Sewage pump multi-parameter intelligent control method based on bidirectional LSTM neural network

Abstract

The invention relates to the technical field of sewage pump control, in particular to a sewage pump multi-parameter intelligent control method based on a bidirectional LSTM neural network. According to the method, firstly, multiple types of sensors for flow, impurity concentration and the like are installed in a sewage conveying system, and the operation parameter acquisition function of a servo motor body is utilized to obtain system operation data in real time, transmit the system operation data to a data processing unit for preprocessing, and arrange the data according to a time sequence to form a sample set; and next, constructing an adaptive neural network model, selecting a proper network structure, determining the number of nodes in each layer, an activation function, a loss function and an optimization algorithm, and performing training and performance evaluation on the model by using the divided training set, verification set and test set, so that the model has good generalization ability. During actual operation of the sewage pump, data collected and preprocessed in real time are input into the trained neural network model, the model outputs sewage pump control instructions such as rotating speed adjustment, start-stop control and fault early warning, a motor controller is driven to regulate and control operation of the sewage pump according to the control instructions, and meanwhile a real-time monitoring and alarming mechanism is set. And the running state accords with prediction. Intelligent and precise control over the sewage pump is achieved, the operation efficiency can be effectively improved, energy consumption is reduced, the service life of equipment is prolonged, the system stability is enhanced, and the system can adapt to various complex working conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of sewage pump control, and particularly to a multi-parameter intelligent control method for sewage pumps based on a bidirectional LSTM neural network. Background Art

[0002] In many fields such as sewage treatment and drainage systems, sewage pumps play a crucial role in transporting sewage from one place to another for subsequent treatment or discharge. Traditional sewage pump control methods mainly rely on PID control, fixed threshold triggering, or manual experience adjustment. There are defects such as insufficient adaptability, high energy consumption, and dependence on manual intervention. The insufficient adaptability is mainly reflected in the dynamic changes of sewage flow, composition, and pipeline pressure. Traditional methods cannot adjust control parameters in real time, resulting in low efficiency. High energy consumption is mainly reflected in that most traditional sewage pump control methods set fixed start-stop thresholds based on simple liquid level sensors, which may cause the sewage pump to start and stop frequently. This not only accelerates equipment wear, shortens the service life of the equipment, but also increases energy consumption and reduces the overall drainage efficiency. Dependence on manual intervention is mainly reflected in that parameter adjustment relies on experience and lacks the ability of autonomous learning and dynamic optimization. This makes it difficult to achieve efficient, accurate, and intelligent control, which is not conducive to the stable operation of the entire sewage transportation and treatment system.

[0003] In terms of sewage flow, in different time periods, such as peak and off-peak residential water use periods, the amount of sewage flowing into the sewage treatment plant will vary significantly. Traditional control methods cannot adjust the operating state of the sewage pump in time according to this flow change, resulting in insufficient treatment capacity and sewage backlog when the flow is large; when the flow is small, the sewage pump still operates at a high power, causing energy waste. Traditional control methods do not monitor the operating state of the motor accurately and comprehensively enough, and it is difficult to obtain key parameters such as the speed, current, and torque of the motor in real time. It is impossible to adjust the operation of the sewage pump in time according to the changes of these parameters, resulting in problems such as excessive energy consumption, untimely maintenance, and frequent failures.

[0004] With the development of artificial intelligence technology, neural network algorithms have shown powerful data processing capabilities and can analyze and predict complex system states. As a special recurrent neural network, the bidirectional long short-term memory network (Bi-LSTM) can effectively process the dependence relationships in long sequence data and has significant advantages in the field of time series prediction and control. Therefore, it is of great practical significance to develop a multi-parameter intelligent control method for sewage pumps based on a bidirectional LSTM neural network. If it is applied to the field of sewage pump control, it is expected to solve the defects existing in the above traditional control methods, improve the intelligent level and operating effect of sewage pump control, thereby reducing energy consumption and extending the service life of the equipment. Summary of the Invention

[0005] In view of the above problems, the object of the present invention is to provide a multi-parameter intelligent control method for sewage pumps based on a bidirectional LSTM neural network, which overcomes the deficiencies of traditional sewage pump control methods. By using the bidirectional LSTM neural network to comprehensively analyze and predict multiple parameters related to the operation of sewage pumps, the intelligent and precise control of sewage pumps is realized, thereby improving the operation efficiency of sewage pumps, reducing energy consumption, and extending the service life of equipment.

