Energy distribution method and system of rail transit station photovoltaic and energy storage system

Abstract

This invention belongs to the field of intelligent energy supply technology for rail transit, specifically involving an energy allocation method and system for photovoltaic and energy storage systems in rail transit stations. It aims to solve the problems of existing technologies, such as single feature learning dimensions, poor balance in multi-objective optimization, and lagging dynamic feedback response. The invention includes: constructing a multi-agent spatiotemporal correlation matrix; employing a deep belief network with spatiotemporal dual-scale attention fusion for feature learning; establishing a three-dimensional energy flow coupled tensor to mine energy supply and demand relationships; constructing a multi-objective optimization model based on the weighted ideal point method to generate a preliminary strategy; and using adaptive gradient feedback to correct the model and dynamically adjust parameters. Through multiple rounds of iterative optimization, the final energy allocation strategy is obtained. This invention solves the problems of single feature learning dimensions, poor balance in multi-objective optimization, and lagging dynamic feedback response in existing technologies, improving the accuracy of energy supply and demand prediction, and achieving synergistic optimization of energy economy, environmental protection, and equipment reliability.

Description

Technical Field

[0001] This invention belongs to the field of smart energy supply technology for rail transit, and specifically relates to an energy distribution method and system for a photovoltaic and energy storage system in rail transit stations. Background Technology

[0002] With the continuous expansion of rail transit operations and the deepening of the "dual-carbon" strategic goals, deploying distributed photovoltaic-energy storage co-generation systems at stations has become an important technological development direction for reducing dependence on the public power grid and improving energy supply flexibility. However, existing energy distribution strategies still have significant shortcomings in meeting the high reliability, high economy, and high environmental protection requirements of rail transit stations, mainly in the following three aspects:

[0003] First, the feature learning dimension is limited. Traditional strategies generally employ a "time-series prediction combined with static rules" approach, primarily focusing on the temporal variation of parameters such as photovoltaic output and load demand, while neglecting energy transmission losses caused by physical distance between devices and the impact of different device spatial layouts on power supply efficiency. For example, within the same time period, the direct power supply efficiency of near-distance photovoltaic power to the load is significantly higher than that of long-distance power supply, but traditional strategies fail to reflect this difference in energy allocation, leading to an increase in overall system energy loss.

[0004] Secondly, the multi-objective optimization suffers from poor balance. Existing methods often employ linear weighting to integrate multiple objectives such as "lowest electricity purchase cost, highest photovoltaic absorption rate, and longest energy storage lifespan," which can easily lead to unintended consequences in the optimization results. For example, excessive use of energy storage discharge in pursuit of minimizing electricity purchase cost will accelerate the degradation of the energy storage unit's cycle life; conversely, forced charging of energy storage to improve photovoltaic absorption may cause overloading during off-peak hours, threatening the operational safety of the regional power grid.

[0005] Secondly, dynamic feedback response is lagging. Traditional feedback correction mechanisms typically employ gradient descent with a fixed learning rate. When the system encounters sudden changes in illumination or a sharp increase in load, the model parameters adjust slowly, often requiring multiple iterations to converge again, leading to a disconnect between the energy allocation strategy and the actual operating state. This lag can cause operational risks such as power supply gaps for critical loads in scenarios like rail transit stations where power supply reliability is extremely important.

[0006] Therefore, there is an urgent need in this field for a comprehensive energy allocation method that can achieve cross-scale feature learning, multi-objective equilibrium optimization and adaptive dynamic feedback, so as to comprehensively improve the operating efficiency, economy and stability of the energy supply system of rail transit stations. Summary of the Invention

[0007] To address the aforementioned problems in existing technologies, namely the single feature learning dimension, poor balance in multi-objective optimization, and lag in dynamic feedback response, this invention provides an energy distribution method and system for a photovoltaic and energy storage system in rail transit stations.

[0008] In a first aspect, the present invention proposes an energy distribution method for a photovoltaic and energy storage system in a rail transit station, the method comprising:

[0009] A multi-entity spatiotemporal correlation matrix integrating photovoltaic system, energy storage system, station load and public power grid is constructed. The attributes, operating characteristics and constraints of each entity are encoded as node vectors, and weighted correlation edges are established based on historical interaction data.

[0010] The multi-agent spatiotemporal correlation matrix is ​​feature extracted by a deep belief network with spatiotemporal dual-scale attention fusion. Based on parameters such as real-time photovoltaic output, remaining energy storage capacity, station load gap rate, and real-time grid load rate, the weights of the correlation edges are adaptively adjusted to achieve cross-scale aggregation of energy supply system state features.

[0011] A three-dimensional energy flow coupled tensor is constructed based on energy flow prediction and constraint verification, which integrates time series data, spatial correlation features and energy parameters.

[0012] The coupled tensor is processed using a combined model of time-series forecasting and spatial constraints to generate predicted sequences of photovoltaic output, load demand, remaining energy storage capacity, and grid interaction power.

[0013] A multi-objective optimization model is constructed with electricity purchase cost, photovoltaic absorption rate and energy storage life as objectives, and a preliminary energy allocation strategy is generated by combining operational constraints.

[0014] Based on the deviation between actual operating data and prediction results, the parameters of the deep belief network and the characteristics of the correlation matrix are dynamically corrected.

[0015] Repeat the prediction, optimization, and correction steps until the iteration termination condition is met, and output the final energy allocation strategy.

[0016] Furthermore, the attention weight calculation in the deep belief network of spatiotemporal dual-scale attention fusion is based on the following factors: physical distance between main nodes, average time difference of historical energy interaction, real-time photovoltaic output, real-time remaining capacity of energy storage, real-time power interaction of the grid, and maximum power purchase allowed by the grid; the attention weight of the associated edge is calculated by combining the spatial scale attention factor, the temporal scale attention factor, and the real-time parameter association weight coefficient.

