SOC estimation method for flywheel energy storage array system

Abstract

The invention relates to the technical field of flywheel energy storage, and discloses an SOC estimation method for a flywheel energy storage array system. According to the method, running parameters such as flywheel rotating speed and bus current are synchronously collected through a multi-source sensing unit, a dynamic parameter observation model is input to estimate dynamic characteristic parameters, and the model dynamically adjusts observation weight based on historical data. And inputting the dynamic characteristic parameters into a multi-model fusion algorithm to generate a fused state feature vector, wherein the algorithm distributes fusion coefficients according to spatial relevance of each flywheel unit. And correcting the state feature vector deviation by using an error correction model to obtain an SOC estimation result, and storing the SOC estimation result. The method also relates to the steps of model construction, data acquisition and processing, algorithm optimization and the like. The method can improve the SOC estimation precision, is suitable for complex working conditions, guarantees the data integrity and safety, and is of great significance to the development of the flywheel energy storage technology.

Description

Technical Field

[0001] The present invention relates to the technical field of flywheel energy storage, and particularly to a method for estimating the SOC of a flywheel energy storage array system. Background Art

[0002] In today's energy field, with the rapid development of renewable energy, the importance of energy storage technology has become increasingly prominent. As a new type of physical energy storage method, flywheel energy storage has shown broad application prospects in many fields such as power grid peak shaving, uninterruptible power supply, and electric vehicles, due to its advantages of high power density, long life, and fast response speed. However, accurately estimating the SOC (state of charge) of a flywheel energy storage array system remains a key problem to be solved urgently.

[0003] For a flywheel energy storage array system, its operating conditions are complex and variable. Under different charge and discharge rates, environmental temperatures, and mechanical stress conditions, the performance of the flywheel shows significant differences. Traditional SOC estimation methods, such as the ampere-hour integration method, mainly calculate the SOC based on the integration of current over time. However, in practical applications, due to factors such as current measurement errors, changes in charge and discharge efficiency, and self-discharge, the estimation accuracy of this method gradually decreases over time and cannot meet the requirements for real-time and accurate monitoring of the flywheel energy storage array system.

[0004] The open-circuit voltage method estimates the SOC by measuring the open-circuit voltage of the flywheel energy storage system. However, the relationship between the open-circuit voltage and the SOC is not a simple linear relationship, and it is difficult to accurately implement this method in an actual operating flywheel energy storage array system due to interference from various factors such as temperature and battery aging. Moreover, during the operation of the flywheel energy storage array system, there is an interaction between each flywheel unit, and traditional methods often ignore this spatial correlation, resulting in an inability to comprehensively and accurately reflect the SOC state of the entire array system.

[0005] In addition, in a complex industrial environment, the sensors of the flywheel energy storage array system are easily affected by factors such as electromagnetic interference and mechanical vibration, resulting in measurement errors. If these errors cannot be processed and corrected in a timely manner, they will further reduce the accuracy of SOC estimation. At the same time, as the scale of the flywheel energy storage array system continues to expand, the amount of data increases exponentially. How to efficiently process and analyze this data to achieve accurate estimation of the SOC has become another challenge in this field. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for estimating the SOC of a flywheel energy storage array system to solve the problems raised in the above background art.

[0007] To achieve the above object, the present invention provides the following technical solutions: A method for estimating the SOC of a flywheel energy storage array system, the method comprising: Synchronously collecting the operating parameters of the flywheel energy storage array through a multi-source sensing unit, wherein the operating parameters include flywheel speed, bus current, winding temperature, and mechanical vibration intensity; Inputting the operating parameters into a preset dynamic parameter observation model to estimate the dynamic characteristic parameters of the flywheel system in real time, wherein the dynamic parameter observation model dynamically adjusts the observation weights based on the historical operating data of the flywheel unit; Inputting the dynamic characteristic parameters into a preset multi-model fusion algorithm to generate a fused state feature vector, wherein the multi-model fusion algorithm allocates fusion coefficients according to the spatial correlation of each flywheel unit; Using a preset error correction model to correct the deviation of the state feature vector to obtain the SOC estimation result, and storing the corrected result in a state database according to the time series.

[0008] Preferably, the steps of constructing the dynamic parameter observation model include: Obtaining a historical operating data set, wherein each piece of data in the historical operating data set is labeled with the error type and error amplitude of the dynamic characteristic parameters; Dividing the training subsets based on the error type and error amplitude, and each training subset corresponds to a dynamic characteristic scenario; Parallelly training an initial observation model using the training subsets until the parameter estimation error rate of the initial observation model for each scenario is less than or equal to a preset first threshold, and then stopping the training to obtain an intermediate observation model; Inputting the historical operating data set into the intermediate observation model, and verifying whether the parameter estimation result output by the intermediate observation model meets the preset accuracy range; If it meets the requirements, determining the intermediate observation model as the dynamic parameter observation model.

