Coordination control method, device and equipment suitable for optical storage hybrid system and medium thereof

Abstract

The invention relates to a coordination control method and device suitable for an optical storage hybrid system, equipment and a medium. The method is applied to an optical storage hybrid system, and comprises the following steps: collecting multi-source data, and generating a real-time data vector; based on the vector, generating a photovoltaic output prediction sequence, a load demand prediction sequence and a battery life attenuation rate prediction sequence, and constructing a time alignment prediction data set; according to an energy storage system charge state value and a power grid real-time electricity price in the vector, combining a battery life attenuation rate prediction sequence in the time alignment prediction data set, and constructing a life-economy collaborative optimization objective function; and performing model prediction control optimization processing through the function and the time alignment prediction data set, constructing a strategy library, starting a corresponding operation mode, and outputting an equipment control instruction. According to the method, through multi-dimensional data acquisition and dynamic optimization control, the prediction precision and the full life cycle benefit of the optical storage hybrid system are improved, and the operation stability and reliability of the system are enhanced.

Description

Technical Field

[0001] This invention belongs to the field of computer technology, and in particular relates to a coordinated control method, apparatus, equipment and medium applicable to hybrid optical-storage systems. Background Technology

[0002] As the global energy structure shifts towards cleaner energy sources, photovoltaic (PV) power generation has been widely adopted due to its renewable resources and environmental friendliness. However, PV output is significantly intermittent and fluctuating due to meteorological factors such as sunlight and temperature, and direct grid connection can impact grid frequency and voltage stability. Hybrid PV-storage systems, by introducing energy storage units such as lithium batteries and supercapacitors, can effectively mitigate PV output fluctuations and enhance the absorption capacity of new energy sources. They have become one of the core application forms in the new energy field, and their coordinated control technology is crucial to ensuring the safe and efficient operation of the system.

[0003] Most existing methods for coordinated control of photovoltaic-storage hybrid systems adopt a hierarchical architecture of prediction-optimization-execution, but these methods still have many technical limitations in practical applications. First, the prediction stage often uses a single time-scale model, relying solely on historical photovoltaic power or meteorological data for forecasting, without compensating for and correcting based on the real-time operating status of the energy storage system. This results in insufficient accuracy in photovoltaic output prediction and battery life prediction. Second, the objective function construction often focuses on single-dimensional optimization, such as pursuing only grid interaction economics or simply emphasizing battery life protection, without establishing a dynamic coordination mechanism between lifespan loss and economic costs, making it difficult to maximize the benefits throughout the system's entire lifecycle. Third, the constraints in the optimization process are mostly statically set, such as fixing the state of charge (SOC) boundary and charge / discharge power threshold of the energy storage system, without dynamically adjusting the constraint range according to battery life protection requirements. This can easily lead to overcharging and over-discharging risks when the battery's state of health (SOH) declines. Furthermore, this method primarily employs offline evaluation for the lifespan management of energy storage systems. It updates model parameters by periodically detecting battery state of health (SOH), failing to quantify lifespan loss online and dynamically correct the model based on real-time operational data. This leads to a disconnect between the control strategy and the actual aging state of the battery, further reducing control accuracy and system reliability. Summary of the Invention

[0004] Therefore, it is necessary to provide coordinated control methods, devices, equipment and media suitable for photovoltaic-storage hybrid systems to address the above-mentioned technical problems, aiming to improve the prediction accuracy and scientific nature of optimization decisions of photovoltaic-storage hybrid systems, and enhance the stability, reliability and economic efficiency of system operation throughout its entire life cycle.

[0005] In a first aspect, this application provides a coordinated control method applicable to a photovoltaic-storage hybrid system. This method, applied to the photovoltaic-storage hybrid system, includes:

[0006] The system collects photovoltaic array output power, energy storage system state of charge, energy storage system health status, battery aging coefficient, DC bus voltage, load power, and real-time grid electricity price to generate real-time data vectors. Based on the real-time data vectors, it generates photovoltaic output prediction sequences, load demand prediction sequences, and battery life degradation rate prediction sequences to construct a time-aligned prediction dataset.

[0007] Based on the energy storage system's state of charge value and the grid's real-time electricity price in the real-time data vector, and combined with the battery life degradation rate prediction sequence in the time-aligned prediction dataset, a life-economic co-optimization objective function is constructed.

[0008] Model predictive control optimization is performed using a life-economic co-optimization objective function and a time-aligned prediction dataset. A strategy library is then constructed. Based on the strategy library and real-time data vectors, the corresponding operating modes are enabled, and device control commands containing the set power values ​​for the supercapacitor and the lithium battery are output. The strategy library also contains charging and discharging commands for future preset time periods.

[0009] In one embodiment, based on real-time data vectors, a photovoltaic power output prediction sequence, a load demand prediction sequence, and a battery life degradation rate prediction sequence are generated to construct a time-aligned prediction dataset, including:

[0010] The actual supercapacitor output in the photovoltaic-storage hybrid system is obtained, and the photovoltaic array output power and the actual supercapacitor output power are algebraically superimposed to obtain the compensated photovoltaic power.

[0011] The compensated photovoltaic power is processed by a pre-set long short-term memory network model to generate a photovoltaic output prediction sequence.

[0012] Based on load power and a pre-set user behavior database, a load demand prediction sequence is generated through a pre-set gradient boosting tree model.

[0013] The battery aging coefficient is processed based on the convolutional neural network-Transformer fusion model to generate a battery life degradation rate prediction sequence.

[0014] The photovoltaic power output prediction sequence, load demand prediction sequence, and battery life degradation rate prediction sequence are time-aligned according to a preset time resolution to generate a time-aligned prediction dataset.

[0015] In one embodiment, based on the energy storage system's state of charge value and the grid's real-time electricity price in the real-time data vector, and combined with the battery life degradation rate prediction sequence in the time-aligned prediction dataset, a lifespan-economic co-optimization objective function is constructed, including:

[0016] Based on the state of charge value of the energy storage system, the lifetime protection weight coefficient is calculated using a piecewise function.

[0017] Based on the real-time electricity price of the power grid and the preset extreme range of electricity price, the economic weight coefficient is calculated through normalization.

[0018] Based on the battery life degradation rate prediction sequence and the preset battery unit price, the economic cost of life loss is calculated.

[0019] The voltage fluctuation is calculated based on the DC bus voltage in the real-time data vector. The economic weight coefficient, the life protection weight coefficient, and the economic cost of life loss are combined to construct a life-economic co-optimization objective function.

[0020] In one embodiment, a policy library is constructed by performing model predictive control optimization processing using a lifetime-economic co-optimization objective function and a time-aligned prediction dataset, including:

[0021] Based on the photovoltaic output prediction sequence and load demand prediction sequence in the time-aligned prediction dataset, power balance constraints are constructed.

[0022] By combining the lifetime protection weight coefficient in the lifetime-economic co-optimization objective function, and adjusting the constraint boundary of the state of charge of the energy storage system, a state of charge constraint condition is constructed.

