Unit combined type compressed air energy storage system and method

Abstract

The invention provides a unit combined type compressed air energy storage system and method, and belongs to the technical field of energy storage. The heat exchanger is connected with the compression unit; the energy storage unit is connected with the heat exchanger; the heat exchange medium box is connected with the heat exchanger; the heat storage unit is connected with the heat exchanger and the heat exchange medium box; the hydroelectric generation unit is connected with the energy storage unit and the heat exchanger; the water tank is connected with the hydroelectric generation unit and the energy storage unit; the air pump is connected with the heat exchanger and the energy storage unit; the water pump is connected with the heat exchanger and the heat storage unit. The invention provides a compressed air energy storage scheme which can be quickly deployed and can flexibly increase or reduce the energy storage capacity according to requirements.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage, and particularly relates to a unit combined compressed air energy storage system and method. Background Art

[0002] In recent years, new energy represented by wind power and photovoltaic power has developed rapidly. However, these power generation methods are greatly affected by the environment, sometimes generating more electricity and sometimes less, with unstable power output, posing a hidden danger to the safe operation of the power grid. Energy storage technology can store the electricity generated in excess by power plants during normal times and release it during peak electricity consumption periods. Peak shaving and valley filling: storing energy during power surplus and releasing energy during power shortage to smooth out supply and demand fluctuations. Reserve capacity: enhancing the resilience of the power system to cope with sudden load changes. Low cost and environmental friendliness: the unit energy storage cost is lower than that of lithium batteries, and there is no chemical pollution. Large capacity and long cycle: the single-unit scale can reach the megawatt level, and the energy storage cycle ranges from several hours to several weeks. Summary of the Invention

[0003] The present invention provides a unit combined compressed air energy storage system and method, aiming to provide a compressed air energy storage solution with low cost, rapid deployability, and the ability to flexibly increase or decrease energy storage capacity according to demand to address the instability of new energy power generation.

[0004] One aspect of this specification discloses a unit combined compressed air energy storage system, including: A compression unit; A heat exchanger, whose first channel inlet is connected to the compression unit; An energy storage unit, whose air port is connected to the first channel outlet of the heat exchanger; A heat transfer medium tank, connected to the second channel inlet of the heat exchanger; A heat storage unit, whose first port is connected to the second channel outlet of the heat exchanger, and whose second port is connected to the second channel inlet of the heat exchanger and the heat transfer medium tank; A hydroelectric power generation unit, connected to the water outlet of the energy storage unit and the first channel inlet of the heat exchanger; A water tank, connected to the water inlets of the hydroelectric power generation unit and the energy storage unit; An air pump, one end of which is connected to the first channel outlet of the heat exchanger, and the other end is connected to the air port of the energy storage unit; A water pump, one end of which is connected to the second channel outlet of the heat exchanger, and the other end is connected to the first port of the heat storage unit; Among them, during the energy storage stage, during the compression process in the first channel, the compressed air after temperature rise exchanges heat with the heat exchange medium in the second channel. The compressed air after temperature drop enters the energy storage unit for energy storage, and the heat exchange medium after temperature rise enters the heat storage unit and exchanges heat with the phase change energy storage material filled therein, thereby achieving energy storage. During the energy release stage, the water in the energy storage unit enters the hydraulic power generation unit for power generation to release energy. When the water in the energy storage unit is drained, the air pump is started to transport the compressed air in the energy storage unit to the heat exchanger, and the water pump is started to transport the heat exchange medium in the heat storage unit to the heat exchanger. The compressed air after temperature rise in the heat exchanger enters the hydraulic power generation unit for auxiliary power generation, thereby achieving energy release.

[0005] In this specification, two air valves are provided between the compression unit and the heat exchanger, and two water valves are provided between the energy storage unit and the hydraulic power generation unit. The compression unit, heat exchanger, energy storage unit, heat storage unit, and hydraulic power generation unit constitute an energy storage mechanism. There are multiple such energy storage mechanisms and they are connected between the two air valves and between the two water valves.

[0006] In this specification, the energy storage unit is internally provided with an air bag for storing compressed air, and the space between the air bag and the inner wall of the energy storage unit is for storing water.

[0007] In this specification, the first channel and the second channel in the heat exchanger are arranged at intervals, and the entrances and exits of the first channel and the entrances and exits of the second channel are arranged opposite to each other, so that the compressed air and the heat exchange medium perform convective heat exchange.

[0008] In this specification, manifolds are provided at the entrances and exits of both the first channel and the second channel.

[0009] In this specification, the heat storage unit is internally provided with coiled pipes, and phase change energy storage materials are filled between the coiled pipes and the inner wall of the heat storage unit.

[0010] On the other hand, an embodiment of this specification discloses a unit combined compressed air energy storage method, which applies the unit combined compressed air energy storage system described in any one of the above. The unit combined compressed air energy storage method includes: S1. Data collection and preprocessing: Use sensors to collect the data of the energy storage system itself, external meteorological data, and real-time grid data, and integrate them to form an initial data set; adopt a method combining isolation forest and generative adversarial network to process outliers, use a filling model based on gradient boosting decision tree to process missing values, and use a combination of principal component analysis and mutual information analysis for feature engineering to obtain input feature vectors; S2. Energy storage stage: The compression unit compresses air to generate high-temperature compressed air. The air pump transports the high-temperature compressed air to the airbag of the energy storage unit according to the pressure and storage capacity of the energy storage unit. The water pump pumps the low-temperature deionized water from the heat exchange medium tank into the second channel of the heat exchanger to exchange heat with the compressed air in the first channel. The heated deionized water is pumped to the heat storage unit to exchange heat with the phase change energy storage material, and then pumped back to the heat exchanger after cooling to form a cycle; S3. Construction of the control algorithm framework: Construct a deep Q-network based on the attention mechanism. Its input layer receives the input feature vector, the hidden layer is processed by a multi-layer perceptron and the attention mechanism, and the output layer calculates the Q value corresponding to each action; Design a hierarchical reinforcement learning framework. The high-level policy network outputs the charge and discharge mode decision according to the macroscopic system state, and the low-level policy network outputs the specific device control action according to the high-level decision and the device state; S4. Design of the reward function: Construct a multi-objective optimization reward function, comprehensively consider power balance, energy storage unit stability, energy conversion efficiency, safety, reliability and sustainability indicators to calculate the reward value, and use meta-learning technology to dynamically adjust the reward function weight coefficient and the reward and punishment intensity according to the system operation state and external environment changes; S5. Model training and optimization: Use the Horovod distributed training framework based on the parameter server architecture to perform distributed training on the deep Q-network and the hierarchical reinforcement learning framework; Integrate the deep Q-network and the Advantage Actor-Critic model based on policy gradient, and use transfer learning to optimize the model; Use progressive neural networks for online continuous learning to achieve real-time update of model parameters; S6. Energy release stage: The air pump extracts and pressurizes the compressed air from the airbag of the energy storage unit according to the real-time power demand, and sends it into the first channel of the heat exchanger; The water pump extracts the heated deionized water from the heat storage unit and pumps it into the second channel of the heat exchanger to exchange heat with the compressed air. The heated compressed air enters the hydraulic power generation unit to generate electricity, and the cooled deionized water is pumped back to the heat storage unit; S7. System execution and feedback: Send the control actions output by the low-level policy network to the air pump and the water pump for execution; After execution, collect system data again, calculate the reward value, and store the new state feature vector, execution action, reward value and the state feature vector at the next moment in the experience replay buffer; S8. Online model update: Randomly extract data samples from the experience replay buffer, and perform online updates on the deep Q-network and the hierarchical reinforcement learning framework according to the model training and optimization method, and dynamically adjust the reward function weight coefficient; S9. Continuous monitoring and optimization: Real-time collect system operation data to evaluate system performance, fine-tune system parameters according to the evaluation results, use machine learning algorithms to predict potential device failures, regularly re-evaluate the effectiveness of control strategies, and update control strategies as needed.