[0009] Preferably, the step of synchronously collecting the operating parameters of the flywheel energy storage array through a multi-source sensing unit includes: Establishing a communication link with a target flywheel unit, wherein the target flywheel unit is deployed at a preset node position of the energy storage array; Continuously reading the real-time operating data stream of the target flywheel unit according to a preset sampling period, and marking the acquisition timestamp based on the timing characteristics of the real-time operating data stream; According to the topological structure of the energy storage array, spatially synchronously aligning the real-time operating data streams of different nodes at the same timestamp to form a spatially correlated operating parameter set.

[0010] Preferably, inputting the operating parameters into a preset dynamic parameter observation model includes: Extracting abnormal fluctuation segments from the operating parameters, where the abnormal fluctuation segments are signal segments with a parameter change rate exceeding a preset dynamic threshold within a continuous time window; Generating a dynamic characteristic evaluation index based on the duration and change slope of the abnormal fluctuation segments; Dynamically selecting a corresponding parameter estimation algorithm according to the dynamic characteristic evaluation index, where the Kalman filtering algorithm is used for short-term high-frequency fluctuations and the least squares fitting algorithm is used for long-term low-frequency fluctuations.

[0011] Preferably, the method further includes: After completing the dynamic characteristic parameter estimation, performing data integrity verification on the state feature vector; If it is found through verification that the data missing rate exceeds a preset second threshold, triggering a preset redundancy compensation model to perform priority compensation on the missing data, where the high-priority missing data is a parameter segment with a continuous missing duration exceeding a preset duration.

[0012] Preferably, the multi-model fusion algorithm includes the following fusion steps: Constructing a distributed state observation model according to the node distribution of the energy storage array, where each node corresponds to the operating state of a flywheel unit; Calculating a fusion weight coefficient based on the dynamic characteristic differences between adjacent nodes; Combining the time series correlation of the operating parameters to perform joint fusion calculation on the multi-node states.

[0013] Preferably, the method further includes: After the fusion calculation is completed, performing consistency verification on the fusion result, where the verification method includes comparing the deviation between the fusion data and the actual data of adjacent nodes; If the deviation exceeds a preset third threshold, readjusting the fusion weight coefficient and iterating the fusion calculation until the deviation is less than the third threshold.

[0014] Preferably, using a preset error correction model to correct the deviation of the state feature vector includes: Dividing correction priority labels according to the error type, where the correction priority labels include transient error types and steady-state error types; Under each type of correction label, dividing correction sub-strategies based on the error amplitude range; Storing the corrected SOC data in an independent storage partition of the state database according to the priority label.

[0015] Preferably, the method further includes: Modify the encryption verification rule of the priority label according to the preset access permission configuration; When receiving a data query request, verify whether the permission identifier provided by the requester matches the encryption rule of the target correction label; If it matches, open the data access channel corresponding to the correction label.

[0016] Preferably, the method further includes optimizing the multi-model fusion algorithm in the following manner: Statistically analyze the fusion error distribution of the SOC estimation results under different working conditions; Determine the weight adjustment amount of the fusion model according to the error distribution, where the fusion weight of adjacent nodes is increased in high-error working conditions, and the fusion weight of historical data is increased in low-error working conditions; Iteratively optimize the multi-model fusion algorithm based on the weight adjustment amount until the fusion error rate is less than a preset fourth threshold.

[0017] Compared with the prior art, the beneficial effects of the present invention are: The SOC estimation method for the flywheel energy storage array system provided by the present invention shows significant advantages in multiple aspects, bringing a positive impact on the development and application of flywheel energy storage technology. In terms of improving the estimation accuracy, the present invention synchronously collects various operating parameters such as flywheel speed, bus current, winding temperature, and mechanical vibration intensity through a multi-source sensing unit, enabling a comprehensive acquisition of system operating state information. Compared with traditional methods that only rely on single or a few parameters for SOC estimation, multi-parameter collection greatly enriches the data dimension. Inputting these parameters into a dynamic parameter observation model that dynamically adjusts the observation weight based on the historical operating data of the flywheel unit can more accurately estimate the dynamic characteristic parameters of the flywheel system. For example, at different ambient temperatures, historical data can help the model automatically adjust the observation weight of the winding temperature parameter, making the estimation of dynamic characteristic parameters more in line with the actual situation. Using the multi-model fusion algorithm, the fusion coefficient is allocated according to the spatial correlation of each flywheel unit to generate a fused state feature vector, fully considering the internal mutual relationship of the system. This avoids the estimation deviation caused by traditional methods ignoring spatial correlation, thereby significantly improving the accuracy of SOC estimation and making the estimation result closer to the true state of charge.