[0023] Based on the lifespan protection weight coefficient, a threshold for the rate of change of lithium battery charging and discharging power is set, and a power change rate constraint condition is constructed.

[0024] Based on the power balance constraint, state of charge constraint, and power change rate constraint, the lifetime-economic co-optimization objective function is transformed into a nonlinear optimization problem with constraints.

[0025] The constrained nonlinear optimization problem is solved iteratively using a preset rolling time window to obtain the optimized power sequence of supercapacitors and the optimized power sequence of lithium batteries within a preset future time period.

[0026] The optimized power sequences of supercapacitors and lithium batteries are packaged into time slices to obtain the charge and discharge instruction groups corresponding to each time slice, and a strategy library is constructed.

[0027] In one embodiment, based on the strategy library and real-time data vectors, a corresponding operating mode is enabled, and device control commands containing the supercapacitor set power value and the lithium battery set power value are output, including:

[0028] Collect the battery temperature value of the lithium battery in the photovoltaic-storage hybrid system;

[0029] Based on the energy storage system health status value, real-time grid price, energy storage system state of charge value, and battery temperature value from the real-time data vector, the corresponding operating mode is activated, and the corresponding charge / discharge command group is extracted from the policy library; the charge / discharge command group is extracted through the following steps:

[0030] When the health status value of the energy storage system is lower than the preset health threshold or the battery temperature value is higher than the preset temperature threshold, the life protection mode is activated, and the charging and discharging instruction group corresponding to the minimum optimized power of the lithium battery is selected from the strategy library.

[0031] When the real-time electricity price of the power grid in the real-time data vector is higher than the preset electricity price threshold and the state of charge value of the energy storage system is higher than the preset state of charge threshold, the economic discharge mode is activated, and the charging and discharging instruction group corresponding to the maximum optimized power of the lithium battery is selected from the strategy library.

[0032] When the activation conditions for life protection mode and economic discharge mode are not met, the default mode is activated, and the charge and discharge instruction group corresponding to the current time slice is extracted from the strategy library.

[0033] Extract the supercapacitor set power value and the lithium battery set power value from the charge and discharge command group, and output the device control command containing the supercapacitor set power value and the lithium battery set power value.

[0034] In one embodiment, the method further includes:

[0035] Obtain the supercapacitor set power value from the equipment control command and use the supercapacitor set power value as the reference control power;

[0036] Based on the DC bus voltage in the real-time data vector and by calling the historical DC bus voltage in the preset historical database, the rate of change of the DC bus voltage is calculated.

[0037] The rate of change of DC bus voltage is converted into power compensation amount by a preset differential compensation formula. The reference control power is adjusted according to the power compensation amount to generate the corrected supercapacitor output.

[0038] The system detects whether the output of the corrected supercapacitor exceeds the preset maximum allowable power of the supercapacitor. If it does, the power is limited according to the preset slope to obtain the output of the corrected supercapacitor after the limit is obtained. The output of the corrected supercapacitor after the limit is used as the actual output of the supercapacitor in the next round and then algebraically superimposed.

[0039] In one embodiment, the compensated photovoltaic power is calculated using the following formula:

[0040]

[0041]

[0042] in, For the first Photovoltaic power after compensation at any time For discrete time steps, For photovoltaic power smoothing weighting coefficients, The moving average window length, For the first The output power of the photovoltaic array at any given time. This is the weighting factor for the output attenuation of the supercapacitor. For the first The actual supercapacitor output at any given moment For the first The power loss of the supercapacitor at any time This represents the basic loss coefficient of the supercapacitor. This refers to the temperature sensitivity coefficient of a supercapacitor. For the first Real-time temperature of the supercapacitor This is the reference temperature for the supercapacitor.

[0043] Secondly, this application also provides a coordination control device suitable for a photovoltaic-storage hybrid system. The device is applied to the photovoltaic-storage hybrid system and includes:

[0044] The data acquisition and prediction dataset construction module is used to collect photovoltaic array output power, energy storage system state of charge, energy storage system health status, battery aging coefficient, DC bus voltage, load power and real-time grid electricity price, and generate real-time data vectors; based on the real-time data vectors, it generates photovoltaic output prediction sequences, load demand prediction sequences and battery life degradation rate prediction sequences, and constructs time-aligned prediction datasets.

[0045] The collaborative optimization objective function construction module is used to construct a lifetime-economic collaborative optimization objective function based on the energy storage system's state of charge value and the grid's real-time electricity price in the real-time data vector, combined with the battery lifetime degradation rate prediction sequence in the time-aligned prediction dataset.

[0046] The optimization decision and operation control module is used to perform model predictive control optimization processing through the lifetime-economic co-optimization objective function and time-aligned prediction dataset, build a strategy library, enable the corresponding operation mode based on the strategy library and real-time data vectors, and output device control instructions containing the set power values ​​of the supercapacitor and the lithium battery; the strategy library contains charging and discharging instructions for future preset time periods.

[0047] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the first aspect.

[0048] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the first aspect.

[0049] The aforementioned coordinated control method, device, equipment, and medium applicable to photovoltaic-storage hybrid systems firstly collect multi-source data to generate real-time data vectors, solving the problems of incomplete data collection and lack of dynamic correlation in traditional methods, thus improving data accuracy. Secondly, based on real-time data, three types of prediction sequences are generated and a time-aligned dataset is constructed, overcoming the shortcomings of traditional predictions that are single-scale and do not incorporate the real-time state of the system, thereby enhancing prediction reliability. Furthermore, a lifetime-economic co-optimization objective function is constructed by combining energy storage status and electricity price, maximizing the benefits throughout the entire life cycle. Finally, a strategy library is built through model predictive control and output commands are adapted to the operating mode, overcoming the shortcomings of traditional static constraints and simple mode switching, thus improving the scientific nature of control decisions and the stability of system operation. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 A flowchart of a coordinated control method for a photovoltaic-storage hybrid system is provided as an exemplary embodiment of the present invention.

[0052] Figure 2 A flowchart illustrating a method for constructing a lifetime-economic co-optimization objective function is provided as an exemplary embodiment of the present invention.

[0053] Figure 3 A schematic diagram of a coordinated control device for a photovoltaic-storage hybrid system is provided as an exemplary embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0055] In one embodiment, such as Figure 1 As shown, a coordinated control method suitable for a photovoltaic-storage hybrid system is provided. This embodiment illustrates the application of this method to a photovoltaic-storage hybrid system. In this embodiment, the method includes the following steps:

[0056] S101: Collect photovoltaic array output power, energy storage system state of charge, energy storage system health status, battery aging coefficient, DC bus voltage, load power and real-time grid electricity price to generate real-time data vectors; based on the real-time data vectors, generate photovoltaic output prediction sequences, load demand prediction sequences and battery life degradation rate prediction sequences to construct a time-aligned prediction dataset.