[0011] In this specification, the data of the energy storage system itself includes the pressure, water level, and temperature data of the energy storage unit, the flow rate and pressure data of the hydropower unit, and the power and voltage data of the grid access terminal.

[0012] In this specification, in S5, the distributed training framework uses the average gradient aggregation algorithm to calculate the global gradient to update the model parameters.

[0013] In this specification, in S9, the machine learning algorithm is used to analyze the device operation data, and the fault prediction model based on the neural network is used to predict the potential faults of the device.

[0014] The embodiments of this specification can at least achieve the following beneficial effects: 1. The unitized compressed air energy storage system of the present invention adopts a unitized combination, which can be quickly deployed and can flexibly increase or decrease the energy storage capacity by increasing or decreasing some units according to needs.

[0015] 2. The unitized compressed air energy storage method of the present invention improves the energy conversion efficiency, system stability, and flexibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a schematic structural diagram of the unitized compressed air energy storage system involved in some embodiments of the present invention.

[0018] Figure 2 It is a schematic structural diagram of the heat exchanger involved in some embodiments of the present invention.

[0019] Figure 3 It is a schematic structural diagram of the manifold distributed in the heat exchanger involved in some embodiments of the present invention.

[0020] Figure 4 It is a schematic structural diagram of the heat storage unit involved in some embodiments of the present invention.

[0021] Reference numerals: 1. Compression unit; 2. Heat exchanger; 21. First channel; 22. Second channel; 23. Header; 3. Energy storage unit; 4. Heat transfer medium tank; 5. Heat storage unit; 51. Coiled pipe; 52. Phase change energy storage material; 6. Hydroelectric power generation unit; 7. Water tank; 8. Air pump; 81. Air valve; 9. Water pump; 91. Water valve; 10. Valve and / or pump. Detailed implementation manners

[0022] In the following text, only some exemplary embodiments are simply described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the present invention. Therefore, the drawings and the description are considered to be exemplary in nature rather than restrictive.

[0023] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "center", "thickness", "upper", "lower", "front", "rear", "left", "right", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the present invention is usually placed during use, or the orientation or positional relationship commonly understood by those skilled in the art. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention.

[0024] The terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality" means two or more unless otherwise specifically defined.

[0025] In addition, terms such as "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or integrated; it can be directly connected, or indirectly connected through an intermediate medium, and can be the communication inside two elements or the interaction relationship between two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0026] The embodiments of the present invention will be described in detail below with reference to the drawings.

[0027] As Figure 1 shown, this embodiment provides a unitized compressed air energy storage system, including: Compression unit 1; Heat exchanger 2, the inlet of its first channel 21 is connected to the compression unit 1; An energy storage unit 3, whose air port is connected to the outlet of the first channel 21 of the heat exchanger 2; A heat exchange medium tank 4, which is connected to the inlet of the second channel 22 of the heat exchanger 2; A heat storage unit 5, whose first port is connected to the outlet of the second channel 22 of the heat exchanger 2, and whose second port is connected to the inlet of the second channel 22 of the heat exchanger 2 and the heat exchange medium tank 4; A hydraulic power generation unit 6 (existing equipment such as a hydraulic power generation unit), which is connected to the water outlet of the energy storage unit 3 and the inlet of the first channel 21 of the heat exchanger 2; A water tank 7, which is connected to the water inlets of the hydraulic power generation unit 6 and the energy storage unit 3; An air pump 8, one end of which is connected to the outlet of the first channel 21 of the heat exchanger 2, and the other end of which is connected to the air port of the energy storage unit 3; A water pump 9, one end of which is connected to the outlet of the second channel 22 of the heat exchanger 2, and the other end of which is connected to the first port of the heat storage unit 5; Wherein, during the energy storage stage, the compressed air heated during the compression process in the first channel 21 exchanges heat with the heat exchange medium in the second channel 22. The cooled compressed air enters the energy storage unit 3 for energy storage, and the heated heat exchange medium enters the heat storage unit 5 to exchange heat with the phase change energy storage material 52 filled therein, thereby realizing energy storage. During the energy release stage, the water in the energy storage unit 3 enters the hydraulic power generation unit 6 for power generation to release energy. When the water in the energy storage unit 3 is drained, the air pump 8 is started to transport the compressed air in the energy storage unit 3 to the heat exchanger 2, and the water pump 9 is started to transport the heat exchange medium in the heat storage unit 5 to the heat exchanger 2. The compressed air heated in the heat exchanger 2 enters the hydraulic power generation unit 6 for auxiliary power generation, thereby realizing energy release.

[0028] In some embodiments, two air valves 81 are provided between the compression unit 1 and the heat exchanger 2, and two water valves 91 are provided between the energy storage unit 3 and the hydraulic power generation unit 6. The compression unit 1, the heat exchanger 2, the energy storage unit 3, the heat storage unit 5 and the hydraulic power generation unit 6 constitute an energy storage mechanism. There are multiple energy storage mechanisms and they are connected between the two air valves 81 and between the two water valves 91. As Figure 1 shown, it is a connection schematic diagram of 2 energy storage mechanisms. The connection of more than 2 energy storage mechanisms can be made with reference to Figure 1 and the relevant description.

