Compressed air energy storage system power grid intelligent guarding method based on big data

Abstract

The invention relates to the technical field of data processing, in particular to a compressed air energy storage system power grid intelligent guarding method based on big data, and the method comprises the steps: collecting operation data in a power grid of a compressed air energy storage system, and carrying out the cleaning, denoising and feature extraction of the operation data, so as to obtain operation features; when a risk occurs in the prediction result, switching the compressed air energy storage system from a shutdown or conventional standby state to a power grid guarding hot standby mode; determining whether the moving average window length needs to be increased; acquiring the transmission delay duration of the operation data to determine whether the suitability of the prediction model update meets the requirement or not; determining whether the dynamic threshold of the minimum training data volume needs to be reduced; and if the dynamic threshold of the minimum training data volume does not need to be reduced, determining the offset of the update triggering time based on the effective prediction duration ratio of the prediction model in unit time. The power grid protection stability of the compressed air energy storage system is improved.

Description

Technical Field

[0001] This invention relates to the field of compressed air energy storage technology, and in particular to a method for intelligent grid protection of compressed air energy storage systems based on big data. Background Technology

[0002] Existing CAES systems operate in a single mode, typically relying solely on electricity price signals for peak shaving and valley filling. Their powerful synchronous generators and rapid start-up capabilities are not utilized to support grid stability. A high proportion of renewable energy in the grid leads to frequent voltage fluctuations, necessitating flexible reactive power compensation methods. Simultaneously, while power fluctuations are highly predictable, traditional generating units lack sufficient response. The thermodynamic processes of CAES themselves exhibit inertia, preventing millisecond-level responses like those of batteries. However, grid power changes (such as cloud movement and evening load ramp-up) are often predictable. Current technologies fail to combine prediction with preparedness to overcome the response delay of CAES.

[0003] Chinese Patent Publication No. CN119965833A discloses a method for optimizing the response time of a compressed air energy storage power station based on time series prediction. The method includes the following steps: S1. Collect local historical environmental and power generation and consumption data, and predict the power generation and consumption data for the current day; S2. Use the GluonTS algorithm to establish a local total power generation model and a local power consumption model based on the data collected in step S1; S3. Use the data from the local total power generation model and the local power consumption model in step S2, along with historical data on the local power grid's dispatch of the local energy storage power plant, to establish a time series prediction model for the dispatch command of a certain power plant; S4. Use the time series prediction model for the dispatch command of a certain power plant in step S3 to predict the probability that the energy storage power plant will be dispatched by the power grid in the next hour; S5. If the probability obtained by the time series prediction model for the dispatch command of a certain power plant exceeds a set threshold, that is, it is considered that it may be dispatched by the power grid in the next hour, then the energy storage and power generation systems are prepared so that they can respond immediately when the dispatch command is received. Therefore, the time-series prediction-based compressed air energy storage power station response time optimization method has the problem that it uses whether a fixed threshold is exceeded as the basis for start-up preparation, without considering the real-time fluctuations such as the transmission delay of grid operation data and data noise. When the grid experiences sudden disturbances, the fixed threshold cannot be dynamically adjusted, which not only causes unnecessary preparations and increases energy consumption, but also poses a risk of untimely response, resulting in insufficient grid protection stability of the compressed air energy storage system. Summary of the Invention

[0004] To address this, the present invention provides a big data-based intelligent grid protection method for compressed air energy storage systems. This method overcomes the problem in existing technologies where a fixed threshold is used as the basis for startup preparation, without considering real-time fluctuations such as transmission delays and data noise in grid operation data. When sudden disturbances occur in the grid, the fixed threshold cannot be dynamically adjusted, which not only causes unnecessary preparations and increased energy consumption but also poses a risk of untimely response, resulting in insufficient grid protection stability of the compressed air energy storage system.

[0005] To achieve the above objectives, this invention provides a method for intelligent grid protection of compressed air energy storage systems based on big data, comprising: The operation data of the compressed air energy storage system in the power grid is collected, and the operation data is cleaned, denoised and feature extracted to obtain operation features. The initial model is trained according to the operation features to obtain a prediction model, and the prediction model is used to predict the operation status of the power grid to obtain the prediction result. When the predicted result indicates a risk, the compressed air energy storage system is switched from shutdown or normal standby state to grid protection hot standby mode. When the risk actually occurs, a loading command is issued to the compressed air energy storage system in hot standby state. The efficiency of obtaining operating data within a single cycle is used to determine whether the grid protection stability of the compressed air energy storage system meets the requirements based on the efficiency of operating data within the single cycle. If the power grid protection stability does not meet the requirements, determine whether it is necessary to increase the moving average window length; If it is not necessary to increase the moving average window length, then obtain the transmission delay of the running data to determine whether the adaptability of the prediction model update meets the requirements; If the adaptability of the prediction model update does not meet the requirements, then determine whether it is necessary to reduce the dynamic threshold of the minimum training data volume. If it is not necessary to reduce the dynamic threshold of the minimum training data volume, the offset of the update trigger time is determined based on the proportion of the effective prediction time of the prediction model per unit time.