[0057] Specifically, the output power of the photovoltaic array is the core energy input source of the photovoltaic-storage hybrid system, and its fluctuations directly affect the direction of energy dispatch. The state of charge (SOC) and state of health (SOH) of the energy storage system reflect the current energy reserves and long-term health level of the photovoltaic-storage hybrid system, respectively. The battery aging factor is a core parameter for quantifying the degree of battery degradation and can provide a basic support for subsequent lifespan prediction. The DC bus voltage reflects whether the voltage of the photovoltaic-storage hybrid system is stable, and its fluctuation range determines whether fluctuation suppression is needed. The load power reflects the energy consumption demand of the photovoltaic-storage hybrid system. The real-time grid price is the core incentive signal for achieving economic dispatch. Illustratively, the output power of the photovoltaic array can be collected through a photovoltaic power sensor. Furthermore, the SOC, SOH, and battery aging factor can be collected synchronously through the battery management system (BMS) of the photovoltaic-storage hybrid system. The SOC can be obtained through the ampere-hour integration method combined with open-circuit voltage calibration, the SOH can be calculated from the capacity decay rate, and the battery aging factor can be pre-calibrated based on historical cycle data and real-time charge / discharge status and stored in the BMS. Further, the DC bus voltage and load power can also be collected through a DC power meter, with the sampling frequency consistent with the photovoltaic power. In addition, real-time electricity prices can be received through the communication interface of the power grid dispatching platform. Subsequently, the aforementioned multi-source data can be synchronized and integrated through timestamp alignment to generate a real-time data vector. This vector is stored in the form of a dimensional array, with each data item corresponding to a unique status identifier and timestamp, thereby ensuring the traceability and relevance of the data.

[0058] Specifically, based on this vector, dynamic prediction and correction can be performed to further generate photovoltaic (PV) output prediction sequences, load demand prediction sequences, and battery life degradation rate prediction sequences. The PV output prediction sequence can be generated by integrating meteorological factors such as irradiance and temperature, as well as historical output power data of the PV array, using models such as machine learning or physical models. The load demand prediction sequence can be obtained by combining historical load data, seasonal variations, and user behavior patterns. The battery life degradation rate prediction sequence can be obtained based on the battery's aging coefficient, current health status, and historical charge / discharge data, thus providing dynamic information related to battery life for subsequent optimized control. Time-aligning these three prediction sequences ensures consistency across time dimensions, constructing a time-aligned prediction dataset. This dataset provides comprehensive and accurate prediction information for subsequent optimized control, enabling control strategies to be formulated based on accurate predictions of future system states.

[0059] S102: Based on the energy storage system's state of charge value and the grid's real-time electricity price in the real-time data vector, and combined with the battery life degradation rate prediction sequence in the time-aligned prediction dataset, construct a life-economic co-optimization objective function.

[0060] Specifically, a lifetime-economic co-optimization objective function can be constructed based on factors such as the charging and discharging costs of the energy storage system, the economic benefits of grid interaction, and the costs of battery lifespan degradation. By incorporating the predicted battery lifespan degradation rate sequence into this objective function, the impact of battery health status on system operating costs can be dynamically reflected. Furthermore, by combining real-time grid electricity prices, the charging and discharging strategies of the energy storage system can be further optimized to maximize the economic efficiency of grid interaction. Constructing a lifetime-economic co-optimization objective function not only effectively extends battery lifespan and reduces long-term system operating costs in subsequent decision-making processes, but also improves the overall performance and reliability of the system while ensuring its economic viability.

[0061] S103: Model predictive control optimization is performed using a lifetime-economic co-optimization objective function and a time-aligned prediction dataset. A strategy library is constructed, and based on the strategy library and real-time data vectors, the corresponding operating mode is enabled. Device control commands containing the set power values ​​of the supercapacitor and the lithium battery are output. The strategy library contains charging and discharging commands for future preset time periods.

[0062] Specifically, model predictive control (MMCC) is a control strategy that, considering the system's dynamic characteristics and constraints, optimizes the control sequence over a future period to achieve optimal system operation. Based on a lifetime-economic co-optimization objective function and a time-aligned prediction dataset, MMCC optimization processes are used to construct a strategy library containing charging and discharging commands for preset future time periods. These commands consider not only the current system state but also predictions of future system states. This dynamic optimization control effectively mitigates fluctuations in photovoltaic output, improves the absorption capacity of new energy sources, and maximizes system economy and battery life while ensuring safe system operation. Subsequently, based on the strategy library and real-time data vectors, corresponding operating modes are activated, and device control commands containing set power values ​​for supercapacitors and lithium batteries are output. These commands precisely guide the charging and discharging behavior of supercapacitors and lithium batteries in the energy storage system, ensuring efficient and stable operation under different operating modes.

[0063] The above method first collects multi-dimensional real-time data and generates three types of time-aligned prediction sequences, solving the problems of single data dimensions and asynchronous prediction in traditional methods, and achieving coordinated adaptation between data collection and prediction. Secondly, it constructs a lifespan-economic co-optimization objective function by combining SOC, electricity price, and lifespan degradation rate, overcoming the one-sidedness of single-objective optimization, improving the overall efficiency of objective optimization, and maximizing benefits throughout the entire life cycle. Finally, it constructs a strategy library through model predictive control and adapts the output instructions to the operating mode, making up for the shortcomings of traditional static constraints and simple mode switching, and improving the scientific nature of control decisions and the stability of system operation.

[0064] In one embodiment, based on real-time data vectors, a photovoltaic power output prediction sequence, a load demand prediction sequence, and a battery life degradation rate prediction sequence are generated to construct a time-aligned prediction dataset, including:

[0065] The actual supercapacitor output in the photovoltaic-storage hybrid system is obtained, and the photovoltaic array output power and the actual supercapacitor output power are algebraically superimposed to obtain the compensated photovoltaic power.

[0066] The compensated photovoltaic power is processed by a pre-set long short-term memory network model to generate a photovoltaic output prediction sequence.

[0067] Based on load power and a pre-set user behavior database, a load demand prediction sequence is generated through a pre-set gradient boosting tree model.

[0068] The battery aging coefficient is processed based on the convolutional neural network-Transformer fusion model to generate a battery life degradation rate prediction sequence.

[0069] The photovoltaic power output prediction sequence, load demand prediction sequence, and battery life degradation rate prediction sequence are time-aligned according to a preset time resolution to generate a time-aligned prediction dataset.

[0070] Specifically, as a high-frequency fluctuation suppression unit in a photovoltaic-storage hybrid system, the attenuation and loss of the supercapacitor's actual output can affect the accuracy of photovoltaic power compensation. Therefore, a comprehensive compensation formula incorporating photovoltaic power smoothing, supercapacitor output attenuation, and temperature-dependent losses can be introduced to more accurately remove interference components from the photovoltaic array's output power and restore the low-to-mid-frequency trend of the photovoltaic output. For example, the compensated photovoltaic power can be calculated using the following comprehensive compensation formula:

[0071]

[0072]

[0073] in, For the first Photovoltaic power after compensation at any time For discrete time steps, For photovoltaic power smoothing weighting coefficients, The moving average window length, For the first The output power of the photovoltaic array at any given time. This is the weighting factor for the output attenuation of the supercapacitor. For the first The actual supercapacitor output at any given moment For the first The power loss of the supercapacitor at any time This represents the basic loss coefficient of the supercapacitor. This refers to the temperature sensitivity coefficient of a supercapacitor. For the first Real-time temperature of the supercapacitor This is the reference temperature for the supercapacitor.