[0029] In some embodiments, valves and / or pumps 10, such as air valve 81, water valve 91, water pump 9, air pump 8, etc., can be added between the corresponding device components (energy storage unit 3, heat exchanger 2, heat exchange medium tank 4, water tank 7, heat storage unit 5, hydroelectric power generation unit, etc.) mentioned in this specification according to actual requirements to control the connection / disconnection between the device components and adjust the corresponding pressure and flow rate. For example, during the energy storage stage, the heat exchange medium is transported from the heat exchange medium tank 4 to the heat exchanger 2 through the water valve 91 and / or the water pump 9; a heat exchange cycle of the heat exchange medium is formed between the heat exchanger 2 and the heat storage unit 5 through the water valve 91 and / or the water pump 9; and so on. By using the valves and / or pumps 10, the compressed air and heat exchange medium circuits during the energy storage / potential energy stage are circulated, and the input and output of water in the energy storage unit are completed.

[0030] In some embodiments, the energy storage unit 3 is internally provided with an airbag for storing compressed air, and the space between the airbag and the inner wall of the energy storage unit 3 is for storing water.

[0031] In some embodiments, as Figure 2 shown, the first channel 21 and the second channel 22 in the heat exchanger 2 are arranged at intervals, and the entrances and exits of the first channel 21 and the entrances and exits of the second channel 22 are arranged opposite to each other, so that the compressed air and the heat exchange medium perform convective heat exchange (the specific convection is shown by the arrows in Figure 2 .

[0032] In some embodiments, as Figure 2 and Figure 3 shown, manifolds 23 are provided at the entrances and exits of the first channel 21 and the entrances and exits of the second channel 22. By providing the manifolds 23, the compressed air can enter and exit multiple first channels 21 more evenly, and the heat exchange medium can enter and exit multiple second channels 22 more evenly.

[0033] In some embodiments, as Figure 4 shown, the heat storage unit 5 is internally provided with a coil 51, and a phase change energy storage material 52 is filled between the coil 51 and the inner wall of the heat storage unit 5.

[0034] This embodiment provides a unit combined compressed air energy storage method, which applies the unit combined compressed air energy storage system described in any one of the above. The unit combined compressed air energy storage method includes: S1. Data collection and preprocessing; S2. Energy storage stage; S3. Control algorithm framework construction; S4. Reward function design; S5. Model training and optimization; S6. Energy release stage; S7. System execution and feedback; S8. Model online update; S9. Continuous monitoring and optimization.

[0035] In some embodiments, S1. Data collection and preprocessing includes: Data collection: Use various sensors to collect data of the energy storage system itself, such as pressure sensor data (water pressure, airbag air pressure), water level sensor data, and temperature sensor data (compressed air temperature of the airbag) of the energy storage unit 3; flow sensor data and pressure sensor data of the hydropower unit 6; power sensor data and voltage sensor data at the grid access end. At the same time, obtain external relevant data, such as temperature, humidity, and light intensity in meteorological data (obtained from the meteorological data interface), and real-time grid electricity price data (obtained from the grid operation platform). Integrate these data to form the initial dataset D.

[0036] Outlier processing: Adopt a method combining Isolation Forest and Generative Adversarial Network (GAN) to process outliers. First, use the Isolation Forest algorithm to detect outliers in the dataset D and find out the abnormal data points. For these abnormal data points, repair them with the trained GAN. The GAN consists of a generator G and a discriminator D. The generator G takes the random noise z as input, learns the normal data distribution, and generates similar normal data G(z). The discriminator D makes true or false judgments on the input data (including real normal data and data generated by the generator). Through adversarial training, the data generated by the generator is made closer to the real normal data. Input the abnormal data points into the generator to obtain the repaired data, and replace the outliers in the original dataset to obtain the processed dataset D1.

[0037] Missing value filling: Based on the processed dataset D1, use the complete feature data as the training set to train the filling model based on Gradient Boosting Decision Tree (GBDT). For data samples with missing values, input them into the trained GBDT model for prediction and filling. The GBDT model predicts the filling values of the missing values by learning the relationship between other features and the feature where the missing values are located, and obtains the filled dataset D2.

[0038] Feature engineering: Use a method combining Principal Component Analysis (PCA) and Mutual Information Analysis for feature engineering. First, use PCA to perform dimensionality reduction on the dataset D2. PCA determines the principal component directions of the data by calculating the covariance matrix of the data, projects the high-dimensional data into a low-dimensional space, and removes redundant information. The feature matrix of the dataset D2 is transformed by PCA to obtain a low-dimensional feature matrix. Then, use Mutual Information Analysis to calculate the mutual information between each low-dimensional feature and the target variable (such as the control decision variable of the energy storage system), and select the features with higher mutual information to construct the final input feature vector.

[0039] In some embodiments, S2. The energy storage stage includes: Air Compression and Transportation: The compression unit 1 starts to work, compressing air to generate high-temperature compressed air. The air pump 8 starts and extracts the high-temperature compressed air from the compression unit 1. Based on the pressure of the energy storage unit 3, the current storage capacity, and the energy required to be stored, the air pump 8 monitors the output in real time through the built-in pressure sensor and flow sensor, and automatically adjusts its own rotation speed and power to accurately control the output air pressure and flow rate, and directly transports the compressed air into the airbag of the energy storage unit 3.

[0040] Heat Exchange and Water Transportation: The high-temperature compressed air enters the first channel 21 of the heat exchanger 2. The water pump 9 pumps the low-temperature heat exchange medium (deionized water, softened water, propylene glycol aqueous solution, mineral oil, silicone fluid) from the heat exchange medium tank 4 into the second channel 22 of the heat exchanger 2. The water pump 9 automatically adjusts the flow rate and flow velocity of the deionized water according to the temperature difference (the difference between the high-temperature compressed air temperature and the low-temperature deionized water temperature) monitored by the temperature sensor in the heat exchanger 2 and the preset heat exchange efficiency target. The deionized water absorbs the heat of the high-temperature compressed air in the heat exchanger 2, cools the compressed air and then enters the energy storage unit 3 to improve the energy storage efficiency. The deionized water that has absorbed heat and increased in temperature is pumped out of the heat exchanger 2 by the water pump 9 and directly pumped to the heat storage unit 5. During the transportation process, the water pump 9 automatically adjusts the water delivery volume and pressure according to the temperature sensor data, pressure sensor data, and water level sensor data in the heat storage unit 5 to ensure that the deionized water efficiently exchanges heat with the phase change energy storage material 52 in the heat storage unit 5. Heat Storage: In the heat storage unit 5, the deionized water exchanges heat with the phase change energy storage material 52 (such as molten salt, sodium acetate trihydrate, hydrated salt, paraffin, fatty acids and their esters and other organic or inorganic composite phase change energy storage materials) through the coil 51. Since the temperature of the deionized water is higher than the phase change temperature of the phase change material, the phase change material absorbs heat and changes from a solid state to a liquid state, thereby storing the heat. The deionized water that has completed heat exchange and cooled down is pumped back to the second channel 22 of the heat exchanger 2 by the water pump 9 again to form a cycle.

