Photovoltaic power generation-energy storage-power grid intelligent scheduling method and system based on multi-mode energy management

Abstract

The invention relates to the technical field of intelligent control of a power system, in particular to a photovoltaic power generation-energy storage-power grid intelligent scheduling method and system based on multi-mode energy management, and the method comprises the steps: 1, collecting and fusing multi-source operation data in real time; 2, establishing a multi-modal feature fusion model; 3, according to the photovoltaic fluctuation amplitude and the frequency deviation component, the energy storage adjusting capacity is formed; 4, constructing a differential control strategy for each mode, and achieving the dynamic matching of the virtual inertia and the power grid demand in the power grid supporting mode; and step 5, executing multi-time scale rolling optimization control, triggering strategy reconstruction according to mode switching, and maintaining power continuity among different modes through a transition process. According to different modal requirements, differentiated control strategies are adopted, so that the service life loss of the energy storage system can be reduced and the service cycle of the energy storage equipment can be prolonged while the energy storage system is ensured to quickly respond to the power grid requirement.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for power systems, and in particular to a method and system for intelligent dispatching of photovoltaic power generation, energy storage, and power grid based on multimodal energy management. Background Technology

[0002] With the increasing penetration rate of renewable energy, photovoltaic power generation-energy storage-grid coordinated dispatch systems play a crucial role in the power system. Such systems need to simultaneously address multi-dimensional challenges such as photovoltaic output volatility, grid frequency stability, and energy storage device lifespan management. The technical difficulties are concentrated in three aspects: multi-source heterogeneous data fusion, dynamic operating condition adaptability, and multi-objective optimization coordination.

[0003] Traditional dispatching systems employ preset charge / discharge thresholds and fixed operating modes, failing to effectively differentiate the impact of varying photovoltaic (PV) output fluctuations (such as sudden cloud cover changes versus gradual temperature variations) on the power grid. For instance, when PV volatility exceeds 15%, existing methods still utilize the same control strategy, leading to frequent switching of charge / discharge states in the energy storage system and accelerated battery aging. Studies show that such systems experience up to 32 charge / discharge cycles per day under abrupt change conditions, a 75% increase compared to steady-state conditions.

[0004] Existing forecasting models often rely on a single dimension of meteorological or historical load data, failing to consider the nonlinear coupling effect between photovoltaic panel backsheet temperature and irradiance. For example, when ambient temperature drops sharply, the increased efficiency of photovoltaic modules may offset the impact of decreased irradiance, but traditional forecasting models, lacking multimodal data correlation, result in output prediction errors exceeding 9%. In a microgrid case, such errors caused the energy storage system to prematurely enter a deep discharge state, resulting in a capacity loss of up to 12%.

[0005] Current dispatch strategies typically employ a constant proportional gain to control the energy storage response rate when addressing grid frequency deviations, without dynamically adjusting for changes in grid inertia. In scenarios with a high proportion of renewable energy integration, reduced system inertia leads to a 42% increase in frequency recovery time, and a fixed adjustment strategy may cause the energy storage's state of charge to exceed safety boundaries, triggering overcharging / over-discharging risks. For example, in one regional grid, due to the lack of a virtual inertia matching mechanism, the energy storage overshoot reached 28% of its rated power during a sudden frequency drop.

[0006] Therefore, there is an urgent need for a photovoltaic power generation-energy storage-grid intelligent dispatching method and system based on multimodal energy management to solve the above problems. Summary of the Invention

[0007] To achieve the above objectives, this invention provides a photovoltaic power generation-energy storage-grid intelligent dispatching method and system based on multimodal energy management. One such method includes the following steps:

[0008] Step 1: Collect and integrate multi-source operational data in real time, including photovoltaic meteorological parameters and module temperature, energy storage system status parameters, grid frequency and load demand parameters, to form a spatiotemporally aligned feature dataset;

[0009] Step 2: Establish a multimodal feature fusion model. By coupling the photovoltaic power output fluctuation characteristics with the grid frequency deviation component, identify the dominant characteristics of load mutation and new energy fluctuation, and generate energy storage dynamic adjustment capability indicators.

[0010] Step 3: Based on the photovoltaic fluctuation amplitude, frequency deviation component composition and energy storage regulation capability, dynamically divide the frequency regulation priority mode, power smoothing mode and grid support mode, and set the boundary conditions for mode switching;

[0011] Step 4: Construct differentiated control strategies for each mode, optimize the energy storage response rate under the frequency regulation priority mode, balance photovoltaic fluctuation suppression and energy storage lifetime loss under the power smoothing mode, and achieve dynamic matching between virtual inertia and grid demand under the grid support mode;

[0012] Step 5: Perform multi-timescale rolling optimization control, reconstruct according to the mode switching trigger strategy, and maintain power continuity between different modes through the transition process;

[0013] Preferably, the generation of the spatiotemporal alignment feature dataset in step 1 includes:

[0014] The irradiance, ambient temperature and wind speed data are collected at a frequency of seconds by a meteorological monitoring device, and the backsheet temperature data of each subarray in the photovoltaic array is obtained simultaneously.

[0015] A spatial interpolation model is established for photovoltaic temperature data, and the estimated temperature values ​​of unmonitored locations are calculated based on the spacing between adjacent subarrays and the spatial gradient of environmental parameters.

[0016] Meteorological and temperature data are aligned with a unified time base to construct a three-dimensional data matrix that includes timestamps, spatial locations, and parameter types.

[0017] The energy storage management system acquires real-time data on the state of charge and health of battery clusters, and a recursive filtering algorithm based on a state-space model is used to eliminate sensor noise.

