Integrated magnetic suspension-superconducting hybrid power flywheel energy storage method

Abstract

The invention discloses an integrated magnetic suspension-superconducting hybrid power flywheel energy storage method, and belongs to the field of intelligent flywheel energy storage. The problems that existing flywheel energy storage depends on fixed control parameters and lacks dynamic optimization capacity are solved, time period triggering and event driving triggering are set, and if the current data feedback time is the same as the preset feedback time of a time period table, data are predicted and analyzed according to a normal process; if the feedback time of the current data is consistent with the triggering condition of a certain event in the event-driven table, independently predicting and analyzing an abnormal event in the current data according to an event-driven attribute; therefore, abnormal events can be timely responded and mainly analyzed, and the processing efficiency of the system is improved. After liquid nitrogen flow or flywheel parameters are adjusted, various data of flywheel energy storage are continuously monitored and fed back, and a new adjusting strategy is implemented when a feedback result has deviation, so that abnormal problems are timely found and solved, and a closed-loop control mechanism is formed.

Description

Technical Field

[0001] The present invention relates to the field of intelligent flywheel energy storage, and in particular to an integrated magnetic levitation-superconducting hybrid flywheel energy storage method. Background Art

[0002] Flywheel energy storage, as an efficient energy storage device, has the advantages of high power density, fast charging and discharging speed, and long cycle life. It is widely used in power systems, transportation and other fields. Traditional flywheel energy storage systems usually use mechanical bearings to support the flywheel rotor, but due to the friction loss of mechanical bearings, it is difficult to achieve efficient and long-term energy storage. Therefore, superconducting magnetic bearings and electromagnetic bearings are used to support the flywheel rotor to achieve low-loss magnetic levitation flywheel energy storage. As an active magnetic levitation bearing, the electromagnetic bearing can adjust its thrust stiffness and damping through the control system, actively suppress the rotor vibration and compensate for imbalance, which is beneficial to improve the dynamic stability of the rotor. It is widely used in power generation and power conversion fields such as high-speed motors and flywheel energy storage.

[0003] However, in existing technologies, traditional suspended flywheels and superconducting bearing cooling systems rely on fixed control parameters, making them difficult to cope with complex operating disturbances and lacking dynamic optimization capabilities.

[0004] Therefore, it does not meet the existing needs. We propose an integrated magnetic levitation-superconducting hybrid flywheel energy storage method. Summary of the Invention

[0005] The object of the present invention is to provide an integrated magnetic levitation-superconducting hybrid flywheel energy storage method. By setting a time period trigger and an event-driven trigger, if the current data feedback time is the same as the feedback time preset in the time period table, the data is predicted and analyzed according to the normal process; if the current data feedback time is consistent with an event trigger condition in the event-driven table, then according to the event-driven attribute, the abnormal event in the current data is predicted and analyzed separately; thereby, it is possible to respond to and focus on analyzing abnormal events in a timely manner, thereby improving the processing efficiency of the system; by continuing to monitor and feedback various flywheel energy storage data after adjusting the liquid nitrogen flow rate or flywheel parameters, and implementing a new adjustment strategy when there is a deviation in the feedback result, so as to timely discover and solve abnormal problems, form a closed-loop control mechanism, and solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An integrated magnetic levitation-superconducting hybrid flywheel energy storage method, the method is based on the influence of the flywheel's rotational speed, temperature, and vibration spectrum, as well as the temperature of the superconducting block on the flywheel energy storage, the method comprising:

[0008] A Transformer hybrid architecture model is constructed based on the flywheel's historical multimodal data and the superconducting block's historical temperature data as the flywheel control model.

[0009] The Transformer encoder is used to extract features from historical multimodal data and historical temperature data, with the flywheel speed, temperature, and vibration spectrum data and features as the first input set, and the historical temperature data and features as the second input set.