[0017] Furthermore, the construction of the three-dimensional energy flow coupling tensor is achieved by coupling the time dimension feature matrix, the spatial dimension feature matrix, and the energy parameter dimension feature matrix. The time dimension feature matrix contains the normalized values ​​of historical time-series data and introduces a time decay factor. The spatial dimension feature matrix contains the spatial correlation degree between each subject and introduces a spatial correlation coefficient. The energy parameter dimension feature matrix contains the normalized values ​​of real-time energy parameters and introduces a parameter influence weight matrix. The three-dimensional energy flow coupling tensor is generated through Hadamard product and Kronecker product operations.

[0018] Furthermore, the combined model incorporates irradiance compensation when predicting photovoltaic output, charge / discharge efficiency and self-discharge compensation when predicting remaining energy storage capacity, passenger flow compensation when predicting load, and dynamic adjustment of the electricity price sensitivity coefficient when predicting grid interaction power.

[0019] Furthermore, the multi-objective optimization model comprehensively optimizes the normalized electricity purchase cost, photovoltaic absorption rate, and energy storage life loss through the weighted ideal point method, and dynamically allocates the weight coefficients of each objective.

[0020] Furthermore, the dynamic correction process includes: calculating the error level based on the prediction loss function, dynamically adjusting the learning rate, updating the network parameters using gradient descent, and adjusting the edge weights of the association matrix according to the actual interaction frequency.

[0021] Furthermore, when constructing the multi-subject spatiotemporal correlation matrix, the node feature vector integrates the device's physical attributes, operating characteristics, and constraint boundaries to form a composite feature code.

[0022] Furthermore, when generating the power grid interaction prediction sequence, a price sensitivity coefficient is introduced in conjunction with the division of peak and valley electricity price periods to dynamically correct the prediction value.

[0023] Furthermore, when constructing the multi-objective optimization model, the load guarantee requirements during peak passenger flow periods and the emergency response time of energy storage when photovoltaic output drops sharply are included in the constraints.

[0024] In a second aspect, the present invention proposes an energy distribution system for a photovoltaic and energy storage system in a rail transit station, based on an energy distribution method for such a system, the system comprising:

[0025] The spatiotemporal correlation matrix construction module is configured to construct a multi-entity spatiotemporal correlation matrix integrating photovoltaic system, energy storage system, station load and public power grid, encode the attributes, operating characteristics and constraints of each entity as node vectors, and establish weighted correlation edges based on historical interaction data;

[0026] The cross-scale feature learning module is configured to extract features from the correlation matrix through a deep belief network that fuses spatiotemporal dual-scale attention, and adaptively adjust the weights of the correlation edges to achieve cross-scale aggregation of energy supply system state features.

[0027] The energy flow coupling tensor construction module is configured to extract features from the multi-subject spatiotemporal correlation matrix through a deep belief network with spatiotemporal dual-scale attention fusion. Based on parameters such as real-time photovoltaic output, remaining energy storage capacity, station load gap rate, and real-time grid load rate, it adaptively adjusts the weights of the correlation edges to achieve cross-scale aggregation of energy supply system state features.

[0028] The multi-parameter prediction module is configured to process the coupled tensor using a combined model of time-series prediction and spatial constraints to generate prediction sequences for photovoltaic output, load demand, remaining energy storage capacity, and grid interaction power.

[0029] The multi-objective strategy generation module is configured to construct a multi-objective optimization model with electricity purchase cost, photovoltaic absorption rate and energy storage lifetime as objectives, and generate a preliminary energy allocation strategy in combination with operational constraints.

[0030] An adaptive feedback correction module is configured to dynamically correct the parameters of the deep belief network and the features of the correlation matrix based on the deviation between the actual running data and the prediction results.

[0031] The iterative optimization module is configured to repeatedly execute the prediction, optimization, and correction steps until the iteration termination condition is met, and output the final energy allocation strategy.

[0032] The beneficial effects of this invention are:

[0033] This invention overcomes the limitations of traditional methods that only focus on changes in the time dimension by introducing a spatiotemporal dual-scale attention fusion mechanism. The deep belief network simultaneously considers the spatial layout of equipment and differences in operating periods, achieving cross-scale aggregation of the state characteristics of the power supply system. This significantly improves the comprehensiveness and accuracy of feature representation, effectively ensuring the prediction accuracy of key parameters such as photovoltaic output and load demand.

[0034] To address the balance issue in multi-objective optimization, this invention employs a weighted ideal point method to construct the optimization model. By normalizing different objective functions and introducing a dynamic weight allocation mechanism, it achieves synergistic improvement of multiple optimization objectives such as electricity purchase cost, photovoltaic grid integration rate, and energy storage lifetime, avoiding the mutual constraints between objectives that are easily caused by traditional linear weighting methods.

[0035] Regarding dynamic response capabilities, this invention designs an adaptive gradient feedback correction model. This model can automatically adjust the learning rate and network parameters according to the degree of deviation between the actual operating state of the system and the prediction results, significantly accelerating the convergence speed of the model under sudden operating conditions and enhancing the system's adaptability to photovoltaic power output fluctuations and load changes.

[0036] Furthermore, this invention fully considers the special operational needs of rail transit stations, incorporating passenger flow load correlation characteristics and emergency power supply response requirements into the optimization constraints. This ensures that the final energy allocation strategy not only meets the theoretical optimization objectives but also has good engineering applicability, directly guiding on-site equipment scheduling and ensuring the power supply reliability of critical loads. Attached Figure Description

[0037] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0038] Figure 1 This is a flowchart of an energy distribution method for a photovoltaic and energy storage system in a rail transit station according to the present invention;

[0039] Figure 2 This is a structural diagram of an energy distribution system for a photovoltaic and energy storage system in a rail transit station according to the present invention;

[0040] Figure 3 A schematic diagram of the structure of a computer system used to implement the methods, systems, and electronic devices of this application. Detailed Implementation

[0041] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0042] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0043] The first embodiment of the present invention provides an energy distribution method for a photovoltaic and energy storage system in a rail transit station, the method comprising:

[0044] Step S10: Construct a multi-entity spatiotemporal correlation matrix integrating photovoltaic system, energy storage system, station load and public power grid, encode the attributes, operating characteristics and constraints of each entity as node vectors, and establish weighted correlation edges based on historical interaction data;

[0045] Step S20: The multi-subject spatiotemporal correlation matrix is ​​feature extracted by a deep belief network with spatiotemporal dual-scale attention fusion. Based on the parameters of real-time photovoltaic output, remaining energy storage capacity, station load gap rate and real-time grid load rate, the weight of the correlation edge is adaptively adjusted to realize cross-scale energy supply system state feature aggregation.

