Water-gas coordinated regulation pumped storage and compressed air coupled energy storage method and system

Abstract

The invention discloses a water-gas coordinated regulation pumped storage and compressed air energy storage coupling method and system, and relates to the technical field of energy storage, and the method comprises the steps: building a dual-medium response characteristic vector after collecting a response data set of water and air dual media, and executing renewable energy input power prediction and power grid load prediction; constructing an external demand tensor in the adjustment window; the external demand tensor and the double-medium response characteristic vector are fused, an adjustment intention tensor is output to a strategy generator, and a water-gas joint output strategy matrix is output; and configuring a joint scheduling instruction set according to the water-gas joint output strategy matrix, issuing the joint scheduling instruction set to an execution terminal, and executing collaborative energy storage control management. The technical problems that in the prior art, water-gas double energy storage media are insufficient in cooperative adjustment capacity and difficult to adapt to power grid loads and renewable energy source fluctuation are solved, and the technical effect of improving the response speed of the energy storage system and the overall energy utilization efficiency is achieved.

Description

Technical Field

[0001] This invention relates to the field of energy storage technology, specifically to a pumped-storage energy storage method and system coupled with compressed air energy storage through water-air coordinated regulation. Background Technology

[0002] In energy storage systems, such as pumped hydro storage or compressed air storage, the potential energy of water and the pressure energy of air are typically operated and controlled as independent energy carriers, each employing a fixed or weakly coupled scheduling method. When the grid load changes rapidly or the output of renewable energy sources such as wind power and photovoltaics exhibits strong fluctuations, the differences in response speed, regulation inertia, and energy conversion characteristics of different media can easily lead to regulation lag or unreasonable energy allocation, making it difficult to achieve coordinated and complementary output. This limits the adaptability of the energy storage system to grid fluctuations and its overall operating efficiency. Summary of the Invention

[0003] This application provides a pumped-storage energy storage coupled with compressed-air energy storage method and system for water-air coordinated regulation, which is used to address the technical problems of insufficient coordinated regulation capability of water-air dual energy storage media in the prior art and difficulty in adapting to grid load and renewable energy fluctuations.

[0004] In view of the above problems, this application provides a method and system for pumped-storage energy storage coupled with compressed air energy storage, which involves coordinated regulation of water and air.

[0005] The first aspect of this application provides a method for pumped-storage energy storage coupled with compressed-air energy storage, the method comprising: After collecting response datasets from both water and air media, a dual-media response characteristic vector is constructed using a multimodal time-series coding network. The response dataset includes the water potential energy change rate, inertial parameters of the pumping / draining process, air cavity pressure decay characteristics, and dynamic response delay of the compression / expansion process. Renewable energy input power prediction and grid load prediction are performed, and the prediction results are used to construct an external demand tensor within the regulation window. The external demand tensor is fused with the dual-media response characteristic vector to output a regulation intent tensor. This regulation intent tensor is sent to a strategy generator to output a water-air joint output strategy matrix. This matrix characterizes the optimal output ratio and coupling strength coefficient of the air expander and water turbine within the current regulation window. A joint scheduling instruction set is configured based on the water-air joint output strategy matrix and then sent to the execution terminal to perform collaborative energy storage control and management.

[0006] A second aspect of this application provides a pumped-storage energy storage system coupled with compressed air energy storage, wherein the system comprises: The system comprises the following modules: a characteristic vector construction module, which constructs a dual-medium response characteristic vector using a multimodal time-series coding network after collecting response datasets from both water and air; a characteristic vector construction module, which includes the water potential energy change rate, inertial parameters of the pumping / draining process, air cavity pressure decay characteristics, and dynamic response delay of the compression / expansion process; a prediction module, which performs renewable energy input power prediction and grid load prediction, and constructs an external demand tensor within the regulation window using the prediction results; a vector fusion module, which fuses the external demand tensor with the dual-medium response characteristic vector to output a regulation intent tensor; a strategy matrix acquisition module, which sends the regulation intent tensor to a strategy generator to output a water-air joint output strategy matrix, which characterizes the optimal output ratio and coupling strength coefficient of the air expander and water turbine within the current regulation window; and a control and management module, which configures a joint scheduling instruction set according to the water-air joint output strategy matrix, sends the joint scheduling instruction set to the execution terminal, and performs collaborative energy storage control and management.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application, after collecting response datasets from both water and air media, constructs a dual-media response characteristic vector using a multimodal time-series coding network. The response dataset includes the water potential energy change rate, inertial parameters of the pumping / draining process, air cavity pressure decay characteristics, and dynamic response delay of the compression / expansion process. It then performs renewable energy input power prediction and grid load prediction, using the prediction results to construct an external demand tensor within the regulation window. This external demand tensor is fused with the dual-media response characteristic vector to output a regulation intent tensor. The regulation intent tensor is sent to a strategy generator to output a water-air joint output strategy matrix. This matrix characterizes the optimal output ratio and coupling strength coefficient of the air expander and water turbine within the current regulation window. Based on the water-air joint output strategy matrix, a joint scheduling instruction set is configured and sent to the execution terminal to perform coordinated energy storage control management. This invention solves the technical problems of insufficient coordinated regulation capability of water and air dual energy storage media in existing technologies, making it difficult to adapt to grid load and renewable energy fluctuations. Through coordinated regulation control of water and air joint output, it achieves the technical effect of improving the response speed and overall energy utilization efficiency of the energy storage system. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0009] Figure 1 A schematic diagram of the pumped-storage coupled compressed-air energy storage method for water-air coordinated regulation provided in the embodiments of this application; Figure 2 This is a schematic diagram of a pumped-storage energy storage system coupled with compressed air energy storage, which provides a water-air coordinated regulation system according to an embodiment of this application.

[0010] Figure labeling: Feature vector construction module 11, prediction module 12, vector fusion module 13, policy matrix acquisition module 14, control management module 15. Detailed Implementation

[0011] This application provides a pumped-storage energy storage coupled with compressed air energy storage method and system with water-air coordinated regulation. It addresses the technical problems of insufficient coordinated regulation capability of water and air dual energy storage media in the prior art, and difficulty in adapting to grid load and renewable energy fluctuations. Through coordinated regulation and control of water and air joint output, it achieves the technical effect of improving the response speed and overall energy utilization efficiency of the energy storage system.

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0013] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0014] Example 1, as Figure 1 As shown, this application provides a method for pumped-storage coupled with compressed-air energy storage with coordinated water and air regulation, the method comprising: Step S100: After collecting the response datasets of water and air as dual media, a dual-media response characteristic vector is constructed through a multimodal temporal coding network. The response dataset includes the water potential energy change rate, inertial parameters of the pumping / draining process, air cavity pressure decay characteristics, and dynamic response delay of the compression / expansion process.

[0015] In this embodiment, when collecting response datasets for both water and air, the operating status information of the water and air media is simultaneously acquired at corresponding positions in the pumped storage operation loop and the compressed air energy storage operation loop. The changes in water level and head of the upper and lower reservoirs are continuously recorded within the same adjustment timescale, and the water potential energy change rate is formed based on the relationship between water level changes at adjacent sampling times. Simultaneously, during the switching between pumping and discharge operations and power adjustment, the changes in pump or turbine speed, guide vane opening, and output power are continuously recorded, and inertial parameters of the pumping / discharge process are extracted to characterize the dynamic hysteresis characteristics of the pumping / discharge process. During air energy storage, the pressure change within the air cavity over time is continuously sampled to form the air cavity pressure decay characteristic, which reflects the pressure change law during the storage or release process. Under compression or expansion regulation, the time difference between the application of the regulation command and the time when the key operating state reaches the response criterion is recorded to form the dynamic response delay of the compression / expansion process, which is used to characterize the hysteresis characteristics of the compression / expansion process. Thus, a response dataset of water and air dual media is obtained, which includes the water potential energy change rate, the inertial parameters of the pumping / draining process, the air cavity pressure decay characteristics, and the dynamic response delay of the compression / expansion process.