[0074] In the above formula, the moving average term for photovoltaic power can reduce the high-frequency noise of the original photovoltaic power; the supercapacitor output attenuation term can correct the impact of its own capacity decay on the compensation accuracy; and the temperature-dependent loss term can quantify the internal resistance loss of the supercapacitor at different temperatures. Combining these three terms can effectively reduce the fluctuation range of photovoltaic output, providing more stable and physically consistent basic data for subsequent photovoltaic output prediction, and avoiding the problem of amplified prediction errors caused by insufficient compensation accuracy.

[0075] Specifically, photovoltaic (PV) power output exhibits significant time-series correlation, with future output closely linked to recent solar irradiance trends and historical output patterns. The pre-defined Long Short-Term Memory (LSTM) network model, through its gated unit mechanism, effectively captures the long-short-term dependencies in time-series data, overcoming the gradient vanishing or exploding problems of traditional recurrent neural networks. For example, the input layer of the pre-defined LSTM model has a dimension of 1, inputting only the PV power time-series data compensated by the aforementioned formula. The hidden layers can be set to 3 layers, each with 64 neurons. The output layer dimension represents the number of predicted PV power output values ​​within a pre-defined future prediction period, such as 60 minutes (12 values, corresponding to 5 minutes per time slice). The training dataset can utilize historical data from the PV-storage hybrid system over the past 6 months, including compensated PV power and corresponding meteorological data (solar intensity, ambient temperature). During training, the root mean square error (RMSE) can be used as the loss function, and the Adam optimizer can be used for iterative parameter updates until the model converges (loss function value is below 0.01). When the prediction is executed, the real-time collected and formula-compensated photovoltaic power time series data (the first 30 time slices, i.e., 150 minutes of data) can be input into the trained LSTM model. The model can output the photovoltaic power output prediction sequence divided into 5-minute time slices within the next 60 minutes by learning the time series pattern.

[0076] Specifically, load demand is not only related to the current load power, but also highly correlated with users' electricity consumption habits, such as the difference in electricity consumption between weekdays and weekends, and the patterns of peak electricity consumption periods. The pre-defined gradient boosting tree model can accurately capture the non-linear mapping relationship between load power and user behavior characteristics through ensemble learning of multiple decision trees. The pre-defined user behavior database can be constructed through statistical analysis of historical system electricity consumption data, containing statistical characteristics (average, maximum, and rate of change) of user load power for different date types and time periods, stored in the form of feature vectors. The pre-defined gradient boosting tree model can have up to 100 decision trees, with a maximum depth of 8 for each tree, a learning rate of 0.1, and a minimum number of sample splits of 20. The input features of this model can include the current load power value from the real-time data vector, statistical feature vectors for the corresponding date type and time period from the user behavior database, and the output is a predicted sequence of load demand divided into 5-minute time slices for the next 60 minutes. Illustratively, the model training process can use the mean absolute percentage error as the loss function and optimize the model parameters through a grid search method, thereby ensuring the model's adaptability to different electricity consumption scenarios.

[0077] Specifically, changes in battery aging coefficients include both short-term local fluctuations, such as small changes caused by a single charge-discharge cycle, and long-term global degradation trends, such as continuous declines due to accumulated cycle counts. Convolutional Neural Network (CNN) models can extract local features through convolutional kernels, while Transformer models can capture global temporal dependencies through self-attention mechanisms. Fusion of these models allows for a comprehensive representation of the battery aging process. For example, the CNN part of a CNN-Transformer fusion model can contain two convolutional layers (kernel sizes of 3 and 5, and output channels of 32 and 64 respectively) and one pooling layer (max pooling, with a kernel size of 2) to extract local features from the battery aging coefficient temporal data. The Transformer part can contain two encoder layers (each with 8 attention heads and a hidden layer dimension of 128) to capture global temporal correlations between local features. The model's input can be temporal data (the first 60 time slices, i.e., 300 minutes of data) composed of battery aging coefficients, SOC, and SOH from a real-time data vector, and the output is a predicted sequence of battery life degradation rates divided into 5-minute time slices for the next 60 minutes. Furthermore, the dataset used for training this model can include battery cycle aging test data (including changes in aging coefficient under different charge / discharge depths and temperature conditions), and the training process employs the cross-entropy loss function and the AdamW optimizer.

[0078] Furthermore, although a target time resolution of 5 minutes is used in the generation of the three types of prediction sequences, slight time deviations may exist due to the influence of model output characteristics and data acquisition cycles. Directly using these deviations for subsequent optimization calculations could lead to power balance constraint failure and objective function calculation errors. Therefore, the preset time resolution can be uniformly set to 5 minutes, and a precise timestamp can be added to each data point of each prediction sequence, using the unified clock signal of the photovoltaic-storage hybrid system as a reference. Subsequently, using 12 target time slots within the next 60 minutes as a reference, time matching is performed on the data points of the three types of prediction sequences. If multiple data points exist within a time slot, the average value can be taken as the prediction value for that time slot. If no data points exist within a time slot, linear interpolation is used to supplement the calculation based on the prediction values ​​of adjacent time slots. The final generated time-aligned prediction dataset can be indexed by time slots. Each index corresponds to a set of data items containing photovoltaic power output predictions, load demand predictions, and battery life degradation rate predictions. The data items are stored in array form to ensure that subsequent steps can directly call them by time slot.

[0079] In one embodiment, such as Figure 2 As shown, based on the energy storage system's state of charge value and the grid's real-time electricity price in the real-time data vector, and combined with the battery life degradation rate prediction sequence in the time-aligned prediction dataset, a lifespan-economic co-optimization objective function is constructed, including:

[0080] S201: Calculate the lifetime protection weight coefficient using a piecewise function based on the state of charge value of the energy storage system;

[0081] S202: Based on the real-time electricity price of the power grid and the preset extreme range of electricity price, the economic weight coefficient is calculated through normalization.

[0082] S203: Calculate the economic cost of life loss based on the battery life degradation rate prediction sequence and the preset battery unit price.

[0083] S204: Calculate the voltage fluctuation based on the DC bus voltage in the real-time data vector, and construct a life-economic co-optimization objective function by combining the economic weight coefficient, the life protection weight coefficient, and the economic cost of life loss.

[0084] Specifically, the aging rate of a battery is strongly correlated with its State of Charge (SOC) level. The lower the SOC, the higher the risk of capacity degradation due to deep discharge; conversely, while a higher SOC carries a lower risk from deep charging, long-term storage at a high SOC will still accelerate aging. Therefore, a piecewise function can be used to dynamically adjust the lifetime protection weight to match the degradation risk. For example, the lifetime protection weight coefficient... The calculation can be performed using the following piecewise function:

[0085]

[0086] In the above formula, when SOC < 20%, the battery is in the deep discharge range, and the risk of irreversible lithium-ion deintercalation increases significantly. At this point, the highest lifetime protection weight of 0.7 can be assigned, forcing the optimization process to prioritize avoiding over-discharge. When 20% ≤ SOC ≤ 50%, the battery degradation risk decreases linearly with increasing SOC. A linear function can be used to dynamically adjust the weights to achieve a precise mapping between risk and weight. When SOC > 50%, the battery is in a relatively safe charge / discharge range with a low degradation risk, and a lower fixed weight of 0.4 can be assigned.

