A flywheel-compressed air hybrid energy storage system power distribution and switching control method, system, device and storage medium based on an isolated forest-CNN fusion algorithm

Abstract

The application discloses a flywheel-compressed air hybrid energy storage system power distribution and switching control method, system, equipment and storage medium based on an isolated forest-CNN fusion algorithm, relates to the cross field of application of artificial intelligence and power transmission in an energy storage system, and comprises the following steps: a dynamic efficiency model and a time sequence characteristic matrix are constructed; the working condition is discriminated by using an isolated forest; a safety distribution strategy is triggered in an abnormal state; the optimal distribution ratio is predicted by a CNN in a normal state, and is corrected through torque feedback. The target torque is calculated according to the distribution ratio, smooth switching is realized through planetary gear speed ratio pre-synchronization and clutch linear gradual engagement. The power grid frequency, temperature, vibration and torque are monitored in real time, emergency support, hierarchical thermal protection, vibration isolation or fault switching are executed; the application can realize intelligent power distribution and abnormal rapid identification, smooth switching without impact, reduce component wear, improve system efficiency and power grid support capability, and guarantee long-term safe operation.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of artificial intelligence and power transmission in energy storage systems, and particularly to a power distribution and switching control method, system, device and storage medium for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm. Background Technology

[0002] With the increasing penetration rate of new energy power generation, the power grid is placing higher demands on the power response speed, energy storage density, and operational stability of energy storage systems. Flywheel energy storage offers millisecond-level response speeds, enabling rapid smoothing of high-frequency fluctuations in the power grid, but its energy storage density is relatively low. Compressed air energy storage boasts high energy storage density, enabling large-capacity, long-term energy storage, but its response speed is slower. The flywheel-compressed air hybrid energy storage system, combining the two, can balance the dual advantages of "rapid response" and "large-capacity energy storage," becoming an important technological direction for power grid frequency regulation and peak shaving.

[0003] Currently, the power allocation and switching control of hybrid energy storage systems suffer from the following technical challenges: Traditional efficiency models for flywheel and compressed air energy storage paths often use fixed loss coefficients, failing to consider the impact of dynamic conditions such as ambient temperature, component wear, and lubrication status. This leads to significant efficiency prediction errors and fails to provide accurate data for power allocation. Existing power allocation methods largely rely on traditional optimization algorithms, achieving only single-objective optimization and exhibiting poor adaptability to complex conditions such as grid fluctuations and component anomalies, easily resulting in unreasonable power allocation and low system efficiency. During the switching between flywheel and pneumatic paths, hybrid energy storage systems are prone to shaft impact and power fluctuations due to low planetary gear ratio matching accuracy and abrupt clutch torque engagement, which can severely damage core components such as clutches and gears. The lack of rapid identification and handling mechanisms for abnormal conditions such as sensor distortion, clutch overheating, and air tank overpressure can easily lead to fault propagation and affect system operational safety. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention provides a power distribution and switching control method, system, device and storage medium for a flywheel-compressed air hybrid energy storage system based on the isolated forest-CNN fusion algorithm.

[0005] Therefore, the technical problem solved by this invention is that, under complex dynamic operating conditions, traditional efficiency models of flywheel-compressed air hybrid energy storage systems have low prediction accuracy, poor adaptability of power distribution to abnormal operating conditions, easy generation of shaft impact and power fluctuations during mode switching, and lack of rapid anomaly identification and safety response mechanisms.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, comprising: Based on the operating state parameters and environmental parameters of the flywheel system and compressed air system, a dynamic efficiency model is constructed, and a time-series feature matrix is ​​constructed based on the multi-dimensional features of the sampling period. The isolated forest is used to determine whether the operating condition is normal based on the feature vector of the current sampling period. For abnormal operating conditions, the corresponding safe power allocation strategy is triggered. For normal operating conditions, the optimal power allocation ratio is predicted by a convolutional neural network based on the time-series feature matrix. After real-time feedback correction by the torque sensor, the final power allocation ratio is output. Calculate the target torque of the flywheel path and the compressed air path based on the final power distribution ratio, perform closed-loop pre-synchronization of the planetary gear system until the speed deviation meets the target, and then control the electromagnetic clutch to engage linearly and gradually according to the pressure-torque mapping. Real-time monitoring of grid frequency, clutch temperature, system vibration, and shaft torque; and execution of emergency power support, graded thermal protection, vibration isolation, or fault path switching based on monitoring results.

[0007] As a preferred scheme for power distribution and switching control of a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, wherein: The dynamic efficiency model is constructed based on the operating state parameters and environmental parameters of the flywheel system and compressed air system. A time-series feature matrix is ​​constructed based on the multi-dimensional features of the sampling period, including: Efficiency models for the flywheel path and the compressed air path are established separately. The efficiency model for each path is expressed as a combination of the product of the efficiencies of the series components and the loss correction coefficient. Among them, the efficiency product of series components represents the transmission efficiency of energy passing through planetary gears, electromagnetic clutches, permanent magnet couplings and compressors or expanders in sequence. The loss correction factor includes wind resistance loss, bearing friction loss, leakage loss and flow resistance loss introduced based on the loss mechanism, and also incorporates dynamic correction factors for ambient temperature, lubrication status, sealing status and pipeline cleanliness.

[0008] As a preferred scheme for power distribution and switching control of a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, wherein: The dynamic efficiency model is constructed based on the operating state parameters and environmental parameters of the flywheel system and compressed air system, and a time-series feature matrix is ​​constructed based on the multi-dimensional features of the sampling period. The method also includes: Multidimensional features are extracted based on the sampling period, and these features cover grid demand, flywheel operating status, compressed air operating status, constraints, and system losses. Multidimensional features from multiple sampling periods are arranged in time sequence to form a temporal feature matrix. Each feature in the temporal feature matrix is ​​normalized, and the normalized two-dimensional feature matrix is ​​converted into a single-channel three-dimensional tensor adapted to the input format of a convolutional neural network.