[0016] When constructing the dual-medium response characteristic vector, the response datasets of water and air are organized into a multivariate time-series input sequence according to a unified time order, so that various types of response data form a correspondence on the same time axis. Then, the multivariate time-series input sequence is input into a multimodal time-series coding network to jointly encode the water response information corresponding to the water potential energy change rate and the inertial parameters of the pumping / draining process, as well as the air response information corresponding to the air cavity pressure decay characteristics and the dynamic response delay of the compression / expansion process. By continuously representing and converging the response characteristics at each moment in the time dimension, the dynamic response behavior of the two media, water and air, is mapped into a unified feature space, and finally a dual-medium response characteristic vector that simultaneously represents the water potential energy change characteristics and the air pressure change characteristics is formed.

[0017] Step S200: Perform renewable energy input power forecasting and grid load forecasting, and use the forecast results to construct the external demand tensor within the adjustment window.

[0018] In this embodiment, when performing renewable energy input power prediction, a historical renewable energy output time series sample set is first constructed based on wind power output data and photovoltaic power output data collected during historical operation. The historical output data is then organized according to a unified time resolution so that each training sample contains renewable energy output observations from multiple consecutive historical moments. Subsequently, the renewable energy historical output time series sample set is input into the renewable energy input power prediction model for training. The renewable energy input power prediction model consists of an input layer, a time series feature extraction layer, and an output layer. The input layer receives the renewable energy historical output time series samples, the time series feature extraction layer recursively represents the changing trend and periodic characteristics of renewable energy output in the time dimension based on a recurrent neural network structure, and the output layer generates the renewable energy input power output value corresponding to the prediction moment. During the training process, the error between the predicted output value and the actual output observation value is used as the optimization objective to iteratively update the model parameters. After the model training is completed, it is used to output the renewable energy input power prediction results corresponding to each moment within the future adjustment window.

[0019] When performing grid load forecasting, a historical load time series sample set is constructed based on grid load data collected during historical operation phases. This historical load data is then processed using the same time resolution as renewable energy input power forecasting, ensuring that each training sample contains grid load observations from multiple consecutive historical moments. This historical load time series sample set is then input into the grid load forecasting model for training. The model consists of an input layer, a time series feature extraction layer, and an output layer. The input layer receives the historical load time series samples. The time series feature extraction layer recursively represents the changing trends and periodic characteristics of grid load over time using a recurrent neural network structure. The output layer generates the grid load output value corresponding to the forecast time. During training, the error between the predicted output value and the actual load observation value is used as the optimization objective to iteratively update the model parameters. After training, the model outputs the grid load forecast results for each moment within the future adjustment window.

[0020] When performing renewable energy input power forecasting, wind power output data and photovoltaic power output data from multiple consecutive time points are used as inputs and fed into a pre-constructed renewable energy input power forecasting model. The model then performs time-series analysis of historical output changes and outputs the renewable energy input power forecast results for each time point within the future adjustment window. Similarly, when performing grid load forecasting, historical grid load data from multiple consecutive time points are used as inputs and fed into a pre-constructed grid load forecasting model. The model then performs time-series analysis of historical load changes and outputs the grid load forecast results for each time point within the future adjustment window.

[0021] After obtaining the renewable energy input power prediction results from the renewable energy input power prediction model and the grid load prediction results from the grid load prediction model, the two types of prediction results are aligned according to a unified time index, so that each prediction moment within the adjustment window corresponds to both a renewable energy input power prediction value and a grid load prediction value. The prediction results of each prediction moment are combined and arranged in the time dimension to construct the external demand tensor within the adjustment window. The external demand tensor is used to uniformly represent the comprehensive external adjustment demand formed by the changes in renewable energy power supply and grid load demand within the adjustment window.

[0022] Step S300: Fuse the external demand tensor with the dual-medium response characteristic vector to output the adjustment intention tensor.

[0023] Furthermore, in the method provided in the application embodiments, fusing the external demand tensor with the dual-medium response characteristic vector to output the adjustment intention tensor further includes: A water body inertial time constant is constructed based on the inertial parameters of the pumping / discharging process. Short-period and long-period time bases are then constructed using this water body inertial time constant and the dynamic response delay of the compression / expansion process, respectively. The external demand tensor is then decomposed into a multi-scale model, generating a short-period disturbance eigenvalue tensor and a long-period trend tensor. The short-period disturbance eigenvalue tensor characterizes the rapid power gap caused by pulsating renewable energy input within the regulation window, while the long-period trend tensor characterizes the steady-state regulation demand corresponding to a slow load trend. A bidirectional gated feedback mechanism is used to establish the mutual feedback relationship between the short-period disturbance eigenvalue tensor, the long-period trend tensor, and the dual-medium vector characteristic vector. The system is structured as follows: a hydraulic-air physical constraint vector is established, which includes the head-pressure cavity equivalent pressure boundary consistency condition, the coupling power conservation condition, and the air-water interface response threshold condition; the external demand sub-tensor after multi-scale modeling using the physical constraint vector is projected and modulated; a local sensitivity map of the water potential energy change rate and the air cavity pressure decay rate is calculated based on the dual-medium response characteristic vector, and a medium sensitivity factor is established; the medium sensitivity factor is used to perform dynamic compression and amplification processing of the external demand tensor expression under attention-driven conditions; based on sequential execution of mutual feedback processing, projection modulation, and dynamic compression and amplification processing, a regulation intention tensor is constructed.

[0024] In this embodiment, when fusing the external demand tensor with the dual-medium response characteristic vector, the water body inertial time constant is first constructed based on the inertial parameters of the pumping / draining process. In this process, multiple power adjustment processes for pumping and draining conditions are selected within the adjustment window. For each power adjustment process, the time of the adjustment command issuance and the corresponding power change sequence are recorded. The moment when the power change first enters the preset stable range is determined, and the time interval between the adjustment command issuance and the moment it enters the stable range is taken as the inertial time corresponding to that pumping or draining process. Subsequently, all inertial times obtained within the adjustment window are statistically processed, and the water body inertial time constant is calculated using an arithmetic mean, thus obtaining the water body inertial time constant characterizing the overall dynamic response characteristics of the water body. Simultaneously, the dynamic response delay of the compression / expansion process is directly read as the lag time scale for the air medium, and a short-period time base is constructed using the time scale corresponding to the water body inertial time constant, while a long-period time base is constructed using the time scale corresponding to the dynamic response delay of the compression / expansion process. Short-cycle and long-cycle time bases do not directly characterize the inherent response speed of water or air media, but are used to depict the changing characteristics of external regulation demand at different time scales; the water body inertial time constant is used to determine the minimum regulation time scale for the disturbance caused by the pulsating input of renewable energy to the system, and the dynamic response delay of the compression / expansion process is used to determine the time span that can be smoothly absorbed in load trend regulation.