[0087] Specifically, fluctuations in grid electricity prices directly determine the economic dispatch space of energy storage systems. Normalization can convert price fluctuations into weighting coefficients of a uniform magnitude, avoiding weight imbalances caused by differences in the absolute value of electricity prices, and ensuring that economic weights and lifetime protection weights coordinate within the same numerical range. For example, the economic weighting coefficient is calculated using the following normalization formula:

[0088]

[0089] in, Let Ct be the real-time electricity price at time t, Cmin be the lowest price within the preset price extreme range, and Cmax be the highest price within the preset price extreme range. Both can be obtained by statistically analyzing the historical electricity price data of the region where the photovoltaic-storage hybrid system is located over the past 12 months. This formula allows economic weights to accurately respond to electricity price fluctuations, providing a core decision-making basis for subsequent peak-valley electricity price arbitrage and reducing grid electricity costs.

[0090] Specifically, based on the predicted battery life degradation rate sequence in the time-aligned prediction dataset and combined with a preset battery unit price, the economic cost of life degradation is calculated. This transforms the abstract concept of battery life degradation into a quantifiable economic cost, giving the life protection objective and the economic benefit objective a unified quantitative dimension and laying the foundation for their synergistic optimization. For example, the economic cost of life degradation can be calculated using the following formula:

[0091]

[0092] in, Here, T represents the prediction time slice index, and T represents the total prediction duration. (This is the predicted battery life degradation rate for the τ-th time slice, ranging from 0 to 1).

[0093] To preset the battery unit price, the value is taken as the actual purchase price of the battery or the industry average replacement unit price, to ensure the accuracy of the economic cost. The rated capacity of the lithium battery is determined by the hardware configuration parameters of the energy storage system.

[0094] Furthermore, voltage stability is a fundamental prerequisite for the safe operation of a photovoltaic-storage hybrid system. Focusing solely on lifespan and economic targets may lead to excessive DC bus voltage fluctuations, triggering equipment shutdowns. Therefore, voltage stability should be incorporated into the overall optimization process, forming a three-dimensional target system encompassing economy, lifespan, and voltage stability. For example, voltage fluctuations can be calculated using the following formula:

[0095]

[0096] in, The rated voltage of the DC bus is determined by the design parameters of the photovoltaic-storage hybrid system. This represents the real-time DC bus voltage. Voltage fluctuations reflect the degree of deviation between the real-time voltage and the rated voltage; the greater the deviation, the worse the system voltage stability. Subsequently, a lifetime-economic co-optimization objective function can be constructed:

[0097]

[0098] Where Jt* is the value of the collaborative optimization objective function at time t, and the optimization direction is minimization. The cost of electricity used by the power grid is calculated using real-time electricity prices and predicted grid interaction power (positive cost when the grid supplies electricity, negative cost, i.e., revenue, when energy storage discharges and connects to the grid). The above formula can utilize dynamic weights (…). , This achieves an adaptive balance between economic and lifespan goals under various operating conditions. Furthermore, a fixed weight of 0.3 ensures basic voltage stability, and the weights of these three factors are matched within their respective ranges. ∈[0,0.5], ∈[0.4,0.7], with a voltage term weight of 0.3, thus avoiding the system from becoming unbalanced due to the excessive dominance of a certain objective.

[0099] In one embodiment, a policy library is constructed by performing model predictive control optimization processing using a lifetime-economic co-optimization objective function and a time-aligned prediction dataset, including:

[0100] Based on the photovoltaic output prediction sequence and load demand prediction sequence in the time-aligned prediction dataset, power balance constraints are constructed.

[0101] By combining the lifetime protection weight coefficient in the lifetime-economic co-optimization objective function, and adjusting the constraint boundary of the state of charge of the energy storage system, a state of charge constraint condition is constructed.

[0102] Based on the lifespan protection weight coefficient, a threshold for the rate of change of lithium battery charging and discharging power is set, and a power change rate constraint condition is constructed.

[0103] Based on the power balance constraint, state of charge constraint, and power change rate constraint, the lifetime-economic co-optimization objective function is transformed into a nonlinear optimization problem with constraints.

[0104] The constrained nonlinear optimization problem is solved iteratively using a preset rolling time window to obtain the optimized power sequence of supercapacitors and the optimized power sequence of lithium batteries within a preset future time period.

[0105] The optimized power sequences of supercapacitors and lithium batteries are packaged into time slices to obtain the charge and discharge instruction groups corresponding to each time slice, and a strategy library is constructed.

[0106] Specifically, in a photovoltaic-storage hybrid system, the photovoltaic output, energy storage charging and discharging power, and load demand must be balanced in real time; otherwise, it will cause drastic fluctuations in the DC bus voltage or even system collapse. For example, photovoltaic output prediction can be performed using time-aligned prediction datasets. and load demand forecast Based on this, the power balance constraint formula is constructed:

[0107]

[0108] in, For discrete time slices, For lithium batteries The charging and discharging power at any given moment (discharging is positive, charging is negative). For supercapacitors in The charging and discharging power at each moment. This constraint ensures the balance of energy supply and demand within the system at each time slice. And the lifetime protection weighting factor... The size of the value directly reflects the strength of the current lifetime protection requirement, therefore, this requirement can be matched by dynamically shrinking the SOC feasible region. The larger the value, the higher the lifetime protection requirement, and the smaller the feasible region of State of Charge (SOC) should be to avoid overcharging and over-discharging. For example, the SOC constraint formula is:

[0109]

[0110] in, This represents the minimum state of charge of the energy storage system. This represents the maximum state of charge, and 0.1 is the constraint adjustment factor.

[0111] Specifically, rapid changes in the charging and discharging power of lithium batteries can cause an imbalance in the internal lithium-ion insertion / extraction, exacerbating polarization and accelerating capacity decay. Therefore, the larger the lifetime protection weighting coefficient, the more stringent the limitation on the power change rate needs to be. For example, the power change rate constraint formula is:

[0112]

[0113] in, This represents the maximum charge and discharge power of the lithium battery, with 0.5 as a reference coefficient. When... At that time, the power change rate threshold is This effectively limits fast charging and discharging behavior. When, the threshold is This allows for greater power adjustments to adapt to economic dispatch requirements. Subsequently, power balance constraints, SOC constraints, power change rate constraints, and lifetime-economic co-optimization objective functions can be integrated, transforming it into a constrained nonlinear optimization problem. Illustratively, the optimization direction is to minimize the lifetime-economic-voltage co-optimization objective function, while simultaneously satisfying four types of constraints: power balance constraints, SOC dynamic constraints, power change rate constraints, and the hardware power upper and lower limits constraints of the supercapacitor and lithium battery themselves. Through this transformation, the actual control problem of the photovoltaic-storage system can be abstracted into a mathematically solvable nonlinear programming problem, providing a theoretical framework for subsequent numerical solutions. By iteratively solving the constrained nonlinear optimization problem within a preset rolling time window (5 minutes in this embodiment), the optimized power sequence of the supercapacitor and the optimized power sequence of the lithium battery within a preset future time period, such as 60 minutes, can be obtained. Each optimization starts from the current moment, predicts 12 time slices within the next 60 minutes, and after obtaining the charging and discharging power commands for each time slice, only the commands for the first time slice are executed. After the first time slice is completed, optimization can be restarted based on the latest collected real-time data, such as actual photovoltaic output and actual load demand, to solve for the next 60-minute sequence.