[0009] As a preferred scheme for power distribution and switching control of a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, wherein: The method of using isolated forests to determine whether the operating condition is normal based on the feature vector of the current sampling period, and triggering corresponding safe power allocation strategies for abnormal operating conditions; for normal operating conditions, predicting the optimal power allocation ratio based on the time-series feature matrix through a convolutional neural network, and then outputting the final power allocation ratio after real-time feedback correction by the torque sensor, includes: The isolated forest is used to determine the operating conditions of the feature vector of the current sampling period. In response to the determination of abnormal operating conditions, the corresponding safe power allocation strategy is triggered according to different abnormality types. In response to the determination of normal operating conditions, the optimal power distribution ratio is predicted by a convolutional neural network based on the time-series feature matrix, and then the final power distribution ratio is output after real-time feedback correction by the torque sensor. The convolutional neural network takes the temporal feature matrix as input and the flywheel power ratio as output. It uses the Pareto optimal solution obtained by the multi-objective optimization algorithm as the supervision label. The loss function adopts the form of combining prediction bias and constraint violation penalty. After the network output, the predicted value is pruned in real time to meet the upper limit of flywheel power, upper limit of compressed air power and power fluctuation constraints. The real-time feedback correction collects the actual output power of the two paths and calculates the deviation from the target power. When the deviation exceeds the threshold, the allocation ratio is corrected and the constraints are re-verified. At the same time, normal operating data is collected periodically to incrementally train the convolutional neural network.

[0010] As a preferred scheme for power distribution and switching control of a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, wherein: The process of calculating the target torque of the flywheel path and compressed air path based on the final power distribution ratio, performing closed-loop pre-synchronization of the planetary gear system until the speed deviation meets the target, and then controlling the electromagnetic clutch to engage linearly and gradually according to the pressure-torque mapping includes: The actual speed ratio between the ring gear and the sun gear is calibrated under no-load conditions. Based on the speed relationship of the planetary gear system, the corresponding target motor speed is determined according to different working modes. Closed-loop regulation is used to control the deviation between the motor speed and the target speed within the allowable range to complete the pre-synchronization. The mapping coefficient between clutch pressure and transmitted torque is calibrated through experiments, and temperature correction is introduced. The pressure command is calculated based on the target torque, and the clutch pressure is controlled to make the transmitted torque gradually change from zero to the target torque in a linear manner over time. During engagement, the deviation between the actual torque and the theoretical gradually changing torque is monitored in real time, and the pressure command is dynamically adjusted according to the deviation.

[0011] As a preferred scheme for power distribution and switching control of a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, wherein: The process of calculating the target torque of the flywheel path and compressed air path based on the final power distribution ratio, performing closed-loop pre-synchronization of the planetary gear system until the speed deviation meets the target, and then controlling the electromagnetic clutch to engage linearly and gradually according to the pressure-torque mapping also includes: In response to the transition conditions of the flywheel speed dropping to the lower limit and the flywheel kinetic energy approaching the minimum allowable value, a seamless transition between the flywheel path and the compressed air path is automatically triggered: the flywheel-side clutch pressure decreases linearly and the compressed air-side clutch pressure increases linearly. At the same time, the compressor is started to pre-charge with low torque. The deviation between the sum of the actual power of the two paths and the power demand of the grid is monitored in real time. The clutch pressure change rate is adjusted according to the deviation until the flywheel-side clutch is completely disengaged and the compressed air-side clutch is fully engaged, and the total output power is always equal to the power demand of the grid.

[0012] As a preferred scheme for power distribution and switching control of a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, wherein: The system monitors grid frequency, clutch temperature, system vibration, and shaft torque in real time, and executes emergency power support, graded thermal protection, vibration isolation, or fault path switching based on the monitoring results, including: Emergency power support includes: in response to the grid frequency falling below a set threshold, triggering the flywheel to output maximum power instantaneously, simultaneously starting the expander to preload and linearly increasing the compressed air output power to maintain total power support until the grid frequency returns to the normal range; The graded thermal protection includes: establishing a dynamic temperature change model based on the clutch sliding friction power and heat dissipation power, and executing forced air cooling and reduced switching frequency in the early warning state, linear reduction of power transmission in the unload state, and clutch disengagement and power transmission termination in the emergency stop state according to the clutch temperature. Vibration isolation includes: isolating micro-vibrations generated by the flywheel through non-contact transmission via a permanent magnet coupling, monitoring vibration acceleration in real time, reducing the flywheel speed and switching to a flexible connection mode when the vibration exceeds a threshold, and disengaging the flywheel-side clutch if the vibration is not relieved. The fault path switching includes: comprehensive monitoring of flywheel shaft torque, speed change rate, kinetic energy decay rate, air tank pressure change rate, expander output torque and compressed air power deviation. If any indicator exceeds the preset threshold, it is judged as a fault. The corresponding clutch is immediately disconnected and the normal path takes over the power alone. After the fault is eliminated, the pre-synchronization operation is automatically performed to gradually restore the dual-path operation.

[0013] Secondly, this invention provides a power distribution and switching control system for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, comprising: The dynamic efficiency modeling and time-series feature construction module is used to construct a dynamic efficiency model based on the operating state parameters and environmental parameters of the flywheel system and compressed air system, and to construct a time-series feature matrix based on the multi-dimensional features of the sampling period. The operating condition discrimination and power allocation decision module is used to determine whether the operating condition is normal by using the feature vector of the current sampling period through isolated forest. For abnormal operating conditions, it triggers the corresponding safe power allocation strategy. For normal operating conditions, it predicts the optimal power allocation ratio based on the time-series feature matrix through a convolutional neural network, and then outputs the final power allocation ratio after real-time feedback correction by the torque sensor. The speed ratio pre-synchronization and torque gradual engagement module is used to calculate the target torque of the flywheel path and the compressed air path according to the final power distribution ratio, perform speed ratio closed-loop pre-synchronization of the planetary gear system until the speed deviation meets the target, and then control the electromagnetic clutch to engage linearly and gradually according to the pressure-torque mapping. The real-time monitoring and graded safety protection module is used to monitor the power grid frequency, clutch temperature, system vibration and shaft torque in real time, and to perform emergency power support, graded thermal protection, vibration isolation or fault path switching according to the monitoring results.