[0025] Next, when performing multi-scale decomposition of the external demand tensor using short-period and long-period time bases, the external demand tensor is first expanded into a continuous time series within the adjustment window according to the time index. Then, using the time span corresponding to the short-period time base as the segment length, the continuous time series is divided into equally spaced segments, and the changes between adjacent moments within each segment are extracted. The components with larger changes are aggregated to form a short-period disturbance feature sub-tensor, which is used to characterize the rapid power gap caused by the pulsating input of renewable energy within the adjustment window. Simultaneously, using the time span corresponding to the long-period time base as the segment length, the continuous time series is segmented, and the overall trend within each segment is extracted. The trend components of each segment are aggregated to form a long-period trend sub-tensor, which is used to characterize the steady-state adjustment demand corresponding to slow load changes.

[0026] Subsequently, a bidirectional gating feedback mechanism was used to establish the mutual feedback relationship between the short-period perturbation feature tensor, the long-period trend tensor, and the dual-medium vector characteristic vector. In this process, forward and backward gating modulation processes were set for the short-period perturbation feature tensor and the long-period trend tensor, respectively. Based on the local variation characteristics reflecting the water potential energy change rate and air cavity pressure attenuation characteristics in the dual-medium response characteristic vector, the intensity of feature information transmission at different time scales was adaptively adjusted. Simultaneously, a dual-medium dependency factor composed of the water potential energy change rate and air cavity pressure attenuation characteristics was introduced to analyze the correlation between the short-period perturbation feature tensor and the long-period trend tensor in the dual-medium response characteristic space, forming a cross-period interactive weight expression. Based on this interactive weight, the forward and backward gating results were updated bidirectionally. Finally, the short-period perturbation feature tensor and the long-period trend tensor updated through bidirectional mutual feedback were jointly encoded to obtain a mutually enhanced characterization tensor used to represent the synergistic effect of demand characteristics at different time scales under dual-medium response constraints.

[0027] After constructing the mutual feedback enhancement characterization tensor, a hydraulic-air physical constraint vector is established to constrain the physical feasibility of multi-scale demand characteristics. This process begins by reading the head variation range corresponding to water body operation and the pressure variation range of the pressure chamber corresponding to air energy storage operation within the adjustment window. These two ranges are then mapped to the same physical quantity dimension, forming a head-pressure chamber equivalent pressure boundary consistency condition to limit the matching relationship between head and pressure chamber equivalent pressure. Subsequently, the power distribution relationship between the turbine output power and the air expander output power under combined operation is read, constructing a coupled power conservation condition to constrain the energy balance relationship between water body output power and air output power. Simultaneously, the allowable response range of the air-water interface under the combined effects of water level and pressure changes is read, constructing an air-water interface response threshold condition to limit the dynamic response amplitude of the air-water interface. Finally, these three types of conditions are combined according to a unified dimension to form the hydraulic-air physical constraint vector.

[0028] Next, the external demand sub-tensor after multi-scale modeling using the aforementioned physical constraint vector is projected and modulated. In this process, mapping rules for constrained demand expression are constructed based on the head-pressure cavity equivalent pressure boundary consistency condition, coupling power conservation condition, and air-water interface response threshold condition. Short-period perturbation feature sub-tensor and long-period trend sub-tensor are input into the mapping rules, and their feature components are adjusted so that, while maintaining their original time-scale characteristic structure, both types of sub-tensors are constrained within a feasible range that satisfies the hydraulic-air physical constraints, thus obtaining a multi-scale external demand expression that meets the physical constraints of joint operation.

[0029] Subsequently, local sensitivity maps of the water potential energy change rate and the air cavity pressure decay rate were calculated based on the dual-medium response characteristic vector to characterize the sensitivity of the water and air media to changes in external demand. In this process, firstly, a continuous local time interval was selected within the adjustment window, and the corresponding water potential energy change rate sequence, air cavity pressure decay characteristic sequence, and external demand sub-tensor change sequence after projection modulation were simultaneously extracted within that time interval. Then, the magnitudes of the changes between the water potential energy change rate and the external demand change, and between the air cavity pressure decay rate and the external demand change, were compared time-by-time to form local sensitivity maps reflecting the differences in the strength of the responses of the water and air media to changes in external demand. Finally, the local sensitivity maps were normalized within the adjustment window to obtain the medium sensitivity factor.

[0030] Subsequently, the attention-driven external demand tensor representation is dynamically compressed and amplified using a medium sensitivity factor. In this process, the medium sensitivity factor is first aligned with the projected short-period perturbation feature tensor and long-period trend tensor according to the time index, so that the medium sensitivity factor at each time step corresponds to the external demand feature at the same time step. Then, at each time step, the medium sensitivity values ​​corresponding to the water potential energy change rate and the air cavity pressure decay rate are read and normalized, ensuring that the sum of the sensitivity values ​​corresponding to the water and air media at the same time step is a preset constant. This generates corresponding attention weight coefficients, with media having higher sensitivity values ​​assigned higher weight coefficients and media having lower sensitivity values ​​assigned lower weight coefficients. Based on this, attention weight coefficients are used to perform time-by-time weighting on each feature component in the short-period perturbation feature sub-tensor and the long-period trend sub-tensor. Feature components with higher weight coefficients are subjected to amplitude enhancement processing to increase their proportion in the expression of external demand, while feature components with lower weight coefficients are subjected to amplitude compression processing to decrease their proportion in the expression of external demand, thereby forming a multi-scale expression of external demand after dynamic compression and amplification processing.

[0031] Finally, a regulation intent tensor is constructed based on sequentially executed mutual feedback processing, projection modulation, and dynamic compression amplification processing. In this process, the short-period perturbation feature sub-tensor and the long-period trend sub-tensor, after dynamic compression amplification, are first concatenated along the time dimension and then fused with the dual-medium response characteristic vector under the same time index. This establishes a one-to-one correspondence between the modulation results of external demand at both short-period and long-period time scales and the dynamic response capabilities of the water and air dual media. Subsequently, the fused results are organized according to the time sequence of the regulation window, forming a unified tensor structure that includes short-period perturbation regulation demand, long-period steady-state regulation demand, medium response sensitivity weights, and physical constraint modulation results. The final output is the regulation intent tensor, which characterizes the comprehensive matching relationship between the external demand characteristics within and outside the regulation window and the response capabilities of the water and air dual media.

[0032] Furthermore, the method provided in the application embodiments, which utilizes bidirectional gating feedback to establish the mutual feedback relationship between the short-period perturbation feature sub-tensor, the long-period trend sub-tensor, and the dual-medium vector characteristic vector, also includes: Forward gating units and backward gating units are established for the short-period perturbation feature tensor and the long-period trend tensor, respectively. The gating units adjust the gating weights by the local gradient change rate of the dual-medium response characteristic vector to modulate the information transmission intensity of different periodic features. A dual-medium dependency factor is generated based on the water potential energy change rate and the air cavity pressure attenuation characteristics. Interactive attention analysis is performed on the dual-medium dependency factor, and a cross-period interactive weight matrix is ​​generated by calculating the correlation scores of the short-period perturbation feature tensor and the long-period trend tensor in the dual-medium response characteristic space. The outputs of the forward gating unit and the backward gating unit are updated bidirectionally based on the cross-period interactive weight matrix. The short-period perturbation feature tensor and the long-period trend tensor updated by bidirectional mutual feedback are stacked and encoded to output a mutual feedback enhanced representation tensor.