[0041] In some embodiments, S3. The construction of the control algorithm framework includes: Construction of the Deep Q-Network based on Attention Mechanism (A-DQN): Build the A-DQN network structure, which includes an input layer, multiple hidden layers, and an output layer. The input layer receives the input feature vector F after feature engineering processing. The hidden layer adopts a multi-layer perceptron (MLP) structure, and feature transformation is performed through a non-linear activation function (such as the ReLU function). An attention mechanism module is introduced between the hidden layers, and this module calculates the attention weights according to the importance of the input features. Assuming the input of the hidden layer is H, the attention mechanism module calculates to obtain the attention weight vector α, then the output after being processed by the attention mechanism is , where "·" represents element-wise multiplication. The output layer calculates the Q-value corresponding to each possible action (such as adjusting the operating parameters of the air pump 8 and the water pump 9, regulating the inlet and outlet valves of the energy storage unit 3 for gas and water, etc.) based on the output of the hidden layer, obtaining the Q-value vector Q.

[0042] Hierarchical Reinforcement Learning (HRL) architecture design: The control task is divided into two levels: a high-level policy and a low-level policy. The high-level policy network receives macroscopic system state information, such as the peak-valley electricity price of the power grid, the predicted trend of power grid load changes (which can be obtained through a time series prediction model), and the overall power state of the energy storage system, etc., and outputs a macroscopic charge-discharge mode decision M, such as "charge first", "discharge first", "balanced charge and discharge", etc. The low-level policy network outputs specific device control actions A, such as the rotational speed adjustment value of the air pump 8, the flow rate adjustment value of the water pump 9, etc., according to the decision M of the high-level policy and the specific device state information (such as the real-time pressure, water level, etc. of the energy storage unit 3).

[0043] In some embodiments, S4. Reward function design includes: Construction of a multi-objective optimization reward function: Construct a reward function R based on multi-objective optimization, comprehensively considering multiple performance indicators of the energy storage system.

[0044] ; where is the weight coefficient of each objective, determined by experience and experiments, and .

[0045] Power balance reward : Measures the matching degree between the system output power and the power grid load demand.

[0046] ; is the actual power of the power grid, is the power grid load demand power. When the system output power is closer to the load demand, is closer to 0, and the higher the reward.

[0047] Stability reward of the energy storage unit 3 : Calculated according to the degree of deviation of the pressure and water level of the energy storage unit 3 from the normal range.

[0048] ; is the current pressure of the energy storage unit 3, is the optimal pressure value, is the maximum allowable pressure value; is the current water level of the energy storage unit 3, is the optimal water level value, is the maximum allowable water level value. The closer the state of the energy storage unit 3 is to the optimal state, the higher it is.

[0049] Energy conversion efficiency reward : Calculated according to the energy conversion efficiency of the energy storage system.

[0050] ; is the current energy conversion efficiency, is the minimum acceptable efficiency, is the maximum theoretical efficiency. The higher the efficiency, the higher it is.

[0051] Safety reward : When the system detects possible safety risks (such as , is the upper limit of safety pressure, is the upper limit of safety temperature, is the current temperature of the energy storage unit 3), and effective measures are taken in time to eliminate the risks, a reward is given, such as ; otherwise it is 0.

[0052] Reliability reward : Evaluated according to the historical failure data and operation stability of the energy storage system. When the system operates stably within a certain time t without failures, ; is the set maximum evaluation time.

[0053] Sustainability reward : Considering the service life of the energy storage device and its impact on the environment, if the system extends the service life of the device through reasonable control strategies (such as reducing the device wear rate , reducing the negative impact on the environment (such as reducing carbon emissions ), then , and are the corresponding weight coefficients.

[0054] Dynamic reward function adjustment mechanism: According to the operating state of the system and changes in the external environment, the weight coefficients of the reward function and the intensity of rewards / punishments are adjusted in real time. For example, during the peak grid load period, increase (the power balance reward weight) to encourage the system to give priority to ensuring power supply to the grid; when the power of the energy storage unit 3 is low, appropriately reduce the punishment intensity for the discharge power (adjust (i.e., the coefficients of relevant penalty terms in the formula). Meanwhile, combining the meta-learning technology in reinforcement learning, the intelligent control system can automatically learn how to adjust the reward function under different working conditions. By setting up a meta-learner, taking the operation state information of the system (such as the current power balance state, the power of energy storage unit 3, etc.) as input, and outputting the adjusted weight coefficients , which are used to update the reward function.

[0055] In some embodiments, S5. Model training and optimization includes: Distributed training acceleration: Adopt the Horovod distributed training framework based on the parameter server architecture. Suppose there are n computing nodes, and each computing node has a local dataset , and these datasets are partitioned from the overall dataset. On each computing node, use the local dataset to train the A-DQN and HRL models and calculate the gradients of the models . The computing nodes upload the gradients to the parameter server. The parameter server calculates the global gradient using an aggregation algorithm (such as average gradient aggregation) based on the gradients uploaded by all computing nodes , and updates the model parameters. The updated model parameters are then sent down to each computing node, and the computing nodes continue the next round of training based on the updated parameters, and iterate continuously until the model converges.

[0056] Model fusion and transfer learning: During the training process, adopt the model fusion technology to combine the A-DQN and the A2C (Advantage Actor-Critic) model based on policy gradients. The A2C model consists of an actor network and a critic network. The actor network outputs action policies, and the critic network evaluates the action values. During training, train the A-DQN and A2C models simultaneously, and the Q-value estimation of the A-DQN and the policy optimization of the A2C complement each other. For example, during the update process of the actor network of the A2C, refer to the Q-value information of the A-DQN to make the action policy more in line with the optimal decision; during the training of the A-DQN, utilize the policy gradient information of the A2C to accelerate the model convergence speed.

[0057] Meanwhile, transfer learning is introduced. For similar energy storage system scenarios (such as energy storage power stations with different scales but similar structures), the model parameters trained in one scenario are used as initialization parameters and applied to the model training in the new scenario. In the new scenario, a small amount of data in the new scenario is used to fine-tune the model, enabling the model to adapt to the characteristics of the new scenario. For example, in the new scenario, according to the differences in the capacity of energy storage unit 3 and equipment performance in the new scenario, the parts of the model related to these parameters are adjusted. For example, the coefficient related to the energy storage capacity in the reward function is adjusted, and the model is retrained to improve the adaptability and generalization ability of the model in the new scenario.