[0018] Preferably, the establishment of the multimodal feature fusion model in step 2 includes:

[0019] The process of extracting the fluctuation characteristics of photovoltaic output: The irradiance data is nonlinearly coupled with the temperature of the photovoltaic panel at the corresponding location, and the rapid fluctuation component caused by cloud movement and the slow fluctuation component caused by temperature change are separated by the variational mode decomposition algorithm.

[0020] The decomposition process of power grid frequency deviation: empirical wavelet transform is used to perform time-frequency analysis on the frequency signal to extract the load change component in the high-frequency band and the new energy fluctuation component in the low-frequency band.

[0021] The calculation process of the dynamic adjustment capability index of energy storage: Based on the historical degradation curves of the current state of charge and health, a dynamic constraint model reflecting the maximum allowable charge and discharge rate is established to generate the adjustable capability coefficient.

[0022] Preferably, the setting of the mode switching boundary conditions in step 3 includes:

[0023] Determination of photovoltaic fluctuation amplitude threshold: Analyze the statistical distribution of output fluctuation under different weather types in historical operation data, and select the fluctuation rate with a cumulative probability reaching a set confidence level as the grading threshold;

[0024] Calculation of frequency deviation component weight: Based on the instantaneous power change rate of the load mutation component and the duration of the new energy fluctuation component, a comprehensive index characterizing the urgency of grid regulation is constructed.

[0025] Dynamic adjustment of energy storage regulation capacity threshold: Based on the current rate of decline in the health status of the energy storage system, predict the downward trend of the adjustable capacity within a set time period in the future, and correct the mode switching threshold in advance.

[0026] Preferably, the optimization process of the frequency-priority mode in step 4 includes:

[0027] Establish a transfer function model of frequency deviation-energy storage power response, and calculate the optimal response time constant based on the current grid inertial time constant;

[0028] A dynamic adjustment range considering the safety boundary of the energy storage state of charge is constructed. A variable proportional coefficient control strategy is adopted in the frequency recovery stage. High gain fast adjustment is adopted in the initial stage, and low gain fine adjustment is adopted when approaching steady state.

[0029] The predicted value of load mutation component is introduced as a feedforward compensation term to adjust the energy storage output in advance to offset the cumulative effect of frequency deviation.

[0030] Preferably, the multi-objective optimization of the power smoothing mode in step 4 includes:

[0031] Define the photovoltaic fluctuation suppression index: calculate the moving root mean square value of the rate of change of output in the predicted time domain, and correlate it with the current temperature change gradient of the photovoltaic module;

[0032] Define the energy storage lifetime degradation index: Based on the current charge-discharge cycle count and health status parameters, establish a lifetime degradation model with the change in state of charge as the variable;

[0033] A dynamic weight allocation method is adopted to balance the dual-objective optimization, and the weight coefficients are adaptively adjusted according to the urgency of grid peak-shaving demand and the predicted value of remaining energy storage lifetime.

[0034] Preferably, the implementation of multi-timescale rolling optimization in step 5 includes:

[0035] On a second-level time scale, feedforward compensation is performed on the energy storage output based on the real-time frequency deviation. The amount of compensation is determined by the instantaneous amplitude and trend of the load mutation component.

[0036] On a minute-level timescale, the distribution characteristics of photovoltaic power output prediction errors are recalculated, and the prediction model parameters are adjusted according to the error type: bias correction is used for systematic errors caused by temperature changes, and variance scaling is used for random errors caused by cloud fluctuations.

[0037] Optimization cycle adaptive adjustment: When the prediction error exceeds the set range for multiple consecutive cycles, the model reconstruction mechanism is activated, and the prediction model is retrained based on the corresponding data features selected from the current dominant error source.

[0038] Preferably, the mode switching transition process control in step 5 includes:

[0039] Determining the transition time window: Based on the difference in control objectives between the preceding and following modes, calculate the minimum transition time to ensure power continuity. The difference is characterized by the cosine of the angle between the weight vectors of the objective function.

[0040] Smooth adjustment method for control parameters: The weight coefficients in the objective function are transitioned using an exponential decay law, and the decay time constant is matched with the transition time window;

[0041] Power mutation suppression mechanism: A power change rate constraint is introduced during the transition phase. The constraint value is determined based on the current maximum allowable charge and discharge rate of the energy storage system and the grid frequency tolerance capability.

[0042] Preferably, the closed-loop control link implementation in steps 2 to 5 includes:

[0043] Establish a data flow tracking mechanism to generate a data lineage graph for each processing step, ensuring the complete transmission path of input and output parameters;

[0044] During mode switching, perform data consistency checks, cache unprocessed data in the middle of the process, and re-inject it into the updated processing flow.

[0045] An exception handling loop is set up so that when a data gap or logical conflict is detected, the system automatically rolls back to the previous stable state and reinitializes the control parameters.

[0046] Accordingly, embodiments of the present invention also provide a photovoltaic power generation-energy storage-grid intelligent dispatching system based on multimodal energy management, including a memory configured to store instructions, a processor configured to call the instructions from the memory, and capable of implementing any of the photovoltaic power generation-energy storage-grid intelligent dispatching methods based on multimodal energy management described in any of the embodiments of the present invention when executing the instructions.

[0047] The beneficial effects of this invention are:

[0048] 1. Through a multimodal data-based energy management approach, the system can identify different types of photovoltaic (PV) output fluctuations in real time and adopt differentiated control strategies for different fluctuation patterns. Specifically, when PV output fluctuations are severe (such as sudden cloud cover changes), the system can predict the fluctuation characteristics in advance and reduce the frequent switching of the energy storage system through optimized charging and discharging strategies, thereby effectively extending battery life and reducing the number of overcharging and discharging cycles. This intelligent scheduling not only improves the lifespan of energy storage equipment but also reduces energy loss in the energy storage system.