[0006] Furthermore, determining whether the grid protection stability of the compressed air energy storage system meets the requirements based on the efficiency of the operating data within the single cycle includes: The efficiency of the data running in a single cycle is compared with the preset second efficiency. If the efficiency of the operating data within a single cycle is greater than or equal to the preset second efficiency, then the grid protection stability of the compressed air energy storage system is determined to meet the requirements. If the efficiency of the operating data within a single cycle is less than the preset second efficiency, then it is determined that the grid protection stability of the compressed air energy storage system does not meet the requirements.

[0007] Further, determine whether the moving average window length needs to be increased, including: The efficiency of the running data within the single cycle is compared with the preset first efficiency and the preset second efficiency, respectively. If the efficiency of the running data within a single period is less than or equal to the preset first efficiency, then it is determined that the moving average window length needs to be increased. If the efficiency of the running data within a single period is greater than a preset first efficiency and less than a preset second efficiency, then it is determined that there is no need to increase the moving average window length.

[0008] Furthermore, the increase in the length of the moving average window is determined by the difference between a preset first efficiency and the efficiency of the running data within a single period.

[0009] Furthermore, the suitability of the prediction model update is determined based on the transmission delay of the running data, including: The transmission delay of the running data is compared with the preset first delay. If the transmission delay of the running data is less than or equal to the preset first delay, then it is determined that the adaptability of the prediction model update meets the requirements, and it is determined whether the moving average window length meets the requirements. If the transmission delay of the running data is greater than the preset first delay duration, then it is determined that the adaptability of the prediction model update does not meet the requirements.

[0010] Further, determine whether the dynamic threshold for reducing the minimum training data size needs to be decreased, including: The transmission delay of the running data is compared with the preset first delay and the preset second delay, respectively; If the transmission delay of the running data is greater than the preset first delay and less than or equal to the preset second delay, then a dynamic threshold for reducing the minimum amount of training data is determined. If the transmission delay of the running data is greater than the preset second delay, then it is determined that there is no need to increase the dynamic threshold of the minimum training data amount.

[0011] Furthermore, the reduction in the dynamic threshold of the minimum training data volume is determined by the difference between the transmission delay of the running data and the preset first delay.

[0012] Furthermore, the offset of the update trigger time is determined based on the proportion of effective prediction time of the prediction model per unit time, including: Compare the percentage of effective prediction time of the prediction model per unit time with the preset percentage; If the effective prediction time ratio of the prediction model within the unit time is greater than or equal to the preset ratio, then it is determined that the scenario adaptability of the prediction model update meets the requirements, and there is no need to increase the offset of the update trigger time. It is also determined whether the dynamic threshold of the minimum training data volume meets the requirements. If the effective prediction time percentage of the prediction model within a unit of time is less than the preset percentage, it is determined that the scenario adaptability of the prediction model update does not meet the requirements, and the offset of the update trigger time needs to be increased.

[0013] Furthermore, the effective prediction time percentage of the prediction model per unit time is the ratio of the time during which the prediction model normally outputs prediction results to the total prediction time per unit time.

[0014] Furthermore, the increase in the offset of the update trigger time is determined by the difference between the preset proportion and the proportion of the effective prediction time of the prediction model per unit time.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: The method of this invention adjusts the moving average window length based on the effectiveness of the running data within a single cycle. Due to prolonged use, some sensors age, leading to insufficient sensor accuracy and resulting in defective collected data. Increasing the moving average window length can smooth out the jumps and noise data from aging sensors, reduce repeatability errors, and minimize the interference of distorted data on the model. Furthermore, the method adjusts the dynamic threshold for the minimum training data volume based on the transmission delay of the running data. Because real-time weather, load, and power grid parameter data cannot quickly form a usable training dataset due to transmission delays and incomplete cleaning, the model cannot update the data in real time. By reducing the dynamic threshold for the minimum amount of training data, the time required to build the training dataset can be shortened without waiting for sufficient data accumulation, thus adapting to the real-time requirements of grid protection. The offset of the update trigger time is adjusted according to the proportion of effective prediction time of the prediction model within a unit of time. Since model updates require a certain amount of time, and the prediction window for grid protection is only 5 minutes to 1 hour, the model may not be able to output prediction results during the update period, affecting scheduling decisions and thus delaying updates. By increasing the offset of the update trigger time, the core decision-making period within the prediction window can be avoided, preventing updates from occupying critical prediction time, ensuring that scheduling decisions continuously obtain effective prediction results, and improving the grid protection stability of the compressed air energy storage system.