[0114] Specifically, by encapsulating the optimized power sequences of the supercapacitor and the lithium battery by time slices, the corresponding charge / discharge instruction sets for each time slice can be obtained, and a strategy library can be constructed. The strategy library can be stored as key-value pairs of time slice indices and charge / discharge instruction sets, with each time slice index corresponding to a tuple containing the power values ​​of the supercapacitor and the lithium battery. When the system triggers a mode switch, such as switching from lifespan protection mode to economic discharge mode, the corresponding instruction set can be quickly retrieved directly based on the current time slice index.

[0115] In one embodiment, based on the policy library and real-time data vectors, a corresponding operating mode is enabled, and device control commands containing the supercapacitor set power value and the lithium battery set power value are output, including:

[0116] Collect the battery temperature value of the lithium battery in the photovoltaic-storage hybrid system;

[0117] Based on the energy storage system health status value, real-time grid price, energy storage system state of charge value, and battery temperature value from the real-time data vector, the corresponding operating mode is activated, and the corresponding charge / discharge command group is extracted from the policy library; the charge / discharge command group is extracted through the following steps:

[0118] When the health status value of the energy storage system is lower than the preset health threshold or the battery temperature value is higher than the preset temperature threshold, the life protection mode is activated, and the charging and discharging instruction group corresponding to the minimum optimized power of the lithium battery is selected from the strategy library.

[0119] When the real-time electricity price of the power grid in the real-time data vector is higher than the preset electricity price threshold and the state of charge value of the energy storage system is higher than the preset state of charge threshold, the economic discharge mode is activated, and the charging and discharging instruction group corresponding to the maximum optimized power of the lithium battery is selected from the strategy library.

[0120] When the activation conditions for life protection mode and economic discharge mode are not met, the default mode is activated, and the charge and discharge instruction group corresponding to the current time slice is extracted from the strategy library.

[0121] Extract the supercapacitor set power value and the lithium battery set power value from the charge and discharge command group, and output the device control command containing the supercapacitor set power value and the lithium battery set power value.

[0122] Specifically, the capacity decay and internal resistance growth rate of lithium batteries increase exponentially with rising temperature. When the temperature exceeds a critical value, such as 45°C, the aging rate accelerates dramatically, and there is a risk of thermal runaway. Illustratively, battery temperature can be collected by the temperature sensor of the battery management system (BMS), with the sampling frequency consistent with SOC and SOH to ensure time synchronization of temperature data with other state parameters. Subsequently, based on the real-time data vector containing SOH, real-time grid price, energy storage system SOC, and battery temperature, the corresponding operating mode can be activated, and charge / discharge command sets can be extracted. This multi-dimensional fusion of SOH, price, SOC, and temperature allows for the determination of the system's most critical priority. A preset health threshold, such as 80%, can be determined through battery cycle aging tests. When SOH falls below this value, battery capacity decay has entered an accelerated phase, requiring the activation of strong lifespan protection. A preset temperature threshold, such as 45°C, can be determined by combining the heat dissipation capacity of the battery thermal management system with the battery thermal runaway risk boundary to ensure that temperature does not trigger thermal runaway and effectively delays aging. The real-time electricity price threshold can be calculated based on the local peak-valley electricity price policy. For example, a charge threshold of 40% can be calculated using the formula: remaining battery capacity × discharge efficiency ≥ peak load demand, to ensure sufficient power support during economical discharge.

[0123] Specifically, the higher the charge / discharge power and the deeper the cycle depth of a lithium battery, the faster its aging rate. Therefore, when lifespan protection is the highest priority, the power output of the lithium battery can be limited, reducing its charge / discharge frequency and depth. By traversing the lithium battery optimized power sequence for each time slice in the strategy library, the instruction group with the smallest absolute power value can be selected (discharge is positive, charging is negative, both are taken as minimum to reduce charge / discharge intensity). For example, when SOH=75% (below the 80% threshold) or temperature=48℃ (above the 45℃ threshold), this mode can be triggered, and the lithium battery power is then limited to within 30% of its rated power, with the supercapacitor undertaking more high-frequency fluctuation suppression tasks. Furthermore, during peak grid electricity price periods, the revenue from discharging the energy storage system to the grid is far higher than the charging cost during off-peak periods. Therefore, when the charging condition (sufficient power to discharge) is met, the lithium battery discharge power can be maximized to obtain economic benefits. The electricity price threshold can be set to the starting value of the peak electricity price, and the charging threshold can be set to 40%. By selecting the instruction group with the maximum optimized power of the lithium battery, the battery can be made to discharge as much as possible during peak periods. Furthermore, when the activation conditions for lifetime protection mode and economic discharge mode are not met, the default mode can be enabled. This mode extracts the charge / discharge instruction set corresponding to the current time slice from the strategy library to strike a balance between lifetime protection and economic benefits, ensuring continuous and stable power supply to the system. For example, the charge / discharge instruction set corresponding to the current time slice in the strategy library can be directly called. This set of instructions represents the optimal solution for model predictive control under the balance between lifetime and economic objectives.

[0124] Specifically, charging and discharging commands need to be translated into power setpoints that can be executed by the supercapacitor controller and lithium battery controller in order to achieve physical implementation of energy dispatch. For example, the supercapacitor power setpoint and lithium battery power setpoint can be sent to the corresponding controllers in digital signal form via an industrial communication bus such as CAN bus. After receiving the commands, the controllers can achieve power tracking through internal power regulation algorithms such as PI control, keeping the deviation between the actual output power and the set power within a preset range.

[0125] In one embodiment, the method further includes:

[0126] Obtain the supercapacitor set power value from the equipment control command and use the supercapacitor set power value as the reference control power;

[0127] Based on the DC bus voltage in the real-time data vector and by calling the historical DC bus voltage in the preset historical database, the rate of change of the DC bus voltage is calculated.

[0128] The rate of change of DC bus voltage is converted into power compensation amount by a preset differential compensation formula. The reference control power is adjusted according to the power compensation amount to generate the corrected supercapacitor output.

[0129] The system detects whether the output of the corrected supercapacitor exceeds the preset maximum allowable power of the supercapacitor. If it does, the power is limited according to the preset slope to obtain the output of the corrected supercapacitor after the limit is obtained. The output of the corrected supercapacitor after the limit is used as the actual output of the supercapacitor in the next round and then algebraically superimposed.