[0010] Inputting the first input set and the second input set into the flywheel control model respectively, outputting the optimal suspension gap and motor power parameters, as well as the predicted temperature value of the superconducting block based on the flywheel control model, for dynamically adjusting the operating state of the flywheel and the liquid nitrogen flow rate;

[0011] The trained flywheel control model is applied to the actual flywheel energy storage system to regularly optimize the operating status of the flywheel energy storage.

[0012] Furthermore, the operating state of the flywheel energy storage is optimized regularly, specifically:

[0013] Based on the operating time of the flywheel energy storage system, a clock periodic activation trigger mechanism and an event-driven trigger mechanism are set, and a time period table and an event-driven table are made; event-driven includes: flywheel speed deviation, temperature gradient mutation or vibration energy entropy value;

[0014] Based on a variety of sensor devices, multi-modal data of the flywheel, temperature data of the superconducting block, and liquid nitrogen flow rate are collected in real time;

[0015] Based on the set trigger mechanism, various data of the flywheel energy storage system are regularly fed back to the flywheel control model for prediction to obtain the current operating status of the flywheel energy storage system;

[0016] Based on the prediction results, the operating status of the flywheel energy storage system is dynamically adjusted.

[0017] Furthermore, when feeding back various data of the flywheel energy storage system to the flywheel control model, it includes:

[0018] Get the set time period table and event-driven table to get the feedback time of the current data;

[0019] Verify whether the feedback time of the current data is the same as the feedback time preset in the time period table. If so, control the flywheel control model to perform prediction analysis on the current data according to the normal process;

[0020] If not, verify whether the feedback time of the current data is consistent with that in the event-driven table. If so, control the flywheel control model to perform prediction analysis on the abnormal events in the current data based on the event-driven attributes.

[0021] If neither is the case, check whether there are any abnormal settings in the clock periodic activation trigger mechanism and the event-driven trigger mechanism, and reset the clock periodic activation trigger mechanism and the event-driven trigger mechanism.

[0022] Furthermore, before building the Transformer hybrid architecture model, the following steps are required:

[0023] Collect data on the speed, temperature, and vibration spectrum of the flywheel during operation, along with their corresponding suspension gap and motor power parameters. Also collect data on the temperature of the superconducting block, the actual flow rate of liquid nitrogen, and its corresponding cooling requirements, as data samples for training the flywheel control model.

[0024] Clean the collected data samples, supplement missing values or remove outliers;

[0025] Perform Z-score normalization on the data samples, scale all features to the same range, eliminate dimensional differences, and make the data samples conform to the input format of the flywheel control model;

[0026] Data samples with different sampling rates are aligned to a unified timestamp through interpolation and sliced according to a fixed time window to form a multi-channel input matrix.

[0027] Furthermore, after forming a multi-channel input matrix, it includes:

[0028] The first input set and the second input set are segmented according to time series, and each segment includes data within a certain time window;

[0029] Add position encoding to the data in the first input set and the second input set based on the sine function to retain time series information;

[0030] The processed data samples are divided into training sets and test sets, so that each training set and test set includes an input set and an output set.

[0031] Furthermore, feature extraction is performed on historical multimodal data and historical temperature data based on the Transformer encoder, specifically:

[0032] Construct a multi-head self-attention layer, input the first and second input sets that form the multi-channel input matrix into the embedding layer of the Transformer architecture, and capture the cross-modal and cross-timestep correlations between the first and second input sets;

[0033] A cross-modal attention mechanism is introduced to interactively fuse the vibration modal keys and temperature modal values of the first and second input sets as feature information of historical multimodal data and historical temperature data.

[0034] Two independent output heads are provided to output the processing results of the first input set and the second input set respectively.