[0033] In this embodiment, when establishing forward gating units and reverse gating units for short-period perturbation feature tensors and long-period trend tensors respectively, at each time point, the short-period perturbation feature tensors, long-period trend tensors, and dual-medium response characteristic vectors are aligned by time index, and the feature vectors at the same time point are extracted as gating calculation inputs. Subsequently, the local gradient change rate of the dual-medium response characteristic vector is calculated. Specifically, the difference between the dual-medium response characteristic vectors at two adjacent time points is used to obtain the change, and the absolute value of the change is taken as the local gradient change rate. Then, the local gradient change rate is normalized to the interval between 0 and 1, and the normalization result is used as the gating weight. The gating weight of the forward gating unit is used to perform element-wise multiplication weighting on the feature vectors of the short-period perturbation feature tensors or long-period trend tensors at that time point, and the gating weight of the reverse gating unit is used to perform element-wise multiplication weighting on the feature vectors of the dual-medium response characteristic vectors at that time point, thereby obtaining the output of the forward gating unit and the output of the reverse gating unit, and realizing the modulation of the transmission intensity of feature information of different periods.

[0034] When generating a dual-medium dependency factor based on the water potential energy change rate and air cavity pressure decay characteristics, at each time point, the water potential energy change rate and air cavity pressure decay characteristics are extracted from the dual-medium response characteristic vector, and their absolute values ​​are taken to eliminate the positive and negative directional influences. Then, the sum of the two is calculated, and the water dependency component is obtained by dividing the absolute value of the water potential energy change rate by the sum of the two, and the air dependency component is obtained by dividing the absolute value of the air cavity pressure decay characteristics by the sum of the two. Thus, at each time point, a dual-medium dependency factor composed of the water dependency component and the air dependency component is formed, so that the dual-medium dependency factor directly represents the relative interaction ratio between the water medium and the air medium at that time point.

[0035] When performing interactive attention analysis on the dual-medium dependency factor and generating a cross-period interactive weight matrix, the eigenvectors of the short-period perturbation eigenvalue tensor and the long-period trend tensor are extracted at each time point, and their inner product is calculated as the correlation score. Subsequently, the correlation score is weighted using the water-dependent and air-dependent components of the dual-medium dependency factor. Specifically, the correlation score is multiplied by the water-dependent and air-dependent components respectively to obtain two weighted scores, which are then summed to obtain the final correlation score for that time point. The final correlation scores for each time point within the adjustment window are then arranged chronologically to form the cross-period interactive weight matrix. This matrix is ​​used to characterize the cross-period correlation strength between the short-period perturbation eigenvalue tensor and the long-period trend tensor in the dual-medium response characteristic space.

[0036] When updating the outputs of the forward and reverse gating units bidirectionally based on the cross-period interactive weight matrix, the weight value of the cross-period interactive weight matrix is ​​read at each time point. The weight value is then used to perform element-wise multiplication weighting on the short-period perturbation feature tensor vector and the long-period trend tensor vector output by the forward gating unit. Subsequently, the weighted short-period perturbation feature tensor vector and the weighted long-period trend tensor vector are added element-wise to obtain the periodic modulation vector. The periodic modulation vector is then added element-wise to the dual-medium response characteristic vector output by the reverse gating unit to form the update vector. This element-wise addition process corresponds to the additive residual path, thereby completing the bidirectional mutual feedback update. This allows the updated short-period perturbation feature tensor and long-period trend tensor to characterize the coupling behavior of the water-air dual media under different regulation time scales.

[0037] When stacking and encoding the updated short-period perturbation feature tensor and long-period trend tensor to output the enhanced representation tensor, at each time point, the updated short-period perturbation feature tensor vector and the updated long-period trend tensor vector are concatenated along the feature dimension to form a stacked vector. Then, the arithmetic mean of the stacked vectors along the feature dimension is calculated to obtain an encoded vector of uniform dimension. Finally, the encoded vectors from each time point within the adjustment window are arranged in chronological order to output the enhanced representation tensor.

[0038] Furthermore, the method provided in the application embodiments also includes: The bidirectional mutual feedback update constructs an additive residual path to jointly update the periodic modulation vector generated by the forward gating unit and the reverse gating unit with the dual-medium response characteristic vector. This allows the short-period perturbation characteristic tensor and the long-period trend tensor to characterize the coupling behavior of the water-air dual media under different regulation time scales after the update.

[0039] In this embodiment, when performing bidirectional mutual feedback updates, the output results of the forward gating unit and the reverse gating unit are first obtained in the adjustment window according to a unified time index. The periodic modulation vector output by the forward gating unit is used to characterize the modulation result of the short-period perturbation feature sub-tensor or the long-period trend sub-tensor at the current time point. The periodic modulation vector output by the reverse gating unit is used to characterize the feedback modulation result of the dual-medium response characteristic vector at the current time point. The periodic modulation vector and the dual-medium response characteristic vector at the corresponding time point are aligned in dimension consistency to ensure that each vector has the condition for element-wise joint processing at the same time point.

[0040] When constructing an additive residual path for joint update, the periodic modulation vector and the dual-medium response characteristic vector are updated by element-wise addition. That is, at each time point, the periodic modulation vector generated by the forward gating unit is first added element-wise to the dual-medium response characteristic vector at the corresponding time point to obtain an intermediate result superimposed with periodic modulation information. Then, the intermediate result is added element-wise to the periodic modulation vector generated by the reverse gating unit to form the jointly updated dual-medium response characteristic vector. Thus, while retaining the original dynamic characteristics of the dual-medium response characteristic vector, modulation information from the short-period perturbation feature sub-tensor and the long-period trend sub-tensor is introduced to realize the residual update of the dual-medium response characteristic vector.

[0041] After jointly updating the dual-medium response characteristic vectors, the updated dual-medium response characteristic vectors are used to update the short-period perturbation eigenvector tensor and the long-period trend tensor. In this process, at each time point, the updated dual-medium response characteristic vector is added element-wise to the corresponding time-point eigenvectors of the short-period perturbation eigenvector tensor and the long-period trend tensor. This allows the updated short-period perturbation eigenvector tensor to incorporate the dual-medium dynamic response information reflected by the water potential energy change rate and air cavity pressure decay characteristics, while the updated long-period trend tensor incorporates the response characteristics of the water and air media to steady-state regulation demands. Thus, the updated short-period perturbation eigenvector tensor and the updated long-period trend tensor respectively characterize the coupling behavior of the water-air dual media at different regulation time scales.

[0042] Furthermore, in the method provided in the application embodiments, the projection modulation of the external demand sub-tensor after multi-scale modeling using the physical constraint vector further includes: A projection operator is constructed based on the physical constraint vector; the short-period perturbation feature tensor and the long-period trend tensor are respectively input into the projection operator, and the short-period perturbation feature tensor and the long-period trend tensor are adjusted to the feasible region that satisfies the constraints according to the linear or nonlinear mapping rules defined by the projection operator.

[0043] In this embodiment, when constructing the projection operator based on physical constraint vectors, the head-pressure cavity equivalent pressure boundary consistency condition, the coupling power conservation condition, and the gas-water interface response threshold condition are first transformed into feasible domain constraint parameters for the short-period perturbation eigenvalue tensor and the long-period trend tensor, respectively. Specifically, these include the lower and upper limits of the allowable eigenvalues ​​corresponding to the head-pressure cavity equivalent pressure boundary consistency condition, the upper limit of the allowable power combination corresponding to the coupling power conservation condition, and the upper limits of the allowable amplitude change and the upper limit of the allowable rate of change corresponding to the gas-water interface response threshold condition. Subsequently, these constraint parameters are written into a process that corrects each eigenvalue individually, enabling the projection operator to perform three types of correction operations—boundary truncation, scaling, and rate limiting—when any eigenvalue is input, thereby forming a complete projection operator.