[0006] Specific objectives include precise speed control, predicting the target speed through multi-dimensional data such as inlet flow rate and impurity concentration, and matching the dynamic response characteristics of the servo motor. Optimization of start-stop decisions, generating start-stop instructions autonomously in combination with the real-time state of the equipment to reduce ineffective energy consumption. Multi-level fault warning, real-time detection of current and torque abnormalities, and triggering a hierarchical protection mechanism.

[0007] To achieve the above object, the present invention provides the following technical solutions. A multi-parameter intelligent control method for sewage pumps based on a bidirectional LSTM neural network, comprising the following steps:

[0008] S1. Data acquisition step: Install multiple sensors in the sewage conveyance system where the sewage pump is located, and use the multiple sensors to collect in real time the parameters related to the operation of the sewage pump, and construct a data set with time series characteristics: X t =[Q t ,C t ,n t ,I t ,T t Q t —— Inlet flow rate, unit m 3 / h C t —— Impurity concentration, unit mg / L n t —— Current speed of the pump, unit rpm I t —— Current, unit A T t —— Torque, unit N·m The sensors include a flow sensor and an impurity concentration sensor arranged at the inlet pipeline of the sewage pump. The motor of the sewage pump uses a servo motor, and the servo motor is equipped with advanced sensors and a feedback system, which can directly, real-time and accurately obtain the speed, current and torque information of the motor. These data can be quickly transmitted to the control system through the communication interface of the servo motor.

[0009] S2. Data preprocessing step: It includes three stages: data cleaning, data normalization and time series feature extraction. The data normalization adopts the minimum-maximum normalization method, and the expression is where x is the original data, x​min , x max are the minimum and maximum values of the original dataset, respectively, and x norm is the normalized data. In the data cleaning stage, for the judgment of outliers, the 3σ rule is used to eliminate the outliers. If the data points exceed the range of the mean ± 3 times the standard deviation, they are regarded as abnormal and deleted or replaced with the mean values of the previous and subsequent moments. For the handling of missing values, the linear interpolation method is used to complete the missing data, and the expression is

[0010] In step S2, the time series feature extraction specifically includes calculating the mean μ t and variance The expression is where N is the window size and x i is the data at the i-th moment.

[0011] S3. Construction and training of the bidirectional LSTM neural network model: Construct a neural network model suitable for sewage pump control. The neural network model selects the bidirectional long short-term memory network (Bi-LSTM) structure. Determine that the number of input layer nodes of the neural network is 5, and the time series length T = 10. The corresponding collected data are the corresponding inlet flow rate, impurity concentration, current pump speed, torque, and current. Set the number of hidden layers to 1, and set the number of neurons in the hidden layer to 128. Determine that the number of full connection layer nodes is 3, which are the target speed S 目标 of the pump, the start / stop B 启停 of the pump, and the fault warning L 故障 . Select appropriate activation functions: For S 目标 , select the Sigmoid function and normalize it to the rated speed range of the servo motor. For B 启停 , select the Softmax function for binary classification, where the pump stops at 0 and starts at 1. For L 故障 , select the Softmax function for three-class classification, where it is in the normal situation at 0, a warning is issued at 1, and a serious fault occurs at 2.

[0012] In step S3, the input time series data X t = [Q t , C t , n t , I t , T t . Calculate the hidden state through the bidirectional LSTM layer. The bidirectional LSTM hidden layer consists of a forward LSTM unit and a backward LSTM unit, which can process the past and future information of the data simultaneously, so as to more comprehensively capture the time series features and long-term dependencies in the data. The output layer generates the predicted control parameters according to the output results of the hidden layer.