[0046] Step S30: Construct a three-dimensional energy flow coupling tensor based on energy flow prediction and constraint verification, and fuse time series data, spatial correlation features and energy parameters;

[0047] Step S40: The coupled tensor is processed using a combined model of time-series prediction and spatial constraints to generate predicted sequences of photovoltaic output, load demand, remaining energy storage capacity and grid interaction power.

[0048] Step S50: Construct a multi-objective optimization model with electricity purchase cost, photovoltaic absorption rate and energy storage life as objectives, and generate a preliminary energy allocation strategy in combination with operational constraints;

[0049] Step S60: Based on the deviation between the actual running data and the prediction results, dynamically correct the parameters of the deep belief network and the characteristics of the correlation matrix;

[0050] Step S70: Repeat the prediction, optimization and correction steps until the iteration termination condition is met, and output the final energy allocation strategy.

[0051] To more clearly illustrate the energy distribution method of a photovoltaic and energy storage system for rail transit stations according to the present invention, the following is in conjunction with... Figure 1 The steps in the embodiments of the present invention are described in detail below:

[0052] Step S10: Construct a multi-entity spatiotemporal correlation matrix integrating photovoltaic system, energy storage system, station load and public power grid, encode the attributes, operating characteristics and constraints of each entity as node vectors, and establish weighted correlation edges based on historical interaction data;

[0053] In this embodiment, when constructing the multi-subject spatiotemporal correlation matrix, the node feature vector integrates the device's physical attributes, operating characteristics, and constraint boundaries to form a composite feature code.

[0054] In practical implementation, the first step is to encode node features to form a composite feature vector with a unified dimension. For photovoltaic system nodes, the parameters that need to be encoded include core attributes such as installed capacity, module orientation, temperature response characteristics, historical output fluctuation patterns, and irradiance response characteristics.

[0055] For energy storage system nodes, the coding parameters include key information such as rated capacity, charge-discharge conversion efficiency, cycle life characteristics, charge-discharge voltage range, battery type, and historical remaining capacity variation range.

[0056] For station load nodes, it is necessary to distinguish between critical loads and non-critical loads. The power consumption priority, correlation characteristics with external factors such as passenger flow, power fluctuation patterns, and peak load percentages of each type of load must be clearly defined. In this embodiment, critical loads may include traction assistance and emergency systems, while non-critical loads include air conditioning and lighting.

[0057] For public power grid nodes, the coding parameters include constraints such as peak-valley electricity price difference rules, regional power supply limits, voltage level standards, allowable frequency fluctuation range, and power supply reliability level. All parameters are normalized and integrated into a unified node feature vector to ensure that physical attributes, operational characteristics, and constraint boundaries are fully represented.

[0058] Secondly, the spatiotemporal interaction intensity is calculated based on long-term accumulated historical energy interaction data, and weighted correlation edges are established. The spatiotemporal interaction intensity is determined by statistically analyzing the energy interaction frequency and spatial relationships between various entities. This includes the direct power supply frequency between photovoltaics and loads, the excess power charging frequency between photovoltaics and energy storage, the emergency power supply frequency between energy storage and critical loads, and the off-peak charging frequency between the grid and energy storage.

[0059] At the same time, the physical distance between devices needs to be considered. The spatial correlation coefficient model is used to calculate the spatial correlation degree. This model adjusts the correlation weight by the ratio of the device spacing to the maximum allowable spacing, thereby strengthening the energy interaction influence of close-range subjects.

[0060] In the specific process of constructing the associated edges, it is necessary to calculate the initial weight of each edge based on historical energy interaction data and spatial relationships. The higher the interaction frequency and the stronger the spatial correlation of the subject pairs, the higher the initial weight. For associated edges related to critical loads, the weight allocation is further adjusted in combination with power supply priority to ensure that the core energy supply relationships are characterized in detail.

[0061] Through the above implementation, the multi-agent spatiotemporal correlation matrix can accurately depict the spatiotemporal dependencies between photovoltaics, energy storage, loads, and the power grid, ensuring the accuracy and adaptability of energy allocation strategies. Those skilled in the art can adjust parameter values ​​according to specific station configurations and automatically calculate interaction strengths and weights using historical data, thus enabling the engineering application of the method.

[0062] Step S20: The multi-subject spatiotemporal correlation matrix is ​​feature extracted by a deep belief network with spatiotemporal dual-scale attention fusion. Based on the parameters of real-time photovoltaic output, remaining energy storage capacity, station load gap rate and real-time grid load rate, the weight of the correlation edge is adaptively adjusted to realize cross-scale energy supply system state feature aggregation.

[0063] In this embodiment, the attention weight calculation in the spatiotemporal dual-scale attention fusion deep belief network is based on the following factors:

[0064] Physical distance between main nodes, average time difference of historical energy interaction, real-time photovoltaic output, real-time remaining capacity of energy storage, real-time power interaction with the grid, and maximum power purchase allowed by the grid.

[0065] The attention weight of the associated edge is calculated by combining the spatial scale attention factor, the temporal scale attention factor, and the real-time parameter association weight coefficient.