[0014] Thirdly, the present invention provides a computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the power distribution and switching control method of the flywheel-compressed air hybrid energy storage system based on the isolated forest-CNN fusion algorithm.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of a power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm.

[0016] The beneficial effects of this invention are as follows: This invention adopts a fusion algorithm architecture of isolated forest and CNN (Convolutional Neural Network), which can quickly identify abnormal operating conditions such as sensor distortion and component overheating and trigger targeted safety strategies to prevent fault propagation; it can also accurately learn the mapping relationship between operating condition characteristics and optimal power allocation under normal operating conditions, achieving synergistic optimization of efficiency and loss, and ensuring the rationality and stability of power allocation; through the coordinated control of planetary gear ratio closed-loop pre-synchronization and clutch torque linear gradual change, the shaft impact and power fluctuation problems during mode switching are eliminated from the root, reducing the wear of core components such as clutches and gears, extending equipment service life, and achieving seamless transition between flywheel and pneumatic path; combined with vibration isolation, rapid fault path cutoff and other technical means, it effectively avoids operational risks such as component overheating, vibration damage, and fault propagation, and improves the long-term reliability of the system; at the same time, it can quickly respond to the grid frequency fluctuation demand, enhancing the hybrid energy storage system's support capability for the grid. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is an overall flowchart of a power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm provided by the present invention.

[0019] Figure 2 This is a flowchart of the working condition discrimination and power allocation decision-making process for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, provided by the present invention.

[0020] Figure 3 This is a flowchart of the mode switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, provided by the present invention. Detailed Implementation

[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0022] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, including: S1: Based on the operating state parameters and environmental parameters of the flywheel system and compressed air system, a dynamic efficiency model is constructed, and a time-series feature matrix is ​​constructed based on the multi-dimensional features of the sampling period; S2: Use isolated forest to determine whether the operating condition is normal based on the feature vector of the current sampling period. For abnormal operating conditions, trigger the corresponding safe power allocation strategy. For normal operating conditions, predict the optimal power allocation ratio based on the time-series feature matrix through a convolutional neural network, and then output the final power allocation ratio after real-time feedback correction by the torque sensor. S3: Calculate the target torque of the flywheel path and the compressed air path according to the final power distribution ratio, perform closed-loop pre-synchronization of the planetary gear system until the speed deviation meets the target, and then control the electromagnetic clutch to engage linearly and gradually according to the pressure-torque mapping. S4: Real-time monitoring of grid frequency, clutch temperature, system vibration and shaft torque, and execution of emergency power support, graded thermal protection, vibration isolation or fault path switching based on monitoring results.

[0023] It should be noted that, through steps S1-S4, this invention constructs an integrated control closed loop encompassing operating condition perception, anomaly identification, optimized power allocation, smooth switching, and safety protection. Based on a dynamic efficiency model and a time-series feature matrix, combined with a fusion architecture of isolated forests and convolutional neural networks, adaptive power allocation for both normal and abnormal operating conditions is achieved, balancing system efficiency and operational safety. The switching control, coupled with planetary gear ratio pre-synchronization and linearly gradual clutch torque changes, significantly suppresses mechanical shocks and power fluctuations during mode transitions. Multi-dimensional real-time monitoring and hierarchical protection mechanisms further enhance the system's response capabilities to grid frequency fluctuations, thermal anomalies, vibrations, and faults, ensuring the efficient, stable, and reliable operation of the flywheel-compressed air hybrid energy storage system under all operating conditions.

[0024] Example 2, refer to Figures 1-3 As an embodiment of the present invention, based on the previous embodiment, a power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm is provided, including: In this embodiment, the dynamic efficiency model constructed in step S1 above, based on the operating state parameters and environmental parameters of the flywheel system and compressed air system, and the time-series feature matrix constructed based on the multi-dimensional features of the sampling period, includes: Efficiency models for the flywheel path and compressed air path are established separately, and loss mechanism terms and dynamic correction coefficients for environment and operating conditions are introduced.

[0025] The flywheel path efficiency model is as follows: in, The overall efficiency of the flywheel energy storage path. For the transmission efficiency of planetary gears, The efficiency of the first electromagnetic clutch. For the efficiency of permanent magnet couplings. This is the flywheel path loss correction factor. As the efficiency product of the series components, when energy passes through the planetary gear, electromagnetic clutch, and permanent magnet coupling in sequence, the transmission efficiency of each link is ≤1. The more links there are, the greater the loss and the lower the overall efficiency. The smaller the meshing clearance and the better the lubrication, the better the lubrication. The closer to 1; Affected by the engagement pressure and the difference in sliding speed, the more stable the pressure and the smaller the speed difference, the better. The higher; Due to the influence of air gap size, when the air gap is ≤0.5mm, the loss of non-contact transmission is extremely low. It is close to 0.999.

[0026] Furthermore, the flywheel path loss correction factor Represented as: in, The power loss is due to the flywheel's wind resistance. The power loss due to friction of the flywheel bearing. The power input to the flywheel path is the electrical power input from the power grid or power source into the flywheel path.

[0027] The flywheel drag loss power is expressed as: in, The flywheel angular velocity (unit: rad / s or r / min, coefficient 0.005 corresponds to the wind resistance loss coefficient under standard operating conditions). This is the air density correction factor. , Ambient temperature (in °C).

[0028] Flywheel bearing friction loss power , The torque transmitted by the flywheel shaft (in N·m) is 0.01, which is the friction loss coefficient under standard lubrication conditions. This is the lubrication condition coefficient, under normal lubrication conditions. When lubrication is low The temperature is determined in real time by a lubricating oil temperature sensor.

[0029] The compressed air path efficiency model is as follows: .