[0049] 2. This invention utilizes multimodal data fusion technology to combine various data sources, including meteorological data, historical load data, photovoltaic panel backsheet temperature, and irradiance, to establish a nonlinear coupling model between photovoltaic module performance and environmental changes. This model can more accurately predict the changing trend of photovoltaic output, thereby significantly improving the accuracy of photovoltaic output prediction. When the system predicts changes in photovoltaic output, it can more precisely adjust the charging and discharging strategies of energy storage devices, avoiding deep discharge of the energy storage system due to prediction errors and reducing unnecessary capacity loss.

[0050] 3. This invention introduces a mechanism for dynamically adjusting the energy storage response rate, enabling real-time adjustments to the energy storage system's regulation strategy based on changes in grid inertia. When grid frequency fluctuates, the system dynamically adjusts the energy storage response rate according to the grid's inertial time constant, ensuring timely grid frequency recovery. This mechanism avoids prolonged frequency recovery time due to excessively slow grid inertial changes and effectively prevents overcharging / over-discharging problems caused by excessively fast or slow responses in the energy storage system, thereby ensuring the safe and stable operation of the energy storage system.

[0051] 4. This invention introduces a virtual inertia matching mechanism. When the grid frequency drops sharply, the energy storage system can adjust its virtual inertia to match the grid frequency recovery. This mechanism simulates inertial response, balances rapid fluctuations in grid frequency, and effectively reduces the overshoot of the energy storage system. This mechanism not only optimizes the grid frequency recovery speed but also prevents overcharging / over-discharging of the energy storage system during frequency fluctuations, ensuring coordinated stability between the grid and the energy storage system. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart of the steps of the method of the present invention;

[0054] Figure 2 This is a flowchart of the mode switching transition process control steps in step 5 of the method of the present invention;

[0055] Figure 3 This is a flowchart illustrating the steps of the closed-loop control link implementation in steps 2 to 5 of the method of the present invention.

[0056] Figure 4 An integrated cabinet provided for a specific implementation method. Detailed Implementation

[0057] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0058] Please see Figures 1-4 This invention provides a photovoltaic power generation-energy storage-grid intelligent dispatching method based on multimodal energy management. This step first involves real-time collection of operational data from various sources using multiple sensors and monitoring devices. Specifically, it includes:

[0059] Photovoltaic side: meteorological parameters (such as solar radiation intensity, air temperature, humidity, wind speed, etc.) and photovoltaic module temperature (reflecting the temperature change of the photovoltaic panel, which affects its output efficiency).

[0060] Energy storage system: Status parameters of energy storage devices (such as battery capacity, charging and discharging power, battery temperature, charging and discharging efficiency, etc.).

[0061] On the grid side: grid frequency (reflecting changes in grid load) and load demand parameters (predicting changes in grid load demand).

[0062] This data is collected through a real-time monitoring system and transmitted to a central processing system. Then, using spatiotemporal alignment technology, this data is fused within the same time window to form a comprehensive feature dataset, facilitating subsequent analysis and processing.

[0063] In step 2, by combining the characteristics of photovoltaic power output fluctuations with the grid frequency deviation, the dominant characteristics of load mutations and new energy fluctuations are identified and extracted through a model coupling method.

[0064] Characteristics of photovoltaic power output fluctuations: These include short-term fluctuations in photovoltaic power output, as well as uncertainties in photovoltaic power generation caused by weather changes.

[0065] Grid frequency deviation: This is usually caused by a sudden increase in load or fluctuations in new energy sources, which leads to grid frequency fluctuations.

[0066] By fusing these features and establishing a feature fusion model based on multimodal data, we can identify and predict the potential impact of photovoltaic fluctuations on grid frequency and generate dynamic regulation capability indicators for energy storage systems (i.e., the ability of energy storage systems to cope with sudden fluctuations).

[0067] In step 3, based on the photovoltaic fluctuation amplitude, frequency deviation component, and energy storage regulation capability, the present invention dynamically divides different scheduling modes.

[0068] When grid frequency deviations are significant, energy storage systems are prioritized for frequency regulation to quickly respond to grid frequency changes and reduce the impact of frequency fluctuations on grid stability. When photovoltaic (PV) output fluctuates significantly, energy storage systems smooth out these fluctuations, maintaining stable grid power output while considering the lifespan of the energy storage systems to avoid degradation issues caused by frequent charging and discharging. When grid load demand increases, energy storage systems provide virtual inertia support based on grid demand, assisting in maintaining grid frequency stability. These modes dynamically switch according to specific system states and changes in the external environment to ensure the optimality of the dispatch strategy.

[0069] In step 4, differentiated control strategies are adopted under different scheduling modes:

[0070] Frequency regulation priority mode: Optimizes the response rate of the energy storage system by responding to changes in grid frequency through rapid charging and discharging, ensuring the accuracy and timeliness of frequency control.

[0071] Power smoothing mode: Balances the relationship between photovoltaic power output fluctuations and energy storage system lifespan, adopts appropriate charging and discharging strategies, reduces unnecessary energy loss, and extends the lifespan of the energy storage system.

[0072] Grid-supported mode: In grid-supported mode, the rapid adjustment capability of the energy storage system and the real-time coordination with grid demand are realized through the dynamic matching of virtual inertia and grid demand.

[0073] Step 5 involves real-time adjustments to control decisions using a multi-timescale rolling optimization method. Within each optimization cycle, the system recalculates the energy storage system's charging and discharging plan based on the current mode state and future load forecasts. The mode switching triggering strategy is also adjusted as the system state changes to ensure power continuity between different modes. During the transition process, a smooth transition method is employed to avoid excessively drastic power changes between different modes, thereby ensuring the stability of the power grid and the energy storage system.