[0018] From the perspective of adapting to complex working conditions, the present invention has strong robustness. When facing complex working conditions such as different charge and discharge rates, environmental temperature changes, and mechanical vibrations, the dynamic parameter observation model can flexibly adjust its own observation weights according to the real-time collected operating parameters and historical data to adapt to the changes in various working conditions. For abnormally fluctuating operating parameters, by extracting abnormal fluctuation segments and generating dynamic characteristic evaluation indicators, a suitable parameter estimation algorithm is dynamically selected (such as the Kalman filtering algorithm for short-term high-frequency fluctuations and the least squares fitting algorithm for long-term low-frequency fluctuations) to ensure accurate estimation of dynamic characteristic parameters under various complex conditions. In a high-temperature environment, when the flywheel speed shows short-term high-frequency fluctuations, the Kalman filtering algorithm can quickly and effectively process noise and interference, accurately estimate dynamic characteristic parameters, and thus ensure the reliability of SOC estimation.

[0019] In terms of data processing and system stability, the present invention has taken a series of effective measures. After completing the estimation of dynamic characteristic parameters, data integrity verification is performed on the state feature vector. When it is found that the data missing rate exceeds the preset threshold, the redundant compensation model is triggered to perform priority compensation on the missing data. This ensures the integrity of the data, avoids estimation errors caused by data missing, and improves the stability and reliability of the system. After the multi-model fusion algorithm completes the fusion calculation, consistency verification is performed on the fusion result. If the deviation exceeds the preset threshold, the fusion weight coefficient is readjusted and the fusion calculation is iterated until the deviation is less than the threshold. This self-optimization and adjustment mechanism can effectively eliminate errors in the fusion process, ensure the accuracy of the fusion result, and further improve the reliability of SOC estimation.

[0020] In addition, the present invention also pays attention to data management and security. The preset error correction model is used to correct the deviation of the state feature vector, and the corrected result is stored in the state database according to the time series. Correction priority labels are divided according to the error type, and correction sub-strategies are divided based on the error amplitude range under different correction labels. The corrected SOC data is stored in an independent storage partition according to the priority label, which is convenient for users to quickly query and analyze different types of error data. By configuring the encryption verification rules for the correction priority label, the data access channel is only opened when the permission identifier of the verification requester matches, effectively ensuring data security and preventing the leakage of sensitive information.

[0021] The multi-model fusion algorithm of the present invention can also statistically analyze the fusion error distribution of SOC estimation results under different working conditions, determine the weight adjustment amount according to the error distribution, and iteratively optimize the algorithm until the fusion error rate is less than the preset threshold. This enables the algorithm to continuously adapt to the changes of the system, continuously improve the estimation accuracy, and provide strong support for the long-term stable operation of the flywheel energy storage array system. Description of the Drawings

[0022] Figure 1 This is the working principle diagram of the SOC estimation method for the flywheel energy storage array system of the present invention; Figure 2 This is the step diagram for collecting the operating parameters of the flywheel energy storage array; Figure 3 This is the step diagram for inputting the operating parameters into the dynamic parameter observation model; Figure 4 This is the step diagram for processing the fusion result. Specific implementation manners

[0023] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0024] Please refer to Figures 1 - 4 , the present invention provides a SOC estimation method for a flywheel energy storage array system, aiming to more accurately and efficiently estimate the SOC (State of Charge) of the flywheel energy storage array system. The following is the detailed implementation process of this method.

[0025] Use a multi-source sensing unit to synchronously collect the operating parameters of the flywheel energy storage array. In actual application scenarios, the flywheel energy storage array may consist of multiple flywheel units distributed at different locations. The multi-source sensing unit is deployed at each key position for real-time monitoring of the operating parameters. Among them, the operating parameters include flywheel speed, bus current, winding temperature, and mechanical vibration intensity. For example, in a large energy storage power station, each flywheel unit is equipped with a high-precision speed sensor to measure the flywheel speed, a current sensor to obtain the bus current, a temperature sensor to monitor the winding temperature, and a vibration sensor to detect the mechanical vibration intensity. These sensors convert the collected analog signals into digital signals for subsequent processing.

[0026] The collected operating parameters are input into a preset dynamic parameter observation model to estimate the dynamic characteristic parameters of the flywheel system in real time. This dynamic parameter observation model is not static, but dynamically adjusts the observation weights based on the historical operating data of the flywheel unit. During the historical operating data accumulation stage, the system continuously records the operating parameters of the flywheel under different working conditions and the true values of the corresponding dynamic characteristic parameters. Through in-depth analysis of these data, the model can automatically adjust the observation weights of each parameter according to the similarity between the current operating state and the historical data, so as to more accurately estimate the dynamic characteristic parameters. For example, when the system detects that the current operating condition is similar to that in a certain section of historical data, the model will increase the weight of the parameters related to this section of historical data, making the estimation result closer to the actual situation.