[0030] in, The overall efficiency of the compressed air energy storage path. For the efficiency of the second electromagnetic clutch. For the efficiency of another permanent magnet coupling, The energy conversion efficiency of a compressor (energy storage) or expander (energy release). This is a correction factor for compressed air path losses. Compressor efficiency. Defined as the pressure energy gained from the gas tank divided by the mechanical energy input to the compressor, it is affected by the pressure ratio (gas tank pressure / atmospheric pressure) and flow rate; expander efficiency. Defined as the mechanical energy output by the expander divided by the gas pressure energy input to the gas tank, it is affected by the gas tank pressure and the expansion ratio.

[0031] The compressed air path loss correction factor is expressed as: in, For leakage loss power, Power loss due to pipeline flow resistance, Input power to the compressed air path.

[0032] Leakage loss power is expressed as: in, The pressure of the gas storage tank is given in MPa, and 0.02 represents the leakage loss coefficient under the new sealing conditions. This is the sealing condition coefficient (1 for new seals, 1.3 for worn seals, monitored by a leak sensor).

[0033] The power loss due to flow resistance in a pipeline is expressed as: in, Gas volumetric flow rate (unit: m³) 3 / s or L / min), 0.008 is the resistance loss coefficient under clean pipeline conditions, The pipeline cleanliness factor is 1.5 (1.5 when blocked, determined by the pipeline differential pressure sensor).

[0034] To calibrate the model coefficients and collect real-world operating data, a modular test system was built in terms of hardware: the power unit uses a high-precision bidirectional power supply to simulate grid power, is equipped with a surge protector and a built-in power analyzer; the load unit uses a magnetic powder brake to simulate a flywheel mechanical load, connected by a flexible coupling, and is equipped with a water cooling system and a torque sensor; the gas circuit unit includes a high-pressure gas storage tank, a high-pressure flange, a high-precision flow control valve, a vortex flow meter, and a pressure transmitter; the data acquisition unit uses a high-speed acquisition card and a "master-slave" distributed architecture to collect signals such as voltage, current, speed, torque, pressure, and temperature through fiber optic communication.

[0035] Multidimensional features are extracted based on the sampling period (10ms), and a temporal feature matrix is ​​constructed. Time dimension The sampling period is set to 10 sampling periods (corresponding to 10×10ms=100ms). This duration can fully capture the short-term trends of time-series characteristics such as power fluctuations and flywheel speed changes, avoiding feature fragmentation. Feature Dimension It comprises 12 dimensions, comprehensively covering five core dimensions: grid demand, flywheel status, compressed air status, constraints, and system losses. Specifically, it includes: grid demand power. Grid power change rate Flywheel speed Flywheel available power limit Temperature of the first electromagnetic clutch Gas tank pressure Upper limit of available power for compressed air system Temperature of the second electromagnetic clutch Flywheel path efficiency Compressed air path efficiency Power fluctuation constraint margin (the maximum allowable power change under current operating conditions), total system losses ; To eliminate the influence of different units of measurement on CNN training and prediction, all features are normalized using the "rated value / range normalization" method, which means dividing each feature value by its corresponding rated value or measurement range, and normalizing the feature values ​​to the [0,1] interval. Normalized Two-dimensional feature matrix converted to The single-channel 3D tensor adapts to the input data format requirements of CNN, ensuring that convolution operations can effectively extract local temporal features.

[0036] In this embodiment, step S2 above utilizes an isolated forest to determine whether the operating condition is normal based on the feature vector of the current sampling period, and triggers a corresponding safe power allocation strategy for abnormal operating conditions; for normal operating conditions, the optimal power allocation ratio is predicted based on the time-series feature matrix using a convolutional neural network, and then the final power allocation ratio is output after real-time feedback correction by the torque sensor, including: The Isolation Forest algorithm is used to quickly identify scenarios deviating from normal operating conditions by randomly partitioning the feature space and isolating outlier data points. 1000 hours of normal operating condition data and 50 hours of typical outlier operating condition data (including 8 categories such as sensor drift, clutch overheating, gas tank overpressure, and power grid peak interference) were collected, generating a total of 3.6 × 10^6 samples. The training and test sets were divided in a 9:1 ratio, and the number of trees, sample subset size, anomaly detection threshold (set to 0.5), and maximum tree depth were calibrated. The model calculates anomaly scores on the current 12-dimensional feature vector. When abnormal scores When, it is determined to be an abnormal operating condition; when When this condition is met, it is considered a normal operating condition.

[0037] After identifying the anomaly, the specific anomaly type is determined by analyzing the deviation dimension of the features, and the corresponding safe power allocation strategy is triggered. If it is a sensor data distortion (a certain feature value exceeds three times its standard deviation), the average of the optimal power allocation ratios of the previous five sampling periods is used. As the current allocation ratio, the power change rate is limited to ≤0.01 / 10ms (i.e., power change does not exceed 1% of rated power per 10ms). If the electromagnetic clutch overheats (temperature ≥120℃), the power proportion of the overheated path is limited to ≤30%. If the air tank is overpressurized (pressure ≥32MPa), the power proportion of the compressed air path is limited to ≤20%. If there are extreme fluctuations in the power grid, i.e., the power change rate... , , The rated power of the system is in kW, which increases the power ratio of the flywheel path to ≥90%.

[0038] For normal operating conditions, a convolutional neural network (CNN) is used to learn the mapping relationship between the temporal feature matrix and the optimal power allocation ratio. CNN structure: Input layer receives... The three-dimensional tensor (10 time steps, 12-dimensional features, 1 channel) is directly connected to the preprocessed temporal feature data; Convolutional layer 1 is configured with eight 3×3 convolutional kernels, with a stride of 1, using the ReLU activation function, and an output dimension of... Its core function is to extract local temporal correlation features from the feature matrix; Max pooling layer 1 uses a 2×2 pooling kernel with a stride of 2 and an output dimension of 1. By performing dimensionality reduction, key features are preserved while reducing the computational load of subsequent layers. Convolutional layer 2 is configured with 16 3×3 convolutional kernels, with a stride of 1, using the ReLU activation function, and an output dimension of... Further extract higher-order fusion features; Max pooling layer 2 uses a 2×2 pooling kernel with a stride of 2 and an output dimension of [missing value]. Dimensionality is reduced again to improve computational efficiency; The flattening layer converts the pooled 3D feature map into a 96-dimensional one-dimensional vector, which is compatible with the input format of the fully connected layer. The fully connected layer 1 is configured with 64 neurons and uses the ReLU activation function to achieve nonlinear fusion and mapping of higher-order features; The output layer is configured with one neuron, using the Sigmoid activation function, and the output range is [0,1], representing the proportion of flywheel power to total power. ; The Pareto optimal solution obtained by the NSGA-II (Nondominated Sorting Genetic Algorithm II) algorithm is used as the supervision label for the CNN.