[0074] In one possible implementation, the system uses meteorological monitoring devices to collect data such as irradiance, ambient temperature, and wind speed at a frequency of seconds. This data reflects the external environmental conditions for photovoltaic power generation. Meteorological sensors are installed near the photovoltaic site, enabling real-time recording of environmental changes and providing a basis for subsequent photovoltaic output forecasting.

[0075] In the photovoltaic array, the backsheet temperature data of each subarray is collected in real time. Since the temperature may vary at different locations, a spatial interpolation model is used to process the temperature data.

[0076] The spatial interpolation model estimates the temperature at unmonitored locations based on the spacing between photovoltaic subarrays and the spatial gradient of surrounding environmental parameters using interpolation algorithms (such as Kriging interpolation or inverse distance weighted interpolation). This process is intended to supplement areas not directly monitored, ensuring that the temperature distribution of the entire photovoltaic array can be estimated through spatial interpolation, thereby obtaining a more accurate photovoltaic output prediction.

[0077] The meteorological and temperature data are aligned to a unified time base to ensure that all data points have the same timestamp. This step is to combine data from different sources (such as meteorological and temperature data) into a multidimensional data matrix, so that each data point corresponds to a specific time and spatial location.

[0078] The system acquires real-time state of charge (SOC) and state of health (SOH) data for battery clusters within the energy storage management system. The SOC represents the current energy reserves of the energy storage system, while the SOH indicates the operational health of the energy storage devices. Due to potential noise in the sensor data, a recursive filtering algorithm based on a state-space model (such as Kalman filtering) is used to filter this data.

[0079] The recursive filtering algorithm, by establishing a state-space model of the system, eliminates noise and errors to obtain more accurate battery state of charge and health data. This process helps to remove sensor measurement errors, improves data accuracy, and thus provides more reliable input for subsequent scheduling decisions.

[0080] Finally, the collected meteorological data, photovoltaic temperature data, and energy storage system status data were integrated into a three-dimensional data matrix, including timestamps, spatial locations, and parameter types. This data matrix reflects complete feature information at a specific time, spatial location, and corresponding parameter type, providing rich raw data for subsequent feature fusion, pattern recognition, and decision control.

[0081] In one possible implementation, photovoltaic output is influenced by multiple factors, particularly irradiance and photovoltaic panel temperature. In this step, irradiance data is first nonlinearly coupled with photovoltaic panel temperature data at the corresponding location, taking into account the combined effects of irradiance variations and temperature on photovoltaic output.

[0082] By coupling irradiance and photovoltaic temperature using a nonlinear method, the interaction between the two is revealed, reflecting the fluctuation characteristics of photovoltaic power output.

[0083] Variational mode decomposition (VMD) is employed to decompose the coupled signal. VMD effectively decomposes the signal into multiple intrinsic mode functions (IMFs), thereby extracting the fast fluctuation components (high-frequency fluctuations) caused by cloud movement and the slow fluctuation components (low-frequency fluctuations) caused by temperature changes. This process helps identify different fluctuation components in photovoltaic power generation output, providing clearer signals for subsequent scheduling decisions.

[0084] Variations in frequency deviation reflect the imbalance between grid load and generation. Empirical wavelet transform (EWT) is employed for time-frequency analysis of the grid frequency signal. EWT can efficiently separate different frequency bands in the signal, extracting load spikes (high-frequency bands) and renewable energy fluctuations (low-frequency bands). Load spikes represent sudden changes in grid load demand, while renewable energy fluctuations reflect fluctuations in renewable energy generation such as photovoltaic and wind power. This time-frequency decomposition method allows for a refined analysis of grid frequency deviation, providing a basis for grid dispatching and energy storage system optimization.

[0085] In this step, the historical degradation curves of the energy storage device are first obtained based on the current state of charge (SOC) and state of health (SOH) of the energy storage system. Based on the historical degradation data of the battery, a dynamic constraint model reflecting the relationship between the maximum charge / discharge rate and the battery's health status is established. This model considers battery performance degradation and aging effects, thus enabling the dynamic calculation of the battery's maximum permissible charge / discharge rate.

[0086] Based on the current state of charge and health, and combined with historical degradation curves, an adjustable capability coefficient is generated. This coefficient reflects the real-time adjustment capability of the energy storage system, providing precise adjustment parameters for scheduling decisions.

[0087] In one possible implementation, in order to achieve precise control of the photovoltaic fluctuation amplitude, it is first necessary to analyze the output fluctuation characteristics under different weather conditions in historical operating data.

[0088] Statistical analysis was performed on historical data, selecting photovoltaic power output fluctuations under different weather types (such as sunny, cloudy, and overcast). By analyzing the statistical distribution of these fluctuations, the fluctuation amplitude and its changing trend under different weather conditions were extracted.

[0089] By statistically analyzing the probability distribution of fluctuation data, a volatility rate reaching a certain confidence level (e.g., 95%) is selected as the grading threshold for fluctuation amplitude. This threshold is used to determine whether the fluctuation of photovoltaic output is within the normal range, thereby determining whether it is necessary to switch to a different dispatch mode.

[0090] The load mutation component reflects the rapid changes in grid load demand, and its instantaneous power change rate reflects the urgency of grid regulation. When calculating the instantaneous power change rate, frequency fluctuations caused by rapid load increases or decreases need to be considered.

[0091] The new energy fluctuation component reflects the fluctuations in power generation from renewable energy sources such as photovoltaics and wind power. This component typically has a long fluctuation period and a long duration.

[0092] A comprehensive index is constructed based on the instantaneous power change rate of the load surge component and the duration of the renewable energy fluctuation component. This index characterizes the urgency of grid regulation at the current moment, with higher weights indicating that the grid needs priority regulation to ensure system stability.

