Intelligent shutdown decision-making method based on looseness prediction of anode saturable reactor

Abstract

The invention discloses an intelligent shutdown decision-making method based on looseness prediction of an anode saturable reactor, and the method comprises the following steps: S1, collecting the historical information of different operation times for the anode saturable reactors of the same model; s2, calculating feature information of the anode saturable reactor according to historical information of different operation times, and converting the feature information into continuous time sequence data with a time label; s3, constructing an LSTM network, training the LSTM network by using the continuous time sequence data with the time label, and obtaining the trained LSTM network as a prediction model of the looseness degree; and S4, based on a reinforcement learning mode, obtaining an intelligent shutdown decision. According to the invention, the shutdown maintenance opportunity can be dynamically decided before a fault occurs, and the production continuity (frequent shutdown is avoided) and the equipment safety (major faults are avoided) are balanced.

Description

Technical Field

[0001] The present invention relates to an anode saturable reactor, and particularly to an intelligent shutdown decision-making method based on the prediction of the looseness of the anode saturable reactor. Background Art

[0002] In the converter valve of a high-voltage direct current transmission system, the anode saturable reactor plays a key role: as a non-linear inductor, it can effectively limit the high current rise rate that appears at the initial stage of thyristor turn-on, thereby realizing the protection of thyristors. However, due to the magnetostrictive effect, this reactor will generate violent vibrations, which may cause the connection bolts between it and the valve tower to gradually loosen. It should be noted that such faults are often difficult to detect in the initial stage, but if the equipment continues to operate with faults, it may eventually lead to catastrophic damage consequences. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art and provide an intelligent shutdown decision-making method based on the prediction of the looseness of the anode saturable reactor, which can dynamically decide the shutdown and maintenance time before the occurrence of faults, and balance production continuity and equipment safety.

[0004] The purpose of the present invention is achieved through the following technical solutions: an intelligent shutdown decision-making method based on the prediction of the looseness of the anode saturable reactor, including the following steps: S1. For anode saturable reactors of the same model, collect historical information at different operating times; S2. According to the historical information at different operating times, calculate the characteristic information of the anode saturable reactor, and convert the characteristic information into continuous time-series data with time tags; S3. Construct an LSTM network, and use the continuous time-series data with time tags to train the LSTM network to obtain a trained LSTM network as a prediction model for the looseness degree; S4. Obtain an intelligent shutdown decision based on the method of reinforcement learning.

[0005] The beneficial effects of the present invention are: the present invention can dynamically decide the shutdown and maintenance time before the occurrence of faults, and balance production continuity (avoiding frequent shutdowns) and equipment safety (avoiding major faults). The designed state space support monitoring system can be flexibly deployed at any time during the operation of the equipment. Brief Description of the Drawings

[0006] Figure 1 is the flowchart of the method of the present invention; Figure 2 is the input-output schematic diagram of the LSTM; Figure 3 is the flowchart of the reinforcement learning; Figure 4is the graph of the sigmoid function. DETAILED DESCRIPTION

[0007] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the following.

[0008] like Figure 1 As shown, an intelligent shutdown decision method based on loosening prediction of anode saturated reactor includes the following steps: S1. For the same type of anode saturated reactor, collect historical information of different operating times; Initial data acquisition: A predictive model is constructed based on historical operating data to anticipate potential faults during online monitoring. For anode saturation reactors of the same model installed at converter stations nationwide, multiple sets of samples are collected for different shutdown intervals (e.g., operating durations of 5, 6, 8, 10, and 12 months). Soundprint data is collected immediately before shutdown using soundprint sensors installed on each reactor, and vibration acceleration data is collected in real time using a laser vibrometer. Furthermore, during shutdown inspections, a digital torque wrench is used to measure the loosening torque of the connecting bolts (i.e., the minimum torque at which the bolts begin to turn). The degree of bolt loosening is expressed as "(tightening torque - loosening torque) / tightening torque × 100%)."

[0009] S2. Calculate characteristic information of the anode saturated reactor based on historical information of different operating times and convert the characteristic information into continuous time series data with time tags; Sensor data: The voiceprint collection sensor obtains the voiceprint data of the day of the outage, and the laser vibrometer obtains the vibration acceleration data of the day of the outage; Degree of looseness: As the increase of looseness leads to accelerated failure, a nonlinear fault propagation model is defined, assuming that the degree of looseness L ( t ) changes with time t (unit: day) following the exponential growth law, , the discrete data can be fitted with nonlinear least squares method to obtain unknowns k The degree of looseness is the output side of the LSTM.