[0013] Bidirectional LSTM Hidden Layer Processing: Forward LSTM Calculation: Forget Gate: Input Gate: Candidate Value: Cell State Update: Output Gate: Output Gate and Hidden State: Among them, the role of the forget gate is to determine how much information in the cell state at the previous moment needs to be forgotten. is the forward hidden state at the previous moment and the input x at the current moment t are concatenated. The input gate is used to determine how much new information in the input x at that time t needs to be added to the forward cell state. is the candidate value of the forward cell state at the current moment, through the output of the forward forget gate performs a forgetting operation on the forward cell state at the previous moment and then through the output of the forward input gate filters the forward candidate cell state and finally adds these two parts to obtain the forward cell state at the current moment The forward output gate determines how much information in the forward cell state at the current moment needs to be output to the forward hidden state controls the output switch of the information, performs a non-linear transformation on the forward cell state, and then through filters to obtain the forward hidden state

[0014] Backward LSTM Calculation: Forget Gate: Input Gate: Candidate Value: Cell State Update: Output Gate: Output Gate and Hidden State: Among them, the gating mechanism of the backward LSTM is symmetric to that of the forward LSTM. The difference is that it processes data in reverse chronological order. The gating signals are calculated using the hidden states at subsequent moments to capture future information.

[0015] Output of the Bidirectional LSTM Hidden Layer: Concatenate the forward hidden state and the backward hidden state to obtain the output of the bidirectional LSTM hidden layer at time t

[0016] Output layer generation: Use a fully connected layer to map the concatenated bidirectional LSTM hidden states to three outputs: Target rotational speed: Start / stop of the pump: Fault warning:

[0017] S4. The loss function adopts a combination of the mean squared error (MSE) loss function and the cross-entropy loss function. The mean squared error loss function performs well in dealing with continuous value prediction problems. The cross-entropy loss function can more accurately measure the difference between the prediction result and the true label for classification and probability prediction problems. For the prediction of sewage pump control parameters, considering these two loss functions comprehensively can improve the training effect of the network. The formula for the mean squared loss function is: where: N is the number of samples, y i is the true value of the i-th sample, is the predicted value of the i-th sample. The formula for the cross-entropy loss function is: where: C is the total number of categories in the classification task, Y c is the true probability distribution, P c is the probability of the c-th category predicted by the model. The total loss function is the weighted sum of multiple tasks:

[0018] S5. Control parameter prediction steps: Input the real-time collected and preprocessed data into the trained bidirectional LSTM network to obtain the predicted control parameters. According to the predicted control parameters, the operation of the sewage pump is adjusted in real time through the actuator. Since a servo motor is used, it has high-precision control performance and fast response ability. The control system can send precise control instructions to the servo driver, and the servo driver adjusts the power supply frequency and voltage of the motor precisely according to the predicted control parameters such as the rotational speed adjustment amount of the pump, so as to achieve precise control of the rotational speed of the sewage pump. At the same time, the control parameters output by the bidirectional LSTM network include the start / stop state (0 / 1). By combining real-time sewage flow, motor load and other data with the equipment energy efficiency model, start / stop decisions are automatically generated to reduce ineffective energy consumption. Finally, the current and torque are detected in real time. When the data indicates an abnormality, a fault warning is issued. The fault levels output by the bidirectional LSTM network are divided into three levels (0 = normal, 1 = warning, 2 = fault), triggering a hierarchical protection mechanism.

[0019] In the control of the servo motor in step S5, the target rotational speed is mapped to the motor control instruction: V 控制 = K p ·(S 目标 - S实际 ) + K i ·∫(S 目标 - S 实际 )dt In the formula, V 控制 is the motor control voltage, and K p , K i are the PID parameters of the servo driver.