[0093] To better regulate the state of the energy storage system, this step dynamically adjusts the health status and adjustability of the energy storage system to cope with potential future fluctuations.

[0094] The degradation rate is calculated based on the health status of the energy storage system (such as SOC, SOH, etc.). Deterioration in battery health leads to a decrease in the maximum charge / discharge rate of the energy storage device, thus affecting its regulation capability.

[0095] By utilizing the current degradation rate of the energy storage system, combined with historical data and environmental factors, the declining trend of the system's adjustable capacity over a future set time period can be predicted. Based on the prediction results, the mode switching threshold can be adjusted in advance to address the decline in energy storage capacity.

[0096] Based on the predicted degradation trend of the energy storage system, the mode switching threshold is adjusted to optimize the operation mode of the energy storage system and ensure that the system can still operate stably when the energy storage capacity is weak.

[0097] In one possible implementation, in order to effectively adjust the frequency deviation, it is necessary to establish an accurate relationship model between the frequency deviation and the energy storage power.

[0098] Based on the relationship between grid frequency variations and energy storage system power response, a transfer function of frequency deviation versus energy storage power response is constructed. This model can describe the response characteristics of energy storage systems to frequency fluctuations and provide a theoretical basis for regulation.

[0099] The inertial time constant of a power grid is closely related to its response speed. Based on the current inertial time constant of the power grid, the optimal response time constant of the energy storage system is calculated. The optimal response time constant ensures that the energy storage system can make appropriate power adjustments in a short time to cope with frequency fluctuations.

[0100] The state of charge (SOC) of an energy storage system is crucial to its ability to regulate energy storage power during frequency regulation. To ensure that the energy storage system does not exceed its safe operating range, it is necessary to consider the limitations of the SOC and construct a dynamic regulation range.

[0101] Based on the state of charge of the energy storage system, a safe adjustment range is calculated in real time. This range is adjusted according to changes in factors such as the charging and discharging status and health condition of the energy storage battery, thereby ensuring the safe operation of the energy storage system.

[0102] During the frequency recovery phase, a variable proportional gain control strategy is employed. When the initial frequency deviation is large, high-gain rapid adjustment quickly reduces the frequency deviation. As the system approaches steady state, a low-gain fine-tuning strategy is switched to avoid over-adjustment and unnecessary fluctuations.

[0103] In actual power grid dispatching, sudden changes in load often lead to frequency fluctuations. In order to anticipate and respond to these sudden changes in advance, the predicted value of the load change component is added as a feedforward compensation term.

[0104] By using load demand forecasting models, we can analyze load change trends and predict sudden load changes in advance. Based on this forecast data, we can adjust energy storage output in advance to offset the impact of these sudden changes on frequency.

[0105] By using a feedforward compensation strategy, the corresponding energy storage power can be adjusted before load changes occur, and the frequency deviation can be corrected in advance, thereby reducing the impact of frequency fluctuations on the power grid.

[0106] In one possible implementation, firstly, the moving average root mean square (RMS) value of the rate of change in power output over the predicted time domain needs to be calculated. This can be done by analyzing historical photovoltaic (PV) power output data to predict power output changes over a future period and calculating the RMS value of the rate of change. Simultaneously, the current temperature gradient of the PV modules needs to be considered, as temperature changes directly affect the output power of the PV modules. For example, in high-temperature environments, the output power of the PV modules may decrease, therefore the PV fluctuation suppression index needs to be adjusted based on the temperature change gradient.

[0107] Based on the current number of charge-discharge cycles and health state parameters, a lifetime degradation model is established, with the magnitude of state of charge (SOC) change as the variable. The more charge-discharge cycles, the greater the lifetime degradation of the energy storage system. Health state parameters reflect the current state of the energy storage system, including capacity decay and increased internal resistance. By establishing the lifetime degradation model, the lifetime degradation of the energy storage system under different SOC change magnitudes can be predicted.

[0108] The weighting coefficients are adaptively adjusted based on the urgency of grid peak-shaving demand and the predicted remaining lifespan of energy storage. The urgency of grid peak-shaving demand reflects the real-time demand for peak-shaving capacity. If the grid peak-shaving demand is urgent, the weight of photovoltaic fluctuation suppression needs to be appropriately increased; if the remaining lifespan of energy storage is low, the weight of energy storage lifespan loss needs to be appropriately increased. For example, during peak grid load periods, the urgency of peak-shaving demand is high, so the weight of photovoltaic fluctuation suppression is increased to prioritize meeting the grid's peak-shaving demand. When the lifespan of the energy storage system is nearing its end, the weight of energy storage lifespan loss is increased to reduce the depth of charge and discharge of the energy storage system and extend its service life.

[0109] In one possible implementation, the frequency deviation is calculated by monitoring changes in the power grid frequency in real time on a second-scale timescale using a power grid frequency sensor. The frequency deviation is the difference between the actual frequency and the rated frequency (e.g., 50Hz or 60Hz).

[0110] Based on the real-time frequency deviation, combined with the instantaneous amplitude and trend of load mutation components, the feedforward compensation amount of energy storage output is calculated. Specifically, when a frequency deviation is detected, the system quickly analyzes the instantaneous amplitude (i.e., the magnitude of the mutation) and trend (i.e., the speed of the mutation) of the load mutation, and calculates the required compensation amount based on this information.

[0111] The calculated feedforward compensation is applied to the output control of the energy storage system, rapidly adjusting its charging and discharging power to offset the impact of sudden load changes on the grid frequency. For example, if the frequency deviation is caused by a sudden increase in load, the energy storage system will immediately increase its discharge power to compensate for the power deficit in the grid, thereby stabilizing the frequency.