[0035] Furthermore, the first input set and the second input set are respectively input into the flywheel control model, and the optimal suspension gap and motor power parameters, as well as the temperature prediction value of the superconducting block are respectively output based on the flywheel control model, specifically:

[0036] The first input set and the second input set are input into the embedding layer respectively, and the long-distance dependency between the data in the first input set and the second input set is captured based on the multi-head self-attention layer and the cross-modal attention mechanism;

[0037] Analyze the dependencies and output the optimal suspension gap and motor power parameters, as well as the predicted temperature of the superconducting block, as the prediction results of the flywheel energy storage system;

[0038] Compare the predicted results with the corresponding historical actual results to verify whether the flywheel control model's predictive ability is accurate; if accurate, save the flywheel control model and put it into practical use; if inaccurate, adjust the hyperparameters of the flywheel control model and re-train the prediction until the predicted results are consistent with the actual results;

[0039] Based on the prediction results, the liquid nitrogen flow rate corresponding to the flywheel and its superconducting block is dynamically adjusted.

[0040] Furthermore, based on the prediction results, the liquid nitrogen flow rate corresponding to the flywheel and its superconducting block is dynamically adjusted, specifically as follows:

[0041] Obtain the prediction results of all historical multimodal data and historical temperature data, and extract the prediction results that are in an abnormal state;

[0042] Analyze the forecast results that are in abnormal state one by one and set the adjustment priority based on the degree of abnormality;

[0043] Set different response strategies based on different exception levels.

[0044] Furthermore, after the first input set and the second input set are respectively input into the flywheel control model, the method includes:

[0045] The mean square error is introduced as the loss function, and the loss function value of the first input set and the loss function value of the second input set are obtained through calculation respectively;

[0046] The loss function values of the two tasks are further weighted and summed as the total loss of the flywheel control model, which is used to balance the weights of the first input set and the second input set in the flywheel control model.

[0047] Furthermore, based on the prediction results, the liquid nitrogen flow rate corresponding to the flywheel and its superconducting block is dynamically adjusted, including:

[0048] After adjusting the liquid nitrogen flow rate or flywheel parameters, the actual temperature of the superconducting block, the actual speed of the flywheel, the temperature and vibration spectrum data are monitored in real time as feedback signals;

[0049] Compare the feedback signal with the predicted result and analyze the deviation between the actual value and the predicted value; if the deviation is small or there is no deviation, it means that the adjustment strategy is effective; if the deviation is large, it means that there is a problem with the adjustment strategy and the abnormal problem has not been solved; re-predict the feedback signal based on the flywheel control model and implement a new adjustment strategy; and continue to perform feedback comparison to form a closed-loop control.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] 1. In this invention, a flywheel control model based on a Transformer hybrid architecture is constructed to analyze and predict various data of the energy storage system, and output the optimal suspension gap and motor power parameters, as well as the temperature prediction value of the superconducting block, thereby dynamically adjusting the operating state of the flywheel and the liquid nitrogen flow rate.

[0052] 2. In the present invention, by setting time period triggering and event-driven triggering, if the feedback time of the current data is the same as the feedback time preset in the time period table, the data is predicted and analyzed according to the normal process; if the feedback time of the current data is inconsistent with the time period table, but is consistent with a certain event trigger condition in the event-driven table, then according to the event-driven attribute, the abnormal event in the current data is predicted and analyzed separately; thereby, it is possible to respond to and focus on analyzing abnormal events in a timely manner, avoid unnecessary data analysis, optimize resource utilization and energy consumption management, and improve the processing efficiency, adaptability and reliability of the system.

[0053] 3. In the present invention, by continuing to monitor and feedback various data of the flywheel energy storage system after adjusting the liquid nitrogen flow rate or flywheel parameters, the system can automatically evaluate the effectiveness of the adjustment strategy and timely understand the current operating status; and when there is a deviation in the feedback results, a new adjustment strategy is implemented, so that it can promptly discover and solve abnormal problems, prevent small problems from evolving into big problems, and thus form a closed-loop control mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a flow chart of the integrated magnetic levitation-superconducting hybrid flywheel energy storage method of the present invention. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0056] To solve the technical problems in existing technologies, such as the traditional suspended flywheel and superconducting bearing cooling systems relying on fixed control parameters, making them difficult to cope with complex working conditions and lacking dynamic optimization capabilities, please refer to Figure 1 , this embodiment provides the following technical solutions:

[0057] An integrated magnetic levitation-superconducting hybrid flywheel energy storage method, the method is based on the influence of the flywheel's rotational speed, temperature, and vibration spectrum, as well as the temperature of the superconducting block on the flywheel energy storage, the method comprising:

[0058] Collect the speed, temperature and vibration spectrum data during the operation of the flywheel, as well as the corresponding suspension gap and motor power parameters during the operation of the flywheel; also include collecting the temperature data of the superconducting block and the actual flow rate of liquid nitrogen and its corresponding cooling requirements as data samples for training the flywheel control model; clean the collected data samples, supplement missing values or remove outliers; perform Z-score normalization on the data samples, scale all features to the same range, eliminate dimensional differences, such as: 0 to 1 or -1 to 1, to improve the training efficiency and stability of the model, so that the data samples conform to the input format of the flywheel control model; align data samples with different sampling rates to a unified timestamp through interpolation, for example: if the sampling rate of some data is once per second, and the sampling rate of other data is once per minute, through the interpolation method Align them to a unified timestamp and slice the aligned data according to a fixed time window, such as 60 seconds, to form a multi-channel input matrix; segment the first input set and the second input set according to the time series, with each segment including data within a certain time window; for example, divide each data sample into segments per hour or per minute so that the flywheel control model can capture the dynamic changes of the time series; add position encoding to the data in the first input set and the second input set based on the sine function to retain the time series information; divide the processed data samples into training sets and test sets, so that each training set and test set includes an input set and an output set; for example, the input set is the flywheel speed, temperature and vibration spectrum data, and the output set is the optimal suspension gap and motor power parameters, thereby improving the training efficiency and stability of the flywheel control model.

[0059] A Transformer hybrid architecture model is constructed based on the flywheel's historical multimodal data and the superconducting block's historical temperature data as the flywheel control model. Feature extraction is performed on the historical multimodal data and historical temperature data using the Transformer encoder. The flywheel's speed, temperature, and vibration spectrum data and features are used as the first input set, and the historical temperature data and features are used as the second input set. Specifically:

[0060] A multi-head self-attention layer is constructed, with each head independently calculating self-attention, and the outputs of all heads are concatenated. The first and second input sets, forming a multi-channel input matrix, are fed into the embedding layer of the Transformer architecture, mapping the raw data from the first and second input sets into a high-dimensional space. This allows for better capture of cross-modal and cross-timestep correlations between the first and second input sets, such as the correlation between temperature rise and vibration harmonic amplitude, leading to a better understanding of the flywheel's operating state. A cross-modal attention mechanism is introduced to interactively fuse the vibration mode keys and temperature mode values in the first and second input sets, enhancing the physical correlation between the two, serving as feature information for historical multimodal data and historical temperature data. For example, the flywheel control model leverages this physical correlation to learn how changes in vibration frequency affect temperature, and vice versa, thereby more accurately capturing the dynamics of the flywheel's operating state. Two independent output heads are provided, each outputting the processed results of the first and second input sets. For example, the first output head predicts the optimal levitation gap and motor power parameters, while the second output head predicts the temperature of the superconducting block.

[0061] The first input set and the second input set are respectively input into the flywheel control model. Based on the flywheel control model, the optimal suspension gap and motor power parameters, as well as the temperature prediction value of the superconducting block are respectively output to dynamically adjust the operating state of the flywheel and the liquid nitrogen flow rate. Specifically,

[0062] The first input set, namely the flywheel's speed, temperature, and vibration spectrum data and their features, and the second input set, namely the superconducting block's historical temperature data and its features, are respectively input into the embedding layer. The embedding layer maps the original data into a high-dimensional space to better capture the data's features and correlations, so that each input set obtains a high-dimensional feature representation after passing through the embedding layer.