[0027] The estimated dynamic characteristic parameters are input into a preset multi-model fusion algorithm to generate a fused state feature vector. The multi-model fusion algorithm assigns fusion coefficients according to the spatial correlation of each flywheel unit. In the flywheel energy storage array, there is a certain spatial relationship between flywheel units at different positions. Flywheel units that are closer may have a higher correlation in operating state. This algorithm takes this spatial correlation into account and assigns different fusion coefficients to different flywheel units. For example, for two adjacent flywheel units, since they are close in physical position and are affected by similar environmental factors, the algorithm will assign relatively high fusion coefficients to them, so that their information can be more fully fused when generating the state feature vector.

[0028] A preset error correction model is used to correct the deviation of the state feature vector, thereby obtaining the SOC estimation result. The error correction model will identify and correct the possible deviations in the state feature vector. For example, the deviation of the state feature vector caused by sensor errors or other factors will be adjusted according to the preset rules in the correction model. After obtaining the corrected result, it is stored in the state database in time series. In the state database, the data is stored in chronological order, which is convenient for subsequent query and analysis. Each piece of data contains an accurate time stamp and the corresponding SOC estimation value, which is convenient for users to view the state of the energy storage system at different times at any time.

[0029] The following further elaborates the specific implementation manners of the present invention in detail through 6 embodiments.

[0030] Embodiment 1: When constructing a dynamic parameter observation model, first obtain the historical operation dataset. The source of the historical operation dataset is crucial. It is usually accumulated during the long-term operation of the flywheel energy storage array. These data record the operation of the flywheel under various different working conditions, including operation parameters and corresponding true values of dynamic characteristic parameters under different load conditions, environmental temperatures, operation times, and other factors. For example, in an actual energy storage system that has been operating for one year, the system continuously records the operation parameters such as the flywheel speed, bus current, winding temperature, and mechanical vibration intensity every hour. At the same time, the true values of the corresponding dynamic characteristic parameters are obtained through high-precision measurement equipment. These data together constitute the historical operation dataset.

[0031] After obtaining the historical operation dataset, divide the training subsets based on the error types and error amplitudes of the dynamic characteristic parameters marked in the data. Different error types may represent different operation abnormal situations, such as errors caused by sensor failures, errors caused by mechanical component wear, etc. The error amplitude reflects the severity of the error. For example, divide the error types into several categories such as sensor errors and mechanical errors. Under the category of sensor errors, according to the size of the error amplitude, the data with an error amplitude in the range of 0 - 5% is divided into one training subset, and the data with an error amplitude in the range of 5% - 10% is divided into another training subset, and so on. Each training subset corresponds to a dynamic characteristic scenario.

[0032] Next, use the training subsets to train the initial observation model in parallel. During the training process, a large amount of computing resources will be used to process multiple training subsets simultaneously. Taking a common neural network model as an example, construct an independent neural network structure for each training subset and train these neural networks at the same time. The training goal is to make the parameter estimation error rate of the initial observation model for each scenario less than or equal to a preset first threshold. The preset first threshold is set according to actual application requirements and system accuracy requirements. For example, in some energy storage systems with high precision requirements, the first threshold may be set to 3%. When, after multiple iterative trainings, the parameter estimation error rate of the model for all scenarios corresponding to the training subsets reaches or is lower than this threshold, stop the training to obtain the intermediate observation model.

[0033] After obtaining the intermediate observation model, the historical operation data set is input into the model again, and it is verified whether the parameter estimation results output by the intermediate observation model meet the preset accuracy range. The setting of the preset accuracy range is related to the first threshold, but more comprehensively considers the error requirements in practical applications. For example, the preset accuracy range may require that under different working conditions, the error between the parameter estimation result and the true value should not only be within a certain percentage range, but also meet a certain absolute value limit. If the output result of the intermediate observation model meets this preset accuracy range, then the intermediate observation model is determined as the final dynamic parameter observation model.

[0034] Embodiment 2: When synchronously collecting the operation parameters of the flywheel energy storage array through the multi-source sensing unit, first, a communication link with the target flywheel unit needs to be established. In an actual flywheel energy storage array, the target flywheel unit is deployed at a preset node position in the energy storage array. The selection of the preset node position is usually based on the overall layout and performance optimization of the system. For example, it is selected at a key node on the power transmission path, or at a position with better heat dissipation conditions, etc. Taking a distributed energy storage system as an example, a communication connection with the target flywheel unit is established through a wireless communication module (such as ZigBee, Bluetooth, etc.) or a wired communication line (such as optical fiber, Ethernet cable, etc.). During the process of establishing the communication link, operations such as device address identification and communication protocol negotiation are required to ensure the stability and accuracy of communication.