[0112] The error of the photovoltaic power output prediction model is recalculated and analyzed on a minute-level timescale. The prediction error is the difference between the actual photovoltaic power output and the predicted power output.

[0113] Data analysis algorithms are used to identify the types of prediction errors, distinguishing between systematic errors caused by temperature changes and random errors caused by cloud fluctuations. Systematic errors are usually caused by changes in the performance parameters of photovoltaic modules due to changes in ambient temperature; random errors are caused by random factors such as cloud movement blocking sunlight.

[0114] Based on the error type, appropriate correction methods are used to adjust the parameters of the prediction model. For systematic errors, a bias correction method is used, which adjusts the output of the prediction model by a fixed bias to compensate for the effects of temperature changes. For random errors, a variance scaling method is used, which adjusts the output variance of the prediction model according to the statistical characteristics of cloud fluctuations to better reflect the range of random error variations.

[0115] Continuously monitor the prediction error to determine if it exceeds the set range. The set range is an error threshold predetermined based on system operating experience and stability requirements.

[0116] When the prediction error exceeds a set range for multiple consecutive periods, the model reconstruction mechanism is triggered. This means that the current prediction model is no longer well adapted to actual operating conditions and needs to be retrained to improve prediction accuracy.

[0117] Analyze the main sources of current prediction errors to determine whether temperature changes, cloud fluctuations, or other factors are the main drivers of these errors.

[0118] Based on the dominant error source, select the corresponding data features, reconstruct the training dataset, and retrain the prediction model using the new dataset. For example, if temperature change is the dominant error source, select temperature-related feature parameters to retrain the model to improve its adaptability to temperature changes and prediction accuracy.

[0119] In one possible implementation, a transition time window needs to be determined to ensure power continuity during mode switching. Specifically, this involves calculating the difference in control objectives between the preceding and following modes, characterized by the cosine of the angle between the weight vectors of the objective function. For example, if the current mode weight vector is [0.6, 0.3, 0.1] and the next mode is [0.2, 0.5, 0.3], the difference is calculated using the cosine similarity formula. The minimum transition time is then determined by looking up the difference in a preset table, ensuring a smooth power transition.

[0120] During the transition, the weight coefficients of the objective function are adjusted according to an exponential decay law. Assume the current weights are [0.6, 0.3, 0.1], and the transition objective is [0.2, 0.5, 0.3]. Using the transition time window as the time base, a decay time constant is set to match the window. For example, if the transition duration is 10 seconds, the time constant is set to 5 seconds, and the weight coefficients decay exponentially to achieve a smooth switch and avoid system oscillations caused by sudden parameter changes.

[0121] During the transition phase, a power change rate constraint is introduced, with the constraint value determined by considering both the current maximum allowable charge / discharge rate of the energy storage system and the grid frequency tolerance capability. For example, if the current maximum allowable charge / discharge rate of the energy storage system is 0.5C, and the grid frequency tolerance capability corresponds to a power change rate of 0.3 pu / s, the constraint value is taken as the smaller of the two, 0.3 pu / s. The power change rate is monitored in real time during the transition; if it exceeds the constraint value, power output is limited to ensure a smooth transition.

[0122] In one possible implementation, when data is collected from the photovoltaic (PV) side, energy storage system, and grid side, a unique identifier is assigned to each data source, recording key information such as the data collection time, location, and type. For example, for PV-side meteorological parameters, the system records which meteorological monitoring device the data originated from, the specific timestamp of the collection, and whether it is irradiance, ambient temperature, or wind speed data. This information serves as the starting node of the data lineage map.

[0123] During the processing of the collected data, such as spatial interpolation of photovoltaic temperature data and time alignment of meteorological and temperature data, the processing flow and transformation relationships are recorded in real time. Each processing step updates the data lineage graph, indicating which processing operations the data underwent, the order of processing, and what new datasets were generated. For example, after aligning meteorological and temperature data according to a unified time base, this alignment operation is recorded as a link in the data lineage graph, clarifying the correspondence between input and output data.

[0124] When establishing a multimodal feature fusion model, the nonlinear coupling operations, the application of variational mode decomposition algorithms, and the empirical wavelet transform operations in the photovoltaic power output fluctuation feature extraction process are recorded. The new feature datasets generated by these operations are all included in the data lineage map, indicating their source and generation method. For example, for the separated fast and slow fluctuation components of photovoltaic power output, the specific variational mode decomposition algorithm from which they were obtained from the original irradiance data and photovoltaic panel temperature data, as well as the corresponding parameter settings, are recorded.

[0125] When dynamically classifying frequency regulation priority mode, power smoothing mode, and grid support mode, the basis for mode classification, the method for setting boundary conditions, and the generation process of the corresponding control strategy for each mode are recorded. For example, it records how the photovoltaic fluctuation amplitude threshold is determined based on the statistical distribution of output fluctuations under different weather types in historical operating data, and how the weight of the frequency deviation component is constructed based on the instantaneous power change rate of the load mutation component and the duration of the renewable energy fluctuation component. These records serve as key nodes in the data lineage graph, ensuring that the data source and basis for mode classification and control strategy generation can be traced during subsequent mode switching and control strategy execution.

[0126] Through the above methods, a complete data lineage map was established, providing a clear input and output parameter transmission path for each processing step, and realizing the transparency and traceability of data flow.

[0127] When a mode switch is triggered, the status of the data being processed is monitored in real time. This includes checking whether the data has completed all processing steps under the current mode, or whether it is still being processed. For example, when switching from frequency regulation priority mode to power smoothing mode, it checks whether the response rate optimization operation of the energy storage system has been completed, and whether the calculation of photovoltaic fluctuation suppression index and energy storage lifetime loss index is in progress.