[0039] The loss function is the mean squared error plus a constraint penalty, and the formula is: in, The penalty coefficient is used to significantly increase the loss value when the predicted value violates the flywheel or compressed air power constraints, forcing the model to learn the allocation ratio that satisfies the constraints.

[0040] The training dataset consists of 500 hours of normal working conditions (1.8 × 10^6 samples in total), which are divided into training, validation, and test sets in a ratio of 7:2:1. The Adam optimizer is used, with an initial learning rate of 1e-4, a batch size of 64, and a total of 100 training rounds. The training is stopped early if the validation set loss does not decrease for 5 consecutive rounds to avoid model overfitting.

[0041] CNN outputs initial prediction values Next, it is necessary to further verify whether all preset constraints are met. If not, real-time cropping will be performed. All preset constraints include: Flywheel power constraint verification: If Then Cut to Ensure the flywheel is not overloaded; Compressed air power constraint verification: If Then Cut to Ensure that the compressed air system is not overloaded; Power fluctuation constraint verification: If ( If so, then trim according to the upper limit of fluctuation. This avoids system shocks caused by sudden changes in power distribution.

[0042] Real-time feedback correction is introduced to offset component wear drift and sensor errors. Torque sensors mounted on the flywheel shaft and compressor / expander shaft are used to acquire the actual output power of both paths in real time. and The sampling frequency is consistent with the sampling period; The formula for calculating the deviation between the target power and the actual power is as follows: in, The power allocation ratio currently being implemented.

[0043] When the absolute value of the flywheel path power deviation Or the absolute value of the power deviation in the compressed air path In the next sampling period, the allocation ratio is corrected using the following formula: in, This indicates the flywheel power distribution ratio after real-time feedback correction, that is, the proportion of total power that the flywheel path should bear after correction.

[0044] After correction, the constraints were re-verified. Simultaneously, normal operating condition data was periodically collected to incrementally train the CNN, adapting to component aging and environmental drift during long-term operation. (Refer to...) Figure 2 This is a flowchart of the working condition discrimination and power allocation decision provided by the present invention.

[0045] In this embodiment, step S3 above, which calculates the target torque of the flywheel path and the compressed air path based on the final power distribution ratio, performs closed-loop pre-synchronization of the planetary gear system until the speed deviation meets the target, and then controls the electromagnetic clutch to engage linearly and gradually according to the pressure-torque mapping, includes: First, perform closed-loop pre-synchronization of the planetary gear system. The planetary gears consist of a sun gear, a ring gear, and a planet carrier. Define the gear ratio between the ring gear and the sun gear. : in, This refers to the number of teeth on the gear ring. This represents the number of teeth on the sun gear.

[0046] Under no-load conditions, the sun gear is driven to run at different speeds, and the speeds of the planet carrier and ring gear are measured. The actual speed ratio is then obtained by fitting the data. If the deviation between the theoretical value and the actual value is ≤1%, otherwise adjust the gear installation clearance and finally use the actual speed ratio to participate in the pre-synchronization calculation to ensure matching accuracy.

[0047] The basic speed relationship of the planetary gear system is as follows: in, For the rotational speed of the sun gear, For gear ring speed, This is the rotational speed of the planetary carrier.

[0048] The pre-synchronization logic under different working modes includes: Charging-Flywheel Priority Mode: Locking the Gear Ring ( The planetary carrier is rigidly connected to the flywheel shaft. The motor speed needs to be adjusted to ; Charging-compressed air priority mode: Lock planetary carrier ( The gear ring is rigidly connected to the compressor shaft. Therefore, the motor speed needs to be adjusted to... ; Hybrid mode: Neither the planetary carrier nor the ring gear is locked. The rotational speeds of the three components need to be monitored in real time to verify that the speed ratio deviation is ≤2% in order to avoid gear jamming.

[0049] Pre-synchronization employs a closed-loop control system of "speed measurement - deviation calculation - motor speed regulation." Specifically, it involves real-time acquisition of the target speed (such as flywheel speed). ) and current motor speed ; Calculate the speed deviation , The target motor speed; Speed ​​control commands are output using a PI controller: The proportionality coefficient Integral coefficient (The settings were adjusted through bench testing to ensure there was no overshoot); When the speed deviation If pre-synchronization is completed, clutch engagement is permitted; if the deviation persists... If the speed exceeds 50 r / min, a synchronization failure alarm is triggered, and gear or motor faults are investigated. When 5 r / min < |Δω| ≤ 50 r / min, the current closed-loop control of the PI regulator is maintained, and the speed deviation is adjusted without further processing.

[0050] Then, the electromagnetic clutch engages with a linear, gradual torque change. Clutch pressure-torque coefficient. (N·m / MPa) Calibrated by test: When the input and output shaft speeds are the same, gradually apply the engagement pressure from 0.1MPa to the rated pressure of 5MPa. Measure the corresponding transmitted torque Linear regression was used to fit the data, resulting in: Goodness of fit .

[0051] To account for the effect of temperature on the coefficient of friction, a temperature correction factor is introduced. ,in Given the current clutch temperature (°C) and 25°C as the reference temperature, the actual coefficient is... This avoids insufficient torque transmission at high temperatures.

[0052] The clutch engagement pressure is controlled by a proportional relief valve (response time ≤ 10ms), and the pressure command is expressed as: The central controller outputs a PWM signal (frequency 1kHz) to adjust the opening of the overflow valve, thereby achieving precise pressure control.