[0058] Online continuous learning and evolution of the model: Online continuous learning technology is adopted to enable the model to continuously receive new real-time data during operation and update the model parameters in real time. Taking the Progressive Neural Network (PNN) as an example, the PNN consists of multiple sub-networks, and each sub-network corresponds to a learning stage. When new data arrives, first, the difference in the data distribution between the new data and the data learned by the existing sub-networks is judged. If the difference is large, a new sub-network is created to learn the new data; if the difference is small, the parameters are updated on the existing sub-networks. During the update process, an incremental learning algorithm is adopted to avoid the learning of new data from covering old knowledge. For example, when new equipment is connected to the energy storage system or a new operation mode appears, the new data triggers the creation of a new sub-network or the update of the existing sub-networks, enabling the model to learn these new changes in a timely manner, adjust the control strategy, and always maintain the best control effect on the energy storage system.

[0059] In some embodiments, S6. The energy release stage includes: Compressed air extraction and pressurization: When the system receives the energy release instruction, the air pump 8 extracts compressed air from the airbag of the energy storage unit 3. As the compressed air in the energy storage unit 3 is discharged, the pressure gradually decreases, and the pressure sensor built in the air pump 8 monitors the pressure of the extracted compressed air in real time. The air pump 8 automatically adjusts its rotation speed and power according to the real-time power demand (this demand can be obtained according to the grid load situation and the set target of the power generation system), and pressurizes the extracted low-pressure compressed air to make the compressed air reach the pressure conditions required by the subsequent power generation equipment.

[0060] Heat Exchange and Power Generation: The pressurized compressed air enters the first channel 21 of the heat exchanger 2. The water pump 9 pumps the deionized water that has been heated and raised in temperature by the phase change material from the heat storage unit 5 into the second channel 22 of the heat exchanger 2. The water pump 9 automatically adjusts the flow rate and velocity of the deionized water according to the temperature difference monitored by the temperature sensor in the heat exchanger 2 (the difference between the temperature of the high-temperature side compressed air and the temperature of the low-temperature side deionized water), as well as the preset heat exchange efficiency target. The deionized water transfers heat to the low-temperature compressed air in the heat exchanger 2, raising the temperature of the compressed air and enhancing its work capacity. The heated compressed air enters the hydraulic power generation unit 6 for power generation. The deionized water that has completed heat exchange and cooled down is pumped back into the heat storage unit 5 by the water pump 9 again, where it exchanges heat with the phase change energy storage material 52, absorbs heat and raises the temperature, preparing for the next heat exchange. During the entire energy release process, the air pump 8 and the water pump 9 work together. The air pump 8 continuously provides compressed air with stable pressure and flow rate to the hydraulic power generation unit 6, and the water pump 9 ensures the efficient circulation of deionized water between the heat storage unit 5 and the heat exchanger 2, realizing the stable release and conversion of energy.

[0061] Among them, during the hydraulic power generation process, compared with low-temperature compressed air, high-temperature compressed air has significant advantages, mainly reflected in aspects such as enhancing work capacity, optimizing heat exchange efficiency, and improving the working conditions of the water turbine.

[0062] Enhancing Work Capacity: High-temperature compressed air contains more internal energy. When it is introduced into the hydraulic power generation unit, it can provide a stronger driving force for the water flow. According to the ideal gas state equation (where P is pressure, V is volume, n is the amount of substance, R is the molar gas constant, and T is temperature), under the condition of the same volume and amount of substance, the higher the temperature T, the greater the gas pressure P. This means that high-temperature compressed air has a higher pressure. After entering the hydraulic power generation unit, it can exert a greater pressure on the water flow, pushing the water turbine to rotate faster and more efficiently, thereby increasing the power generation power of the generator and the power generation amount. For example, within the same time, high-temperature compressed air can increase the rotation speed of the water turbine by 10% - 20%, and the power generation power increases accordingly.

[0063] Optimizing Heat Exchange Efficiency: In the entire energy storage and power generation system, heat exchange is an important link. High-temperature compressed air itself has a higher temperature, and when it exchanges heat with other media (such as water or other cooling media in the power generation system), it has a larger temperature difference. According to the heat transfer principle, the larger the temperature difference, the higher the rate and efficiency of heat exchange. This enables high-temperature compressed air to transfer its heat to the surrounding media more quickly and fully. On the one hand, it can improve the working efficiency of other components in the power generation system, such as preheating related equipment and reducing energy loss; on the other hand, efficient heat exchange helps to maintain the temperature balance within the system, ensure the stable operation of the system, and reduce the risk of equipment failure caused by temperature changes.

[0064] Improve the working conditions of the water turbine: After the high-temperature compressed air enters the hydroelectric power unit, it can improve the working environment of the water turbine to a certain extent. On the one hand, the high-temperature compressed air can increase the temperature inside the water turbine, reduce the condensation of water vapor on the surface of the water turbine blades, avoid blade corrosion and erosion problems caused by condensed water, and extend the service life of the water turbine. On the other hand, due to the relatively high pressure of the high-temperature compressed air, it can form a certain pressure environment inside the water turbine, which helps to reduce the cavitation phenomenon generated during the operation of the water turbine. The cavitation phenomenon will cause serious damage to the water turbine blades, and the presence of high-temperature compressed air can effectively reduce the occurrence probability of cavitation, ensure the stable operation of the water turbine, and improve the reliability of the power generation system.

[0065] In some embodiments, S7. System execution and feedback include: Execute control actions: The specific device control actions A output by the low-level policy network (such as the rotation speed adjustment value of the air pump 8, the flow adjustment value of the water pump 9, etc.) are sent to the corresponding devices (the air pump 8 and the water pump 9). The air pump 8 and the water pump 9 adjust their working parameters according to the received control instructions, so as to achieve precise control of the energy storage system and the power generation system.

[0066] Data collection and feedback: After the system executes the control actions, various sensors are used again to collect the data of the energy storage system itself, such as the pressure sensor data, water level sensor data, and temperature sensor data of the energy storage unit 3; the flow sensor data and pressure sensor data of the hydroelectric power unit; the power sensor data and voltage sensor data of the grid access end, etc. At the same time, external relevant data is obtained, such as the temperature, humidity, and light intensity in the meteorological data (obtained from the meteorological data interface), and the real-time grid electricity price data (obtained from the grid operation platform). These data are integrated to form a new data set.

[0067] Reward value calculation and storage: According to the newly collected data, calculate the reward value at this moment according to the calculation method of the reward function R. Store the new state feature vector (the feature vector obtained from the newly collected data through the same data preprocessing and feature engineering as in S1), the executed action A, the obtained reward value, and the state feature vector at the next moment (the feature vector corresponding to the data collected and processed again) into the experience replay buffer for subsequent model update training.

[0068] In some embodiments, S8. Model online update includes: Sample extraction: Randomly extract a batch of data samples from the experience replay buffer. These samples include state feature vectors, executed actions, obtained reward values, and state feature vectors at the next moment.