[0010] Vibration acceleration peak value A: As the mechanical structure loosens, the system stiffness decreases, resulting in a significant increase in vibration.

[0011] Growth is slow in the early stages and accelerates (non-linearly) in the later stages, consistent with exponential growth: a, b, c are fitting constants, and t is time.

[0012] Vibration acceleration high-frequency and low-frequency energy ratio B: , Ai 100* i Hz corresponds to the amplitude of the acceleration waveform. As the mechanical structure loosens, the proportion of high-frequency energy increases, and this feature gradually increases over time, showing a nonlinear growth trend. a, b, c are fitting constants, and t is time.

[0013] Sound pressure level C: As the bolts become looser, the sound pressure level of the anode saturated reactor noise will increase, usually showing a smooth increase over time. a, b, c are fitting constants, and t is time.

[0014] Voiceprint spectrum complexity D: , , A i 100* i Hz corresponds to the amplitude of the sound wave, R i 100* i Hz corresponds to the energy density of the sound wave. As the bolt loosens more, the structure of the spectrum becomes more complex, so this feature will gradually increase over time. a, b, c are fitting constants, and t is time.

[0015] S3. Build an LSTM network and train it using continuous time series data with time labels to obtain the trained LSTM network as a prediction model for the degree of looseness; Figure 2 As shown: LSTM input side: The input of LSTM is multimodal time series data, including the peak value and mean value of vibration acceleration (m / s 2 ), vibration acceleration high- and low-frequency energy ratio, sound pressure level (dB), and voiceprint spectrum complexity. Assume the time window is a sliding time window, with each time step corresponding to one day's data.

[0016] LSTM output side: looseness L S4. Based on reinforcement learning, intelligent shutdown decision is obtained, such as Figure 3 As shown: Building a reinforcement learning environment 4.1 State Space Design: The state space includes the current looseness and the rate of change of the looseness.

[0017] Status indication: s =( , );in L is the current predicted looseness, d L / d t is the rate of change of looseness; dL is the difference between the current predicted loosening degree and the loosening degree obtained from the previous prediction, d t is a time step, i.e., one day; 4.2 Action Space Design: Action a = 0 (Continue to run): In this action, the model decides to continue running the device until the next decision moment. The interval of the decision moment can be configured according to the device monitoring frequency, safety requirements, and maintenance management needs.

[0018] Action a = 1 (Stop for maintenance): In this action, the model decides to pause the device for necessary maintenance.

[0019] 4.3 Reward Function Design: Since a discontinuous and mutant reward function may cause the final decision of the agent to be bounded by a threshold, the reward function is designed as a continuous function. The Sigmoid function is introduced and used S to represent.

[0020] As Figure 4 shown, the mathematical expression of the Sigmoid function is: Reward function: The first line: Rate of change penalty (the first term): If the rate of change of the device loosening is large, the operation will be penalized to prevent failure to stop in time during rapid changes; Balance reward (the second term): When the loosening degree of the device is low and the rate of change is small, the device should be encouraged to continue running and a reward should be given; The second line: Device failure penalty: When the loosening degree of the device exceeds the threshold of 0.8, but the system fails to stop in time, the device fails and a penalty of -100 is given; The third line: Reward for timely shutdown (the first term): When the loosening degree is high, a reward should be given for timely shutdown to prevent failure; Penalty for frequent shutdowns (the second term): If the loosening degree is low, the system will be penalized when shutting down; Reflect the engineering concept of "timely maintenance" (the third and fourth terms): Form an optimal shutdown interval, the third term: Penalty for premature shutdown ( L <0.5), the fourth term: Reward for the best shutdown timing ( L close to 0.6).

[0021] 4.4 Selection of Reinforcement Learning Algorithm: Since the state space sThe dimension is small (three-dimensional) and it is a continuous space. The deep Q-network (DQN) is used in the process of training the model. The traditional Q-learning algorithm stores the state-action value function in the form of a Q-table, which is difficult to handle continuous state spaces. DQN can handle continuous states by means of neural network approximation. Therefore, the DQN algorithm is selected as the reinforcement learning algorithm for this study.

[0022] Carry out model training: 4.41 Initialize the deep Q-network structure Construct a neural network that takes the state vector as input and outputs the Q-values corresponding to two actions. The network contains 2 hidden layers, uses the ReLU activation function, and the output layer uses a linear activation to represent the Q-value. The initial weights are initialized with random numbers.

[0023] 4.42 The agent interacts with the environment to generate experiences Run multiple training cycles in the simulation environment. In each cycle, the agent selects actions according to the ε -greedy policy, and records the state transition (current state, executed action, obtained reward, next state) to form experience data.