[0053] To eliminate engagement shock, the torque is linearly increased from 0 to the target transmitted torque. The target torque is determined by the power distribution ratio: in, For the path target power, This corresponds to the shaft speed.

[0054] The torque gradient curve is represented as follows: ( ) in, To integrate gradual time, It is determined by the power allocation ratio.

[0055] Real-time acquisition of actual torque of clutch output shaft , compared with theoretical value Compare and calculate the deviation. ;like , Immediately adjust the pressure command to the clutch's rated torque: This ensures the accuracy of torque gradient.

[0056] Achieving a seamless transition between the flywheel and the compressed air path: when the flywheel speed... Reduce to rated speed 30% of the flywheel's kinetic energy hour, The minimum permissible kinetic energy of the flywheel automatically triggers a seamless transition process.

[0057] The flywheel-side clutch pressure decreases gradually along the curve, while the compressed air-side clutch pressure increases gradually along the curve. At the same time, the compressor is started for pre-charging (low torque mode, torque ≤ 20% of rated value).

[0058] The actual power of the two paths is monitored in real time by a power sensor. Adjust the clutch pressure change rate; the flywheel side clutch pressure drops to 0 (disengaged), the compressed air side pressure rises to the target value, the power transfer is completed, and the system switches to compressed air independent response mode.

[0059] Full satisfaction To ensure that the total output power matches the grid demand, the power change curve is as follows: , ,in This refers to the handover time.

[0060] Reference Figure 3 This is a flowchart of the mode switching control provided by the present invention.

[0061] In this embodiment, the real-time monitoring of grid frequency, clutch temperature, system vibration, and shaft torque in step S4, and the execution of emergency power support, graded thermal protection, vibration isolation, or fault path switching based on the monitoring results, includes: For emergency grid frequency control: when the grid frequency is less than or equal to 49.5Hz for two consecutive sampling periods, it is determined to be a sudden drop in grid frequency, triggering emergency power support. The flywheel-side clutch completes pre-synchronization and engagement, and the flywheel instantaneously outputs maximum power. This quickly mitigates the frequency decline trend; simultaneously, the expander preloads, and the compressed air power increases linearly. Total power ; With the expander intake valve fully open, the compressed air system outputs maximum power. Total power maintained at It continuously provides power support to the power grid until the grid frequency recovers to above 49.8Hz for five consecutive sampling cycles, then gradually reduces the power to normal levels. It monitors shaft vibration and temperature in real time, and immediately adjusts power output if any abnormalities are detected.

[0062] Establishing a clutch thermal model and graded protection: sliding friction power ,in The speed of the clutch input shaft. The output shaft speed is the difference in speed, which originates from the pre-synchronization residual deviation or the slippage during the torque gradual change process. All of this power is converted into heat. The dynamic model for temperature change is as follows: in, Ambient temperature; This refers to the clutch's heat capacity. For heat dissipation power, during natural heat dissipation , The heat dissipation coefficient; This represents the heat dissipation area of ​​the clutch.

[0063] The overheat protection tiered strategy is set as follows: Warning stage (80℃≤ <100℃): Activate forced air cooling and reduce clutch switching frequency; Unloading stage (100℃≤ <120℃): Triggers linear load reduction; the load reduction formula is: Where 100 is the load reduction initiation temperature (unit: °C), corresponding to the clutch temperature reaching this value when load reduction begins, at which point the limiting factor is 1 (i.e., no load reduction); 120 is the emergency stop temperature ( °C), corresponding to the clutch temperature reaching this value when power must be completely cut off. The denominator 20 is the load reduction temperature range width (120℃-100℃=20℃), ensuring that when... At 100℃, the value in parentheses is 1, and power is not limited; when At 120℃, the value in parentheses is 0, and the power limit is 0. This linear load reduction factor is calibrated through a clutch bench thermal ablation test to ensure that within the range of 100℃~120℃, the power limit is linearly related to the temperature rise, thus avoiding secondary impacts on the power grid caused by sudden power drops.

[0064] Emergency shutdown phase ( ≥120℃): Immediately disconnect the corresponding clutch, stop power transmission in that path, send an overheat alarm signal, and restart only when the temperature drops below 90℃.

[0065] For vibration isolation, the stiffness of the permanent magnet coupling is calibrated through modal testing. , For excitation force, Similarly, for displacement, the stiffness of the mechanical shaft is calibrated. .

[0066] Since the permanent magnet coupling is a non-contact transmission, it can effectively isolate the micro-vibrations generated by the high-speed rotation of the flywheel and prevent the vibration from being transmitted to the planetary gears and the compressed air system. The vibration acceleration is monitored in real time by acceleration sensors installed on the planetary carrier and the frame. When the acceleration exceeds the threshold, it is determined to be abnormal vibration. The flywheel speed is immediately reduced and the flexible connection mode is switched at the same time. If the vibration is still not relieved, the flywheel side clutch is disengaged.

[0067] For rapid switching of fault paths, the monitoring indicators include: flywheel shaft torque, flywheel speed change rate, flywheel kinetic energy decay rate, air tank pressure change rate, expander output torque, and compressed air power deviation. If any indicator exceeds the preset threshold, it is judged as a fault.

[0068] In one possible implementation, the preset thresholds corresponding to each monitoring indicator can be calibrated through offline system simulation and bench fault simulation tests. For example, the flywheel shaft torque threshold is: exceeding 150% of the rated torque, or the instantaneous torque change rate exceeds 500 N·m / s; the flywheel speed change rate threshold is: a speed change exceeding 5% of the rated speed per millisecond; the flywheel kinetic energy decay rate threshold is: a kinetic energy decrease exceeding 15% per 10 milliseconds; the air tank pressure change rate threshold is: a pressure change exceeding 0.5 MPa per millisecond; the expander output torque threshold is: torque fluctuation exceeding ±25% of the set target value and lasting for more than 20 ms; the compressed air power deviation threshold is: the deviation between the actual power and the target power exceeds 15% of the rated power and lasts for more than 3 sampling cycles (30 ms). The calibration method for these thresholds is as follows: simulate typical faults on the system bench, collect the dynamic response curves of each indicator when a fault occurs, take the fluctuation range of the indicator under normal operating conditions (3 times the standard deviation) as the initial value of the threshold, and then adjust it through robustness testing to ensure no false alarms when there are no faults and no missed alarms when there are faults.