[0066] In this embodiment, the attention weight calculation model of the deep belief network (ST-DBN) with spatial dual-scale attention fusion is as follows:

[0067] ;

[0068] Where, ω ij d represents the attention weight of the edge connecting node i (e.g., photovoltaic) and node j (e.g., load); ij D represents the physical distance between nodes i and j (in meters). max The maximum equipment spacing for the station-side power supply system (default value is 500m); ΔT ij P represents the average time difference (in hours) of historical energy interactions between nodes i and j. pv , i The real-time power output (in kW) of the photovoltaic system corresponding to node i; SOC j P represents the real-time remaining capacity of the energy storage corresponding to node j (in %). grid,i P represents the real-time interactive power between node i and the power grid (unit: kW, positive values ​​are purchased electricity); grid,max P represents the maximum power purchase capacity allowed by the power grid (unit: kW). pv,max Maximum installed capacity of photovoltaic system (unit: kW); SOC max The energy storage full capacity threshold (taken as 100%); α, β, δ These are the weighting coefficients for spatial, temporal, and real-time parameter associations (the sum of the three is 1, with default values ​​of 0.25, 0.25, and 0.5); M(i) is the set of valid associated nodes of node i (physical distance ≤ D). max And the historical interaction frequency is ≥5 times / week); d ik ΔT is the physical distance between node i and node k. ik The average time difference of historical energy interaction between the main node i and node k; SOC k This represents the real-time remaining capacity of the energy storage system corresponding to the main node k.

[0069] In practice, the system's real-time operating parameters are first collected at fixed time granularities, including real-time photovoltaic output and fluctuation rate, current remaining energy storage capacity, station load gap, and real-time grid load carrying status, to ensure the timeliness of the data.

[0070] In the calculation of attention weights, the spatial scale attention factor mainly considers the influence of physical distance between devices. Spatial correlation is quantified by a distance decay function; the temporal scale attention factor focuses on the temporal distribution characteristics of historical energy interactions and calculates the average time difference of historical interactions between nodes; the real-time parameter correlation weight coefficient comprehensively considers dynamic parameters such as real-time photovoltaic output, remaining energy storage capacity, and grid interaction power, and integrates them into a unified dimension influence factor after normalization.

[0071] Based on the above model, the weight values ​​of each associated edge are dynamically adjusted. For example, the attention weight of the photovoltaic and critical load edge is calculated as follows:

[0072] For high-priority correlations (such as the correlation between critical loads and emergency power supply of energy storage), the weight is adaptively increased based on real-time parameters; for low-priority correlations (such as the correlation between peak-hour grid and energy storage charging), the weight is appropriately reduced to suppress unnecessary energy interactions.

[0073] Deep belief networks aggregate the state features of energy supply systems across scales through a multi-layer hidden layer structure. The network output includes feature vectors and state labels of each node, providing accurate feature support for the generation of subsequent energy allocation strategies.

[0074] Through the above implementation process, the system can adaptively adjust the correlation relationship according to the real-time operating status, accurately capture the dynamic characteristics of the energy supply system, and provide precise feature support for the generation of subsequent energy allocation strategies. Those skilled in the art can adjust the network structure and parameter settings according to specific application scenarios to ensure the effective application of the method in practical engineering.

[0075] Step S30: Construct a three-dimensional energy flow coupling tensor based on energy flow prediction and constraint verification, and fuse time series data, spatial correlation features and energy parameters;

[0076] In this embodiment, the three-dimensional energy flow coupling tensor is constructed by coupling the time-dimensional feature matrix, the spatial-dimensional feature matrix, and the energy parameter-dimensional feature matrix. The time-dimensional feature matrix contains normalized values ​​of historical time-series data and incorporates a time decay factor; the spatial-dimensional feature matrix contains the spatial correlation between various entities and incorporates a spatial correlation coefficient; and the energy parameter-dimensional feature matrix contains normalized values ​​of real-time energy parameters and incorporates a parameter influence weight matrix. The three-dimensional energy flow coupling tensor is generated through Hadamard and Kronecker product operations.

[0077] The specific construction process adopts the following model:

[0078] ;

[0079] Where T is the three-dimensional energy flow coupling tensor; X time T is the time-dimensional feature matrix; λ X is the time decay factor matrix; space S is the spatial dimension feature matrix; μ This is a spatial correlation coefficient matrix, and the elements in the spatial correlation coefficient matrix are... Strengthen the association weight of nearby subjects). X represents the maximum equipment spacing for the station-side power supply system. param The energy parameter dimension feature matrix (elements are normalized values ​​of real-time parameters); P v represents the weight matrix affecting the parameters; ⊙ represents the Hadamard product (element-wise multiplication). This is the Kronecker product (tensor dimension extension).

[0080] More specifically, the matrices are defined as follows:

[0081] The time dimension feature matrix organizes historical time-series data at a fixed time step, covering the normalized results of operational data from four main entities: photovoltaic, energy storage, load, and power grid. The time decay factor matrix strengthens the weight of recent data and weakens the impact of long-term data on current decisions through preset rules.

[0082] The spatial dimension feature matrix quantifies the spatial correlation between various subjects, and the spatial correlation coefficient matrix adjusts the weights by the ratio of the device spacing to the maximum allowable spacing, highlighting the energy interaction influence of subjects in close proximity.

[0083] The energy parameter dimension feature matrix integrates the normalized values ​​of key parameters such as power output, remaining capacity, load demand, electricity price, and conversion efficiency. The parameter influence weight matrix assigns higher weights to core parameters and corresponding weights to secondary parameters, ensuring that key information receives focused attention.

[0084] By using the Hadamard product (element-level multiplication) and the Kronecker product (tensor dimension expansion), the feature matrices of the three dimensions of time, space, and parameters are coupled and fused to generate a three-dimensional energy flow coupled tensor, thus achieving the organic integration of multi-dimensional information.

[0085] In practical applications, taking a subway transfer station as an example, this station is equipped with a 120kW rooftop photovoltaic array and a 250kWh lithium iron phosphate energy storage system, with an average daily load of 150kW. The three-dimensional energy flow coupling tensor constructed using the above method can accurately reflect the energy supply and demand dependence at different spatiotemporal scales, providing a reliable data foundation for subsequent energy allocation strategies. Those skilled in the art can adjust the dimensions and element values ​​of each matrix according to the specific station's configuration parameters to achieve flexible application of the method.