[0044] When inputting the short-period perturbation eigentensor into the projection operator, the eigenvalues ​​of the short-period perturbation eigentensor are first read sequentially according to the time order of the adjustment window. Boundary correction is then performed on each eigenvalue based on the head-pressure chamber equivalent pressure boundary consistency condition. For example, if the power demand corresponding to the short-period perturbation eigentensor at a certain moment is higher than the allowable equivalent upper limit of head and pressure chamber pressure, the power demand value is adjusted to the upper limit value; if it is lower than the allowable lower limit, it is adjusted to the lower limit value. Subsequently, at the same moment, the power demand corresponding to the boundary-corrected short-period perturbation eigentensor and the long-period trend eigentensor is summed and compared with the upper limit of power given by the coupling power conservation condition. When the summation result exceeds the upper limit of power, the eigenvalues ​​of the short-period perturbation eigentensor are proportionally reduced so that the sum of the adjusted short-period perturbation power demand and the long-period trend power demand is exactly equal to the upper limit of power. Finally, based on the gas-water interface response threshold condition, the change of the short-period perturbation eigentensor between two adjacent time points is restricted. For example, when the power change between a certain time point and the previous time point exceeds the maximum allowable change of the gas-water interface, the change is adjusted to the maximum change, thereby obtaining a short-period perturbation eigentensor that is smooth in the time dimension and meets the interface response capability.

[0045] When the long-period trend sub-tensor is input into the projection operator, projection modulation is performed time-by-time according to the same processing order as the short-period perturbation eigenvalue tensor. First, the eigenvalues ​​of the long-period trend sub-tensor are corrected for upper and lower limits based on the head-pressure cavity equivalent pressure boundary consistency condition to ensure that its long-term power demand level does not exceed the equivalent energy carrying capacity of the water and air media. Subsequently, combined with the coupling power conservation condition, the power demands of the long-period trend sub-tensor and the short-period perturbation eigenvalue tensor at the same time are coordinated. When the sum of the two exceeds the allowable power limit, the eigenvalues ​​of the long-period trend sub-tensor are proportionally adjusted to ensure consistency in power allocation between steady-state regulation demand and rapid regulation demand. Finally, based on the air-water interface response threshold condition, the variation amplitude of the long-period trend sub-tensor between consecutive time points is limited so that its variation process conforms to the stable response capability of the air-water interface over a longer time scale, thus obtaining a long-period trend sub-tensor that meets the physical constraints.

[0046] Through the above-mentioned stepwise projection modulation process, the short-period perturbation feature tensor and the long-period trend tensor, while maintaining their respective time-scale semantic features, are mapped to feasible regions that simultaneously satisfy the head-pressure cavity equivalent pressure boundary consistency condition, the coupling power conservation condition, and the air-water interface response threshold condition, forming a multi-scale external demand expression that meets the physical constraints of the joint operation of hydraulic and air media.

[0047] Step S400: Send the adjustment intention tensor to the strategy generator and output the water-air joint output strategy matrix. The water-air joint output strategy matrix is ​​used to characterize the optimal output ratio and coupling strength coefficient of the air expander and water turbine in the current adjustment window.

[0048] In this embodiment, during the process of sending the regulation intention tensor to the strategy generator to output the water-air joint output strategy matrix, the strategy generator first receives the regulation intention tensor and analyzes the feature information representing short-period disturbance regulation demand and long-period steady-state regulation demand in the regulation intention tensor. It then performs separation processing on the regulation features at different time scales to form corresponding periodized sub-feature representations. Subsequently, it combines the dual-medium response characteristic vector and the mutual feedback enhancement representation tensor formed in the regulation intention tensor to analyze the cooperative output relationship between the turbine and the air expander under joint operation, and calculates the joint output coupling coefficient that reflects the strength of the cooperative effect between the turbine and the air expander within the current regulation window.

[0049] Based on this, the periodic sub-feature representation is fused with the joint output coupling coefficient, and the output results are uniformly scaled through normalization constraints to form a water-air joint output strategy matrix describing the power distribution ratio and coupling strength between the air expander and the water turbine at each adjustment time.

[0050] Furthermore, in the method provided in the application embodiment, sending the adjustment intention tensor to the policy generator and outputting the water-air joint output policy matrix further includes: The strategy generator includes a feature encoding layer, a coupling weight calculation layer, and an output fusion layer. The feature encoding layer receives the adjustment intention tensor and performs channel separation encoding on short-period perturbations and long-period trends to form a periodized sub-feature representation. The coupling weight calculation layer uses the mutual feedback enhancement of the dual-medium response characteristic vector and the adjustment intention tensor to calculate the joint output coupling coefficient of the water turbine and the air expander. The output fusion layer integrates the joint output coupling coefficient and the periodized sub-feature representation through nonlinear weighting and normalization operations to output a water-air joint output strategy matrix.

[0051] In this embodiment, the policy generator includes a feature encoding layer, a coupled weight calculation layer, and an output fusion layer. When the feature encoding layer performs channel-separated encoding on the adjustment intention tensor to form a periodic sub-feature representation, the adjustment intention tensor is first input into the feature encoding layer in the time order of the adjustment window. In the channel dimension, channel data corresponding to short-cycle perturbations are extracted as short-cycle perturbation channel sequences, and channel data corresponding to long-cycle trends are extracted as long-cycle trend channel sequences. Subsequently, a linear transformation is performed on the short-cycle perturbation channel sequences and the long-cycle trend channel sequences at each time step. Specifically, the channel vector at that time step is multiplied by a preset weight matrix to obtain the corresponding encoding vector. Finally, the short-cycle perturbation encoding vectors at each time step within the adjustment window are arranged in time order to form a periodic sub-feature representation of the short-cycle perturbation, and the long-cycle trend encoding vectors at each time step are arranged in time order to form a periodic sub-feature representation of the long-cycle trend, thereby forming the periodic sub-feature representation.

[0052] When calculating the coupling strength coefficient between the turbine and the air expander in the coupling weight calculation layer, the dual-medium response characteristic vector and the mutual feedback enhancement representation tensor formed in the adjustment intention tensor are read simultaneously at each adjustment moment, and the two are aligned in the feature dimension. Then, by calculating the inner product between the dual-medium response characteristic vector and the mutual feedback enhancement representation tensor, the correlation value reflecting the degree of coordinated response between the water and air media is obtained. The correlation value is then normalized within the adjustment window to map it to a preset range, thereby obtaining the coupling strength coefficient corresponding to each adjustment moment.

[0053] Finally, when integrating and generating the water-air joint output strategy matrix at the output fusion layer, the periodic sub-feature representations of short-period disturbances and long-period trends output by the feature encoding layer are read at each regulation time. A summation operation is then performed on each periodic sub-feature representation along its feature dimension. The values ​​of each feature component at the same time are summed to obtain the short-period regulation intensity value representing the intensity of rapid regulation demand and the long-period regulation intensity value representing the intensity of steady-state regulation demand. Subsequently, a coupling strength coefficient corresponding to the regulation time is introduced to weight the short-period and long-period regulation intensity values. The short-period regulation intensity value is multiplied by the coupling strength coefficient to obtain the weighted regulation intensity on the air expander side, and the long-period regulation intensity value is multiplied by a factor minus the coupling strength coefficient to obtain the weighted regulation intensity on the turbine side. This establishes a regulation correlation where the air expander is primarily responsible for short-period disturbance characteristics, and the turbine is primarily responsible for long-period trend characteristics.