[0016] Furthermore, the method of the present invention adjusts the length of the moving average window by setting a preset first efficiency and a preset second efficiency. Due to the aging of some sensors caused by long-term use, the accuracy of the sensors is insufficient, resulting in defects in the collected data. By increasing the length of the moving average window, the jumps and noise data of the aging sensors can be smoothed, the repeatability error can be reduced, and the interference of distorted data on the model can be reduced, thereby further improving the grid protection stability of the compressed air energy storage system.

[0017] Furthermore, the method of the present invention adjusts the dynamic threshold of the minimum training data volume by setting a preset first delay duration and a preset second delay duration. Since real-time weather, load, grid parameters and other data cannot quickly form a usable training dataset due to transmission delays and incomplete cleaning, the model cannot update the data in real time. By reducing the dynamic threshold of the minimum training data volume, it is not necessary to wait for sufficient data accumulation, shortening the training dataset construction time, adapting to the real-time requirements of grid protection, and further improving the grid protection stability of the compressed air energy storage system.

[0018] Furthermore, the method described in this invention adjusts the offset of the update trigger time by setting a preset percentage. Since model updates require a certain amount of time, and the prediction window for grid protection is only 5 minutes to 1 hour, the model may be unable to output prediction results during the update period, affecting scheduling decisions and thus delaying the update. By increasing the offset of the update trigger time, the core decision-making period within the prediction window can be avoided, preventing updates from occupying critical prediction time, ensuring that scheduling decisions continuously obtain effective prediction results, and further improving the grid protection stability of the compressed air energy storage system. Attached Figure Description

[0019] Figure 1 This is an overall flowchart of the smart grid protection method for compressed air energy storage systems based on big data, according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the logic of determining whether to increase the moving average window length in the big data-based smart grid protection method for compressed air energy storage systems according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating the logic of determining whether the dynamic threshold for reducing the minimum training data volume is needed in the big data-based smart grid protection method for compressed air energy storage systems according to an embodiment of the present invention. Figure 4 This is a flowchart illustrating the logic of determining the offset of the update trigger time in the big data-based smart grid protection method for compressed air energy storage systems according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0021] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0022] Please see Figure 1 As shown, it is an overall flowchart of the smart grid protection method for compressed air energy storage system based on big data according to an embodiment of the present invention.

[0023] This invention provides a method for intelligent grid protection of compressed air energy storage systems based on big data, comprising: Step S1: Collect the operating data of the compressed air energy storage system in the power grid, and clean, denoise and extract features from the operating data to obtain operating features. Train the initial model according to the operating features to obtain a prediction model, and use the prediction model to predict the operating state of the power grid to obtain the prediction result. Step S2: When the predicted result shows a risk, the compressed air energy storage system is switched from shutdown or normal standby state to grid protection hot standby mode. When the risk actually occurs, a loading command is sent to the compressed air energy storage system in hot standby state. Step S3: Obtain the efficiency of the operating data within a single cycle, and determine whether the grid protection stability of the compressed air energy storage system meets the requirements based on the efficiency of the operating data within the single cycle. Step S4: If the power grid protection stability does not meet the requirements, determine whether it is necessary to increase the moving average window length. Step S5: If it is not necessary to increase the moving average window length, obtain the transmission delay of the running data to determine whether the adaptability of the prediction model update meets the requirements. Step S6: If the adaptability of the prediction model update does not meet the requirements, determine whether it is necessary to reduce the dynamic threshold of the minimum training data volume. Step S7: If it is not necessary to reduce the dynamic threshold of the minimum training data volume, then determine the offset of the update trigger time based on the proportion of the effective prediction time of the prediction model per unit time.

[0024] Specifically, operational data includes meteorological data, load forecasting data, power grid dispatching plans, and real-time electricity prices.

[0025] Specifically, operational characteristics include average wind speed over several periods, total electricity price over a single statistical period, and priority ranking of grid dispatch plans.

[0026] Specifically, the initial model is a prototype model architecture with basic prediction / classification capabilities that has not been trained with compressed air energy storage system and power grid operation data.

[0027] Specifically, the prediction model can be a long short-term memory network, a temporal attention mechanism model, or a data-mechanism fusion model, with the preferred embodiment being a long short-term memory network.

[0028] Specifically, the process of training an initial model to obtain a prediction model based on operational features involves dividing the operational features into a training set, a validation set, and a test set. The initial model is trained using the training set, and then validated and optimized using the validation set and the test set to finally obtain the prediction model.

[0029] Specifically, the operating status includes the power deficit of the power grid, the surplus trend of the power grid, weak voltage nodes, and risky periods.

[0030] Specifically, the forecast results include the grid frequency for the next 5 to 60 minutes, the maximum frequency regulation / peak shaving capacity that the compressed air energy storage system can provide, and mode switching recommendations.

[0031] Specifically, the grid protection hot standby mode is a high-readiness standby state set up by the compressed air energy storage system to cope with potential risks to the grid, such as voltage instability, frequency fluctuations, and load gaps. It is between the normal standby state and the charging and discharging operation state. The core objective is to shorten the response delay and ensure that it can quickly switch into the grid regulation process.

