A power level magnetic suspension flywheel energy storage control scheduling method

Abstract

The application provides a power level magnetic suspension flywheel energy storage control scheduling method and relates to the technical field of electric power control.The application adopts a sliding window statistics and LSTM hybrid prediction model, improves the prediction accuracy of short-term power demand fluctuation, makes the flywheel charging and discharging strategy be able to accurately respond to the power grid demand in advance, and reduces the energy waste or power shortage caused by prediction lag; secondly, the double closed loop adaptive control generates a global optimal torque current reference value through the outer ring model predictive control, combines the inner ring recursive least square method to identify the moment of inertia and the damping coefficient in real time, dynamically adjusts the proportional gain, and solves the overshoot and oscillation problems caused by the fixed parameters of the traditional PID; in view of the nonlinear disturbance of the magnetic suspension bearing, the sliding mode controller and the extended state observer are introduced, the unmodeled dynamics are compensated, and the rotor displacement deviation is inhibited, so that the application realizes the coordinated breakthrough of response speed, economy, reliability and safety.

Description

Technical Field

[0001] The present invention relates to the field of power control technology, and in particular to a power-level magnetic levitation flywheel energy storage control and scheduling method. Background Art

[0002] As renewable energy penetration increases and grid load fluctuations intensify, the power system is increasingly in need of fast-response energy storage technologies. Magnetic levitation flywheel energy storage, with its high power density, long cycle life, and millisecond-level response characteristics, is an ideal choice for smoothing short-term grid fluctuations and providing frequency support.

[0003] However, existing technologies have significant limitations: Traditional forecasting models lack the accuracy to capture minute-by-minute fluctuations in electricity demand, causing flywheel charging and discharging strategies to lag behind actual demand; speed control often uses fixed-parameter PID, which struggles to adapt to dynamic scenarios such as time-varying flywheel inertia and frequent load disturbances, leading to overshoot or oscillation; and existing linear control methods have limited ability to suppress nonlinear disturbances. These issues have severely restricted the large-scale application of magnetically levitated flywheel energy storage in power-level scenarios. Summary of the Invention

[0004] To solve the above technical problems, the technical solution adopted by the present invention is a power-grade magnetic levitation flywheel energy storage control and scheduling method, comprising the following steps:

[0005] S1. Data collection and preprocessing: Collect grid load, flywheel status, bearing temperature, electromagnetic parameters and ambient temperature, perform Kalman filtering to reduce noise, and extract effective features; S2. Power demand forecasting: Use sliding window statistics and LSTM neural network to build a hybrid forecasting model, and calculate the mean of the sliding window. and variance , LSTM input contains 、 , load change rate , output short-term power demand fluctuation , the objective function is optimized jointly with mean square error and L1 regularization; S3. Speed ​​trajectory generation: According to the prediction results, the predicted power fluctuation is converted into the speed target value , and apply acceleration constraints ,in For current energy storage, Is the charge and discharge efficiency , To control the cycle, Flywheel moment of inertia, acceleration constraint is the set value; S4. Double closed-loop control: the outer loop uses model predictive control to generate the torque current reference value , prediction time domain , control time domain ,weight The inner loop uses the recursive least squares method to identify parameters online and adjust the control gain. S5. Magnetic bearing compensation: The rotor displacement disturbance is suppressed by a sliding mode controller, and an extended state observer is introduced to compensate for unmodeled dynamics. S6. Dynamic energy scheduling: A multi-objective optimization problem model is constructed based on real-time electricity prices, flywheel life loss, and grid frequency deviation. Solve for the optimal charge and discharge power, where is the real-time electricity price, is the flywheel life loss coefficient, is the speed change.

[0006] Furthermore, data collection is specifically as follows: grid load data is obtained through smart meters at a frequency of 10Hz to obtain the instantaneous values ​​of three-phase voltage and current; flywheel status data is measured by a magnetic encoder to measure the speed, and an optical fiber temperature sensor to measure the bearing temperature.