[0086] Step S40: The coupled tensor is processed using a combined model of time-series prediction and spatial constraints to generate predicted sequences of photovoltaic output, load demand, remaining energy storage capacity and grid interaction power.

[0087] In this embodiment, the combined model introduces irradiance compensation when predicting photovoltaic output, charging and discharging efficiency and self-discharge compensation when predicting remaining energy storage capacity, passenger flow compensation when predicting load, and dynamic adjustment of electricity price sensitivity coefficient when predicting grid interaction power.

[0088] When generating the power grid interaction prediction sequence, a price sensitivity coefficient is introduced in conjunction with the division of peak and valley electricity price periods to dynamically correct the prediction value.

[0089] In step S40, a combined model of time-series prediction and spatial constraints is used to process the three-dimensional energy flow coupling tensor, generating predicted sequences for photovoltaic output, load demand, remaining energy storage capacity, and grid interaction power. This combined model employs a collaborative architecture of an LSTM time-series prediction submodule and a Conv1D spatial constraint submodule to achieve accurate prediction of energy flow characteristics.

[0090] In this embodiment, the combined model incorporates irradiance compensation when predicting photovoltaic output, charge / discharge efficiency and self-discharge compensation when predicting remaining energy storage capacity, passenger flow compensation when predicting load, and dynamic adjustment of the electricity price sensitivity coefficient when predicting grid interconnection power. The specific methods for generating each prediction sequence are as follows:

[0091] Photovoltaic power output prediction uses an LSTM time-series prediction function to process historical photovoltaic time-series data and incorporates an irradiance compensation coefficient. The prediction formula is:

[0092] ;

[0093] in X represents the predicted photovoltaic power output at time t, in kW; LSTM(·) is the LSTM time series prediction function; pv,time This represents historical time-series photovoltaic data; ε is the irradiance compensation coefficient, defaulting to 0.3; G t The real-time illumination intensity at time t is expressed in W / m². 2 Gavg This represents the historical average irradiance. For example, during the peak irradiance period from 12:00 to 14:00, the baseline predicted value is 115 kW. t =1100W / m 2 G avg =800W / m 2 At that time, the corrected prediction value reached 127kW.

[0094] The prediction of remaining energy storage capacity takes into account both charging and discharging power and efficiency factors, and adopts the following update formula:

[0095] ;

[0096] in Let t be the predicted SOC of the energy storage. n charge η discharge These represent the energy storage charge / discharge efficiency, with default values ​​of 0.92 and 0.9 respectively; P charge,t P discharge,t These represent the energy storage charging and discharging power at time t, in kW; C ess,rated The rated energy storage capacity is expressed in kWh; Δt is the time step of 0.5 h; ζ is the SOC correction factor, defaulted to 0.98, used to correct for the effects of battery self-discharge. For example, if the SOC is 38% at the start of the off-peak period at 22:00, and the battery is charged at 25kW for 10 hours, the predicted SOC can reach 64%.

[0097] The load demand forecasting formula combines the Conv1D spatial constraint function and passenger flow compensation mechanism, and is as follows:

[0098] ;

[0099] in represents the predicted load demand at time t, in kW; Conv1D(·) is the one-dimensional convolutional spatial constraint function; ST-DBN_feat is the spatial feature extracted by ST-DBN; γ is the passenger flow compensation coefficient, defaulting to 0.4; Q t Q represents the real-time passenger flow at the station at time t, in persons; avg This represents the historical average passenger flow. For example, if the passenger flow at 18:00 during the evening peak is 12,000 people, and the historical average is 8,000 people, the load forecast after compensation will reach 142kW.

[0100] When generating the power grid interaction prediction sequence, a price sensitivity coefficient is introduced based on the grid peak-valley electricity price period division to dynamically correct the predicted value. The grid peak-valley period is divided into peak hours (8:00-22:00) and valley hours (22:00-8:00 the next day), and a corresponding price sensitivity coefficient ξ is set. t The peak value is 1.2, and the trough value is 0.8. The corrected formula is:

[0101] ;

[0102] in The revised power grid interaction forecast is in kW. This is the base forecast value. For example, the base forecast for grid power purchase at 2:00 PM during off-peak hours is 65kW, which is adjusted to 52kW after being corrected by the electricity price sensitivity factor.

[0103] The combined model ensures that the prediction accuracy meets the requirements of engineering applications by reasonably configuring the input sequence length, number of hidden layer units, and dropout rate of the temporal prediction submodule, and spatially constraining the convolutional kernel size, stride, and number of output channels of the submodule. The prediction time range covers the entire future power supply scheduling cycle, generating continuous prediction sequences with a fixed time step.

[0104] Through the above implementation methods, the system can generate accurate prediction sequences for photovoltaic power output, load demand, remaining energy storage capacity, and grid interaction power, providing reliable input data for subsequent multi-objective optimization. Those skilled in the art can adjust the compensation coefficients and model parameters according to specific application scenarios to ensure that the prediction accuracy meets the requirements of engineering applications.

[0105] Step S50: Construct a multi-objective optimization model with electricity purchase cost, photovoltaic absorption rate and energy storage life as objectives, and generate a preliminary energy allocation strategy in combination with operational constraints;

[0106] In this embodiment, the multi-objective optimization model comprehensively optimizes the normalized electricity purchase cost, photovoltaic absorption rate and energy storage life loss through the weighted ideal point method, and dynamically allocates the weight coefficients of each objective.

[0107] Furthermore, when constructing the multi-objective optimization model, the load guarantee requirements during peak passenger flow periods and the emergency response time of energy storage when photovoltaic output drops sharply are included in the constraints.