[0069] Upon detecting a fault, the system immediately outputs a clutch disengagement command to physically disconnect the corresponding clutch. Simultaneously, it adjusts the power distribution strategy, allowing the normal path to solely handle the grid's power demand. After fault isolation, the system continuously monitors the status of the faulty path. When all fault criteria are no longer met, it automatically performs pre-synchronization operations to gradually restore dual-path operation, avoiding the impact caused by sudden loading.

[0070] It should be noted that the preset thresholds involved in the above fault criteria can be calibrated based on the actual system design parameters and bench test results.

[0071] Example 3: The above is an illustrative scheme of a power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on the Isolation Forest-CNN fusion algorithm in this embodiment. It should be noted that the technical solution of a power allocation and switching control system for a flywheel-compressed air hybrid energy storage system based on the Isolation Forest-CNN fusion algorithm belongs to the same concept as the above-described power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on the Isolation Forest-CNN fusion algorithm. Details not described in detail in the technical solution of the power allocation and switching control system for a flywheel-compressed air hybrid energy storage system based on the Isolation Forest-CNN fusion algorithm in this embodiment can be found in the description of the above-described power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on the Isolation Forest-CNN fusion algorithm.

[0072] This embodiment also provides a power distribution and switching control system for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, including: The dynamic efficiency modeling and time-series feature construction module is used to construct a dynamic efficiency model based on the operating state parameters and environmental parameters of the flywheel system and compressed air system, and to construct a time-series feature matrix based on the multi-dimensional features of the sampling period. The operating condition discrimination and power allocation decision module is used to determine whether the operating condition is normal by using the feature vector of the current sampling period through isolated forest. For abnormal operating conditions, it triggers the corresponding safe power allocation strategy. For normal operating conditions, it predicts the optimal power allocation ratio based on the time-series feature matrix through a convolutional neural network, and then outputs the final power allocation ratio after real-time feedback correction by the torque sensor. The speed ratio pre-synchronization and torque gradual engagement module is used to calculate the target torque of the flywheel path and the compressed air path according to the final power distribution ratio, perform speed ratio closed-loop pre-synchronization of the planetary gear system until the speed deviation meets the target, and then control the electromagnetic clutch to engage linearly and gradually according to the pressure-torque mapping. The real-time monitoring and graded safety protection module is used to monitor the power grid frequency, clutch temperature, system vibration and shaft torque in real time, and to perform emergency power support, graded thermal protection, vibration isolation or fault path switching according to the monitoring results.

[0073] This embodiment also provides a computer device applicable to a power distribution and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a power distribution and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, as proposed in the above embodiments.

[0074] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements a power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm as proposed in the above embodiment.

[0075] The storage medium proposed in this embodiment belongs to the same inventive concept as the power distribution and switching control method of the flywheel-compressed air hybrid energy storage system based on the isolated forest-CNN fusion algorithm proposed in the above embodiment. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0076] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, characterized in that, include: Based on the operating state parameters and environmental parameters of the flywheel system and compressed air system, a dynamic efficiency model is constructed, and a time-series feature matrix is ​​constructed based on the multi-dimensional features of the sampling period. The isolated forest is used to determine whether the operating condition is normal based on the feature vector of the current sampling period. For abnormal operating conditions, the corresponding safe power allocation strategy is triggered. For normal operating conditions, the optimal power allocation ratio is predicted by a convolutional neural network based on the time-series feature matrix. After real-time feedback correction by the torque sensor, the final power allocation ratio is output. Calculate the target torque of the flywheel path and the compressed air path based on the final power distribution ratio, perform closed-loop pre-synchronization of the planetary gear system until the speed deviation meets the target, and then control the electromagnetic clutch to engage linearly and gradually according to the pressure-torque mapping. Real-time monitoring of grid frequency, clutch temperature, system vibration, and shaft torque; and execution of emergency power support, graded thermal protection, vibration isolation, or fault path switching based on monitoring results.

2. The power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm as described in claim 1, characterized in that, The dynamic efficiency model is constructed based on the operating state parameters and environmental parameters of the flywheel system and compressed air system. A time-series feature matrix is ​​constructed based on the multi-dimensional features of the sampling period, including: Efficiency models for the flywheel path and the compressed air path are established separately. The efficiency model for each path is expressed as a combination of the product of the efficiencies of the series components and the loss correction coefficient. Among them, the efficiency product of series components represents the transmission efficiency of energy passing through planetary gears, electromagnetic clutches, permanent magnet couplings and compressors or expanders in sequence. The loss correction factor includes wind resistance loss, bearing friction loss, leakage loss and flow resistance loss introduced based on the loss mechanism, and also incorporates dynamic correction factors for ambient temperature, lubrication status, sealing status and pipeline cleanliness.

3. The power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm as described in claim 2, characterized in that, The dynamic efficiency model is constructed based on the operating state parameters and environmental parameters of the flywheel system and compressed air system, and a time-series feature matrix is ​​constructed based on the multi-dimensional features of the sampling period. The method also includes: Multidimensional features are extracted based on the sampling period, and these features cover grid demand, flywheel operating status, compressed air operating status, constraints, and system losses. Multidimensional features from multiple sampling periods are arranged in time sequence to form a temporal feature matrix. Each feature in the temporal feature matrix is ​​normalized, and the normalized two-dimensional feature matrix is ​​converted into a single-channel three-dimensional tensor adapted to the input format of a convolutional neural network.