[0063] The multi-head self-attention layer and cross-modal attention mechanism capture long-range dependencies between data in the first and second input sets. The multi-head self-attention mechanism can simultaneously process attention in multiple subspaces, thereby better capturing complex relationships in the data. For example, it can capture the correlation between temperature changes and vibration spectrum changes, and the impact of these changes on the flywheel's operating state. Another example: through the cross-modal attention mechanism, the flywheel control model learns how changes in vibration frequency affect changes in temperature, and vice versa, thereby strengthening the physical connection between the two.

[0064] The mean square error is introduced as the loss function. After calculation, the loss function value of the first input set and the loss function value of the second input set are obtained respectively; the loss function values of the two tasks are further weighted and summed as the total loss of the flywheel control model, which is used to balance the weights of the first input set and the second input set in the flywheel control model.

[0065] Analyze the dependencies and output the optimal suspension gap and motor power parameters for dynamically adjusting the flywheel's operating state. Output the predicted temperature of the superconducting block for dynamically adjusting the liquid nitrogen flow rate, which serves as the prediction result for the flywheel energy storage system.

[0066] The prediction results are compared with the corresponding historical actual results to verify whether the prediction ability of the flywheel control model is accurate; if it is accurate, the flywheel control model is saved and put into practical use; if it is inaccurate, the hyperparameters of the flywheel control model, such as the learning rate, number of layers, number of hidden units, etc., are adjusted to optimize the model performance, and the prediction training is re-performed until the prediction results are consistent with the actual results, thereby providing an accurate basis for the dynamic adjustment of the flywheel energy storage system.

[0067] Based on the prediction results, the liquid nitrogen flow rate corresponding to the flywheel and its superconducting block is dynamically adjusted as follows:

[0068] Obtain the prediction results of all historical multimodal data and historical temperature data, and extract the prediction results in abnormal states; analyze the prediction results in abnormal states one by one, and set adjustment priorities based on the degree of abnormality; set different response strategies based on different abnormality levels.

[0069] In one embodiment, it is assumed that the abnormal state is defined as the deviation between the predicted value and the actual value exceeds a preset threshold; for example: the deviation between the temperature predicted value and the actual value exceeds 5°C, the deviation between the suspension gap predicted value and the actual value exceeds 0.1mm, and the deviation between the motor power parameter predicted value and the actual value exceeds 10%.

[0070] The degree of abnormality is measured by the size and duration of the deviation; for example: mild abnormality: the deviation is small, such as a temperature deviation of 2-5°C, and the duration is short, such as within 5 minutes; moderate abnormality: the deviation is large, such as a temperature deviation of 5-10°C, and the duration is long, such as 10-30 minutes; severe abnormality: the deviation is extremely large, such as a temperature deviation of more than 10°C, and the duration is very long, such as more than 30 minutes.

[0071] Set different response strategies based on different exception levels, for example:

[0072] Severe abnormality: The adjustment measures are: immediately increase the liquid nitrogen flow rate to quickly reduce the temperature of the superconducting block; adjust the flywheel's suspension gap and motor power parameters to stabilize the flywheel operation.

[0073] Moderate abnormality: Adjustment measures include gradually increasing the liquid nitrogen flow rate and observing temperature changes; adjusting flywheel parameters to optimize operating conditions. The goal is to adjust the temperature and flywheel parameters to the normal range within 10-30 minutes.

[0074] Mild abnormality: Adjustment measures are: record the abnormality, conduct regular monitoring; and conduct a detailed inspection during the next maintenance.

[0075] The trained flywheel control model is applied to the actual flywheel energy storage system to regularly optimize the operating status of the flywheel energy storage. Specifically:

[0076] Based on the operating time of the flywheel energy storage system, a clock periodic activation trigger mechanism and an event-driven trigger mechanism are set up, and a time period table and event-driven table are created. Event-driven events include: flywheel speed deviation, temperature gradient mutation, or vibration energy entropy value. Multiple sensor devices are used to collect multimodal data of the flywheel, temperature data of the superconducting block, and liquid nitrogen flow in real time.