[0108] The optimization model is constructed using the following method:

[0109] The multi-objective optimization function is defined as:

[0110] ;

[0111] ;

[0112] The parameters are defined as follows:

[0113] F is a multi-objective optimization function, a normalized value, ranging from 0 to 1, with values ​​closer to 0 indicating better performance. ω1, ω2, and ω3 are the weighting coefficients for cost, photovoltaic grid integration, and energy storage lifetime, respectively, and their sum is 1, with default values ​​of 0.4, 0.3, and 0.3. C gridC represents the total cost of purchasing electricity from the power grid, expressed in yuan. grid,min C grid,max These represent the theoretical minimum and maximum values ​​of electricity purchase cost, respectively; η pv Photovoltaic grid integration rate, in percentages (%). η pv,min η pv,max These are the minimum and maximum target values ​​for photovoltaic grid integration rate, respectively; L ess This is the energy storage lifespan loss factor, expressed as % / year, L. ess,min L ess,max These represent the minimum and maximum values ​​of energy storage lifetime loss, respectively; P grid,t The grid interaction power at time t is expressed in kW; SOC ess,t Let P be the state of energy stored at time t; pv,t P represents the photovoltaic output at time t, in kW. supply,critical,t P represents the critical load power supplied at time t, in kW. load,critical,t ΔP represents the critical load demand at time t, in kW. ess,t Let t be the change in energy storage charging and discharging power at time t, in kW;

[0114] When constructing the multi-objective optimization model, the load guarantee requirements during peak passenger flow periods and the emergency response time of energy storage in the event of a sudden drop in photovoltaic output are included in the constraints. Specifically, these include:

[0115] The optimization model is constructed with the core objectives of minimizing electricity purchase costs, maximizing photovoltaic (PV) grid integration rate, and minimizing energy storage lifespan losses. Multi-objective collaborative optimization is achieved by dynamically allocating weight coefficients for each objective. The objective function, after normalization, ranges from 0 to 1, with values ​​closer to 0 indicating better optimization performance.

[0116] When constructing the multi-objective optimization model, the load guarantee requirements during peak passenger flow periods and the energy storage emergency response time when photovoltaic output suddenly drops are included in the constraints. Among them, the load guarantee constraint requires that during the peak passenger flow period of the station, the power supply reliability of critical loads should be ensured to meet the preset standard through a preset load guarantee coefficient; the energy storage emergency response constraint requires that when the sudden drop in photovoltaic output exceeds a preset threshold, the energy storage emergency discharge response time should not exceed a preset duration to ensure power supply continuity.

[0117] Equipment operation constraints include the safe operating range of remaining energy storage capacity, charging and discharging power limits, grid interaction power limits, energy storage cycle life constraints, and photovoltaic absorption rate targets. All constraints are reasonably set according to equipment characteristics and operating requirements to ensure the feasibility and safety of optimization results.

[0118] The optimization solution adopts an intelligent optimization algorithm adapted to multi-objective optimization scenarios. By pre-setting the population size and number of iterations, it searches for the optimal solution within the constraints and generates a preliminary energy allocation strategy solution set that satisfies all constraints.

[0119] Through the above implementation methods, the system can generate a preliminary energy allocation strategy that balances economy, environmental protection, and equipment reliability, laying the foundation for subsequent feedback correction and iterative optimization. Those skilled in the art can adjust the weighting coefficients and constraint parameters according to the specific operating characteristics and needs of each station to achieve personalized applications of the method.

[0120] Step S60: Based on the deviation between the actual running data and the prediction results, dynamically correct the parameters of the deep belief network and the characteristics of the correlation matrix;

[0121] The dynamic correction process includes: calculating the error level based on the prediction loss function, dynamically adjusting the learning rate, updating the network parameters using gradient descent, and adjusting the edge weights of the association matrix according to the actual interaction frequency.

[0122] The prediction loss function is used to quantify the degree of deviation between the actual operating data and the prediction results. Its calculation formula is as follows:

[0123] ;

[0124] Where L is the normalized prediction loss function; T is the prediction time step; and N is the number of subjects. , These are the actual power and predicted power of subject i at time t, respectively, in kW; , , respectively, represent the actual SOC and predicted SOC of subject i (energy storage) at time t, in %; L is the prediction loss function (normalized value, ranging from 0 to 10%).

[0125] The error level coefficient E is dynamically calculated based on the loss function value. t Specifically, it is divided into three levels:

[0126] ;

[0127] When the loss function value L does not exceed 2%, the error rating coefficient Et is 0.2; when L is between 2% and 5%, Et is... t Take 0.4; when L exceeds 5%, E t We set it to 0.6. This grading mechanism allows the system to take corrective measures of appropriate strength according to the severity of the error.

[0128] Network parameter correction employs gradient descent with an adaptive learning rate. The parameter update formula is as follows:

[0129] ;

[0130] Where θ t θ t+1 η represents the network parameter set of ST-DBN at times t and t+1, including the weight matrix and bias vector; η is the basic learning rate, initially set to 0.0015, and dynamically adjusted according to the rule of decaying by 15% every 30 iterations; κ is the error amplification factor, with a default value of 0.3. This represents the gradient of the loss function with respect to the parameter θ.

[0131] In practical applications, the system collects actual operational data after the initial strategy is implemented, compares and analyzes the differences between the actual operational data and the predicted data, calculates the prediction loss, and identifies the sources of error. Based on the loss value, the network weight parameters of ST-DBN are dynamically adjusted, and the weights of associated edges are also dynamically adjusted based on the actual energy interaction frequency, ensuring that the association matrix accurately reflects the main relationships under the actual operating state of the system.

[0132] Through this dynamic correction mechanism, the system can quickly respond to changes in its operating status, continuously improving prediction accuracy and strategy effectiveness. Those skilled in the art can adjust the loss function threshold and learning rate parameters according to specific application scenarios to optimize the correction effect and meet different application needs.

[0133] Step S70: Repeat the prediction, optimization and correction steps until the iteration termination condition is met, and output the final energy allocation strategy.