[0024] 4.43 Experience replay and target network update Store the experience data generated during the interaction process in the replay buffer. Every set number of steps, randomly sample experiences from the replay pool for training the main Q-network. Use the mean squared error loss function to optimize the Q-value approximation process, and periodically synchronize the target network weights.

[0025] 4.44 Training convergence and policy extraction Repeat the above process until the average reward change of the model is <0.1 for 100 consecutive rounds. Finally, extract the trained model as the intelligent shutdown decision controller for online judgment of whether to perform equipment maintenance operations.

[0026] The present invention has been described and proven in detail here, enabling those skilled in the art to understand and apply the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. And as long as the changes and modifications made by those skilled in the art do not depart from the spirit and scope of the present invention, they should all be within the protection scope of the appended claims of the present invention.

Claims

1. An intelligent shutdown decision-making method based on the prediction of anode saturable reactor looseness, characterized in that: It includes the following steps: S1. For the anode saturation reactor of the same model, collect historical information at different operating times; S2. According to the historical information at different operating times, calculate the characteristic information of the anode saturation reactor, and convert the characteristic information into continuous time-series data with time tags; S3. Construct an LSTM network, and use the continuous time-series data with time tags to train the LSTM network to obtain a trained LSTM network as a prediction model for the degree of looseness; S4. Obtain an intelligent shutdown decision based on the method of reinforcement learning.

2. The intelligent shutdown decision-making method for looseness prediction based on an anode saturable reactor according to claim 1, wherein: The operating time is in days, and each day is a time step. The historical information at different operating times, that is, the historical information at different time steps, includes: The sound pattern data collected by the sound pattern acquisition sensor, the vibration acceleration data collected by the laser vibrometer, and the loosening torque of the reactor connection bolts collected by the digital display torque wrench; The voiceprint data includes the effective value of the sound pressure, taking the root mean square value of the sound pressure within one time step, and the amplitude of the sound wave corresponding to 100* i Hz when i = 1, 2, …, 30; The vibration acceleration data includes: the peak values of vibration acceleration within one time step, and the amplitudes of the vibration acceleration waveforms corresponding to 100* i Hz when i = 1, 2, …, 30.

3. An intelligent shutdown decision-making method for predicting looseness based on an anode saturable reactor according to claim 1, characterized in that: In step S1, when collecting the historical information of the anode saturation reactor of the same model at different operating times, it needs to be collected during a fixed period before its shutdown, and the fixed period is one day.

4. The intelligent shutdown decision-making method for looseness prediction based on an anode saturable reactor according to claim 2, characterized in that: The step S2 includes: S201. Calculate the characteristic information of the anode saturation reactor, including the average peak value of acceleration, the high-low frequency energy ratio of vibration acceleration, the sound pressure level, the complexity of the sound pattern spectrum, and the degree of looseness; According to the vibration acceleration data collected by the laser vibrometer, calculate the average peak value A of the vibration acceleration and the high-low frequency energy ratio B of the vibration acceleration: Among them, the average peak value A of the vibration acceleration is obtained by averaging the peak values of the vibration acceleration; The calculation method of the high-low frequency energy ratio B of the vibration acceleration is as follows: ； A i is the amplitude of the acceleration waveform corresponding to 100* i Hz; According to the sound pattern data collected by the sound pattern acquisition sensor, calculate the sound pressure level C and the complexity D of the sound pattern spectrum; Among them, the calculation method of the sound pressure level C is as follows: C= 20*lg(P / P 0 ) where P is the effective value of the measured sound pressure, P 0 is the reference sound pressure, taken as 20 μPa; The calculation method of the voiceprint spectrum complexity D is as follows: , , A i is the amplitude of the sound wave corresponding to 100* i Hz, R i is the energy proportion of the sound wave corresponding to 100* i Hz; Calculate the degree of looseness based on the loosening torque of the reactor connection bolts collected by the digital display torque wrench L : L = (Tightening torque - Loosening torque) / Tightening torque * 100% Among them, when the connection bolts of the reactor are in a completely tightened state, the torque of the reactor connection bolts is collected in advance by using a digital display torque wrench, and the obtained torque is the tightening torque; S202. Fit the relationships between the average peak value of acceleration, the high-low frequency energy ratio of vibration acceleration, the sound pressure level, the complexity of the sound pattern spectrum, the degree of looseness and the operating time t; A1. Define the mean peak value of vibration acceleration at the running time of t as :[[-END]] At different operating times t, the corresponding average peak acceleration is obtained according to step S201, denoted as , through multiple and the corresponding operating time t, the fitting constants a 1, b1, c1 are obtained, so as to obtain the variation relationship of the average peak vibration acceleration with the operating time; A2. Define the high-low frequency energy ratio of acceleration at the running time of t : At different operating times t, the corresponding high-low frequency energy ratio of acceleration is obtained according to step S201, denoted as , through multiple and the corresponding operating time t, fitting constants a 2, b2, c2, so as to obtain the variation relationship of the high-low frequency energy ratio of acceleration with the operating time; A3. Define the sound pressure level at the running time of t : At different operating times t, the corresponding sound pressure levels are obtained according to step S201, denoted as , through multiple and the corresponding operating time t, fitting constants a 3, b3, c3, so as to obtain the variation relationship of the sound pressure level with the operating time; A4. Define the voiceprint spectral complexity at runtime t : At different running times t, the corresponding voiceprint spectrum complexity is obtained according to step S201, denoted as , through multiple and the corresponding running time t, fitting constants a 4, b4, c4, so as to obtain the variation relationship of the voiceprint spectrum complexity with the running time; A5. Define the degree of looseness at the running time of t : At different operating times t, the corresponding degree of looseness is obtained according to step S201, denoted as , through multiple and the corresponding operating time t, fit k , so as to obtain the variation relationship of the degree of looseness with the operating time; S203. According to the transformation relationships fitted from A1 to A5, obtain the continuous time-series data of the average peak value of acceleration, the high-low frequency energy ratio of vibration acceleration, the sound pressure level, the complexity of the sound pattern spectrum, and the degree of looseness with time tags. In the continuous time-series data, each day is a time step.