[0077] Based on the set trigger mechanism, various data of the flywheel energy storage system are regularly fed back to the flywheel control model for prediction to obtain the current operating status of the flywheel energy storage system; when the various data of the flywheel energy storage system are fed back to the flywheel control model, it includes:

[0078] Obtain the set time period table and event-driven table, and obtain the feedback time of the current data; verify whether the feedback time of the current data is the same as the feedback moment preset in the time period table. If so, control the flywheel control model to perform predictive analysis on the current data according to the normal process; if not, verify whether the feedback time of the current data is consistent with the event-driven table. If so, control the flywheel control model to perform predictive analysis on the abnormal events in the current data separately according to the event-driven attributes; if neither is the case, check whether there are any setting abnormalities in the clock periodic activation trigger mechanism and the event-driven trigger mechanism, and reset the clock periodic activation trigger mechanism and the event-driven trigger mechanism; based on the prediction results, dynamically adjust the operating status of the flywheel energy storage system.

[0079] Continuing with the above embodiment, the clock periodicity is such as triggering a system optimization process every 10 minutes; event-driven includes but is not limited to: flywheel speed deviation: the deviation between the speed and the set value exceeds a certain threshold (such as ±5%); temperature gradient mutation: the temperature change rate exceeds a certain threshold (such as a change of more than 5°C per minute); vibration energy entropy: the energy entropy value of the vibration spectrum exceeds a preset threshold, indicating abnormal vibration. Then, multiple sensor devices are used to collect multimodal data of the flywheel in real time. The base clock periodic trigger mechanism regularly feeds back various data of the flywheel energy storage system to the flywheel control model and determines whether the current time is consistent with the time point in the time period table. If so, the operating condition data is predicted according to the normal process; or if the detection is consistent with an abnormal event in the event-driven table, the abnormal part is analyzed separately based on the attributes of the abnormal event, thereby improving the system's processing efficiency for abnormal events.

[0080] The beneficial effects achieved by the above content are: by setting time period triggering and event-driven triggering, if the feedback time of the current data is the same as the feedback time preset in the time period table, the data is predicted and analyzed according to the normal process; if the feedback time of the current data is inconsistent with the time period table, but is consistent with an event trigger condition in the event-driven table, then according to the event-driven attribute, the abnormal events in the current data are predicted and analyzed separately; thereby, it is possible to respond to and focus on analyzing abnormal events in a timely manner, avoid unnecessary data analysis, optimize resource utilization and energy consumption management, and improve the processing efficiency, adaptability and reliability of the system.

[0081] After adjusting the liquid nitrogen flow rate or flywheel parameters, the actual temperature of the superconducting block, the actual speed of the flywheel, the temperature and vibration spectrum data are monitored in real time as feedback signals; the feedback signal is compared with the predicted result, and the deviation between the actual value and the predicted value is analyzed; if the deviation is small or there is no deviation, it means that the adjustment strategy is effective; if the deviation is large, it means that there is a problem with the adjustment strategy and the abnormal problem has not been solved; the feedback signal is re-predicted based on the flywheel control model, and a new adjustment strategy is implemented; and feedback comparison is continued to form a closed-loop control.

[0082] The beneficial effects achieved by the above content are: by continuing to monitor and feedback the actual temperature of the superconducting block, the actual speed of the flywheel, the temperature and vibration spectrum data after adjusting the liquid nitrogen flow rate or flywheel parameters, the system can automatically evaluate the effectiveness of the adjustment strategy and timely understand the current operating status; and when there is a deviation in the feedback results, a new adjustment strategy is implemented, so that it can timely discover and solve abnormal problems, avoid small problems from turning into big problems, and thus form a closed-loop control mechanism.