[0035] After establishing the communication link, the real-time operation data stream of the target flywheel unit is continuously read according to the preset sampling period. The determination of the preset sampling period needs to comprehensively consider multiple factors, including the system's requirement for real-time performance, data processing capabilities, and the response speed of the sensor, etc. For example, in an energy storage system with high requirements for real-time performance, the preset sampling period may be set to 100 milliseconds. When reading the real-time operation data stream, the acquisition timestamp is marked based on the timing characteristics of the real-time operation data stream. The marking of the timestamp can accurately record the acquisition time of each group of data, providing a time reference for subsequent data processing and analysis. For example, a high-precision clock chip is used to add a timestamp accurate to the microsecond level to each piece of collected data.

[0036] Finally, according to the topological structure of the energy storage array, the real-time operation data streams of different nodes at the same timestamp are spatially synchronized and aligned to form a set of spatially correlated operation parameters. There are various forms of the topological structure of the energy storage array, such as series, parallel, star, mesh, etc. Taking the star topological structure as an example, the central node is connected to each edge node (i.e., the node where the target flywheel unit is located). When performing spatial synchronization and alignment, the connection relationship between nodes is determined according to the topological structure, and the real-time operation data streams collected from different nodes at the same timestamp are integrated to ensure that each data point corresponds to the operation parameters at different positions at the same moment, thereby forming a set of spatially correlated operation parameters and providing a comprehensive data basis for subsequent analysis and processing.

[0037] Embodiment 3: When inputting the operation parameters into the preset dynamic parameter observation model, first extract the abnormal fluctuation segments in the operation parameters. The abnormal fluctuation segment refers to the signal segment whose parameter change rate exceeds the preset dynamic threshold within a continuous time window. The setting of the preset dynamic threshold needs to be determined according to the parameter change range of the flywheel energy storage array under normal operating conditions. For example, for the parameter of flywheel speed, during normal operation, its speed change is relatively stable. According to the statistical analysis of historical data, it is determined that a speed change exceeding 100 revolutions per minute within 1 second is an abnormal fluctuation. Then this 100 revolutions per minute is the preset dynamic threshold. In actual operation, the operation parameters are monitored by sliding the time window. Once it is found that the parameter change rate exceeds the preset dynamic threshold, the signal segment within that time period is marked as an abnormal fluctuation segment.

[0038] Generate a dynamic characteristic evaluation index based on the duration and change slope of the abnormal fluctuation segment. The duration reflects the time length of the abnormal fluctuation, and the change slope reflects the severity of the parameter change. For example, for an abnormal fluctuation segment, if its duration is 5 seconds and the change slope is 200 revolutions per minute² per minute, these data will be used to calculate the dynamic characteristic evaluation index. The calculation of the dynamic characteristic evaluation index can adopt the method of weighted summation. For example, the dynamic characteristic evaluation index = 0.6×duration + 0.4×change slope. Here, the weights 0.6 and 0.4 are determined based on actual experience and a large amount of experimental data to balance the influence of the duration and change slope on the evaluation index.

[0039] Dynamically select the corresponding parameter estimation algorithm according to the dynamic characteristic evaluation index. For short-term high-frequency fluctuations, the Kalman filter algorithm is adopted. Short-term high-frequency fluctuations usually mean that the system is affected by sudden and rapidly changing interference factors. The Kalman filter algorithm can use the state equation and observation equation of the system to perform an optimal estimation of the system state through two steps: prediction and update. In the present invention, it is assumed that the state equation of the system is , where denotes the state vector of the system at time k, is the state transition matrix, which is used to describe the state transition relationship of the system from time k - 1 to time k, is the control input matrix, is the control input vector, is the process noise vector, representing the uncertain factors inside the system. The observation equation is , is the observation vector at time k, is the observation matrix, which is used to map the system state to the observation space, is the observation noise vector, reflecting the error in the observation process. The Kalman filter algorithm can accurately estimate the system state in the presence of noise through continuous iterative calculations.

[0040] For long - term low - frequency fluctuations, the least - squares fitting algorithm is adopted. Long - term low - frequency fluctuations generally indicate a relatively slow and long - term change trend in the system. The principle of the least - squares fitting algorithm is to find the best function match for the data by minimizing the sum of the squares of the errors. Suppose we have a set of observed data , , and the function to be fitted is , by minimizing to determine the values of the parameters and , so as to obtain the best fitting curve for estimating the dynamic characteristic parameters of the system.

[0041] Example 4: After completing the estimation of the dynamic characteristic parameters, it is necessary to perform data integrity verification on the state feature vector. The purpose of data integrity verification is to ensure that the data in the state feature vector is not missing or damaged, so as to guarantee the accuracy of subsequent calculations and analyses. In practical applications, due to reasons such as communication failures and sensor failures, some data may be lost. For example, during data transmission, due to signal interference, the flywheel speed data at a certain moment fails to be successfully transmitted to the processing unit, which will result in data missing in the state feature vector.