[0054] After weighting, nonlinear transformations are performed on the weighted regulation intensity on the air expander side and the weighted regulation intensity on the turbine side, respectively. Exponential values ​​are taken for both to enhance the distinction between different regulation intensities. Then, the results after exponential transformation are normalized so that the sum of the output ratio of the air expander and the output ratio of the turbine is equal to the preset total power distribution constraint.

[0055] Finally, the output ratios of the air expander and the turbine, as well as the corresponding coupling strength coefficients at each adjustment time within the adjustment window, are combined and arranged in chronological order to form a water-air joint output strategy matrix. The water-air joint output strategy matrix is ​​used to characterize the optimal output ratios of the air expander and the turbine and their coupling strength coefficients within the current adjustment window.

[0056] Step S500: Configure a joint scheduling instruction set according to the water-gas joint output strategy matrix, and send the joint scheduling instruction set to the execution terminal to perform collaborative energy storage control and management.

[0057] In this embodiment, when configuring the joint scheduling instruction set according to the water-air joint output strategy matrix, the water-air joint output strategy matrix is ​​first analyzed time-by-time within the adjustment window. The output ratio of the air expander, the output ratio of the water turbine, and the coupling strength coefficient corresponding to each adjustment time are read according to the time index. Based on the total adjustment power demand at the corresponding time within the adjustment window, the target output power of the air expander and the target output power of the water turbine are generated using a proportional decomposition method. The total adjustment power demand is then multiplied by the output ratio of the air expander and the output ratio of the water turbine, respectively, to obtain the initial target output power of the two types of equipment at that adjustment time.

[0058] Subsequently, a coupling strength coefficient is introduced to perform coordinated correction processing on the target output power of the air expander and the target output power of the turbine. During this process, the difference between the target output power of the two types of equipment is adjusted according to the magnitude of the coupling strength coefficient. When the coupling strength coefficient is large, the mutual constraint between the target output power of the two types of equipment is increased, making the air expander and turbine tend to be consistent in the direction and magnitude of power change. When the coupling strength coefficient is small, the constraint is weakened, allowing the air expander and turbine to adjust their power more independently according to their respective regulation proportions, thus achieving coordinated power correction based on the coupling strength coefficient. After completing the coordinated correction of the target output power, the corrected target output power of the air expander, the corrected target output power of the turbine, and the corresponding regulation timing information at each regulation time are combined to form a joint scheduling instruction set containing power setpoints and time constraint relationships.

[0059] Finally, the joint scheduling instruction set is sent to the execution terminal. The joint scheduling instructions are sent one by one according to the time sequence of the adjustment window, so that the execution terminal adjusts the operating condition of the expander according to the target output power of the air expander and adjusts the operating condition of the water turbine according to the target output power of the water turbine. At the same time, the power change process of the two types of equipment is synchronously constrained by the coupling strength coefficient. Thus, the coordinated adjustment of air pressure energy and water potential energy in the charging and releasing process is realized within the adjustment window, and the coordinated energy storage control and management is completed.

[0060] Furthermore, the method provided in the application embodiments, in performing coordinated energy storage control management, further includes: The system uses an execution terminal to collect operating status data of the turbine and air expander, including head change rate, pump speed, chamber pressure, output power, and response delay. The operating status data is then compared synchronously with the adjustment intention tensor to establish a control error vector. Control compensation management is performed based on the control error vector.

[0061] In this embodiment of the application, during the collaborative energy storage control process, the execution terminal first continuously collects the operating status of the turbine and the air expander. The collected operating status data includes the head change rate, which reflects the rate of change of water conditions; the pump speed, which reflects the mechanical operating status of the turbine or pump group; the cavity pressure, which reflects the air energy storage conditions; the output power, which reflects the actual energy release or absorption level; and the response delay, which characterizes the time lag between the control command and the actual execution. The above operating status data are recorded according to a uniform sampling period, thereby forming an operating status data sequence consistent with the time scale of the adjustment window.

[0062] Then, the operating status data and the adjustment intention tensor are synchronized and aligned according to the time index, so that the actual operating status at each sampling moment corresponds one-to-one with the target adjustment requirement at the corresponding moment in the adjustment intention tensor. Subsequently, the head change rate, pump speed, cavity pressure, output power, and response delay are compared item by item, and the differences between each operating status data and the corresponding target value in the adjustment intention tensor are calculated. The differences are then combined in a preset order to form a control error vector containing head change rate deviation, speed deviation, pressure deviation, power deviation, and response delay deviation.

[0063] Finally, control compensation management is performed based on the control error vector. In this process, the control error vector is decomposed to distinguish the influence direction of each error component on the adjustment deviation of the turbine and air expander. Subsequently, based on the changes in the control error vector, the output proportion coefficient and coupling strength coefficient in the water-air combined output strategy matrix are iteratively adjusted step by step. This ensures that the updated output proportion coefficient and coupling strength coefficient continuously reduce the overall amplitude of the control error vector, thereby achieving adaptive optimization management of control during the coordinated adjustment process of the turbine and air expander.

[0064] Furthermore, in the method provided in the application embodiments, the control compensation management based on the control error vector further includes: After decomposing the control error vector, gradient descent is used to iteratively adjust the output proportion coefficient and coupling strength coefficient in the joint output strategy matrix space to perform adaptive optimization management of control.

[0065] In this embodiment of the application, when performing adaptive optimization management of control, the control error vector is first decomposed into errors, and the head change rate error, pump speed error, cavity pressure error, output power error and response delay error contained therein are respectively mapped to the source of adjustment deviation of the turbine and the air expander, so as to clarify the direction of influence of each error component on the joint adjustment effect.

[0066] Subsequently, using the control error vector as the optimization basis, the output proportion coefficient, which describes the output distribution relationship between the air expander and the water turbine, and the coupling strength coefficient, which describes the degree of synergy between the two, are taken as adjustable variables in the joint output strategy matrix. The above coefficients are iteratively updated in the manner of gradient descent. That is, the direction of coefficient adjustment is determined according to the changing trend of the control error vector, and the output proportion coefficient and coupling strength coefficient are slightly modified in each iteration, so that the operating state of the modified joint output strategy matrix in the next adjustment cycle is closer to the adjustment intention requirements.

[0067] By continuously iterating and updating the output proportion coefficient and coupling strength coefficient, the overall amplitude of the control error vector is gradually reduced, thereby achieving adaptive optimization management of control during the joint operation of the water turbine and the air expander.

[0068] Furthermore, the method provided in the application embodiments also includes: Determine whether the control error vector meets the preset deviation threshold. If it does, establish a control anomaly warning and simultaneously execute shutdown processing and warning issuance management.

[0069] In this embodiment of the application, the real-time generated control error vector is first determined, and the head change rate error, pump speed error, cavity pressure error, output power error and response delay error contained in the control error vector are compared with the corresponding preset deviation thresholds. When any error component in the control error vector reaches or exceeds its preset deviation threshold, the current operating state is determined to meet the abnormal conditions.

[0070] After completing the anomaly determination, a control anomaly early warning is established and used as an anomaly status identifier. Simultaneously, a shutdown process is executed, issuing a shutdown command to the execution terminal to bring the turbine and air expander into a safe shutdown state. Early warning dispatch management is also implemented, sending control anomaly early warning information to the monitoring unit for recording and alarming, thereby achieving timely shutdown and anomaly notification when control deviations exceed limits.