[0032] Specifically, the efficiency rate of data within a single cycle is the ratio of the amount of valid data within a single cycle to the total amount of data that should theoretically be collected.

[0033] Specifically, the moving average window length is the time span corresponding to a fixed number of data points contained in a single moving window during the processing of running data.

[0034] Specifically, the transmission delay of the running data is the total time interval from when the running data is sent from the acquisition end to when the data processing end successfully receives it and completes the preliminary verification.

[0035] Specifically, the dynamic threshold for minimum training data volume refers to the minimum critical value of data volume that is flexibly adjusted in combination with dynamic factors such as system operating status and data characteristics to ensure the training accuracy and reliability of the prediction model for intelligent protection of the compressed air energy storage system and the power grid.

[0036] Specifically, the offset of the update trigger time is the time difference between the actual trigger time and the preset reference trigger time when setting the periodic update trigger rules for the compressed air energy storage system and the grid-related models, data or instructions.

[0037] In implementation, the method of this invention adjusts the moving average window length based on the effectiveness of the running data within a single cycle. Due to prolonged use, some sensors age, leading to insufficient sensor accuracy and resulting in defective collected data. Increasing the moving average window length smooths out the jumps and noise in aged sensors, reduces repeatability errors, and minimizes the interference of distorted data on the model. The dynamic threshold for the minimum training data size is adjusted based on the transmission delay of the running data. Because real-time weather, load, and power grid parameter data cannot quickly form a usable training dataset due to transmission delays and incomplete data cleaning, the model cannot update data in real time. By reducing the minimum training data size... The dynamic threshold for training data volume eliminates the need to wait for sufficient data accumulation, shortening the training dataset construction time and adapting to the real-time requirements of power grid protection. The offset of the update trigger time is adjusted based on the proportion of effective prediction time of the prediction model within a unit of time. Since model updates require a certain amount of time, and the prediction window for power grid protection is only 5 minutes to 1 hour, the model may be unable to output prediction results during the update period, affecting scheduling decisions and thus delaying updates. By increasing the offset of the update trigger time, the core decision-making period within the prediction window can be avoided, preventing updates from occupying critical prediction time and ensuring that scheduling decisions continuously obtain effective prediction results, thereby improving the stability of the compressed air energy storage system for power grid protection.

[0038] Please continue reading. Figure 2 As shown, it is a flowchart illustrating the logic of determining whether to increase the moving average window length in the big data-based smart grid protection method for compressed air energy storage systems according to an embodiment of the present invention.

[0039] Specifically, determining whether the grid protection stability of the compressed air energy storage system meets the requirements based on the efficiency of the operating data within the single cycle includes: The efficiency of the data running in a single cycle is compared with the preset second efficiency. If the efficiency of the operating data within a single cycle is greater than or equal to the preset second efficiency, then the grid protection stability of the compressed air energy storage system is determined to meet the requirements. If the efficiency of the operating data within a single cycle is less than the preset second efficiency, then it is determined that the grid protection stability of the compressed air energy storage system does not meet the requirements.

[0040] The reasons why the grid protection stability of the compressed air energy storage system fails to meet requirements may include inadequate adaptability of the prediction model update or an inappropriate moving average window length. The next step is to determine the specific cause, which is essentially the process of deciding whether to increase the moving average window length.

[0041] Specifically, determining whether the moving average window length needs to be increased includes: The efficiency of the running data within the single cycle is compared with the preset first efficiency and the preset second efficiency, respectively. If the efficiency of the running data within a single period is less than or equal to the preset first efficiency, then it is determined that the moving average window length needs to be increased. If the efficiency of the running data within a single period is greater than a preset first efficiency and less than a preset second efficiency, then it is determined that there is no need to increase the moving average window length.

[0042] Specifically, if the efficiency of the operating data within a single cycle is less than or equal to the preset first efficiency, it is determined that the reason for the non-compliance of the grid protection stability of the compressed air energy storage system is that the moving average window length is not up to standard, and therefore the moving average window length needs to be increased. If the efficiency of the operating data within a single cycle is greater than the preset first efficiency but less than the preset second efficiency, it can be preliminarily determined that the adaptability of the prediction model update is not up to standard. Next, it is necessary to make a final determination on whether the adaptability of the prediction model update meets the requirements based on the transmission delay time of the operating data, that is, to determine whether the reason for the non-compliance of the grid protection stability of the compressed air energy storage system is the non-compliance of the prediction model update.