[0069] Model Update: Using these sample data, the deep Q-network with attention mechanism (A-DQN) and hierarchical reinforcement learning (HRL) model are updated online according to the methods in the model training and optimization (S5) phase. For example, the parameters of the A-DQN network are updated through the backpropagation algorithm, and the weights of neurons in each layer are adjusted to optimize the model's estimation of state-action value; at the same time, according to the policy gradient algorithm, the parameters of the high-level and low-level policy networks in the HRL model are updated to enable the model to better adapt to the dynamic changes of the system and continuously improve the control performance. During the update process, combined with the dynamic reward function adjustment mechanism, according to the real-time operating state of the system and changes in the external environment, the weight coefficients of the reward function are dynamically adjusted to guide the model to learn better control strategies.

[0070] In some embodiments, S9. Continuous Monitoring and Optimization includes: Real-time Monitoring: During the continuous operation of the system, various sensors collect data in real time, including the pressure, water level, and temperature of the energy storage unit 3, the flow rate and pressure of the hydroelectric power generation unit, the power and voltage at the grid access end, as well as external meteorological data and real-time electricity prices of the grid. These data are continuously transmitted to the data processing center for real-time evaluation of the system's operating state.

[0071] Performance Evaluation: According to the real-time data collected, the system is evaluated according to a preset performance index system. In addition to the power balance, stability of the energy storage unit 3, energy conversion efficiency, safety, reliability, and sustainability indicators involved in the reward function in S4, the overall operating cost of the system is also calculated, including equipment energy consumption cost, maintenance cost, etc. By comparing the actual operating data with the ideal performance indicators, the current advantages and existing problems of the system are determined. For example, if the energy conversion efficiency is lower than expected, analyze whether there are problems in the heat exchange link, compressed air transmission link, or power generation link.

[0072] Parameter Fine-tuning and Optimization: Based on the performance evaluation results, the key parameters of the system are fine-tuned. If it is found that the large pressure fluctuation of the energy storage unit 3 affects stability, by adjusting the working parameters (such as rotation speed, flow rate, etc.) of the air pump 8 during the energy storage and energy release stages, the pressure is maintained within a better range. If the heat exchange efficiency decreases, fine-tune the flow rate of the water pump 9 to optimize the flow rate of deionized water in the heat exchanger 2 and improve the heat exchange effect. These parameter adjustment instructions are sent by the intelligent control system to the corresponding equipment for execution.

[0073] Equipment Maintenance Prediction: Analyze the operation data of equipment using machine learning algorithms to predict potential equipment failures. For example, by monitoring the vibration data, temperature data, and operating duration of air pump 8 and water pump 9, and adopting a fault prediction model based on neural networks, determine in advance whether the equipment is likely to fail. If it is predicted that air pump 8 has a high risk of failure in the future, arrange a maintenance plan in advance, prepare corresponding spare parts, reduce the impact of sudden equipment failures on system operation, and ensure the reliability and stability of the system.

[0074] Strategy Update: As the system operating environment changes (such as changes in grid load characteristics, gradual degradation of energy storage device performance, etc.), the intelligent control system regularly re-evaluates the effectiveness of control strategies. If the current strategy cannot meet the system performance requirements, re-build the control algorithm framework in S3 and design the reward function in S4, train the model using the latest operation data, and generate a control strategy that is more adaptable to the new environment to ensure that the system always maintains an efficient operating state.

[0075] In some embodiments, accurately model the energy conservation of the heat exchange process (i.e., add heat management content in S2 / S6): Heat Exchange Equation during Energy Storage Phase: ; Where, : Compressed air mass flow rate (kg / s), : Specific heat capacity of air at constant pressure (kJ / (kg・K)), / : Air temperature at the inlet / outlet of heat exchanger 2 (K), : Deionized water mass flow rate (kg / s), : Specific heat capacity of water at constant pressure (kJ / (kg・K)), / : Water temperature at the inlet / outlet of heat exchanger 2 (K), : Heat dissipation loss of heat exchanger 2 (kJ / s), through equipment thermal resistance Calculated as: ( is the ambient temperature).

[0076] Phase Change Energy Storage Equation of Heat Storage Unit 5: ; During the energy storage phase, when the phase change material absorbs heat, if it is in the solid state and heating up, the sensible heat part is ; if it has reached the liquid state and continues to heat up, the sensible heat part is ; During the energy release phase, when the phase change material releases heat, if it is in the liquid state and cooling down, the sensible heat part is ; if it has solidified into a solid state and continues to cool down, the sensible heat part is ; By clarifying the sensible heat calculation methods in different stages and states, the thermal change law of the phase change material during the energy storage and release processes is clearly presented; Among them, : mass of the phase change material (kg), / : specific heat capacity of the phase change material in solid / liquid state (kJ / (kg·K)), : solid state temperature of the phase change material, : liquid state temperature of the phase change material, : phase change temperature of the phase change material (K), : latent heat of phase change (kJ / kg). The left side of the equation is the heat absorption / release amount of the phase change material, and the right side is the heat transfer amount of deionized water. The essence of the equation is sensible heat change + latent heat change = heat transfer amount of deionized water.

[0077] The role after modeling lies in: 1. Core role of the heat exchange equation in the energy storage stage (1) Precise quantification of heat exchange efficiency to guide real-time control Model input-output relationship: The left side of the equation is the sensible heat released by compressed air (high temperature → low temperature), and the right side is the sum of the sensible heat absorbed by deionized water and heat loss. By measuring in real time , , , (sensor data from S1 data acquisition), the actual heat exchange amount can be calculated and compared with the theoretical value.

[0078] Control application: If the measured heat exchange amount is lower than the theoretical value, it indicates that the heat loss is too large or the flow rate of water pump 9 is insufficient. The control system (S3 / S8) can automatically increase the rotation speed of water pump 9 (adjust ) or optimize the thermal insulation of heat exchanger 2 (reduce ) to ensure effective cooling of compressed air (improve energy storage efficiency).

[0079] (2) Optimization of heat exchanger 2 design and selection parameter sensitivity analysis: Through , the influence of thermal resistance on heat loss can be quantified. For example, if the environmental temperature is relatively low (such as in winter), a heat exchanger 2 material with a lower thermal resistance needs to be selected (such as increasing the thickness of the thermal insulation layer) to reduce and avoid heat waste.

[0080] Flow rate matching calculation: According to , the minimum deionized water flow rate required during the energy storage stage can be pre-calculated , ensuring that the selection of the water pump 9 (such as head and power) meets the actual requirements and avoiding energy consumption waste like using a sledgehammer to crack a nut.

[0081] (3) Provide physical constraints for the energy flow during the energy storage stage Data closed-loop verification: During the S2 energy storage stage, the temperature rise value of the deionized water calculated by the model ( ) can be used as a feedback signal to verify the rationality of the sensor data (such as whether there is abnormal temperature rise caused by sensor failure). If the deviation between the measured value and the model calculation value exceeds the threshold (such as ±5%), trigger the outlier detection mechanism in S1 to repair or calibrate the data again.