[0083] Working Principle: By constructing a flywheel control model based on a Transformer hybrid architecture, it analyzes and predicts various data of the energy storage system and outputs the optimal suspension gap and motor power parameters, as well as the temperature prediction value of the superconducting block, thereby dynamically adjusting the operating status of the flywheel and the liquid nitrogen flow rate. At the same time, combined with a clock periodic activation trigger mechanism and an event-driven trigger mechanism, intelligent control of the flywheel energy storage system is achieved, ensuring the stable operation and efficient performance of the system.

[0084] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "including," "having," or any other variations thereof are intended to cover non-exclusive possessors, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or includes elements that are inherent to such process, method, article, or apparatus.

[0085] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions, and alterations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An integrated magnetic levitation-superconducting hybrid flywheel energy storage method, characterized in that: The method is based on the influence of the speed, temperature and vibration spectrum of the flywheel and the temperature of the superconducting block on the flywheel energy storage, and the method includes: A Transformer hybrid architecture model is constructed based on the flywheel's historical multimodal data and the superconducting block's historical temperature data as the flywheel control model. The Transformer encoder is used to extract features from historical multimodal data and historical temperature data, with the flywheel speed, temperature, and vibration spectrum data and features as the first input set, and the historical temperature data and features as the second input set. Inputting the first input set and the second input set into the flywheel control model respectively, outputting the optimal suspension gap and motor power parameters, as well as the predicted temperature value of the superconducting block based on the flywheel control model, for dynamically adjusting the operating state of the flywheel and the liquid nitrogen flow rate; The trained flywheel control model is applied to the actual flywheel energy storage system to regularly optimize the operating status of the flywheel energy storage.

2. The integrated magnetic levitation-superconducting hybrid flywheel energy storage method according to claim 1, characterized in that: Regularly optimize the operating status of the flywheel energy storage, specifically: Based on the operating time of the flywheel energy storage system, a clock periodic activation trigger mechanism and an event-driven trigger mechanism are set, and a time period table and an event-driven table are made; event-driven includes: flywheel speed deviation, temperature gradient mutation or vibration energy entropy value; Based on a variety of sensor devices, multi-modal data of the flywheel, temperature data of the superconducting block, and liquid nitrogen flow rate are collected in real time; Based on the set trigger mechanism, various data of the flywheel energy storage system are regularly fed back to the flywheel control model for prediction to obtain the current operating status of the flywheel energy storage system; Based on the prediction results, the operating status of the flywheel energy storage system is dynamically adjusted.

3. The integrated magnetic levitation-superconducting hybrid flywheel energy storage method according to claim 2, characterized in that: When feeding back various data of the flywheel energy storage system into the flywheel control model, including: Get the set time period table and event-driven table to get the feedback time of the current data; Verify whether the feedback time of the current data is the same as the feedback time preset in the time period table. If so, control the flywheel control model to perform prediction analysis on the current data according to the normal process; If not, verify whether the feedback time of the current data is consistent with that in the event-driven table. If so, control the flywheel control model to perform prediction analysis on the abnormal events in the current data based on the event-driven attributes. If neither is the case, check whether there are any abnormal settings in the clock periodic activation trigger mechanism and the event-driven trigger mechanism, and reset the clock periodic activation trigger mechanism and the event-driven trigger mechanism.

4. The integrated magnetic levitation-superconducting hybrid flywheel energy storage method according to claim 1, characterized in that: Before building the Transformer hybrid architecture model, include: Collect data on the speed, temperature, and vibration spectrum of the flywheel during operation, along with their corresponding suspension gap and motor power parameters. Also collect data on the temperature of the superconducting block, the actual flow rate of liquid nitrogen, and its corresponding cooling requirements, as data samples for training the flywheel control model. Clean the collected data samples, supplement missing values or remove outliers; Perform Z-score normalization on the data samples, scale all features to the same range, eliminate dimensional differences, and make the data samples conform to the input format of the flywheel control model; Data samples with different sampling rates are aligned to a unified timestamp through interpolation and sliced according to a fixed time window to form a multi-channel input matrix.