[0043] It is understandable that the preset first efficiency is less than the preset second efficiency, and the three intervals divided by the preset first efficiency and the preset second efficiency correspond to three different situations: The first interval is when the efficiency of the running data in a single period is less than or equal to the preset first efficiency. The corresponding situation is: due to long-term use, some sensors have aged, resulting in insufficient sensor accuracy, which leads to defects in the collected data. In this case, it is necessary to adjust the moving average window length. The second interval is where the efficiency of the running data within a single cycle is greater than the first preset efficiency and less than the second preset efficiency. The corresponding situation is that due to transmission delays and incomplete cleaning of real-time weather, load, power grid parameters and other data, a usable training dataset cannot be quickly formed, which causes the model to be unable to update the data in real time. At this time, it is necessary to further judge whether the adaptability of the prediction model update meets the requirements. The third interval is when the efficiency of the operating data in a single cycle is greater than or equal to the preset second efficiency. The corresponding situation is: the grid protection stability of the compressed air energy storage system meets the requirements, and no adjustment is required.

[0044] Understandably, the preset first efficiency and preset second efficiency can be set according to actual operating conditions. The setting of the preset first efficiency and preset second efficiency aims to ensure the stability and practicality of grid protection for the compressed air energy storage system. Optionally, the preset first efficiency and preset second efficiency are determined through a limited number of tests by evaluating the grid protection effect of different operating data efficiencies on the compressed air energy storage system. The determined preset first efficiency and preset second efficiency should be neither too low nor cause excessive interference to the grid protection process of the compressed air energy storage system. For example, the preset first efficiency is generally selected within the range of [83%, 87%], and the preset second efficiency is generally selected within the range of [88%, 92%].

[0045] Preferably, the first efficiency is 85% in the preferred embodiment, and the second efficiency is 90% in the preferred embodiment.

[0046] Specifically, the increase in the length of the moving average window is determined by the difference between a preset first efficiency and the efficiency of the running data within a single period.

[0047] Specifically, when the difference between the preset first efficiency and the efficiency of the running data within a single period is within 5%, the moving average window length is increased to 1.1 times the original value. When the difference between the preset first efficiency and the efficiency of the running data within a single period exceeds 5%, the moving average window length is increased by 2 minutes for every 1% increase beyond the original value, in addition to increasing to 1.1 times the original value. For example, if the difference between the preset first efficiency and the efficiency of the running data within a single period is 7%, and the current moving average window length is 20 minutes, the increased moving average window length will be 20 × 1.1 + 2 × 2 = 26 minutes.

[0048] In practice, the method of the present invention adjusts the length of the moving average window by setting a preset first efficiency and a preset second efficiency. Due to long-term use, some sensors age, resulting in insufficient sensor accuracy and thus defects in the collected data. By increasing the length of the moving average window, the jumps and noise data of the aging sensors can be smoothed, the repeatability error can be reduced, the interference of distorted data on the model can be reduced, and the grid protection stability of the compressed air energy storage system can be further improved.

[0049] Please continue reading. Figure 3 As shown, it is a flowchart illustrating the logic of determining whether the dynamic threshold for reducing the minimum training data volume is needed in the big data-based intelligent grid protection method for compressed air energy storage systems according to an embodiment of the present invention.

[0050] Specifically, the suitability of the prediction model update is determined based on the transmission delay of the running data, including: The transmission delay of the running data is compared with the preset first delay. If the transmission delay of the running data is less than or equal to the preset first delay, then it is determined that the adaptability of the prediction model update meets the requirements, and it is determined whether the moving average window length meets the requirements. If the transmission delay of the running data is greater than the preset first delay duration, then it is determined that the adaptability of the prediction model update does not meet the requirements.

[0051] Specifically, if the transmission delay of the operating data is less than or equal to the preset first delay duration, it is determined that the adaptability of the prediction model update meets the requirements. However, if it has been previously determined that the grid protection stability of the compressed air energy storage system does not meet the requirements, then it is necessary to further determine whether the moving average window length meets the requirements.

[0052] In practice, the moving average window length is determined to meet the requirements based on the comparison between the actual moving average window length and the predetermined window length threshold. If the actual moving average window length is less than the predetermined window length threshold, the moving average window length is determined to not meet the requirements. The predetermined window length threshold is the average value of the moving average window length monitored in the previous three months of the historical period.

[0053] If the moving average window length does not meet the requirements, the moving average window length is increased; if the moving average window length meets the requirements, the efficiency of the operating data within a single cycle is re-collected, and the grid protection stability of the compressed air energy storage system is re-evaluated to determine whether it meets the requirements.

[0054] When the transmission delay of the operating data exceeds a preset first delay duration, it can be determined that the reason for the unsatisfactory grid protection stability of the compressed air energy storage system is that the adaptability of the prediction model update is not up to standard. The reasons for this could be that the dynamic threshold for the minimum training data size is not met, or that the scenario adaptability of the prediction model update is not up to standard. The next step is to determine which specific cause it is, which is essentially the process of deciding whether to reduce the dynamic threshold for the minimum training data size.