[0082] 2. The core role of the phase change energy storage equation of the heat storage unit 5 (1) Accurately simulate the energy storage / release process of the phase change material Quantification of the phase change process: The left side of the equation describes the "sensible heat change + latent heat change" of the phase change material: During the energy storage stage: the phase change material changes from solid state to liquid state, absorbing sensible heat + latent heat; during the energy release stage: the phase change material changes from liquid state to solid state, releasing heat.

[0083] By real-time monitoring the temperature of the heat storage unit 5 (from the temperature sensor in S1), the energy storage state of the current phase change material (solid proportion / liquid proportion) can be accurately calculated, avoiding the traditional "one-size-fits-all" control (such as judging whether to charge or discharge only based on temperature).

[0084] (2) Guide the heat release strategy during the energy release stage Dynamic adjustment of the energy release power: During the S6 energy release stage, calculate the remaining available latent heat according to the phase change equation , combined with the requirements of the power generation system , dynamically adjust the flow rate of the water pump 9 : if the remaining latent heat is insufficient, increase to accelerate heat release (sacrificing part of the heat exchange efficiency to give priority to ensuring the power generation power); if the remaining latent heat is sufficient, reduce to reduce the energy consumption of the water pump 9 (optimize the overall efficiency of the system).

[0085] Avoid supercooling / superheating of the phase change material: By restricting (ΔT is the safe temperature range), prevent the performance decay of the phase change material caused by excessive temperature fluctuations (such as long-term over-temperature decomposition of molten salt), and extend the life of the heat storage unit 5.

[0086] (3) Linked with the heat exchange equation to form a closed loop of “energy storage-heat storage” energy flow Cross-stage energy tracing: The total heat absorbed by deionized water during the energy storage stage (from the right side of the heat exchange equation) should be equal to the total heat stored by the phase change material of the heat storage unit 5 (on the right side of the phase change equation). The deviation between the two can reflect pipeline heat loss or measurement error, which can be used for system fault diagnosis (such as whether there is a reduction in water volume due to pipeline leakage).

[0087] Multiphysics coupling optimization: Combining the heat exchange equation and the phase transition equation The control algorithm (S3 / S8) can construct a three-dimensional control space of "flow-temperature-phase change state". For example, in a high temperature environment, the Improve the heat storage speed by reducing Reduce the risk of pipe freezing and cracking.

[0088] Improvement of the overall system after modeling (combined with S1-S9 process): 1. Deep integration of data-driven and physical models S1. Data preprocessing: The temperature and flow data collected by the sensor are directly used as the input of the heat exchange / phase change equation. After calculation by the model, derived features such as "effective heat exchange amount" and "phase change material state" are generated and added to the input feature vector F (replacing the original single sensor data) to improve the state characterization accuracy of A-DQN / GNN.

[0089] S3. Control algorithm: When formulating the "charging and discharging mode decision M", the high-level strategy network (HRL) can predict the remaining capacity of the heat storage unit 5 through the phase change equation to avoid a sudden drop in power during the energy release phase due to insufficient heat storage; the low-level strategy network adjusts the water pump 9 and the air pump 8 in real time according to the heat exchange equation, so that the control action is more in line with physical laws (rather than relying solely on a data-driven black box model).

[0090] 2. Quantitative improvement of energy conversion efficiency Reduce heat loss: Accurate calculation through heat exchange equations The system can dynamically optimize the insulation measures of heat exchanger 2 (such as starting the electric heating tape to prevent condensation on the pipes in low temperature environments), and the heat loss is expected to be reduced by 10%-15%.

[0091] Improved utilization of phase change materials: The phase change equation ensures that the heat of the phase change material is fully utilized during the solid / liquid conversion, avoiding the waste of latent heat caused by extensive temperature control in traditional control. The effective energy storage density of the heat storage unit 5 can be increased by 8%-12%.

[0092] 3. Enhancement of System Stability and Reliability Clarification of safety boundaries: In the heat exchange equation it is necessary to satisfy the safe temperature of the energy storage unit 3 ( ), and T in the phase change equation should be within the safe range of the material, providing hard constraint conditions for the control algorithm to prevent equipment damage caused by overheating.

[0093] Advance fault prediction: If the value calculated by the heat exchange equation continuously exceeds the rated flow rate of the water pump 9, and does not meet the expectation, it can be predicted that the heat exchanger 2 is fouled (resulting in a decrease in the heat transfer coefficient), triggering the equipment maintenance prediction mechanism of S9 and arranging a cleaning plan in advance.

[0094] In some embodiments, an energy correction model considering the humidity of compressed air (improving the power calculation in the energy release stage): Humidity correction factor ; Where: : Actual water vapor density of compressed air (kg / m³), : Saturated water vapor density at the current temperature (kg / m³), calculated by the Antoine equation: ; A, B, C are Antoine constants, determined according to water quality parameters; Corrected power generation ; Where: : Efficiency of the water turbine / generator, : Volume flow rate of compressed air (m³ / s), : Pressure of compressed air (Pa), : Expansion ratio, ( / , is the atmospheric pressure), : Air adiabatic index. It can quantify the influence of humidity on the work capacity of compressed air, accurately reflect the actual work capacity, avoid the deviation between theory and measurement, guide the optimization of the control strategy in the energy release stage (such as adaptive humidity compensation, equipment protection), and inversely guide the energy storage link (insufficient drying of compressed air during the energy storage stage, dehumidification is required); in the energy release stage of S6, when the heat exchange equation increases the temperature of compressed air , the Antoine equation synchronously calculates the saturated vapor density to increase, making increase, forming a positive cycle of "heating → increase in humidity correction factor → increase in power generation". For example: when compressed air is heated from 50°C to 100°C, increases from 8.3 kg / m³ to 59.5 kg / m³. If the actual , then it is increased from 0.39 to 0.92, and the corrected power generation power increases by 133%.

[0095] In summary, multiple specific embodiments of the present invention are disclosed. Without self - contradiction, each embodiment can be freely combined to form a new embodiment, that is, the embodiments belonging to the alternative solutions can be freely replaced with each other, but cannot be combined with each other; the embodiments that do not belong to the alternative solutions can be combined with each other, and these new embodiments also belong to the substantial content of the present invention.

[0096] The above embodiments describe multiple specific implementation manners of the present invention. However, those skilled in the art should understand that, without departing from the principles and essence of the present invention, various changes or modifications can be made to these implementation manners, but these changes and modifications all fall within the protection scope of the present invention.