[0020] S6. Feedback adjustment step: Continuously monitor the actual operating status of the sewage pump, compare the actual operating data with the prediction results, and perform online update and adjustment on the trained bidirectional LSTM network according to the error situation. During the feedback adjustment process, use the error analysis module to calculate the error between the actual operating data and the prediction results, such as the mean square error, mean absolute error, etc. If the error exceeds the preset threshold, it is considered that the prediction result is inaccurate and the bidirectional LSTM network needs to be updated. The update method can adopt the incremental learning method, that is, use the newly collected data to fine-tune the network to improve the adaptability and prediction accuracy of the network.

[0021] Compared with the prior art, the present invention has the following advantages:

[0022] Intelligent and precise control: The bidirectional LSTM network can simultaneously utilize the past and future information of the sewage pump operation data, and more comprehensively capture the long-term dependence relationship between the data. Combining the mean square error loss function and the cross-entropy loss function can more accurately measure the difference between the prediction result and the true value, so as to more accurately predict the control parameters of the sewage pump. Combining with the servo motor, accurate speed, current, and torque information can be directly obtained to achieve precise control of the sewage pump. For example, when the sewage flow and impurity concentration change, the pump speed can be adjusted in a timely manner according to the actual operating status of the motor to achieve intelligent control and ensure the efficiency and quality of sewage treatment; Improve the service life of the equipment and reduce energy consumption: By avoiding the frequent and unreasonable start and stop of the sewage pump, the wear of the motor and mechanical components is reduced, and the service life of the equipment is extended. At the same time, according to the real-time working conditions, the operating parameters of the sewage pump are dynamically adjusted. The servo motor can automatically adjust the output power according to the feedback of torque and current, avoiding the energy consumption waste caused by fixed parameter settings in traditional control methods and improving the energy utilization efficiency; Enhance system stability and reliability: Through real-time monitoring and feedback adjustment, it can promptly detect and handle abnormal situations during the operation of sewage pumps, reduce the probability of failures, and improve the stability and reliability of the sewage treatment system. For example, when detecting an abnormal increase in the torque of the motor or an excessive current, the pump speed is promptly reduced or other protective measures are taken to avoid pump damage; Adapt to different working conditions: The neural network model can continuously learn and adapt to different complex working conditions such as sewage composition and flow rate changes based on continuously collected data, has good generalization ability, and can be widely applied to various sewage pump control scenarios. Description of the Drawings

[0024] Figure 1 is the control system architecture diagram of the present invention.

[0025] Figure 2 is the schematic diagram of the bidirectional LSTM neural network structure of the present invention.

[0026] Figure 3 is the data flow diagram of the sewage pump control method of the present invention. Detailed Embodiments

[0028] The following will clearly and completely describe a multi-parameter intelligent control method for sewage pumps based on a bidirectional LSTM neural network of the present invention in combination with embodiments. It should be noted that the described embodiments are only representative embodiments of the present invention, rather than all possible embodiments. Based on the technical solutions and embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work by adjusting parameters, replacing hardware, or expanding application scenarios fall within the scope of protection of the present invention.

[0029] Apply the method of the present invention to the sewage pump drainage system of a small sewage treatment plant. There is a sewage pump responsible for transporting sewage from the sump to the subsequent treatment unit. At fixed time intervals (once every 10 minutes), relevant parameters are collected using sensors set at different positions. A high-precision electromagnetic flowmeter and an optical impurity concentration sensor are installed at the inlet pipe of the sewage pump to monitor the sewage flow rate and impurity concentration entering the sewage pump in real time. The servo motor directly obtains speed, current, and torque information through its interior and quickly transmits this data to the control system through the communication interface of the servo motor. The collected data is recorded in chronological order to form a dataset with time series characteristics.