[0042] The calculation method of the data missing rate is: data missing rate = number of missing data / total number of data × 100%. The preset second threshold is set according to the system's requirement for data integrity. For example, in some application scenarios with high requirements for data integrity, the second threshold may be set to 5%. When the verification finds that the data missing rate exceeds the preset second threshold, the preset redundancy compensation model is triggered to perform priority compensation on the missing data. In the redundancy compensation model, high-priority missing data is defined as a parameter segment with a continuous missing duration exceeding the preset duration. The setting of the preset duration also needs to be determined according to the actual application requirements. For example, in an energy storage system with high requirements for real-time performance, the preset duration may be set to 10 seconds.

[0043] For high-priority missing data, backup sensor data is preferentially used for compensation. If there is no backup sensor data, interpolation calculation is performed based on historical data and data at adjacent moments. For example, for continuously missing flywheel speed data, the speed data at the previous moment and the next moment can be used to estimate it by using the method of linear interpolation. Assume that the speed at the previous moment is , the speed at the next moment is , the missing moment is , the previous moment is , the next moment is , then the estimated value of the speed at the missing moment . For low-priority missing data, the method of statistical averaging can be used for compensation. For example, calculate the average value of data under other similar working conditions in the same time period to fill in the missing value.

[0044] Example 5: The specific fusion steps of the multi-model fusion algorithm are as follows: First, a distributed state observation model is constructed according to the node distribution of the energy storage array. In an actual flywheel energy storage array, each node corresponds to the operating state of a flywheel unit. For example, in an energy storage array composed of 10 flywheel units, it is divided into 10 nodes, and corresponding sensors are installed at each node to monitor the operating parameters of the flywheel unit, such as speed, current, temperature, etc. Based on the operating parameters of each node, an independent state observation model is constructed, and these models can reflect the changes in the operating state of each flywheel unit in real time.

[0045] Next, the fusion weight coefficient is calculated based on the dynamic characteristic differences between adjacent nodes. The dynamic characteristic differences between adjacent nodes can be measured by calculating the similarity of their operating parameters. For example, calculate the Euclidean distance of parameters such as flywheel speed and bus current between two adjacent nodes. The Euclidean distance calculation formula is , where and are the values of the th operating parameter of the two nodes respectively, is the number of operating parameters. The smaller the Euclidean distance, the more similar the dynamic characteristics of the two nodes, and the higher the fusion weight coefficient. Assume that the Euclidean distance between node A and node B is , set a reference distance , the fusion weight coefficient , which can ensure that the smaller the distance, the larger the weight coefficient, and the weight coefficient is between 0 and 1.

[0046] Finally, combining the time series correlation of the operating parameters, joint fusion calculation is performed on the multi-node states. The time series correlation of the operating parameters can be determined by calculating the autocorrelation function. The calculation formula of the autocorrelation function is , where is the time series data, is the data length, is the delay time. According to the value of the autocorrelation function, the correlation strength of the data at different times is determined. During the joint fusion calculation, higher weights are given to the time series data with stronger correlations. For example, when calculating the fusion state feature vector at a certain moment, the weights of the data at the previous and next moments with strong correlations are increased, so that the fusion result can better reflect the true state change trend of the system.

[0047] After the fusion calculation is completed, consistency verification is performed on the fusion result. The verification methods include comparing the deviation between the fusion data and the actual data of adjacent nodes. The actual data of adjacent nodes can be obtained in real time through sensors. For example, calculate the difference between the fused flywheel speed and the actually measured flywheel speed of adjacent nodes. If the difference exceeds the preset third threshold, it is considered that there is a deviation in the fusion result. The preset third threshold is set according to the accuracy requirements of the system. For example, in an energy storage system with high accuracy requirements, the third threshold may be set to 5 revolutions per minute. If the deviation exceeds the preset third threshold, the fusion weight coefficient is readjusted and the fusion calculation is iterated until the deviation is less than the third threshold. When adjusting the fusion weight coefficient, optimization algorithms such as gradient descent can be used to gradually find the optimal combination of weight coefficients to make the fusion result more accurate.

[0048] Embodiment 6: When using the preset error correction model to correct the deviation of the state feature vector, first divide the correction priority labels according to the error type. The error types are divided into transient error types and steady-state error types. Transient errors are usually caused by sudden interference factors, such as momentary voltage fluctuations, external shocks, etc. Such errors are characterized by shortness and suddenness. Steady-state errors are the errors that exist during the long-term stable operation of the system, which may be caused by the inherent deviation of the sensor, the inaccuracy of the model, etc.

[0049] Under each type of correction label, correction sub-strategies are divided based on the error amplitude range. For example, in the transient error category, when the error amplitude is within the range of 0 - 10%, a simple filtering algorithm is used for correction; when the error amplitude is within the range of 10% - 20%, a more complex model prediction correction is performed by combining historical data and the current operating state. In the steady-state error category, for small error amplitudes, correction can be achieved by periodically calibrating the sensor; for large error amplitudes, model parameters need to be readjusted.