Claims

1. A unitized compressed air energy storage system, characterized in that, Comprising: A compression unit; A heat exchanger, the inlet of its first channel being connected to the compression unit; An energy storage unit, the air port thereof being connected to the outlet of the first channel of the heat exchanger; A heat exchange medium tank, connected to the inlet of the second channel of the heat exchanger; A heat storage unit, the first port thereof being connected to the outlet of the second channel of the heat exchanger, and the second port thereof being connected to the inlet of the second channel of the heat exchanger and the heat exchange medium tank; A hydropower unit, connected to the water outlet of the energy storage unit and the inlet of the first channel of the heat exchanger; A water tank, connected to the water inlets of the hydropower unit and the energy storage unit; An air pump, one end being connected to the outlet of the first channel of the heat exchanger, and the other end being connected to the air port of the energy storage unit; A water pump, one end being connected to the outlet of the second channel of the heat exchanger, and the other end being connected to the first port of the heat storage unit; Wherein, during the energy storage stage, the compressed air heated during the compression process in the first channel exchanges heat with the heat exchange medium in the second channel, and the cooled compressed air enters the energy storage unit for energy storage, and the heated heat exchange medium enters the heat storage unit to exchange heat with the phase change energy storage material filled therein, thereby realizing energy storage; during the energy release stage, the water in the energy storage unit enters the hydropower unit to generate electricity for energy release, and when the water in the energy storage unit is drained, the air pump is started to transport the compressed air in the energy storage unit to the heat exchanger, and the water pump is started to transport the heat exchange medium in the heat storage unit to the heat exchanger, and the compressed air heated in the heat exchanger enters the hydropower unit for auxiliary power generation, thereby realizing energy release.

2. The modular compressed air energy storage system according to claim 1, wherein, Two air valves are provided between the compression unit and the heat exchanger, and two water valves are provided between the energy storage unit and the hydropower unit. The compression unit, the heat exchanger, the energy storage unit, the heat storage unit and the hydropower unit constitute an energy storage mechanism, and there are multiple energy storage mechanisms connected between the two air valves and between the two water valves.

3. The unitized compressed air energy storage system according to claim 1, wherein The energy storage unit is internally provided with an airbag for storing compressed air, and the space between the airbag and the inner wall of the energy storage unit is for storing water.

4. The unitized compressed air energy storage system according to claim 1, wherein The first channel and the second channel in the heat exchanger are arranged at intervals, and the inlets and outlets of the first channel and the inlets and outlets of the second channel are arranged opposite to each other, so that the compressed air and the heat exchange medium perform convective heat exchange.

5. The unitized compressed air energy storage system according to claim 1, characterized in that, Manifolds are provided at the inlets and outlets of both the first channel and the second channel.

6. The unitized compressed air energy storage system according to claim 1, wherein The heat storage unit is internally provided with a coil, and phase change energy storage material is filled between the coil and the inner wall of the heat storage unit.

7. A unitized compressed air energy storage method, characterized in that, Applying the unit combined compressed air energy storage system according to any one of claims 1 to 6, the unit combined compressed air energy storage method comprises: S1. Data acquisition and preprocessing: Using sensors to collect the data of the energy storage system itself, external meteorological data and real-time grid data, and integrating them to form an initial data set; adopting a method combining isolation forest and generative adversarial network to process outliers, using a filling model based on gradient boosting decision tree to process missing values, and performing feature engineering by combining principal component analysis and mutual information analysis to obtain an input feature vector; S2. Energy storage stage: The compression unit compresses air to generate high-temperature compressed air. The air pump transports the high-temperature compressed air to the airbag of the energy storage unit according to the pressure and storage capacity of the energy storage unit. The water pump pumps the low-temperature heat exchange medium from the heat exchange medium tank into the second channel of the heat exchanger to conduct heat exchange with the compressed air in the first channel. The heated heat exchange medium is pumped to the heat storage unit to conduct heat exchange with the phase change energy storage material, and then pumped back to the heat exchanger after cooling to form a cycle; S3. Control algorithm architecture construction: Construct a deep Q network based on the attention mechanism. Its input layer receives the input feature vector, the hidden layer is processed by a multi-layer perceptron and the attention mechanism, and the output layer calculates the Q value corresponding to each action. Design a hierarchical reinforcement learning architecture. The high-level policy network outputs the charge and discharge mode decision according to the macroscopic system state, and the low-level policy network outputs the specific device control actions according to the high-level decision and the device state; S4. Reward function design: Construct a multi-objective optimization reward function, comprehensively consider power balance, energy storage unit stability, energy conversion efficiency, safety, reliability and sustainability indicators to calculate the reward value, and use meta-learning technology to dynamically adjust the reward function weight coefficient and the reward and punishment intensity according to the system operation state and external environment changes; S5. Model training and optimization: Adopt the Horovod distributed training framework based on the parameter server architecture to perform distributed training on the deep Q network and the hierarchical reinforcement learning architecture; fuse the deep Q network and the Advantage Actor-Critic model based on policy gradient, and use transfer learning to optimize the model; adopt progressive neural network for online continuous learning to realize real-time update of model parameters; S6. Energy release stage: The air pump extracts and pressurizes the compressed air from the airbag of the energy storage unit according to the real-time power demand and sends it into the first channel of the heat exchanger; the water pump extracts the heated heat exchange medium from the heat storage unit and pumps it into the second channel of the heat exchanger to conduct heat exchange with the compressed air. The heated compressed air enters the hydraulic power generation unit to generate electricity, and the cooled heat exchange medium is pumped back to the heat storage unit; S7. System execution and feedback: Send the control actions output by the low-level policy network to the air pump and the water pump for execution; after execution, collect system data again, calculate the reward value, and store the new state feature vector, executed actions, reward value and the state feature vector at the next moment in the experience replay buffer; S8. Model online update: Randomly extract data samples from the experience replay buffer, and perform online update on the deep Q network and the hierarchical reinforcement learning architecture according to the model training and optimization method, and dynamically adjust the reward function weight coefficient; S9. Continuous monitoring and optimization: Continuously collect system operation data to evaluate system performance, fine-tune system parameters according to the evaluation results, use machine learning algorithms to predict potential device failures, regularly re-evaluate the effectiveness of control strategies, and update control strategies as needed.

8. The unitized compressed air energy storage method according to claim 7, wherein The self-data of the energy storage system includes the pressure, water level, and temperature data of the energy storage unit, the flow rate and pressure data of the hydraulic power generation unit, and the power and voltage data of the grid access end.

9. The unitized compressed air energy storage method according to claim 7, characterized in that In S5, the distributed training framework uses the average gradient aggregation algorithm to calculate the global gradient to update the model parameters.

10. The unitized compressed air energy storage method according to claim 7, characterized in that, In S9, machine learning algorithms are used to analyze the device operation data, and a fault prediction model based on neural network is adopted to predict potential device faults.

Images