[0134] In step S70, the system repeatedly performs prediction, optimization, and correction steps to obtain the final energy allocation strategy through multiple rounds of iterative optimization. This iterative optimization process continues until a preset iteration termination condition is met, ensuring that the output strategy achieves an optimal balance among multiple objectives.

[0135] The iteration termination condition is set with two key indicators: the rate of change of the objective function value in multiple consecutive iterations is less than a preset threshold, and the prediction error is controlled within an acceptable range. These two conditions must be met simultaneously to ensure that the system maintains high prediction accuracy while achieving convergence stability. In this embodiment, the rate of change of the objective function value in multiple consecutive iterations is preferably less than the preset threshold of 0.5%, with an acceptable range of no more than 3%. Those skilled in the art can set appropriate values ​​according to actual conditions, and this invention does not impose specific limitations here. In the specific implementation process, the system performs iterative optimization through the following steps: First, multi-parameter prediction is performed based on current parameters to generate prediction sequences for photovoltaic output, load demand, energy storage SOC, and grid interaction power; then, multi-objective optimization is performed using the weighted ideal point method to generate a preliminary energy allocation strategy; then, the strategy is applied to the actual system, and operational data is collected to calculate the prediction loss; finally, network parameters and correlation matrix characteristics are dynamically adjusted based on the loss value.

[0136] After multiple rounds of iterative optimization, the final energy allocation strategy has achieved significant improvements in terms of time-series adaptability, spatial coordination, emergency response capabilities, and equipment protection characteristics: it can dynamically adjust the energy allocation scheme according to the photovoltaic output characteristics, load demand patterns, and electricity price differences at different times; it fully considers the spatial correlation characteristics and energy transmission efficiency among various entities to optimize energy flow paths; it can quickly adjust the energy storage scheduling and grid interaction strategies in response to sudden situations such as sudden drops in photovoltaic output and load changes; and it reduces energy storage lifespan loss and achieves long-term stable operation of equipment by reasonably controlling the depth and frequency of energy storage charging and discharging.

[0137] See Figure 2 The second embodiment of the present invention provides an energy distribution system for a photovoltaic and energy storage system in a rail transit station, based on an energy distribution method for such a system. The system includes:

[0138] The spatiotemporal correlation matrix construction module is configured to construct a multi-entity spatiotemporal correlation matrix integrating photovoltaic system, energy storage system, station load and public power grid, encode the attributes, operating characteristics and constraints of each entity as node vectors, and establish weighted correlation edges based on historical interaction data;

[0139] The cross-scale feature learning module is configured to extract features from the correlation matrix through a deep belief network that fuses spatiotemporal dual-scale attention, and adaptively adjust the weights of the correlation edges to achieve cross-scale aggregation of energy supply system state features.

[0140] The energy flow coupling tensor construction module is configured to extract features from the multi-subject spatiotemporal correlation matrix through a deep belief network with spatiotemporal dual-scale attention fusion. Based on parameters such as real-time photovoltaic output, remaining energy storage capacity, station load gap rate, and real-time grid load rate, it adaptively adjusts the weights of the correlation edges to achieve cross-scale aggregation of energy supply system state features.

[0141] The multi-parameter prediction module is configured to process the coupled tensor using a combined model of time-series prediction and spatial constraints to generate prediction sequences for photovoltaic output, load demand, remaining energy storage capacity, and grid interaction power.

[0142] The multi-objective strategy generation module is configured to construct a multi-objective optimization model with electricity purchase cost, photovoltaic absorption rate and energy storage lifetime as objectives, and generate a preliminary energy allocation strategy in combination with operational constraints.

[0143] An adaptive feedback correction module is configured to dynamically correct the parameters of the deep belief network and the features of the correlation matrix based on the deviation between the actual running data and the prediction results.

[0144] The iterative optimization module is configured to repeatedly execute the prediction, optimization, and correction steps until the iteration termination condition is met, and output the final energy allocation strategy.

[0145] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related explanations of the methods described above can be found in the corresponding processes in the foregoing system embodiments, and will not be repeated here.

[0146] It should be noted that the energy distribution system of the photovoltaic and energy storage system for rail transit stations provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the various modules or steps and are not considered as an improper limitation of the present invention.

[0147] A device according to a third embodiment of the present invention includes:

[0148] At least one processor;

[0149] and a memory communicatively connected to at least one of the processors;

[0150] The memory stores instructions that can be executed by the processor to implement the above-described energy distribution method for a photovoltaic and energy storage system in a rail transit station.

[0151] A fourth embodiment of the present invention provides a computer-readable storage medium storing computer instructions, which are executed by the computer to implement the above-described energy distribution method for a photovoltaic and energy storage system in a rail transit station.

[0152] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the storage device and processing device described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0153] The following is for reference. Figure 3 It shows a schematic diagram of the structure of a computer system for implementing embodiments of the systems, methods, and electronic devices of this application. Figure 3 The server shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0154] like Figure 3 As shown, the computer system includes a Central Processing Unit (CPU) 301, which can perform various appropriate actions and processes based on programs stored in Read Only Memory (ROM) 302 or programs loaded from storage section 308 into Random Access Memory (RAM) 303. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An Input / Output (I / O) interface 305 is also connected to the bus 304.