[0071] In summary, the embodiments of this application have at least the following technical effects: This application, after collecting response datasets from both water and air media, constructs a dual-media response characteristic vector using a multimodal time-series coding network. The response dataset includes the water potential energy change rate, inertial parameters of the pumping / draining process, air cavity pressure decay characteristics, and dynamic response delay of the compression / expansion process. It then performs renewable energy input power prediction and grid load prediction, using the prediction results to construct an external demand tensor within the regulation window. This external demand tensor is fused with the dual-media response characteristic vector to output a regulation intent tensor. The regulation intent tensor is sent to a strategy generator to output a water-air joint output strategy matrix. This matrix characterizes the optimal output ratio and coupling strength coefficient of the air expander and water turbine within the current regulation window. Based on the water-air joint output strategy matrix, a joint scheduling instruction set is configured and sent to the execution terminal to perform coordinated energy storage control management. This invention solves the technical problems of insufficient coordinated regulation capability of water and air dual energy storage media in existing technologies, making it difficult to adapt to grid load and renewable energy fluctuations. Through coordinated regulation control of water and air joint output, it achieves the technical effect of improving the response speed and overall energy utilization efficiency of the energy storage system.

[0072] Example 2, based on the same inventive concept as the water-air coordinated regulation pumped storage coupled with compressed air energy storage method in the foregoing examples, such as... Figure 2 As shown, this application provides a pumped-storage energy storage system coupled with compressed air energy storage, characterized by coordinated water and air regulation. The system and method embodiments in this application are based on the same inventive concept. The system includes: The feature vector construction module 11 is used to construct a dual-medium response feature vector through a multimodal time-series coding network after collecting response datasets of water and air. The response dataset includes the water potential energy change rate, inertial parameters of the pumping / draining process, air cavity pressure decay characteristics, and dynamic response delay of the compression / expansion process. The prediction module 12 is used to perform renewable energy input power prediction and grid load prediction, and construct an external demand tensor within the regulation window using the prediction results. The vector fusion module 13 is used to fuse the external demand tensor with the dual-medium response feature vector to output a regulation intent tensor. The strategy matrix acquisition module 14 is used to send the regulation intent tensor to the strategy generator to output a water-air joint output strategy matrix. The water-air joint output strategy matrix is ​​used to characterize the optimal output ratio and coupling strength coefficient of the air expander and water turbine within the current regulation window. The control and management module 15 is used to configure a joint scheduling instruction set according to the water-air joint output strategy matrix, and send the joint scheduling instruction set to the execution terminal to perform collaborative energy storage control and management.

[0073] Furthermore, the system is also used to implement the following functions: A water body inertial time constant is constructed based on the inertial parameters of the pumping / discharging process. Short-period and long-period time bases are then constructed using this water body inertial time constant and the dynamic response delay of the compression / expansion process, respectively. The external demand tensor is then decomposed into a multi-scale model, generating a short-period disturbance eigenvalue tensor and a long-period trend tensor. The short-period disturbance eigenvalue tensor characterizes the rapid power gap caused by pulsating renewable energy input within the regulation window, while the long-period trend tensor characterizes the steady-state regulation demand corresponding to a slow load trend. A bidirectional gated feedback mechanism is used to establish the mutual feedback relationship between the short-period disturbance eigenvalue tensor, the long-period trend tensor, and the dual-medium vector characteristic vector. The system is structured as follows: a hydraulic-air physical constraint vector is established, which includes the head-pressure cavity equivalent pressure boundary consistency condition, the coupling power conservation condition, and the air-water interface response threshold condition; the external demand sub-tensor after multi-scale modeling using the physical constraint vector is projected and modulated; a local sensitivity map of the water potential energy change rate and the air cavity pressure decay rate is calculated based on the dual-medium response characteristic vector, and a medium sensitivity factor is established; the medium sensitivity factor is used to perform dynamic compression and amplification processing of the external demand tensor expression under attention-driven conditions; based on sequential execution of mutual feedback processing, projection modulation, and dynamic compression and amplification processing, a regulation intention tensor is constructed.

[0074] Furthermore, the system is also used to implement the following functions: Forward gating units and backward gating units are established for the short-period perturbation feature tensor and the long-period trend tensor, respectively. The gating units adjust the gating weights by the local gradient change rate of the dual-medium response characteristic vector to modulate the information transmission intensity of different periodic features. A dual-medium dependency factor is generated based on the water potential energy change rate and the air cavity pressure attenuation characteristics. Interactive attention analysis is performed on the dual-medium dependency factor, and a cross-period interactive weight matrix is ​​generated by calculating the correlation scores of the short-period perturbation feature tensor and the long-period trend tensor in the dual-medium response characteristic space. The outputs of the forward gating unit and the backward gating unit are updated bidirectionally based on the cross-period interactive weight matrix. The short-period perturbation feature tensor and the long-period trend tensor updated by bidirectional mutual feedback are stacked and encoded to output a mutual feedback enhanced representation tensor.

[0075] Furthermore, the system is also used to implement the following functions: The bidirectional mutual feedback update constructs an additive residual path to jointly update the periodic modulation vector generated by the forward gating unit and the reverse gating unit with the dual-medium response characteristic vector. This allows the short-period perturbation characteristic tensor and the long-period trend tensor to characterize the coupling behavior of the water-air dual media under different regulation time scales after the update.

[0076] Furthermore, the system is also used to implement the following functions: A projection operator is constructed based on the physical constraint vector; the short-period perturbation feature tensor and the long-period trend tensor are respectively input into the projection operator, and the short-period perturbation feature tensor and the long-period trend tensor are adjusted to the feasible region that satisfies the constraints according to the linear or nonlinear mapping rules defined by the projection operator.

[0077] Furthermore, the system is also used to implement the following functions: The strategy generator includes a feature encoding layer, a coupling weight calculation layer, and an output fusion layer. The feature encoding layer receives the adjustment intention tensor and performs channel separation encoding on short-period perturbations and long-period trends to form a periodized sub-feature representation. The coupling weight calculation layer uses the mutual feedback enhancement of the dual-medium response characteristic vector and the adjustment intention tensor to calculate the joint output coupling coefficient of the water turbine and the air expander. The output fusion layer integrates the joint output coupling coefficient and the periodized sub-feature representation through nonlinear weighting and normalization operations to output a water-air joint output strategy matrix.

[0078] Furthermore, the system is also used to implement the following functions: The system uses an execution terminal to collect operating status data of the turbine and air expander, including head change rate, pump speed, chamber pressure, output power, and response delay. The operating status data is then compared synchronously with the adjustment intention tensor to establish a control error vector. Control compensation management is performed based on the control error vector.

[0079] Furthermore, the system is also used to implement the following functions: Determine whether the control error vector meets the preset deviation threshold. If it does, establish a control anomaly warning and simultaneously execute shutdown processing and warning issuance management.

[0080] Furthermore, the system is also used to implement the following functions: After decomposing the control error vector, gradient descent is used to iteratively adjust the output proportion coefficient and coupling strength coefficient in the joint output strategy matrix space to perform adaptive optimization management of control.