[0050] The corrected SOC data is stored in an independent storage partition of the state database according to the priority label. In the state database, independent storage areas are opened for the data corrected for transient errors and the data corrected for steady-state errors respectively. The advantage of this is that it is convenient for users to quickly query and analyze data according to different needs. For example, when a user wants to understand the SOC change of the system under sudden interference, data can be directly obtained from the storage partition of the transient error correction data; when the user is concerned about the long-term operation stability of the system, data can be queried from the storage partition of the steady-state error correction data.

[0051] In addition, encryption verification rules for correction priority labels are configured according to the preset access permissions. In an energy storage system management platform used by multiple users, different users may have different access permissions. For example, the system administrator has the highest authority and can access data of all correction labels; ordinary operators may only be able to access the steady-state error correction data. When a data query request is received, it is verified whether the permission identifier provided by the requester matches the encryption rule of the target correction label. The permission identifier can be the user's account password, digital certificate, etc. If it matches, the data access channel for the corresponding correction label is opened to ensure the security and confidentiality of the data.

[0052] At the same time, the multi-model fusion algorithm is optimized in the following way: statistically analyze the fusion error distribution of the SOC estimation results under different working conditions. Different working conditions include different load conditions, ambient temperatures, operating times, etc. For example, under high-load working conditions, the error between the SOC estimation result and the actual value is statistically analyzed; under low-temperature ambient working conditions, the corresponding error is also statistically analyzed. Through a large amount of experimental and actual operation data accumulation, the fusion error distribution under different working conditions is obtained.

[0053] Determine the weight adjustment amount of the fusion model according to the error distribution. For high-error working conditions, increase the fusion weight of adjacent nodes. This is because in high-error working conditions, it may mean that the measurement data of a single node is more affected by interference. At this time, increasing the fusion weight of adjacent nodes can utilize the relatively reliable data of adjacent nodes to improve the estimation accuracy. Assume that under high-error working conditions, the original fusion weight of adjacent nodes is , and the weight adjustment ratio is set as ( ), the adjusted adjacent node fusion weight is .

[0054] For low-error working conditions, increase the fusion weight of historical data. Low-error working conditions indicate that the current system is running relatively stably, and historical data can reflect the normal operation law of the system. Increasing its fusion weight helps to improve the estimation accuracy by using the accumulated experience. Let the original fusion weight of historical data be , and the weight adjustment ratio is ( ), and the adjusted fusion weight of historical data is .

[0055] Based on the above weight adjustment amount, the multi-model fusion algorithm is iteratively optimized until the fusion error rate is less than the preset fourth threshold. In each iteration process, the multi-model fusion calculation is re-performed according to the adjusted weights, and the fusion error rate is statistically calculated again. If the fusion error rate is still greater than the preset fourth threshold, continue to adjust the weights according to the above rules and perform the next iteration; if the fusion error rate is less than the fourth threshold, it is considered that the optimization is completed. The preset fourth threshold is determined according to the requirements of the actual application for the estimation accuracy. For example, in an energy storage system with extremely high accuracy requirements, the fourth threshold may be set to 2%.

[0056] In actual operation, first collect a large amount of operation data under different working conditions, including operation parameters such as flywheel speed, bus current, winding temperature, mechanical vibration intensity, etc. and the corresponding actual SOC values. Use these data to calculate the initial fusion error distribution and determine the initial weight adjustment amount. Then, continuously repeat the processes of weight adjustment, fusion calculation, and error rate statistics to gradually optimize the multi-model fusion algorithm, making the SOC estimation result more and more accurate. Through such an optimization process, it can adapt to different operating conditions, improve the overall performance of the SOC estimation of the flywheel energy storage array system, and provide strong support for the efficient management and reliable operation of the energy storage system.

[0057] It should be noted that in this article, relational terms such as first and second are only used 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 term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0058] Although embodiments of the present invention have been shown and described, those of ordinary skill in the art will appreciate that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A SOC estimation method for a flywheel energy storage array system, characterized in that: include: The operating parameters of the flywheel energy storage array are synchronously collected through a multi-source sensing unit, wherein the operating parameters include flywheel speed, bus current, winding temperature and mechanical vibration intensity; Inputting the operating parameters into a preset dynamic parameter observation model to estimate the dynamic characteristic parameters of the flywheel system in real time, wherein the dynamic parameter observation model dynamically adjusts the observation weight based on the historical operating data of the flywheel unit; Inputting the dynamic characteristic parameters into a preset multi-model fusion algorithm to generate a fused state feature vector, wherein the multi-model fusion algorithm allocates fusion coefficients according to the spatial correlation of each flywheel unit; The state characteristic vector is corrected for deviation using a preset error correction model to obtain an SOC estimation result, and the corrected result is stored in a state database in a time series.