[0030] Perform data preprocessing. It includes the following steps: 1) Data cleaning; 2) Data normalization; 3) Data format conversion; 4) Temporal feature extraction; In step 1): Check the collected data, identify and remove outliers and missing values. Based on the data at previous and current moments and the actual situation, mark the data that deviates from the normal range as outliers and correct or delete them; for missing values, use linear interpolation to fill them. In step 2): Use the min-max normalization method to map the cleaned data to the interval [0, 1] to eliminate the dimensional differences between different parameters and improve the training effect of the bidirectional LSTM network. In step 3): Convert the normalized data into a three-dimensional tensor format (number of samples, time steps, number of features) suitable for input to the bidirectional LSTM network. For example, select the time step as 10, that is, use the data of the past 10 time points to predict the output at the current moment. Take the data of 5 features (inlet flow rate, impurity concentration, rotational speed, current, torque) at 10 time points as a sample, and the dimension of each sample is [10, 5]. As time goes by, continuously generate new samples, and finally form a three-dimensional tensor data set. In step 4): Specifically, it includes calculating the mean μ t and variance

[0031] to train the bidirectional LSTM network. It includes the following steps: 1) Data set division; 2) Network construction; 3) Training process; In step 1): Divide the preprocessed data set into a training set, a validation set and a test set according to the ratio of 70%, 20%, 10%. In step 2): Construct a bidirectional LSTM network. The number of neurons in the input layer is 5 (corresponding to 5 features of inlet flow rate, impurity concentration, rotational speed, current, torque), the number of neurons in the output layer is 3 (corresponding to the target rotational speed of the pump, the start / stop of the pump and fault warning), and set one layer of bidirectional LSTM hidden layer with 128 neurons in the hidden layer. In step 3): Use the training set to train the bidirectional LSTM network, and set the number of training epochs and the batch size. During the training process, use the total loss function for optimization. Use the Adam optimizer to update the network parameters, and set the initial learning rate to 0.001. At the same time, use the validation set to monitor the performance of the network to avoid overfitting.

[0032] Control parameter prediction and pump control execution. First, predict the control parameters. During the real-time operation of the sewage pump, collect the operation parameters in real time and perform preprocessing. Then input the processed data into the trained bidirectional LSTM network to obtain the predicted control parameters, thus realizing multi-parameter control. When performing the execution control operation, send the predicted control parameters to the servo driver. The servo driver accurately adjusts the power supply frequency and voltage of the motor according to parameters such as the rotational speed adjustment amount, thereby realizing the accurate control of the rotational speed of the sewage pump. At the same time, compare the inlet flow rate in the current time period with the set minimum flow rate threshold. If the inlet flow rate is less than the set minimum flow rate threshold, send a signal of 0 to the servo driver and the pump stops working, otherwise. Generate start-stop commands autonomously in combination with the real-time state of the equipment to reduce ineffective energy consumption. Finally, detect the current and torque in real time. When the current and torque exceed 10% of the set threshold, reduce the rotational speed by 50% and give a fault alarm. When the current and torque exceed 30% of the set threshold, stop the machine immediately and trigger emergency protection.

[0033] Feedback adjustment. Continuously monitor the actual operation state of the sewage pump, compare the actual operation data with the prediction results, and calculate the error. If the error exceeds the preset threshold, online update and adjust the trained bidirectional LSTM network according to the error situation to improve the prediction accuracy and control effect of the network.

[0034] The above are only the specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. An intelligent control method for multi-parameters of sewage pumps based on a bidirectional LSTM neural network, characterized in that, It includes the following steps: S1. Data acquisition step: Use multiple sensors to collect sewage parameters and parameters related to the operation of sewage pumps in real time. The multiple sensors include a flow sensor, an impurity concentration sensor, etc. The sewage parameters include inlet flow rate and impurity concentration, and the sewage pump operation parameters include rotational speed, torque, and current; S2. Data preprocessing step: Preprocess the collected data. The preprocessing includes removing outliers, data normalization processing, and time series feature extraction; S3. Neural network model construction step: Construct a neural network model for sewage pump control, determine the number of input layer nodes of the neural network model, set the number of hidden layers and the number of neurons in each hidden layer, select a suitable activation function, and determine the number of output layer nodes for outputting results related to sewage pump control; S4. Neural network model training step: Divide the preprocessed data sample set into a training set, a validation set, and a test set according to a certain ratio, and use the training set to train the constructed neural network model. During the training process, use a loss function for optimization; S5. Real-time control step of sewage pump: During the actual operation of the sewage pump, input the real-time collected data after preprocessing into the trained neural network model, and output the optimal control parameters of the corresponding sewage pump. The control parameters include target rotational speed, start / stop of the pump, and fault warning. Adjust the operation state of the sewage pump in real time according to the control parameters output by the model. At the same time, set up a real-time monitoring mechanism to continuously monitor the actual operation state of the sewage pump, and feedback the actual operation data back to the data processing unit. If a large deviation is found between the actual operation state and the model prediction control result, a fault warning signal is triggered in a timely manner.