[0007] Furthermore, the data preprocessing adopts the improved Kalman filter: , ,in is the estimated value of the state, is the Kalman gain, To observe the noise covariance matrix, the filter bandwidth is dynamically adjusted to match the signal characteristics; the window length of the sliding window statistics method corresponding to 1 minute data is set to , the hidden layer of the LSTM neural network has 32 neurons, is the fluctuation amount for the next 5 minutes; the objective function is optimized as follows: ,in is the regularization coefficient.

[0008] Furthermore, in the inner loop identification parameters, the least squares method is used to update the moment of inertia and damping coefficient , forgetting factor , the proportional coefficient of the control gain ,in , The ideal speed reference value generated based on power demand forecast and energy scheduling, is the current speed of the flywheel measured in real time, It is the deviation between the target value and the actual value, which is used for feedback control adjustment.

[0009] Furthermore, in the magnetic bearing compensation step, the sliding mode controller is designed as follows: , ,in is the rotor displacement deviation, the integral coefficient , switch gain ; The disturbance compensation formula of the extended state observer is: , is the torque current after compensation, is the total disturbance of the extended state observer, is the motor torque constant, observer bandwidth .

[0010] Furthermore, in the dynamic energy scheduling step, the constraints of the multi-objective optimization model of dynamic energy scheduling are: ,in and The speed safety margin is 20%-95% of the rated speed. is the rated power of the flywheel, The solution algorithm uses improved dynamic programming for the bearing temperature, and the discount factor .

[0011] Furthermore, the hierarchical protection strategy includes: Level 1 protection: when the speed exceeds When the charging and discharging power is limited to 50%; Secondary protection: the bearing temperature exceeds When the vibration amplitude exceeds 50μm, the emergency brake is triggered and the flywheel stops within 3 seconds.

[0012] Compared with the existing technology, the beneficial effects of the present invention include: using a sliding window statistics and LSTM hybrid prediction model to improve the prediction accuracy of short-term power demand fluctuations, so that the flywheel charging and discharging strategy can respond to the grid demand accurately in advance, reducing energy waste or power shortage caused by prediction lag; secondly, the dual closed-loop adaptive control generates a global optimal torque and current reference value through the outer loop model predictive control, combined with the inner loop recursive least squares method to identify the moment of inertia and damping coefficient in real time, dynamically adjust the proportional gain, and solve the overshoot and oscillation problems caused by the fixed parameters of traditional PID; for the nonlinear disturbance of the magnetic levitation bearing, a sliding mode controller and an extended state observer are introduced to compensate for the unmodeled dynamics and suppress the rotor displacement deviation. The present invention achieves a synergistic breakthrough in response speed, economy, reliability and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The disclosure of the present invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. In the accompanying drawings, the same reference numerals are used to refer to the same components. Among them:

[0014] Figure 1 The flowchart of a power-level magnetic levitation flywheel energy storage control and scheduling method proposed according to one embodiment of the present invention is schematically shown. DETAILED DESCRIPTION

[0015] It is easy to understand that according to the technical solution of the present invention, without changing the essential spirit of the present invention, a person skilled in the art can propose a variety of interchangeable structural modes and implementation modes. Therefore, the following specific embodiments and drawings are only exemplary descriptions of the technical solution of the present invention and should not be regarded as the entire invention or as a limitation or restriction of the technical solution of the present invention.

[0016] According to the embodiment of the present invention, Figure 1 A method for controlling and scheduling power-level magnetic levitation flywheel energy storage includes the following steps.

[0017] Step 1. Data Acquisition and Preprocessing: Grid load, flywheel status, bearing temperature, electromagnetic parameters, and ambient temperature are collected, and Kalman filtering is performed to reduce noise and extract effective features. Grid load data is obtained through a smart meter at a frequency of 10 Hz to obtain the instantaneous values ​​of three-phase voltage and current. Flywheel status data is measured using a magnetic encoder to measure speed, and a fiber optic temperature sensor to measure bearing temperature. Data preprocessing uses an improved Kalman filter: , ,in For the The estimated state value at time t, is the predicted value based on the previous moment, is the actual observation value measured by the sensor, To map the state vector to the observation matrix of the observation space, is the forecast error covariance matrix, is the Kalman gain, To observe the noise covariance matrix, the filter bandwidth is dynamically adjusted to match the signal characteristics; by dynamically adjusting the filter bandwidth, high-frequency noise is filtered out, effective features are extracted, and high-precision data is provided for subsequent prediction and control.