5. The integrated magnetic levitation-superconducting hybrid flywheel energy storage method according to claim 4, characterized in that: After forming the multi-channel input matrix, it includes: The first input set and the second input set are segmented according to time series, and each segment includes data within a certain time window; Add position encoding to the data in the first input set and the second input set based on the sine function to retain time series information; The processed data samples are divided into training sets and test sets, so that each training set and test set includes an input set and an output set.

6. The integrated magnetic levitation-superconducting hybrid flywheel energy storage method according to claim 1, characterized in that: Feature extraction of historical multimodal data and historical temperature data is performed based on the Transformer encoder, specifically: Construct a multi-head self-attention layer, input the first and second input sets that form the multi-channel input matrix into the embedding layer of the Transformer architecture, and capture the cross-modal and cross-timestep correlations between the first and second input sets; A cross-modal attention mechanism is introduced to interactively fuse the vibration modal keys and temperature modal values of the first and second input sets as feature information of historical multimodal data and historical temperature data. Two independent output heads are provided to output the processing results of the first input set and the second input set respectively.

7. The integrated magnetic levitation-superconducting hybrid flywheel energy storage method according to claim 6, characterized in that: The first input set and the second input set are respectively input into the flywheel control model. Based on the flywheel control model, the optimal suspension gap and motor power parameters, as well as the temperature prediction value of the superconducting block are respectively output. Specifically, The first input set and the second input set are input into the embedding layer respectively, and the long-distance dependency between the data in the first input set and the second input set is captured based on the multi-head self-attention layer and the cross-modal attention mechanism; Analyze the dependencies and output the optimal suspension gap and motor power parameters, as well as the predicted temperature of the superconducting block, as the prediction results of the flywheel energy storage system; Compare the prediction results with the corresponding historical actual results to verify whether the prediction ability of the flywheel control model is accurate; if it is accurate, save the flywheel control model and put it into practical use; If it is inaccurate, adjust the hyperparameters of the flywheel control model and re-train the prediction until the prediction results are consistent with the actual results; Based on the prediction results, the liquid nitrogen flow rate corresponding to the flywheel and its superconducting block is dynamically adjusted.

8. The integrated magnetic levitation-superconducting hybrid flywheel energy storage method according to claim 7, characterized in that: Based on the prediction results, the liquid nitrogen flow rate corresponding to the flywheel and its superconducting block is dynamically adjusted as follows: Obtain the prediction results of all historical multimodal data and historical temperature data, and extract the prediction results that are in an abnormal state; Analyze the forecast results that are in abnormal state one by one and set the adjustment priority based on the degree of abnormality; Set different response strategies based on different exception levels.

9. The integrated magnetic levitation-superconducting hybrid flywheel energy storage method according to claim 7, characterized in that: After the first input set and the second input set are input into the flywheel control model respectively, it includes: The mean square error is introduced as the loss function, and the loss function value of the first input set and the loss function value of the second input set are obtained through calculation respectively; The loss function values of the two tasks are further weighted and summed as the total loss of the flywheel control model, which is used to balance the weights of the first input set and the second input set in the flywheel control model.

10. The integrated magnetic levitation-superconducting hybrid flywheel energy storage method according to claim 7, characterized in that: Based on the prediction results, the liquid nitrogen flow rate corresponding to the flywheel and its superconducting block is dynamically adjusted, including: After adjusting the liquid nitrogen flow rate or flywheel parameters, the actual temperature of the superconducting block, the actual speed of the flywheel, the temperature and vibration spectrum data are monitored in real time as feedback signals; Compare the feedback signal with the predicted result and analyze the deviation between the actual value and the predicted value; if the deviation is small or there is no deviation, it means that the adjustment strategy is effective; if the deviation is large, it means that there is a problem with the adjustment strategy and the abnormal problem has not been solved; re-predict the feedback signal based on the flywheel control model and implement a new adjustment strategy; and continue to perform feedback comparison to form a closed-loop control.

Images