2. The multi-parameter intelligent control method for sewage pumps based on a bidirectional LSTM neural network according to claim 1, characterized in that, In the step S2, the data normalization adopts the minimum-maximum normalization method. The time series feature extraction specifically includes calculating the mean value through a sliding window and variance .

3. A multi-parameter intelligent control method for sewage pumps based on a bidirectional LSTM neural network according to claim 1, characterized in that, In step S3, the neural network model selects the network structure of a bidirectional long short-term memory network (Bi-LSTM). The number of input layer nodes of the Bi-LSTM network model is 5, the number of hidden layers is set to 1, the number of neurons in the hidden layer is 128, and the number of output layer nodes is determined to be 3.

4. The multi-parameter intelligent control method for sewage pumps based on a bidirectional LSTM neural network according to claim 3, wherein, The bidirectional LSTM hidden layer calculates through a forget gate, an input gate, cell state update, and an output gate. Its calculation process satisfies the following formula: The calculation formula of the forget gate is \(f_t = \sigma (W_f\cdot [h_{t - 1}, x_t]+b_f)\) The switching control formula of the input gate is {i}_{t}=\sigma \left ( {{W}_{i}\cdot \left [ {{h}_{t-1},{x}_{t}} \right ]+{b}_{i}} \right ) The candidate value of the input gate is calculated as {N}_{t}=tanh\left ( {{W}_{c}\cdot \left [ {{h}_{t - 1},{x}_{t}} \right ]+{b}_{c}} \right ) The cell state update formula is The switching control formula of the output gate is {o}_{t}=\sigma \left ( {{W}_{o}\cdot \left [ {{h}_{t - 1},{x}_{t}} \right ]+{b}_{o}} \right ) The hidden state calculation formula is Among them is the sigmoid activation function, which maps the input value to the interval \(\left [ {0,1} \right ]\), such that the output of the forget gate can control the degree of information retention and discard. Tanh is the hyperbolic tangent activation function, 、 、 、 are the weight matrices of the corresponding gates respectively, 、 、 、 are the bias vectors of the corresponding gates respectively.

5. A multi-parameter intelligent control method for sewage pumps based on a bidirectional LSTM neural network according to claim 1, characterized in that, In step S4, the loss function adopts a combination of mean squared error (MSE) loss function and cross-entropy loss function.

6. The multi-parameter intelligent control method for sewage pumps based on a bidirectional LSTM neural network according to claim 1, characterized in that, In step S5, the real-time monitoring mechanism triggers a fault warning signal by comparing the actual rotational speed, current, and torque of the sewage pump with the corresponding parameters predicted by the model. The preset threshold is preset according to the performance parameters and actual operation of the sewage pump.

7. A multi-parameter intelligent control method for sewage pumps based on a bidirectional LSTM neural network according to claim 1, characterized in that, In step S5, the drive motor of the sewage pump adopts a servo motor. The servo driver controls the rotational speed of the motor by adjusting the power supply frequency and voltage of the motor, and then controls the drainage flow rate of the sewage pump to make the sewage pump operate in an optimal state. The optimal state is comprehensively determined according to the sewage inflow situation, the current system state, and the preset operation standard. Meanwhile, start-stop instructions are autonomously generated in combination with the real-time status of the equipment to reduce ineffective energy consumption. The current and torque are detected in real time, and when the data indicates an anomaly, a fault warning is issued to trigger a hierarchical protection mechanism.

Images