[0018] Step 2. Power demand forecast: Use sliding window statistics and LSTM neural network to build a hybrid forecasting model, and calculate the mean of the sliding window and variance , LSTM input contains 、 , load change rate , output short-term power demand fluctuation The objective function is optimized jointly with mean square error and L1 regularization; the window length of the sliding window statistics method corresponding to 1 minute data is set to , the hidden layer of the LSTM neural network has 32 neurons, is the fluctuation amount for the next 5 minutes; the objective function is optimized as follows: ,in is the load change rate, is the LSTM neural network weight parameter, is the regularization coefficient; the LSTM neural network predicts future power demand fluctuations by learning the time series characteristics of historical data and combining the mean and variance of the sliding window. The objective function includes mean square error and L1 regularization to ensure that the prediction results have both accuracy and generalization ability.

[0019] Step 3. Speed ​​trajectory generation: Based on the prediction results, the predicted power fluctuation is converted into the speed target value , and apply acceleration constraints ,in For current energy storage, , is the current speed, Is the charge and discharge efficiency , To control the cycle, Flywheel moment of inertia, The maximum permissible acceleration is to prevent excessive mechanical stress; according to the predicted power fluctuations , calculate the target speed .

[0020] Step 4. Dual closed-loop control: The outer loop uses model predictive control to generate torque current reference values , prediction time domain , control time domain ,weight ; The inner loop uses the recursive least squares method to identify parameters online and adjust the control gain; in the inner loop identification parameters, the least squares method is used to update the moment of inertia and damping coefficient , forgetting factor , the proportional coefficient of the control gain ,in , The ideal speed reference value generated based on power demand forecast and energy scheduling, is the current speed of the flywheel measured in real time, It is the deviation between the target value and the actual value, which is used for feedback control adjustment.

[0021] Step 5. Magnetic bearing compensation: The rotor displacement disturbance is suppressed by a sliding mode controller, and an extended state observer is introduced to compensate for the unmodeled dynamics. The sliding mode controller is designed as follows: ,in is the rotor displacement deviation, the integral coefficient , switch gain , Represents the sliding surface, combined with the displacement deviation and its integral to enhance robustness; the disturbance compensation formula of the extended state observer is: , is the torque current after compensation, is the total disturbance of the extended state observer, is the motor torque constant, observer bandwidth .

[0022] Step 6. Dynamic Energy Scheduling: Build a multi-objective optimization model based on real-time electricity prices, flywheel life loss, and grid frequency deviation. Solve for the optimal charge and discharge power, where is the real-time electricity price, is the flywheel life loss coefficient, is the speed change, which reflects the degree of mechanical wear; the constraints of the multi-objective optimization model of dynamic energy scheduling are ,in and The speed safety margin is 20%-95% of the rated speed. is the rated power of the flywheel, The solution algorithm uses improved dynamic programming for the bearing temperature, and the discount factor .

[0023] Step 7. Safety protection: Real-time monitoring of speed, temperature and vibration data, triggering the hierarchical protection strategy; the hierarchical protection strategy includes: Level 1 protection: speed exceeds When the charging and discharging power is limited to 50%; Secondary protection: the bearing temperature exceeds When the vibration amplitude exceeds 50μm, the emergency brake is triggered and the flywheel stops within 3 seconds.

[0024] The technical scope of the present invention is not limited to the contents of the above description. Those skilled in the art can make various deformations and modifications to the above embodiments without departing from the technical idea of ​​the present invention, and these deformations and modifications should all fall within the protection scope of the present invention.