[0055] Specifically, determining whether the dynamic threshold for reducing the minimum training data size needs to be adjusted includes: The transmission delay of the running data is compared with the preset first delay and the preset second delay, respectively; If the transmission delay of the running data is greater than the preset first delay and less than or equal to the preset second delay, then a dynamic threshold for reducing the minimum amount of training data is determined. If the transmission delay of the running data is greater than the preset second delay, then it is determined that there is no need to increase the dynamic threshold of the minimum training data amount.

[0056] Specifically, when the transmission delay of the running data is greater than a preset first delay but less than or equal to a preset second delay, it is determined that the reason for the unsuitability of the prediction model update is that the dynamic threshold for the minimum training data size does not meet the requirements. Therefore, the dynamic threshold for the minimum training data size needs to be reduced. When the transmission delay of the running data is greater than the preset second delay, it can be preliminarily determined that the scenario suitability of the prediction model update does not meet the requirements. Next, it is necessary to make a final determination of the scenario suitability of the prediction model update based on the proportion of effective prediction time of the prediction model per unit time, that is, to determine whether the reason for the unsuitability of the prediction model update is that the scenario suitability of the prediction model update does not meet the requirements.

[0057] It is understandable that the preset first delay duration is shorter than the preset second delay duration, and the three intervals divided by the preset first delay duration and the preset second delay duration correspond to three different scenarios: The first interval is when the transmission delay of the running data is less than or equal to the preset first delay duration. The corresponding situation is: the adaptability of the prediction model update is determined to meet the requirements. At this time, it is necessary to further determine whether the moving average window length meets the requirements. The second interval is when the transmission delay of the running data is greater than the preset first delay and less than or equal to the preset second delay. The corresponding situation is: due to transmission delay and incomplete cleaning of real-time weather, load, power grid parameters and other data, a usable training dataset cannot be quickly formed, which causes the model to be unable to update the data in real time. In this case, it is necessary to adjust the dynamic threshold of the minimum training data volume. The third interval is when the transmission delay of the running data is longer than the preset second delay duration. The corresponding situation is: since the model update takes a certain amount of time, and the prediction window of the power grid protection is only 5 minutes to 1 hour, the model may not be able to output prediction results during the update period, affecting the scheduling decision and thus delaying the update. At this time, it is necessary to further determine whether the scenario adaptability of the prediction model update meets the requirements.

[0058] Understandably, the preset first delay duration and the preset second delay duration can be set according to actual operating conditions. The setting of the preset first delay duration and the preset second delay duration aims to ensure the stability and practicality of grid protection for the compressed air energy storage system. Optionally, the preset first delay duration and the preset second delay duration are determined through a limited number of tests by evaluating the grid protection effect of different delay durations on the compressed air energy storage system. The determined preset first delay duration and preset second delay duration should satisfy the condition that they are neither too small nor cause excessive interference to the grid protection process of the compressed air energy storage system. For example, the preset first delay duration is generally selected in the range of [6ms, 8ms], and the preset second delay duration is generally selected in the range of [9ms, 10ms].

[0059] Preferably, the first delay duration is 7ms in a preferred embodiment, and the second delay duration is 10ms in a preferred embodiment.

[0060] Specifically, the reduction in the dynamic threshold of the minimum training data volume is determined by the difference between the transmission delay of the running data and the preset first delay.

[0061] Specifically, when the difference between the data transmission delay and the preset first delay is within 3ms, the dynamic threshold for the minimum training data size is reduced to 0.9 times its original value. When the difference exceeds 3ms, in addition to reducing it to 0.9 times its original value, the dynamic threshold for the minimum training data size decreases by 50 data points for every 1ms exceeding 3ms. For example, if the difference between the data transmission delay and the preset first delay is 5ms, and the current dynamic threshold for the minimum training data size is 6000 data points, then the reduced dynamic threshold for the minimum training data size is 6000 × 0.9 - 50 × 2 = 5300 data points.

[0062] In practice, the method of the present invention adjusts the dynamic threshold of the minimum training data volume by setting a preset first delay duration and a preset second delay duration. Because real-time weather, load, grid parameters and other data cannot quickly form a usable training dataset due to transmission delays and incomplete cleaning, the model cannot update the data in real time. By reducing the dynamic threshold of the minimum training data volume, it is not necessary to wait for sufficient data accumulation, shortening the training dataset construction time, adapting to the real-time requirements of grid protection, and further improving the grid protection stability of the compressed air energy storage system.

[0063] Please continue reading. Figure 4 As shown, it is a logical flowchart of the method for determining the offset of the update trigger time in the smart grid protection method for compressed air energy storage system based on big data according to an embodiment of the present invention.

[0064] Specifically, the offset of the update trigger time is determined based on the proportion of effective prediction time of the prediction model per unit time, including: Compare the percentage of effective prediction time of the prediction model per unit time with the preset percentage; If the effective prediction time ratio of the prediction model within the unit time is greater than or equal to the preset ratio, then it is determined that the scenario adaptability of the prediction model update meets the requirements, and there is no need to increase the offset of the update trigger time. It is also determined whether the dynamic threshold of the minimum training data volume meets the requirements. If the effective prediction time percentage of the prediction model within a unit of time is less than the preset percentage, it is determined that the scenario adaptability of the prediction model update does not meet the requirements, and the offset of the update trigger time needs to be increased.