2. The SOC estimation method for a flywheel energy storage array system according to claim 1, characterized in that: The steps of constructing the dynamic parameter observation model include: Acquire a historical operation data set, wherein each data in the historical operation data set is marked with an error type and an error amplitude of a dynamic characteristic parameter; Divide the training subsets based on the error type and error amplitude, each training subset corresponds to a dynamic characteristic scenario; The initial observation model is trained in parallel using the training subset, and the training is stopped when the parameter estimation error rate of the initial observation model for each scene is less than or equal to a preset first threshold, thereby obtaining an intermediate observation model; Inputting the historical operation data set into the intermediate observation model, and verifying whether the parameter estimation result output by the intermediate observation model meets the preset accuracy range; If it is consistent, the intermediate observation model is determined as the dynamic parameter observation model.

3. The SOC estimation method for a flywheel energy storage array system according to claim 1, characterized in that: The synchronous acquisition of the operating parameters of the flywheel energy storage array by the multi-source sensing unit includes: Establishing a communication link with a target flywheel unit, wherein the target flywheel unit is deployed at a preset node position of the energy storage array; Continuously reading the real-time operation data stream of the target flywheel unit according to a preset sampling period, and marking a collection timestamp based on a timing feature of the real-time operation data stream; According to the topological structure of the energy storage array, the real-time operation data streams of different nodes at the same timestamp are spatially synchronized to form a set of spatially associated operation parameters.

4. The SOC estimation method for a flywheel energy storage array system according to claim 1, characterized in that: Inputting the operating parameters into a preset dynamic parameter observation model comprises: Extracting abnormal fluctuation segments from the operating parameters, wherein the abnormal fluctuation segments are signal segments in which the parameter change rate exceeds a preset dynamic threshold within a continuous time window; Generating a dynamic characteristic evaluation index based on the duration and change slope of the abnormal fluctuation segment; The corresponding parameter estimation algorithm is dynamically selected according to the dynamic characteristic evaluation index, wherein the Kalman filtering algorithm is used for short-term high-frequency fluctuations, and the least squares fitting algorithm is used for long-term low-frequency fluctuations.

5. The SOC estimation method for a flywheel energy storage array system according to claim 4, characterized in that: The method further comprises: After the dynamic characteristic parameter estimation is completed, performing data integrity check on the state characteristic vector; If the check finds that the data missing rate exceeds the preset second threshold, the preset redundancy compensation model is triggered to perform priority compensation on the missing data, wherein the high-priority missing data is a parameter segment whose continuous missing duration exceeds the preset duration.

6. The SOC estimation method for a flywheel energy storage array system according to claim 1, characterized in that: The multi-model fusion algorithm includes the following fusion steps: According to the node distribution of the energy storage array, a distributed state observation model is constructed, wherein each node corresponds to the operating state of a flywheel unit; Calculate the fusion weight coefficient based on the dynamic characteristics difference of adjacent nodes; Combined with the time series correlation of the operating parameters, a joint fusion calculation is performed on the multi-node states.

7. The SOC estimation method for a flywheel energy storage array system according to claim 6, characterized in that: The method further comprises: After the fusion calculation is completed, the fusion result is verified for consistency, wherein the verification method includes comparing the deviation between the fusion data and the actual data of the adjacent nodes; If the deviation exceeds a preset third threshold, the fusion weight coefficient is readjusted and the fusion calculation is iterated until the deviation is less than the third threshold.

8. The SOC estimation method for a flywheel energy storage array system according to claim 1, characterized in that: Using a preset error correction model to correct the deviation of the state characteristic vector includes: Classifying correction priority labels according to error types, wherein the correction priority labels include transient error categories and steady-state error categories; Under each type of correction label, the correction sub-strategies are divided based on the error amplitude range; The corrected SOC data is stored in an independent storage partition of the state database according to the priority tags.

9. The SOC estimation method for a flywheel energy storage array system according to claim 8, characterized in that: The method further comprises: Modify the encryption verification rules of the priority tag according to the preset access rights configuration; Upon receiving a data query request, verify whether the permission identifier provided by the requester matches the encryption rule of the target revision tag; If there is a match, the data access channel corresponding to the correction tag is opened.

10. The SOC estimation method for a flywheel energy storage array system according to claim 1, characterized in that: The method further includes optimizing the multi-model fusion algorithm in the following manner: Counting the fusion error distribution of the SOC estimation result under different working conditions; Determining a weight adjustment amount of the fusion model according to the error distribution, wherein a high error condition increases the fusion weight of adjacent nodes, and a low error condition increases the fusion weight of historical data; The multi-model fusion algorithm is iteratively optimized based on the weight adjustment amount until the fusion error rate is less than a preset fourth threshold.

Images