[0081] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0082] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for pumped-storage energy storage coupled with compressed air energy storage with coordinated water and air regulation, characterized in that, The method includes: After collecting response datasets from both water and air, a dual-medium response characteristic vector is constructed using a multimodal temporal coding network. The response dataset includes the water potential energy change rate, inertial parameters of the pumping / draining process, air cavity pressure decay characteristics, and dynamic response delay of the compression / expansion process. Perform renewable energy input power forecasting and grid load forecasting, and use the forecast results to construct an external demand tensor within the adjustment window; The external demand tensor is fused with the dual-medium response characteristic vector to output the adjustment intention tensor; The adjustment intention tensor is sent to the strategy generator to output the water-air joint output strategy matrix. The water-air joint output strategy matrix is ​​used to characterize the optimal output ratio and coupling strength coefficient of the air expander and water turbine in the current adjustment window. Configure a joint scheduling instruction set according to the water-gas joint output strategy matrix, and send the joint scheduling instruction set to the execution terminal to perform collaborative energy storage control and management.

2. The pumped-storage energy storage coupled with compressed air energy storage method for coordinated water and air regulation as described in claim 1, characterized in that, The external demand tensor is fused with the dual-medium response characteristic vector to output a modulation intent tensor, including: The inertial time constant of the water body is constructed based on the inertial parameters of the pumping / draining process. Short-period time base and long-period time base are constructed using the inertial time constant of the water body and the dynamic response delay of the compression / expansion process, respectively. The external demand tensor is decomposed into multiple scales to generate short-period disturbance characteristic sub-tensor and long-period trend sub-tensor. The short-period disturbance characteristic sub-tensor is used to characterize the rapid power gap caused by the pulsating input of renewable energy within the adjustment window, and the long-period trend sub-tensor characterizes the steady-state adjustment demand corresponding to the slow load trend. A mutual feedback relationship between the short-period perturbation feature tensor, the long-period trend tensor, and the dual-medium vector characteristic vector is established using bidirectional gating feedback. Establish a hydraulic-air physical constraint vector, which includes the head-pressure chamber equivalent pressure boundary consistency condition, the coupling power conservation condition, and the air-water interface response threshold condition; The external demand subtensor after multi-scale modeling using the physical constraint vector is projected and modulated. Based on the local sensitivity map of the water potential energy change rate and air cavity pressure decay rate calculated by the dual-medium response characteristic vector, a medium sensitivity factor is established, and the medium sensitivity factor is used to perform dynamic compression and amplification processing of the external demand tensor expression under attention-driven conditions. Based on sequentially executed mutual feedback processing, projection modulation, and dynamic compression amplification processing, an adjustment intention tensor is constructed.

3. The pumped-storage energy storage coupled with compressed air energy storage method for coordinated water and air regulation as described in claim 2, characterized in that, A mutual feedback relationship is established between the short-period perturbation characteristic subtensor, the long-period trend subtensor, and the dual-medium vector characteristic vector using bidirectional gating feedback, including: Forward gating units and backward gating units are established for short-period perturbation feature subtensors and long-period trend subtensors, respectively. The gating units adjust the gating weights by the local gradient change rate of the dual-medium response characteristic vector, which is used to modulate the information transmission intensity of different periodic features. Based on the water potential energy change rate and air cavity pressure attenuation characteristics, a dual-medium dependency factor is generated. Interactive attention analysis is performed on the dual-medium dependency factor. By calculating the correlation scores of the short-period perturbation eigenvalue tensor and the long-period trend tensor in the dual-medium response characteristic space, a cross-period interactive weight matrix is ​​generated. Based on the cross-cycle interactive weight matrix, the outputs of the forward gating unit and the reverse gating unit are updated bidirectionally. The short-period perturbation feature tensor and the long-period trend tensor updated by bidirectional mutual feedback are stacked and encoded to output the mutual feedback enhanced representation tensor.

4. The pumped-storage energy storage coupled with compressed air energy storage method for coordinated water and air regulation as described in claim 3, characterized in that, The bidirectional mutual feedback update constructs an additive residual path to jointly update the periodic modulation vector generated by the forward gating unit and the reverse gating unit with the dual-medium response characteristic vector. This allows the short-period perturbation characteristic tensor and the long-period trend tensor to characterize the coupling behavior of the water-air dual media under different regulation time scales after the update.

5. The pumped-storage energy storage coupled with compressed air energy storage method for coordinated water and air regulation as described in claim 2, characterized in that, Projection modulation of the external demand subtensor after multi-scale modeling using the aforementioned physical constraint vector includes: Construct a projection operator based on the physical constraint vector; The short-period perturbation eigenvalue tensor and the long-period trend tensor are respectively input into the projection operator. According to the linear or nonlinear mapping rules defined by the projection operator, the short-period perturbation eigenvalue tensor and the long-period trend tensor are adjusted to the feasible region that satisfies the constraints.

6. The water-air synergistic regulation method for pumped-storage coupled with compressed-air energy storage as described in claim 1, characterized in that, The adjustment intent tensor is sent to the policy generator to output a water-air joint output policy matrix, including: The policy generator includes a feature encoding layer, a coupled weight calculation layer, and an output fusion layer. The feature encoding layer is used to receive the adjustment intention tensor, perform channel separation encoding on short-period perturbations and long-period trends, and form a periodic sub-feature representation. The coupling weight calculation layer is used to enhance the characterization tensor by utilizing the mutual feedback between the dual-medium response characteristic vector and the adjustment intention tensor to calculate the joint output coupling coefficient of the turbine and the air expander. The output fusion layer is used to integrate the joint output coupling coefficients and periodic sub-feature representations into an output water-gas joint output strategy matrix through nonlinear weighting and normalization operations.

7. The pumped-storage energy storage coupled with compressed air energy storage method for coordinated water and air regulation as described in claim 1, characterized in that, Perform coordinated energy storage control and management, including: The operating status data of the turbine and air expander are collected using the execution terminal. The operating status data includes head change rate, pump speed, chamber pressure, output power and response delay. The operating status data is synchronously compared with the adjustment intention tensor to establish a control error vector; Control compensation management is performed based on the control error vector.

8. The pumped-storage energy storage coupled with compressed air energy storage method for coordinated water and air regulation as described in claim 7, characterized in that, Determine whether the control error vector meets the preset deviation threshold. If it does, establish a control anomaly warning and simultaneously execute shutdown processing and warning issuance management.

9. The pumped-storage energy storage coupled with compressed air energy storage method for coordinated water and air regulation as described in claim 7, characterized in that, Control compensation management based on the control error vector includes: After decomposing the control error vector, gradient descent is used to iteratively adjust the output proportion coefficient and coupling strength coefficient in the joint output strategy matrix space to perform adaptive optimization management of control.

10. A pumped-hydro storage coupled with compressed-air energy storage system with coordinated water and air regulation, characterized in that, The system is used to execute the pumped-storage energy storage coupled with compressed air energy storage method according to any one of claims 1-9, the system comprising: The feature vector construction module is used to construct dual-medium response feature vectors through a multimodal temporal coding network after collecting response datasets of water and air. The response datasets include the water potential energy change rate, inertial parameters of the pumping / draining process, air cavity pressure decay characteristics, and dynamic response delay of the compression / expansion process. The forecasting module is used to perform renewable energy input power forecasting and grid load forecasting, and to construct an external demand tensor within the adjustment window using the forecast results; The vector fusion module is used to fuse the external demand tensor with the dual-medium response characteristic vector to output the adjustment intention tensor; The strategy matrix acquisition module is used to send the adjustment intention tensor to the strategy generator and output the water-air joint output strategy matrix. The water-air joint output strategy matrix is ​​used to characterize the optimal output ratio and coupling strength coefficient of the air expander and water turbine in the current adjustment window. The control and management module is used to configure a joint scheduling instruction set according to the water-gas joint output strategy matrix, and to send the joint scheduling instruction set to the execution terminal to perform collaborative energy storage control and management.

Images