[0065] Specifically, when the proportion of effective prediction time of the prediction model within a unit of time is greater than or equal to the preset proportion, it is determined that the scenario adaptability of the prediction model update meets the requirements. However, if it has been previously determined that the adaptability of the prediction model update does not meet the requirements, then it is necessary to further determine whether the dynamic threshold of the minimum training data volume meets the requirements.

[0066] In implementation, the dynamic threshold of the minimum training data volume is compared with the predetermined dynamic threshold to determine whether the dynamic threshold of the minimum training data volume meets the requirements. If the dynamic threshold of the actual minimum training data volume is greater than the predetermined dynamic threshold, the dynamic threshold of the minimum training data volume is determined to be unacceptable. The predetermined dynamic threshold is the average value of the dynamic threshold of the minimum training data volume monitored in the previous three months of the historical period.

[0067] If the dynamic threshold for the minimum amount of training data does not meet the requirements, then the dynamic threshold for the minimum amount of training data is reduced; if the dynamic threshold for the minimum amount of training data meets the requirements, then the transmission delay of the running data is re-collected, and the adaptability of the prediction model update is re-evaluated.

[0068] When the proportion of effective prediction time of the prediction model within a unit of time is less than the preset proportion, it can be determined that the reason for the failure of the prediction model to adapt to the scene is that the prediction model to adapt to the scene is not suitable. Therefore, it is necessary to increase the offset of the update trigger time.

[0069] It is understandable that the two intervals defined by the preset proportions correspond to two different scenarios: The first interval is when the effective prediction time of the prediction model per unit time is less than the preset percentage. The corresponding situation is: since the model update takes a certain amount of time, and the prediction window of the power grid protection is only 5 minutes to 1 hour, it may cause the model to be unable to output prediction results during the update period, affecting the scheduling decision and thus delaying the update. In this case, it is necessary to adjust the offset of the update trigger time. The second interval is when the proportion of effective prediction time of the prediction model per unit time is greater than or equal to the preset proportion. The corresponding situation is: the scenario adaptability of the prediction model update meets the requirements. At this time, it is necessary to further determine whether the dynamic threshold of the minimum training data volume meets the requirements.

[0070] Understandably, the preset percentage can be set according to actual operating conditions. The preset percentage is designed to ensure the stability and practicality of grid protection for the compressed air energy storage system. Optionally, the preset percentage is determined through a limited number of tests by evaluating the grid protection effect of different prediction duration percentages on the compressed air energy storage system. The determined preset percentage should be neither too small nor cause excessive interference to the grid protection process of the compressed air energy storage system. For example, the preset percentage is generally selected within the range of [88%, 92%].

[0071] Preferably, the preset percentage in the preferred embodiment is 90%.

[0072] Specifically, the effective prediction time percentage of the prediction model per unit time is the ratio of the time during which the prediction model normally outputs prediction results to the total prediction time per unit time.

[0073] Specifically, the increase in the offset of the update trigger time is determined by the difference between a preset percentage and the percentage of the effective prediction time of the prediction model per unit time.

[0074] Specifically, when the difference between the preset percentage and the percentage of effective prediction time of the prediction model within a unit time is within 3%, the offset of the update trigger time is increased to 1.2 times the original value. When the difference between the preset percentage and the percentage of effective prediction time of the prediction model within a unit time exceeds 3%, in addition to increasing to 1.2 times the original value, the offset of the update trigger time increases by 5ms for every 1% exceeding the original value. For example, if the difference between the preset percentage and the percentage of effective prediction time of the prediction model within a unit time is 5%, and the current offset of the update trigger time is 30ms, the increased offset of the update trigger time will be 30×1.2+5×2=46ms.

[0075] In practice, the method described in this invention adjusts the offset of the update trigger time by setting a preset percentage. Since model updates require a certain amount of time, and the prediction window for grid protection is only 5 minutes to 1 hour, the model may not be able to output prediction results during the update period, affecting scheduling decisions and thus delaying the update. By increasing the offset of the update trigger time, the core decision period within the prediction window can be avoided, preventing updates from occupying critical prediction time, ensuring that scheduling decisions continuously obtain effective prediction results, and further improving the grid protection stability of the compressed air energy storage system.