4. The power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm as described in claim 3, characterized in that, The method of using isolated forests to determine whether the operating condition is normal based on the feature vector of the current sampling period, and triggering corresponding safe power allocation strategies for abnormal operating conditions; for normal operating conditions, predicting the optimal power allocation ratio based on the time-series feature matrix through a convolutional neural network, and then outputting the final power allocation ratio after real-time feedback correction by the torque sensor, includes: The isolated forest is used to determine the operating conditions of the feature vector of the current sampling period. In response to the determination of abnormal operating conditions, the corresponding safe power allocation strategy is triggered according to different abnormality types. In response to the determination of normal operating conditions, the optimal power distribution ratio is predicted by a convolutional neural network based on the time-series feature matrix, and then the final power distribution ratio is output after real-time feedback correction by the torque sensor. The convolutional neural network takes the temporal feature matrix as input and the flywheel power ratio as output. It uses the Pareto optimal solution obtained by the multi-objective optimization algorithm as the supervision label. The loss function adopts the form of combining prediction bias and constraint violation penalty. After the network output, the predicted value is pruned in real time to meet the upper limit of flywheel power, upper limit of compressed air power and power fluctuation constraints. The real-time feedback correction collects the actual output power of the two paths and calculates the deviation from the target power. When the deviation exceeds the threshold, the allocation ratio is corrected and the constraints are re-verified. At the same time, normal operating data is collected periodically to incrementally train the convolutional neural network.

5. The power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm as described in claim 4, characterized in that, The process of calculating the target torque of the flywheel path and compressed air path based on the final power distribution ratio, performing closed-loop pre-synchronization of the planetary gear system until the speed deviation meets the target, and then controlling the electromagnetic clutch to engage linearly and gradually according to the pressure-torque mapping includes: The actual speed ratio between the ring gear and the sun gear is calibrated under no-load conditions. Based on the speed relationship of the planetary gear system, the corresponding target motor speed is determined according to different working modes. Closed-loop regulation is used to control the deviation between the motor speed and the target speed within the allowable range to complete the pre-synchronization. The mapping coefficient between clutch pressure and transmitted torque is calibrated through experiments, and temperature correction is introduced. The pressure command is calculated based on the target torque, and the clutch pressure is controlled to make the transmitted torque gradually change from zero to the target torque in a linear manner over time. During engagement, the deviation between the actual torque and the theoretical gradually changing torque is monitored in real time, and the pressure command is dynamically adjusted according to the deviation.

6. The power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm as described in claim 5, characterized in that, The process of calculating the target torque of the flywheel path and compressed air path based on the final power distribution ratio, performing closed-loop pre-synchronization of the planetary gear system until the speed deviation meets the target, and then controlling the electromagnetic clutch to engage linearly and gradually according to the pressure-torque mapping also includes: In response to the transition conditions of the flywheel speed dropping to the lower limit and the flywheel kinetic energy approaching the minimum allowable value, a seamless transition between the flywheel path and the compressed air path is automatically triggered: the flywheel-side clutch pressure decreases linearly and the compressed air-side clutch pressure increases linearly. At the same time, the compressor is started to pre-charge with low torque. The deviation between the sum of the actual power of the two paths and the power demand of the grid is monitored in real time. The clutch pressure change rate is adjusted according to the deviation until the flywheel-side clutch is completely disengaged and the compressed air-side clutch is fully engaged, and the total output power is always equal to the power demand of the grid.

7. The power allocation and switching control method for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm as described in claim 6, characterized in that, The system monitors grid frequency, clutch temperature, system vibration, and shaft torque in real time, and executes emergency power support, graded thermal protection, vibration isolation, or fault path switching based on the monitoring results, including: Emergency power support includes: in response to the grid frequency falling below a set threshold, triggering the flywheel to output maximum power instantaneously, simultaneously starting the expander to preload and linearly increasing the compressed air output power to maintain total power support until the grid frequency returns to the normal range; The graded thermal protection includes: establishing a dynamic temperature change model based on the clutch sliding friction power and heat dissipation power, and executing forced air cooling and reduced switching frequency in the early warning state, linear reduction of power transmission in the unload state, and clutch disengagement and power transmission termination in the emergency stop state according to the clutch temperature. Vibration isolation includes: isolating micro-vibrations generated by the flywheel through non-contact transmission via a permanent magnet coupling, monitoring vibration acceleration in real time, reducing the flywheel speed and switching to a flexible connection mode when the vibration exceeds a threshold, and disengaging the flywheel-side clutch if the vibration is not relieved. The fault path switching includes: comprehensive monitoring of flywheel shaft torque, speed change rate, kinetic energy decay rate, air tank pressure change rate, expander output torque and compressed air power deviation. If any indicator exceeds the preset threshold, it is judged as a fault. The corresponding clutch is immediately disconnected and the normal path takes over the power alone. After the fault is eliminated, the pre-synchronization operation is automatically performed to gradually restore the dual-path operation.

8. A power distribution and switching control system for a flywheel-compressed air hybrid energy storage system based on an isolated forest-CNN fusion algorithm, wherein the method described in any one of claims 1 to 7 is applied, characterized in that, include: The dynamic efficiency modeling and time-series feature construction module is used to construct a dynamic efficiency model based on the operating state parameters and environmental parameters of the flywheel system and compressed air system, and to construct a time-series feature matrix based on the multi-dimensional features of the sampling period. The operating condition discrimination and power allocation decision module is used to determine whether the operating condition is normal by using the feature vector of the current sampling period through isolated forest. For abnormal operating conditions, it triggers the corresponding safe power allocation strategy. For normal operating conditions, it predicts the optimal power allocation ratio based on the time-series feature matrix through a convolutional neural network, and then outputs the final power allocation ratio after real-time feedback correction by the torque sensor. The speed ratio pre-synchronization and torque gradual engagement module is used to calculate the target torque of the flywheel path and the compressed air path according to the final power distribution ratio, perform speed ratio closed-loop pre-synchronization of the planetary gear system until the speed deviation meets the target, and then control the electromagnetic clutch to engage linearly and gradually according to the pressure-torque mapping. The real-time monitoring and graded safety protection module is used to monitor the power grid frequency, clutch temperature, system vibration and shaft torque in real time, and to perform emergency power support, graded thermal protection, vibration isolation or fault path switching according to the monitoring results.

9. A computer device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores computer-executable instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images