[0130] Specifically, the supercapacitor set power value in the equipment control command is generated based on predicted data and can match long-term energy dispatch requirements. However, in actual operation, uncertainties such as photovoltaic output fluctuations and sudden load changes can cause the DC bus voltage to deviate from the rated value. Therefore, the supercapacitor set power value can be used as a benchmark for real-time fine-tuning to avoid excessive compensation deviating from the optimization target. For example, the supercapacitor set power value can be extracted from the equipment control command through the controller's command parsing module. The extraction frequency is consistent with the control command issuance frequency to ensure that the benchmark power is synchronized with the command execution rhythm. At the same time, the benchmark power is temporarily stored in the real-time database and associated with the corresponding timestamp and DC bus voltage value to provide a traceability basis for subsequent compensation calculation. Subsequently, the voltage change rate can be calculated based on the DC bus voltage in the real-time data vector and the historical DC bus voltage in the preset historical database. The voltage change rate can directly reflect the severity and development direction of the fluctuation, providing a basis for early compensation. The historical DC bus voltage can be retrieved from the preset historical database, and the data of the previous sampling point adjacent to the real-time voltage can be retrieved to ensure that the change rate calculation can reflect the latest fluctuation trend. Furthermore, supercapacitors have a charge / discharge response time down to the millisecond level, enabling them to rapidly absorb or release power to suppress voltage fluctuations. The differential compensation formula quantifies the voltage change rate (fluctuation trend) into a precise power compensation requirement, and then adjusts the reference control power based on the compensation amount to generate a corrected supercapacitor output. The preset differential compensation formula can be:

[0131]

[0132] In the above formula This represents the power compensation amount. A positive value indicates that the supercapacitor needs to increase its discharge power (to suppress voltage drop), while a negative value indicates that it needs to increase its charging power (to suppress voltage rise). The rate of change of DC bus voltage. The differential gain coefficient can be calibrated through supercapacitor charge-discharge tests. This is a correction factor used to compensate for the inherent internal resistance losses of supercapacitors. For example, in supercapacitor charge-discharge tests, the optimal compensation power can be recorded under different voltage change rates, and then obtained through linear fitting. This is to ensure that the compensation amount matches the fluctuation trend.

[0133] Furthermore, the maximum allowable power of a supercapacitor is determined by its rated voltage, equivalent internal resistance, and other hardware parameters. Exceeding these parameters for an extended period can lead to electrolyte decomposition and a drastically shortened lifespan. Directly cutting off the power would trigger a sudden power surge, feeding back into the DC bus and causing new voltage fluctuations. Therefore, this embodiment employs a slope-based limiting method for a smooth transition. For example, the maximum allowable power of the supercapacitor is preset based on hardware parameters; in this embodiment, this could be ±50kW, with positive for discharging and negative for charging. The preset slope can be calibrated through supercapacitor charge / discharge characteristic tests, such as a value of 10kW / ms, ensuring that the power change rate matches the capacitor's energy storage / release rate. If the supercapacitor output is corrected to 55kW (exceeding the 50kW upper limit), it is reduced from 55kW to 50kW at a slope of 10kW / ms, taking 0.5ms. If the output is corrected to -58kW (below the -50kW lower limit), it is increased from -58kW to -50kW at a slope of 10kW / ms, avoiding sudden power surges. The corrected supercapacitor output obtained after limiting can be stored in a real-time database and used as the actual supercapacitor output when performing the "algebraic superposition of photovoltaic array output power and actual supercapacitor output" in the next round. This ensures that the photovoltaic output prediction and objective function optimization in the next round are based on actual compensation data, avoiding the accumulation of prediction-execution errors.

[0134] Based on the same inventive concept, this application also provides a coordinated control device for a photovoltaic-storage hybrid system, which implements the coordinated control method for the photovoltaic-storage hybrid system described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the coordinated control device for a photovoltaic-storage hybrid system provided below can be found in the limitations of the coordinated control method for the photovoltaic-storage hybrid system described above, and will not be repeated here.

[0135] In one exemplary embodiment, such as Figure 3 As shown, a coordination control device 300 suitable for a photovoltaic-storage hybrid system is provided. This device, applied to a photovoltaic-storage hybrid system, includes:

[0136] The data acquisition and prediction dataset construction module 301 is used to collect photovoltaic array output power, energy storage system state of charge, energy storage system health status, battery aging coefficient, DC bus voltage, load power and real-time grid electricity price, and generate real-time data vectors; based on the real-time data vectors, it generates photovoltaic output prediction sequence, load demand prediction sequence and battery life decay rate prediction sequence, and constructs time-aligned prediction datasets.

[0137] The collaborative optimization objective function construction module 302 is used to construct a lifetime-economic collaborative optimization objective function based on the energy storage system state of charge value and real-time grid electricity price in the real-time data vector, combined with the battery lifetime degradation rate prediction sequence in the time-aligned prediction dataset.

[0138] The optimization decision and operation control module 303 is used to perform model predictive control optimization processing through the life-economic co-optimization objective function and time-aligned prediction dataset, build a strategy library, enable the corresponding operation mode based on the strategy library and real-time data vector, and output device control instructions containing the set power values ​​of the supercapacitor and the lithium battery; the strategy library contains charging and discharging instructions for future preset periods.

[0139] In one exemplary embodiment, the present invention also provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the coordinated control method applicable to the optical-storage hybrid system of this application. A multi-core processor is preferred to improve the system's parallel processing capability. The memory provides sufficient temporary storage space to support program execution and data processing. The memory capacity should be large enough to accommodate large amounts of data and computational tasks.

[0140] In one exemplary embodiment, the present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the coordinated control method applicable to the optical-storage hybrid system of this application. The computer-readable storage medium may include: a read-only memory, a random access memory (RAM), a solid-state drive (SSD), or an optical disk, etc.

[0141] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A coordinated control method suitable for a photovoltaic-storage hybrid system, characterized in that, The method is applied to a photovoltaic-storage hybrid system, and the method includes: The system collects photovoltaic array output power, energy storage system state of charge, energy storage system health status, battery aging coefficient, DC bus voltage, load power, and real-time grid electricity price to generate a real-time data vector. Based on the real-time data vector, it generates photovoltaic output prediction sequence, load demand prediction sequence, and battery life degradation rate prediction sequence to construct a time-aligned prediction dataset. Based on the energy storage system state of charge value and the real-time electricity price of the grid in the real-time data vector, and combined with the battery life degradation rate prediction sequence in the time-aligned prediction dataset, a life-economic co-optimization objective function is constructed. The model predictive control optimization process is performed using the lifetime-economic co-optimization objective function and the time-aligned prediction dataset to construct a strategy library. Based on the strategy library and the real-time data vector, the corresponding operating mode is enabled, and device control commands containing the set power values ​​of the supercapacitor and the lithium battery are output. The strategy library contains charging and discharging commands for a future preset time period.