Claims

1. A method for controlling and scheduling power-grade magnetic levitation flywheel energy storage, characterized in that: The following steps are involved: S1. Data acquisition and preprocessing: Collect grid load, flywheel status, bearing temperature, electromagnetic parameters, and ambient temperature, perform Kalman filtering to reduce noise, and extract effective features; S2. Power demand forecasting: A hybrid forecasting model is constructed using sliding window statistics and LSTM neural network, and the sliding window calculates the mean. and variance , LSTM input contains 、 , load change rate , output short-term power demand fluctuation , the objective function is optimized jointly with mean square error and L1 regularization; S3. Speed ​​trajectory generation: According to the prediction results, the predicted power fluctuation is converted into the speed target value , and apply acceleration constraints ,in For current energy storage, Charge and discharge efficiency , To control the cycle, Flywheel moment of inertia, acceleration constraint is the set value; S4. Double closed-loop control: the outer loop uses model predictive control to generate the torque current reference value , prediction time domain , control time domain ,weight The inner loop uses the recursive least squares method to identify parameters online and adjust the control gain. S5. Magnetic bearing compensation: The rotor displacement disturbance is suppressed by a sliding mode controller, and an extended state observer is introduced to compensate for unmodeled dynamics. S6. Dynamic energy scheduling: A multi-objective optimization problem model is constructed based on real-time electricity prices, flywheel life loss, and grid frequency deviation. Solve for the optimal charge and discharge power, where is the real-time electricity price, is the flywheel life loss coefficient, is the speed change; S7. Safety protection: Real-time monitoring of speed, temperature and vibration data to trigger hierarchical protection strategies.

2. A power-grade magnetic levitation flywheel energy storage control and scheduling method according to claim 1, characterized in that: The data collection is specifically as follows: the grid load data is obtained by a smart meter at a frequency of 10Hz to obtain the instantaneous values ​​of three-phase voltage and current; the flywheel status data is measured by a magnetic encoder to measure the speed, and the optical fiber temperature sensor to measure the bearing temperature.

3. A power-grade magnetic levitation flywheel energy storage control and scheduling method according to claim 1, characterized in that: The data preprocessing adopts improved Kalman filtering: ,in is the estimated value of the state, is the Kalman gain, The noise covariance matrix is ​​observed and the filter bandwidth is dynamically adjusted to match the signal characteristics.

4. A power-grade magnetic levitation flywheel energy storage control and scheduling method according to claim 3, characterized in that: In the power demand forecasting step, the window length of the sliding window statistics method corresponding to 1 minute data is set to , the hidden layer of the LSTM neural network has 32 neurons.

5. A power-grade magnetic levitation flywheel energy storage control and scheduling method according to claim 4, characterized in that: The objective function is optimized as follows: ,in is the regularization coefficient.

6. A power-grade magnetic levitation flywheel energy storage control and scheduling method according to claim 1, characterized in that: In the inner loop identification parameters, the least square method is used to update the moment of inertia and damping coefficient , forgetting factor , the proportional coefficient of the control gain ,in , The ideal speed reference value generated based on power demand forecast and energy scheduling, is the current speed of the flywheel measured in real time, It is the deviation between the target value and the actual value, which is used for feedback control adjustment.

7. The method for controlling and dispatching power-grade magnetic levitation flywheel energy storage according to claim 1, characterized in that: In the magnetic bearing compensation step, the sliding mode controller is designed as follows: ,in is the rotor displacement deviation, the integral coefficient , switch gain .

8. A power-grade magnetic levitation flywheel energy storage control and scheduling method according to claim 7, characterized in that: The disturbance compensation formula of the extended state observer is: , is the torque current after compensation, is the total disturbance of the extended state observer, is the motor torque constant, observer bandwidth .

9. The method for controlling and dispatching power-grade magnetic levitation flywheel energy storage according to claim 1, characterized in that: In the dynamic energy scheduling step, the constraints of the multi-objective optimization model of dynamic energy scheduling are: ,in and The speed safety margin is 20%-95% of the rated speed. is the rated power of the flywheel, The solution algorithm uses improved dynamic programming for the bearing temperature, and the discount factor .

10. The method for controlling and dispatching power-grade magnetic levitation flywheel energy storage according to claim 1, characterized in that: In step S7, the hierarchical protection strategy includes: Level 1 protection: when the speed exceeds When the charging and discharging power is limited to 50%; Secondary protection: the bearing temperature exceeds When the vibration amplitude exceeds 50μm, the emergency brake is triggered and the flywheel stops within 3 seconds.

Images