[0155] The following components are connected to I / O interface 305: an input section 306 including a keyboard, mouse, etc.; an output section 307 including a cathode ray tube (CRT), liquid crystal display (LCD), and speakers, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN (Local Area Network) card and a modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0156] Specifically, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by central processing unit (CPU) 301, it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0157] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0158] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0159] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0160] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0161] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for energy distribution in a photovoltaic and energy storage system for rail transit stations, characterized in that, The method includes: A multi-entity spatiotemporal correlation matrix integrating photovoltaic system, energy storage system, station load and public power grid is constructed. The attributes, operating characteristics and constraints of each entity are encoded as node vectors, and weighted correlation edges are established based on historical interaction data. The multi-agent spatiotemporal correlation matrix is ​​feature extracted by a deep belief network with spatiotemporal dual-scale attention fusion. Based on parameters such as real-time photovoltaic output, remaining energy storage capacity, station load gap rate, and real-time grid load rate, the weights of the correlation edges are adaptively adjusted to achieve cross-scale aggregation of energy supply system state features. A three-dimensional energy flow coupled tensor is constructed based on energy flow prediction and constraint verification, which integrates time series data, spatial correlation features and energy parameters. The coupled tensor is processed using a combined model of time-series forecasting and spatial constraints to generate predicted sequences of photovoltaic output, load demand, remaining energy storage capacity, and grid interaction power. A multi-objective optimization model is constructed with electricity purchase cost, photovoltaic absorption rate and energy storage life as objectives, and a preliminary energy allocation strategy is generated by combining operational constraints. Based on the deviation between actual operating data and prediction results, the parameters of the deep belief network and the characteristics of the correlation matrix are dynamically corrected. Repeat the prediction, optimization, and correction steps until the iteration termination condition is met, and output the final energy allocation strategy.

2. The energy distribution method of a photovoltaic and energy storage system for rail transit stations according to claim 1, characterized in that, The attention weights in the deep belief network with spatiotemporal dual-scale attention fusion are calculated based on the following factors: Physical distance between main nodes, average time difference of historical energy interaction, real-time photovoltaic output, real-time remaining capacity of energy storage, real-time power interaction with the grid, and maximum power purchase allowed by the grid. The attention weight of the associated edge is calculated by combining the spatial scale attention factor, the temporal scale attention factor, and the real-time parameter association weight coefficient.

3. The energy distribution method of a photovoltaic and energy storage system for rail transit stations according to claim 1, characterized in that, The construction of the three-dimensional energy flow coupling tensor is achieved by coupling the time dimension feature matrix, the spatial dimension feature matrix, and the energy parameter dimension feature matrix. The time dimension feature matrix contains the normalized values ​​of historical time-series data and introduces a time decay factor. The spatial dimension feature matrix contains the spatial correlation degree between each subject and introduces a spatial correlation coefficient. The energy parameter dimension feature matrix contains the normalized values ​​of real-time energy parameters and introduces a parameter influence weight matrix. The three-dimensional energy flow coupling tensor is generated by Hadamard product and Kronecker product operations.

4. The energy distribution method of a photovoltaic and energy storage system for rail transit stations according to claim 1, characterized in that, The combined model incorporates irradiance compensation when predicting photovoltaic output, charge / discharge efficiency and self-discharge compensation when predicting remaining energy storage capacity, passenger flow compensation when predicting load, and dynamic adjustment of the electricity price sensitivity coefficient when predicting grid interaction power.

5. The energy distribution method of a photovoltaic and energy storage system for rail transit stations according to claim 1, characterized in that, The multi-objective optimization model comprehensively optimizes the normalized electricity purchase cost, photovoltaic absorption rate and energy storage life loss through the weighted ideal point method, and dynamically allocates the weight coefficients of each objective.

6. The energy distribution method of a photovoltaic and energy storage system for rail transit stations according to claim 1, characterized in that, The dynamic correction process includes: calculating the error level based on the prediction loss function, dynamically adjusting the learning rate, updating the network parameters using gradient descent, and adjusting the edge weights of the association matrix according to the actual interaction frequency.

7. The energy distribution method of a photovoltaic and energy storage system for rail transit stations according to claim 1, characterized in that, When constructing the multi-subject spatiotemporal correlation matrix, the node feature vector integrates the physical attributes, operating characteristics and constraint boundaries of the equipment to form a composite feature code.

8. The energy distribution method of a photovoltaic and energy storage system for rail transit stations according to claim 1, characterized in that, When generating the power grid interaction prediction sequence, a price sensitivity coefficient is introduced in conjunction with the division of peak and valley electricity price periods to dynamically correct the prediction value.

9. The energy distribution method of a photovoltaic and energy storage system for rail transit stations according to claim 1, characterized in that, When constructing the multi-objective optimization model, the load guarantee requirements during peak passenger flow periods and the emergency response time of energy storage when photovoltaic output drops sharply are included in the constraints.

10. An energy distribution system for a photovoltaic and energy storage system in a rail transit station, based on the energy distribution method for a photovoltaic and energy storage system in a rail transit station as described in any one of claims 1-9, characterized in that, The system includes: The spatiotemporal correlation matrix construction module is configured to construct a multi-entity spatiotemporal correlation matrix integrating photovoltaic system, energy storage system, station load and public power grid, encode the attributes, operating characteristics and constraints of each entity as node vectors, and establish weighted correlation edges based on historical interaction data; The cross-scale feature learning module is configured to extract features from the correlation matrix through a deep belief network that fuses spatiotemporal dual-scale attention, and adaptively adjust the weights of the correlation edges to achieve cross-scale aggregation of energy supply system state features. The energy flow coupling tensor construction module is configured to extract features from the multi-subject spatiotemporal correlation matrix through a deep belief network with spatiotemporal dual-scale attention fusion. Based on parameters such as real-time photovoltaic output, remaining energy storage capacity, station load gap rate, and real-time grid load rate, it adaptively adjusts the weights of the correlation edges to achieve cross-scale aggregation of energy supply system state features. The multi-parameter prediction module is configured to process the coupled tensor using a combined model of time-series prediction and spatial constraints to generate prediction sequences for photovoltaic output, load demand, remaining energy storage capacity, and grid interaction power. The multi-objective strategy generation module is configured to construct a multi-objective optimization model with electricity purchase cost, photovoltaic absorption rate and energy storage lifetime as objectives, and generate a preliminary energy allocation strategy in combination with operational constraints. An adaptive feedback correction module is configured to dynamically correct the parameters of the deep belief network and the features of the correlation matrix based on the deviation between the actual running data and the prediction results. The iterative optimization module is configured to repeatedly execute the prediction, optimization, and correction steps until the iteration termination condition is met, and output the final energy allocation strategy.

Images