2. The method according to claim 1, characterized in that, Based on the real-time data vector, a photovoltaic power output prediction sequence, a load demand prediction sequence, and a battery life degradation rate prediction sequence are generated to construct a time-aligned prediction dataset, including: The actual supercapacitor output in the photovoltaic-storage hybrid system is obtained, and the photovoltaic array output power and the actual supercapacitor output are algebraically superimposed to obtain the compensated photovoltaic power. The compensated photovoltaic power is processed by a preset long short-term memory network model to generate the photovoltaic output prediction sequence. Based on the load power combined with a preset user behavior database, the load demand prediction sequence is generated by processing it through a preset gradient boosting tree model. The battery aging coefficient is processed based on the convolutional neural network-Transformer fusion model to generate the battery life decay rate prediction sequence. The photovoltaic power output prediction sequence, the load demand prediction sequence, and the battery life degradation rate prediction sequence are time-aligned according to a preset time resolution to generate the time-aligned prediction dataset.

3. The method according to claim 1, characterized in that, The step involves constructing a lifespan-economic co-optimization objective function based on the energy storage system's state of charge value and the grid's real-time electricity price in the real-time data vector, combined with the battery lifespan degradation rate prediction sequence in the time-aligned prediction dataset. This includes: Based on the state of charge value of the energy storage system, the lifetime protection weight coefficient is calculated using a piecewise function. Based on the real-time electricity price of the power grid and the preset extreme range of electricity price, the economic weight coefficient is calculated through normalization. Based on the battery life degradation rate prediction sequence and the preset battery unit price, the economic cost of life loss is calculated. The voltage fluctuation is calculated based on the DC bus voltage in the real-time data vector. The life-economic co-optimization objective function is then constructed by combining the economic weight coefficient, the life protection weight coefficient, and the life loss economic cost.

4. The method according to claim 1, characterized in that, The process of performing model predictive control optimization using the lifetime-economic co-optimization objective function and the time-aligned prediction dataset to construct a policy library includes: Based on the photovoltaic output prediction sequence and load demand prediction sequence in the time-aligned prediction dataset, power balance constraints are constructed. Combining the lifetime protection weight coefficient in the lifetime-economic co-optimization objective function, the state of charge constraint conditions are constructed by adjusting the constraint boundary of the state of charge of the energy storage system. Based on the lifespan protection weighting coefficient, a threshold for the rate of change of lithium battery charging and discharging power is set, and a power change rate constraint condition is constructed. Based on the power balance constraint, the state of charge constraint, and the power change rate constraint, the lifetime-economic co-optimization objective function is transformed into a nonlinear optimization problem with constraints. The constrained nonlinear optimization problem is iteratively solved using a preset rolling time window to obtain the optimized power sequence of supercapacitors and the optimized power sequence of lithium batteries within the preset time period in the future. The supercapacitor optimized power sequence and the lithium battery optimized power sequence are packaged into time slices to obtain the charge and discharge instruction groups corresponding to each time slice, and the strategy library is constructed.

5. The method according to claim 1, characterized in that, Based on the strategy library and the real-time data vector, the corresponding operating mode is activated, and device control commands containing the set power values ​​for the supercapacitor and the lithium battery are output, including: Collect the battery temperature value of the lithium battery in the photovoltaic-storage hybrid system; Based on the energy storage system health status value, real-time grid price, energy storage system state of charge value, and battery temperature value in the real-time data vector, the corresponding operating mode is activated, and the corresponding charge / discharge instruction group is extracted from the strategy library; wherein, the charge / discharge instruction group is extracted through the following steps: When the health status value of the energy storage system is lower than the preset health threshold or the battery temperature value is higher than the preset temperature threshold, the life protection mode is activated, and the charging and discharging instruction group corresponding to the minimum optimized power of the lithium battery is selected from the strategy library. When the real-time electricity price of the power grid in the real-time data vector is higher than the preset electricity price threshold and the state of charge value of the energy storage system is higher than the preset state of charge threshold, the economic discharge mode is activated, and the charging and discharging instruction group corresponding to the maximum optimized power of the lithium battery is selected from the strategy library. When the activation conditions of the lifetime protection mode and the economic discharge mode are not met, the default mode is enabled, and the charge and discharge instruction group corresponding to the current time slice in the strategy library is extracted. Extract the supercapacitor set power value and the lithium battery set power value from the charge and discharge command group, and output the device control command containing the supercapacitor set power value and the lithium battery set power value.

6. The method according to claim 2, characterized in that, The method further includes: Obtain the supercapacitor set power value in the device control command, and use the supercapacitor set power value as the reference control power; Based on the DC bus voltage in the real-time data vector and by calling the historical DC bus voltage in the preset historical database, the rate of change of the DC bus voltage is calculated. The rate of change of the DC bus voltage is converted into a power compensation amount by a preset differential compensation formula. The reference control power is adjusted according to the power compensation amount to generate the corrected supercapacitor output. The system detects whether the output of the corrected supercapacitor exceeds the preset maximum allowable power of the supercapacitor. If it does, the system performs power limiting processing according to a preset slope to obtain the output of the corrected supercapacitor after limiting. The output of the corrected supercapacitor after limiting is used as the actual output of the supercapacitor in the next round and then subjected to algebraic superposition processing.

7. The method according to claim 2, characterized in that, The compensated photovoltaic power is calculated using the following formula: in, For the first The compensated photovoltaic power at the specified time, For discrete time steps, For photovoltaic power smoothing weighting coefficients, The moving average window length, For the first The output power of the photovoltaic array at that time. This is the weighting factor for the output attenuation of the supercapacitor. For the first The actual supercapacitor output at that moment, For the first The power loss of the supercapacitor at any time This represents the basic loss coefficient of the supercapacitor. This refers to the temperature sensitivity coefficient of a supercapacitor. For the first Real-time temperature of the supercapacitor This is the reference temperature for the supercapacitor.

8. A coordinated control device suitable for a photovoltaic-storage hybrid system, characterized in that, The device is used in a photovoltaic-storage hybrid system, and the device includes: The data acquisition and prediction dataset construction module is used to collect photovoltaic array output power, energy storage system state of charge, energy storage system health status, battery aging coefficient, DC bus voltage, load power and real-time grid electricity price, and generate real-time data vectors; based on the real-time data vectors, it generates photovoltaic output prediction sequence, load demand prediction sequence and battery life degradation rate prediction sequence, and constructs time-aligned prediction datasets. The collaborative optimization objective function construction module is used to construct a lifetime-economic collaborative optimization objective function based on the energy storage system state of charge value and real-time grid electricity price in the real-time data vector, combined with the battery lifetime degradation rate prediction sequence in the time-aligned prediction dataset. The optimization decision and operation control module is used to perform model predictive control optimization processing through the lifetime-economic co-optimization objective function and the time-aligned prediction dataset, construct a strategy library, enable the corresponding operation mode based on the strategy library and the real-time data vector, and output device control instructions containing the set power values ​​of the supercapacitor and the lithium battery; the strategy library contains charging and discharging instructions for a future preset period.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images