[0128] For intermediate data that has not been fully processed, the system immediately activates a caching mechanism. This cached data includes various intermediate calculation results under the current mode, incomplete control commands, and related status parameters. For example, if the power regulation command of the energy storage system is triggered to switch modes before it is fully executed in frequency priority mode, then the incomplete power regulation command and its related calculation parameters (such as the current energy storage state of charge, frequency deviation, etc.) will be cached in a dedicated storage area.

[0129] Under the new mode, cached intermediate data is re-injected into the processing flow according to the new mode requirements. This requires necessary data transformation and adaptation to ensure it conforms to the processing specifications of the new mode. For example, in the power smoothing mode, cached energy storage power regulation commands are re-evaluated. Based on the photovoltaic fluctuation suppression and energy storage lifetime loss balancing strategy under the power smoothing mode, the commands are adjusted and optimized before execution continues. This may involve recalculating the charging and discharging power of the energy storage system, adjusting the parameters of the photovoltaic output prediction model, etc., to ensure that intermediate data can be correctly processed under the new mode, thereby guaranteeing the continuity and consistency of the entire scheduling process.

[0130] The system continuously monitors the data flow and control process in real time to check for data gaps or logical conflicts. Data gaps may manifest as sudden data loss, inconsistent timestamps, or abnormal data values ​​(such as photovoltaic output or energy storage state of charge exceeding reasonable ranges). Logical conflicts may include contradictions between control commands (such as issuing commands to increase and decrease energy storage power simultaneously) or mismatches between data and control objectives (such as the energy storage system's output power not meeting the dynamic matching requirements of virtual inertia under grid-supported modes).

[0131] Once a data gap or logical conflict is detected, a fault diagnosis procedure should be initiated immediately to quickly locate the source of the fault. This may involve tracing back the data lineage graph, examining the flow of data through each processing step, and analyzing the execution status of each stage in the control process. For example, if an abnormal energy storage power regulation command is detected, the data lineage graph can be used to trace back to the mode in which the command was generated, and what data and parameters were used as the basis for its generation, thereby determining whether the problem occurred in the data acquisition, feature fusion, mode segmentation, or control strategy generation stage.

[0132] After identifying the source of the fault, the system automatically reverts to the previous stable state. This involves restoring previously cached normal data and valid control parameters, and reloading reliable system state information. Then, based on the current system state and new data input, the control parameters are reinitialized. For example, in power smoothing mode, if an anomaly is detected in the calculation of energy storage lifetime loss indicators due to data errors, the system will revert to the previous stable power smoothing mode state, reload the parameters required for calculating the correct photovoltaic fluctuation suppression indicators and energy storage lifetime loss indicators, and reinitialize the control model to ensure that the system can continue to operate in the correct state and maintain stable grid dispatch.

[0133] Accordingly, embodiments of the present invention also provide a photovoltaic power generation-energy storage-grid intelligent dispatching system based on multimodal energy management, including a memory configured to store instructions, a processor configured to call the instructions from the memory, and capable of implementing any of the photovoltaic power generation-energy storage-grid intelligent dispatching methods based on multimodal energy management described in any of the embodiments of the present invention when executing the instructions.

[0134] It is understood that the main operating device of the method of the present invention can be installed and fixed on, for example, Figure 4 The integrated cabinet shown includes an AC / DC side lightning protection system and an access point protection circuit breaker. It also includes a liquid cooling system consisting of a liquid chiller, liquid cooling pipes, and liquid cooling plates, as well as a secondary aerosol fire extinguishing device, to provide the necessary environmental guarantee for the operation of the method.

[0135] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0136] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A photovoltaic power generation-energy storage-grid intelligent dispatching method based on multimodal energy management, characterized in that, Includes the following steps: Step 1: Collect and integrate multi-source operational data in real time, including photovoltaic meteorological parameters and module temperature, energy storage system status parameters, grid frequency and load demand parameters, to form a spatiotemporally aligned feature dataset; Step 2: Establish a multimodal feature fusion model. By coupling the photovoltaic power output fluctuation characteristics with the grid frequency deviation component, identify the dominant characteristics of load mutation and new energy fluctuation, and generate energy storage dynamic adjustment capability indicators. Step 3: Based on the photovoltaic fluctuation amplitude, frequency deviation component composition and energy storage regulation capability, dynamically divide the frequency regulation priority mode, power smoothing mode and grid support mode, and set the boundary conditions for mode switching; Step 4: Construct differentiated control strategies for each mode, optimize the energy storage response rate under the frequency regulation priority mode, balance photovoltaic fluctuation suppression and energy storage lifetime loss under the power smoothing mode, and achieve dynamic matching between virtual inertia and grid demand under the grid support mode; Step 5: Perform multi-timescale rolling optimization control, reconstruct according to the mode switching trigger strategy, and maintain power continuity between different modes through the transition process.

2. The photovoltaic power generation-energy storage-grid intelligent dispatching method based on multimodal energy management according to claim 1, characterized in that, The generation of the spatiotemporal alignment feature dataset in step 1 includes: The irradiance, ambient temperature and wind speed data are collected at a frequency of seconds using a meteorological monitoring device, and the backsheet temperature data of each subarray in the photovoltaic array is obtained simultaneously. A spatial interpolation model is established for photovoltaic temperature data, and the estimated temperature values ​​of unmonitored locations are calculated based on the spacing between adjacent subarrays and the spatial gradient of environmental parameters. Meteorological and temperature data are aligned with a unified time base to construct a three-dimensional data matrix that includes timestamps, spatial locations, and parameter types. The energy storage management system acquires real-time data on the state of charge and health of battery clusters, and a recursive filtering algorithm based on a state-space model is used to eliminate sensor noise.

