq (t), represents the center and standard deviation of the qth neuron in the RBF layer for the pth input in the input layer at time t, w q (t) represents the connection weight between the qth neuron in the normalized layer and the output, and the input sample vector at time t is determined to be X(t) = [x1(t),...,x p (t),...,x o (t)] T , the corresponding model output is the blower motor bearing temperature y(t), the network topology and calculation method are: ① Input layer: The neurons in this layer are used to input the characteristic variables extracted from the blower into the prediction model. The output of the neurons in this layer is X(t) = [x1(t),...,x p (t),...,x o (t)] T ; ②RBF layer: used to simulate the relationship between characteristic variables and axle temperature. The output u of the qth neuron in this layer is q (t) is ③ Normalization layer: This layer of neurons includes Ω, and the output of the qth neuron is φ q (t) is ④ Output layer: This layer has one neuron, and the output y(t) is calculated as Among them, w q (t) is the qth weight in the normalized layer, and y(t) represents the output of the output neuron at the tth moment; (3) The fuzzy rules of the agent model for the event of adding lubricant to the bearing are designed as the operating condition vectors at different times, expressed as m(τ) = [m1(τ), ..., m i (τ),...,m k (τ)], i=1,...,k,k=2, m1(τ) represents the main motor current, m2(τ) ​​represents the main motor bearing temperature, τ is the time after the event occurs, represented by the difference between the current time and the time when the event occurs; Rule 1: When the motor current is large, the motor shaft temperature is high, that is, m1 is greater than I up , m2 is greater than T up When up Indicates the conditional threshold of large current, T up The condition threshold for the high shaft temperature is 70% of the warning value of the bearing operating parameters specified by the factory. At this time, the temperature rise caused by the refueling event is expressed as Among them, l 11 ,l 12 ,l 13 , is the function parameter of rule 1, obtained by fitting the data of bearing temperature rise under the refueling event; Rule 2: When the motor current is normal and the motor shaft temperature is normal, that is, m1 is less than I nor , m2 is less than T nor , I nor Indicates the normal current condition threshold, T nor The condition threshold for normal shaft temperature is 50% of the warning value of the bearing operating parameters specified by the factory, and the temperature rise at this time is expressed as In the formula, l 21 ,l 22 ,l 23 ,l 24 , is the function parameter of rule 2, D represents the temperature rise effect when the influence of the refueling event is stable, and each parameter is obtained by fitting the bearing temperature rise value data under the operating conditions corresponding to the refueling event; After obtaining the condition input, the membership function is used for fuzzy reasoning, and the trapezoidal membership function is used to score the membership of the premise condition, as shown in the formula In the formula, m i is the value of the i-th condition, a and b are the two inflection points of the trapezoidal membership function, a1 and a2 are the values ​​of I nor ,T nor ,b1,b2 are I up ,T up ,θ 1i represents the membership of the i-th condition of the first rule; among them, the membership of rule one is θ1(τ)=min(θ 11 (t),i 12 (t)) (7) The rule membership of rule 2 is θ2(τ)=min((1-θ 11 (τ)),(1-θ 12 (t))) (8) In formulas (7) and (8), min(.) means taking the value of the smallest element in the brackets; The total output of the proxy model is N(τ)=f1(τ)β1(τ)+f2(τ)β2(τ) (9) Where N(τ) is the total output of the proxy model, β1(τ) and β2(τ) represent the normalized membership of the rules. (4) Training the main fuzzy neural network model The model parameters are updated using the gradient descent algorithm to predict the motor bearing temperature when no refueling event occurs. ①Define the loss function as Among them, e1(t) represents the difference between the actual temperature value at time t and the predicted output of the main model, y(t) is the expected output of the neural network model at time t, is the actual output; ② Initialize the number of training rounds Lp1 = 1, and set the total number of training rounds MLp1 = 50; ③ Set the initial training step number to Lp2 = 1, input samples for training, use dynamic sliding window to dynamically update model parameters, set the window size to P = 18, take the training samples of window size to update model parameters, and the sliding window matrix is ​​represented as X1 = [x1,...,x p ,...,x o ] T ,x p =[x p (t ζ -P+1),…,x p (t ζ -1),x p (t ζ )],t ζ Represents the last moment in the model sliding window; ④ Input the sample at time t, calculate the model output loss L(t) according to equations (1), (2), (3), (12), and (13), and use the gradient descent method to adjust the model parameters at time t+1. The update rules for the connection weights of the output layer and the mean and standard deviation of the RBF layer are as follows: Where λ is the learning rate. When the number of training rounds is 50, a value between 0.001 and 0.1 is selected, which has little effect on the experimental results. C(t)=[c1(t),...,c q (t),...,c Ω (t)],c q (t) = [c 1q (t),...,c pq (t),...,c oq (t)],Θ(t)=[σ1(t),...,σ q (t),...,σ Ω (t)],σ q (t)=[σ 1q (t),...,σ pq (t),...,σ oq (t)],w(t)=[w1(t),...,w q (t),...,w Ω (t)],ΔC(t)=[Δc1(t),...,Δc q (t),...,Δc Ω (t)],Δc q (t) = [Δc 1q (t),...,Δc pq (t),...,Δc oq (t)],ΔΘ(t)=[Δσ1(t),...,Δσ q (t),...,Δσ Ω (t)],Δσ q (t) = [Δσ 1q (t),...,Δσ pq (t),...,Δσ oq (t)],Δw(t)=[Δw1(t),...,Δw q (t),...,Δw Ω (t)], where the partial derivative is calculated as follows ⑤ Determine whether the maximum number of training rounds has been reached. If Lp1 = 50, the training ends and the trained model is used for the tth round. ζ +1 time; otherwise, determine whether the maximum number of training steps has been reached. If Lp2=P, then Lp1=Lp1+1, and return to ③ to continue training; otherwise, the number of training steps increases, Lp2=Lp2+1, t=t+1, and return to ④ to continue training; (5) Use the proxy model to correct the output of the main prediction model To achieve the prediction of the motor bearing temperature when adding lubricating oil to the blower bearing temperature, if lubricating oil is added at time t1; ① Through the steps in Section (4), the uncorrected main model training is realized, where t ζ =t-1, t is the current moment, the trained model predicts the fan bearing temperature at the current moment under normal working conditions through equations (1), (2), and (3); ② Through the steps in Section (3), according to the corresponding conditions of the fan operation time, use formula (9) to calculate the temperature difference caused by the refueling event at the sampling time τ after the event occurs, τ = t-t1; ③ In order to improve the adaptability of the rules, the proxy model output correction weight v is set, and v is initialized to a value between 1 and 2. The total output of the blower motor bearing temperature prediction model at time t under the influence of the refueling event is ④ The proxy model corrects the weight update and defines the loss function as The parameter v is updated once every moment after the event occurs. The sample and rule conditions at time t are input. The model output loss L2(t) is calculated according to equations (1), (2), (3), (9), and (18). The weight parameter at time t+1 is adjusted using the gradient descent method. The learning rate is twice the learning rate of the main model. The update rule is: ⑤Use ultrasonic or infrared thermal imaging to observe the time T1 required for a uniform oil film to form inside the bearing. When τ>T1, stop the proxy model output and use the trained main model output formula (3) as the temperature prediction value; otherwise, t=t+1, return to step ① and use formula (18) as the bearing temperature prediction output.

Images