[0076] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for intelligent grid protection of compressed air energy storage systems based on big data, characterized in that, include: The operation data of the compressed air energy storage system in the power grid is collected, and the operation data is cleaned, denoised and feature extracted to obtain operation features. The initial model is trained according to the operation features to obtain a prediction model, and the prediction model is used to predict the operation status of the power grid to obtain the prediction result. When the predicted result indicates a risk, the compressed air energy storage system is switched from shutdown or normal standby state to grid protection hot standby mode. When the risk actually occurs, a loading command is issued to the compressed air energy storage system in hot standby state. The efficiency of obtaining operating data within a single cycle is used to determine whether the grid protection stability of the compressed air energy storage system meets the requirements based on the efficiency of operating data within the single cycle. If the power grid protection stability does not meet the requirements, determine whether it is necessary to increase the moving average window length; If it is not necessary to increase the moving average window length, then obtain the transmission delay of the running data to determine whether the adaptability of the prediction model update meets the requirements; If the adaptability of the prediction model update does not meet the requirements, then determine whether it is necessary to reduce the dynamic threshold of the minimum training data volume. If it is not necessary to reduce the dynamic threshold of the minimum training data volume, the offset of the update trigger time is determined based on the proportion of the effective prediction time of the prediction model per unit time.

2. The method for intelligent grid protection of compressed air energy storage systems based on big data as described in claim 1, characterized in that, Determining whether the grid protection stability of the compressed air energy storage system meets the requirements based on the efficiency of the operating data within the single cycle includes: The efficiency of the data running in a single cycle is compared with the preset second efficiency. If the efficiency of the operating data within a single cycle is greater than or equal to the preset second efficiency, then the grid protection stability of the compressed air energy storage system is determined to meet the requirements. If the efficiency of the operating data within a single cycle is less than the preset second efficiency, then it is determined that the grid protection stability of the compressed air energy storage system does not meet the requirements.

3. The method for intelligent grid protection of compressed air energy storage systems based on big data as described in claim 2, characterized in that, Determine whether the moving average window length needs to be increased, including: The efficiency of the running data within the single cycle is compared with the preset first efficiency and the preset second efficiency, respectively. If the efficiency of the running data within a single period is less than or equal to the preset first efficiency, then it is determined that the moving average window length needs to be increased. If the efficiency of the running data within a single period is greater than a preset first efficiency and less than a preset second efficiency, then it is determined that there is no need to increase the moving average window length.

4. The method for intelligent grid protection of compressed air energy storage systems based on big data according to claim 3, characterized in that, The increase in the length of the moving average window is determined by the difference between a preset first efficiency and the efficiency of the running data within a single period.

5. The method for intelligent grid protection of compressed air energy storage systems based on big data according to claim 4, characterized in that, The suitability of the prediction model update is determined based on the transmission delay of the running data, including: The transmission delay of the running data is compared with the preset first delay. If the transmission delay of the running data is less than or equal to the preset first delay, then it is determined that the adaptability of the prediction model update meets the requirements, and it is determined whether the moving average window length meets the requirements. If the transmission delay of the running data is greater than the preset first delay duration, then it is determined that the adaptability of the prediction model update does not meet the requirements.

6. The method for intelligent grid protection of compressed air energy storage systems based on big data according to claim 5, characterized in that, Determining whether a dynamic threshold for reducing the minimum training data size needs to be lowered includes: The transmission delay of the running data is compared with the preset first delay and the preset second delay, respectively; If the transmission delay of the running data is greater than the preset first delay and less than or equal to the preset second delay, then a dynamic threshold for reducing the minimum amount of training data is determined. If the transmission delay of the running data is greater than the preset second delay, then it is determined that there is no need to increase the dynamic threshold of the minimum training data amount.

7. The method for intelligent grid protection of compressed air energy storage systems based on big data as described in claim 6, characterized in that, The reduction in the dynamic threshold of the minimum training data volume is determined by the difference between the transmission delay of the running data and the preset first delay.

8. The method for intelligent grid protection of compressed air energy storage systems based on big data according to claim 7, characterized in that, The offset of the update trigger time is determined based on the proportion of effective prediction time of the prediction model per unit time, including: Compare the percentage of effective prediction time of the prediction model per unit time with the preset percentage; If the effective prediction time ratio of the prediction model within the unit time is greater than or equal to the preset ratio, then it is determined that the scenario adaptability of the prediction model update meets the requirements, and there is no need to increase the offset of the update trigger time. It is also determined whether the dynamic threshold of the minimum training data volume meets the requirements. If the effective prediction time percentage of the prediction model within a unit of time is less than the preset percentage, it is determined that the scenario adaptability of the prediction model update does not meet the requirements, and the offset of the update trigger time needs to be increased.

9. The method for intelligent grid protection of compressed air energy storage systems based on big data according to claim 8, characterized in that, The effective prediction time percentage of the prediction model per unit time is the ratio of the time during which the prediction model normally outputs prediction results to the total prediction time per unit time.

10. The method for intelligent grid protection of compressed air energy storage systems based on big data according to claim 9, characterized in that, The increase in the offset of the update trigger time is determined by the difference between the preset percentage and the percentage of the effective prediction time of the prediction model per unit time.

Images