3. The photovoltaic power generation-energy storage-grid intelligent dispatching method based on multimodal energy management according to claim 1, characterized in that, The establishment of the multimodal feature fusion model in step 2 includes: The process of extracting the fluctuation characteristics of photovoltaic output: The irradiance data is nonlinearly coupled with the temperature of the photovoltaic panel at the corresponding location, and the rapid fluctuation component caused by cloud movement and the slow fluctuation component caused by temperature change are separated by the variational mode decomposition algorithm. The decomposition process of power grid frequency deviation: empirical wavelet transform is used to perform time-frequency analysis on the frequency signal to extract the load change component in the high-frequency band and the new energy fluctuation component in the low-frequency band. The calculation process of the dynamic adjustment capability index of energy storage: Based on the historical degradation curves of the current state of charge and health, a dynamic constraint model reflecting the maximum allowable charge and discharge rate is established to generate the adjustable capability coefficient.

4. The photovoltaic power generation-energy storage-grid intelligent dispatching method based on multimodal energy management according to claim 1, characterized in that, The setting of the mode switching boundary conditions in step 3 includes: Determination of photovoltaic fluctuation amplitude threshold: Analyze the statistical distribution of output fluctuation under different weather types in historical operation data, and select the fluctuation rate with a cumulative probability reaching a set confidence level as the grading threshold; Calculation of frequency deviation component weight: Based on the instantaneous power change rate of the load mutation component and the duration of the new energy fluctuation component, a comprehensive index characterizing the urgency of grid regulation is constructed. Dynamic adjustment of energy storage regulation capacity threshold: Based on the current rate of decline in the health status of the energy storage system, predict the downward trend of the adjustable capacity within a set time period in the future, and correct the mode switching threshold in advance.

5. The photovoltaic power generation-energy storage-grid intelligent dispatching method based on multimodal energy management according to claim 1, characterized in that, The optimization process for the frequency-priority mode in step 4 includes: Establish a transfer function model of frequency deviation-energy storage power response, and calculate the optimal response time constant based on the current grid inertial time constant; A dynamic adjustment range considering the safety boundary of the energy storage state of charge is constructed. A variable proportional coefficient control strategy is adopted in the frequency recovery stage. High gain fast adjustment is adopted in the initial stage, and low gain fine adjustment is adopted when approaching steady state. The predicted value of load mutation component is introduced as a feedforward compensation term to adjust the energy storage output in advance to offset the cumulative effect of frequency deviation.

6. The photovoltaic power generation-energy storage-grid intelligent dispatching method based on multimodal energy management according to claim 1, characterized in that, The multi-objective optimization of the power smoothing mode in step 4 includes: Define the photovoltaic fluctuation suppression index: calculate the moving root mean square value of the rate of change of output in the predicted time domain, and correlate it with the current temperature change gradient of the photovoltaic module; Define the energy storage lifetime degradation index: Based on the current charge-discharge cycle count and health status parameters, establish a lifetime degradation model with the change in state of charge as the variable; A dynamic weight allocation method is adopted to balance the dual-objective optimization, and the weight coefficients are adaptively adjusted according to the urgency of grid peak-shaving demand and the predicted value of remaining energy storage lifetime.

7. The photovoltaic power generation-energy storage-grid intelligent dispatching method based on multimodal energy management according to claim 1, characterized in that, The implementation of multi-timescale rolling optimization in step 5 includes: On a second-level time scale, feedforward compensation is performed on the energy storage output based on the real-time frequency deviation. The amount of compensation is determined by the instantaneous amplitude and trend of the load mutation component. On a minute-level timescale, the distribution characteristics of photovoltaic power output prediction errors are recalculated, and the prediction model parameters are adjusted according to the error type: bias correction is used for systematic errors caused by temperature changes, and variance scaling is used for random errors caused by cloud fluctuations. Optimization cycle adaptive adjustment: When the prediction error exceeds the set range for multiple consecutive cycles, the model reconstruction mechanism is activated, and the prediction model is retrained based on the corresponding data features selected from the current dominant error source.

8. The photovoltaic power generation-energy storage-grid intelligent dispatching method based on multimodal energy management according to claim 1, characterized in that, The mode switching transition process control in step 5 includes: Determining the transition time window: Based on the difference in control objectives between the preceding and following modes, calculate the minimum transition time to ensure power continuity. The difference is characterized by the cosine of the angle between the weight vectors of the objective function. Smooth adjustment method for control parameters: The weight coefficients in the objective function are transitioned using an exponential decay law, and the decay time constant is matched with the transition time window; Power mutation suppression mechanism: A power change rate constraint is introduced during the transition phase. The constraint value is determined based on the current maximum allowable charge and discharge rate of the energy storage system and the grid frequency tolerance capability.

9. The photovoltaic power generation-energy storage-grid intelligent dispatching method based on multimodal energy management according to claim 1, characterized in that, The closed-loop control link implementation in steps 2 to 5 includes: Establish a data flow tracking mechanism to generate a data lineage graph for each processing step, ensuring the complete transmission path of input and output parameters; During mode switching, perform data consistency checks, cache unprocessed data in the middle of the process, and re-inject it into the updated processing flow. An exception handling loop is set up so that when a data gap or logical conflict is detected, the system automatically rolls back to the previous stable state and reinitializes the control parameters.

10. A photovoltaic power generation-energy storage-grid intelligent dispatching system based on multimodal energy management, characterized in that, The system includes a memory configured to store instructions, a processor configured to retrieve the instructions from the memory, and, when executing the instructions, to implement a photovoltaic power generation-storage-grid intelligent scheduling method based on multimodal energy management as described in any one of claims 1-9.

Images