5. The intelligent shutdown decision-making method for looseness prediction based on an anode saturable reactor according to claim 4, wherein: When training the LSTM network in step S3, at each running time t, the data ( , , , ) is used as the input feature of the LSTM network, and the corresponding loosening degree is used as the expected output to implement the training of the LSTM network, obtaining a trained and mature LSTM network as a prediction model for the loosening degree.

6. The intelligent shutdown decision-making method for looseness prediction based on an anode saturable reactor according to claim 4, wherein: In step S4, when the operating time of the equipment is unknown, starting from any time point, collect the average peak value A of acceleration, the high-low frequency energy ratio B of vibration acceleration, the sound pressure level C, and the complexity D of the sound pattern spectrum according to step S201, and send them into the trained LSTM network to predict the corresponding degree of looseness. And starting from the first prediction, perform a prediction of the degree of looseness every other time step. Starting from the second prediction as the decision starting point, obtain an intelligent shutdown decision based on the method of reinforcement learning.

7. An intelligent shutdown decision-making method for predicting looseness based on an anode saturable reactor according to claim 6, characterized in that: The obtaining of the intelligent shutdown decision based on the method of reinforcement learning starting from the second prediction as the decision starting point includes: S401. State space design: Status representation: s = ( , ); where L is the degree of looseness currently predicted, dL / dt is the change rate of the degree of looseness; dL is the difference between the degree of looseness currently predicted and the degree of looseness predicted last time, and dt is a time step, that is, one day; S402. Action space design: Action a = 0: Under this action, the model decides to continue running the device until the next decision point; Action a = 1: Under this action, the model decides to pause the device for maintenance; S403. Reward function design: Since the final decision of the agent is bounded by a threshold due to the discontinuous and mutated reward function, the reward function is designed as a continuous function, and the Sigmoid function is introduced and represented by S as follows: The mathematical expression of the sigmoid function is: The reward function is designed as: S404. Model training: The deep Q-network, i.e., DQN network, is adopted in the process of training the model: S4041. Initialize the deep Q-network structure: Construct a neural network with the state vector as the input and the Q-values corresponding to two actions as the output; the network contains 2 hidden layers, the ReLU activation function is used, and the linear activation is used in the output layer to represent the Q-values; the initial weights are initialized with random numbers; S4042. The agent interacts with the environment to generate experiences Run multiple training cycles in the simulation environment. In each cycle, the agent selects actions according to ε - the greedy policy, and records the state transitions, including the current state, the executed action, the obtained reward, and the next state, to form experience data; S4043. Experience replay and target network update: Store the experience data generated during the interaction in the replay buffer. Every set number of steps, randomly sample experiences from the replay pool for training the main Q-network. The mean squared error loss function is used to optimize the Q-value approximation process, and the target network weights are synchronized periodically; S4044. Training convergence and policy extraction: Repeat the above process until the average reward change of the model is <0.1 for 100 consecutive rounds. Finally, extract the trained model as the intelligent shutdown decision controller for online judgment of whether to perform equipment maintenance operations.

Images