Optical storage off-grid system suitable for tropical grassland terrain

Abstract

The invention belongs to the technical field of new energy power generation, and discloses an optical storage off-grid system suitable for tropical grassland terrains. Environmental data are acquired through the microclimate acquisition module, a dust accumulation model is constructed in combination with the deposition prediction module, shadow dynamic influence is evaluated by using the terrain shielding analysis module, and the temperature gradient problem is solved by relying on the energy storage heat management module. A three-dimensional correction curved surface is adopted to accurately predict photovoltaic available power, and energy optimization distribution is realized based on a multi-time scale scheduling strategy. According to the invention, an abnormal current drop mode is identified in real time through a ripple monitoring technology, and a maintenance period and operation parameters are dynamically adjusted. The technologies work cooperatively, the influence of the severe environment on the system efficiency is effectively overcome, and the stability and reliability of the off-grid system under the resource fluctuation condition are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of new energy power generation, more particularly, the present application relates to a photovoltaic and energy storage off-grid system suitable for tropical savanna terrain. BACKGROUND

[0002] The existing photovoltaic and energy storage off-grid system generally ignores the unique environmental characteristics of the tropical savanna terrain environment, such as strong dry-wet season alternation, high concentration of airborne particulate matter, complex undulating terrain, and extreme temperature difference conditions. In actual application, the long dry season without rain and frequent sandstorms cause rapid dust accumulation on the photovoltaic surface. The existing system lacks accurate prediction capability for dust deposition rate, and the cleaning cycle setting often relies on experience rather than data-driven, resulting in waste of cleaning resources or serious efficiency decline. At the same time, the hilly terrain of the tropical savanna causes the photovoltaic array to be subjected to complex shadow shielding at different times. The traditional system uses static shielding analysis and ignores the dynamic evolution characteristics of the shadow boundary, resulting in a significant deviation between the actual power generation and the theoretical prediction value. The tropical savanna environment has large diurnal temperature difference and strong sunlight, and the efficiency of the energy storage unit decays seriously in high temperature environment. However, the existing system lacks fine monitoring of the temperature gradient field and heat decay compensation mechanism. At the energy scheduling level, the traditional strategy is based on a single time scale and cannot effectively cope with the seasonal load demand changes and intermittent resource characteristics in the tropical savanna area. In addition, the lack of real-time micro-characteristics monitoring capability of the system operation state makes it difficult for the system to maintain energy balance under large resource fluctuations, resulting in unstable power supply, energy waste, and intensified equipment wear and tear, which seriously affects the popularization and application of off-grid systems in remote tropical savanna areas and their long-term reliability.

[0003] In view of this, the present application proposes a photovoltaic and energy storage off-grid system suitable for tropical savanna terrain to solve the above problems. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application provides the following technical scheme: a photovoltaic and energy storage off-grid system suitable for tropical savanna terrain, comprising:

[0005] A microclimate acquisition module is configured to acquire microclimate time series data of a region where the photovoltaic array is located.

[0006] A deposition prediction module is configured to construct a dust deposition rate prediction model of the photovoltaic surface according to the wind speed vector, humidity fluctuation and particulate matter concentration in the microclimate time series data. The dust deposition rate prediction model includes a dry season acceleration factor and a wet season self-cleaning coefficient.

[0007] A terrain shading analysis module is configured to acquire terrain elevation data and photovoltaic array layout coordinates, calculate terrain shadow projection paths at different times, and generate a shading influence matrix containing dynamic evolution characteristics of the shadow boundary.

[0008] a thermal management module for collecting multi-point temperature distribution data of the energy storage unit housing, calculating an internal temperature gradient field of the energy storage unit, and establishing a thermal decay compensation model of energy storage efficiency according to the correlation between the temperature gradient field and the diurnal temperature difference of the environment;

[0009] a power correction module for combining the dust deposition rate prediction model, the shading influence matrix and the thermal decay compensation model to construct a three-dimensional correction surface of the available power of the photovoltaic, the dimensions of the three-dimensional correction surface being time, spatial position and environmental factors, respectively;

[0010] a strategy generation module for generating a multi-time scale energy scheduling strategy including hour-level, day-level and week-level based on the three-dimensional correction surface and the seasonal decomposition data of historical load demand, the energy scheduling strategy achieving peak clipping and valley filling through non-uniform segmentation of the energy storage charging and discharging window;

[0011] a ripple monitoring module for extracting micro-fluctuation characteristics of the output of the photovoltaic array through high-frequency sampling of the direct-current side current ripple of the inverter, and identifying an abnormal current drop mode associated with the dust shading gradient in the micro-fluctuation characteristics;

[0012] a dynamic correction module for dynamically correcting the deposition rate parameter in the dust deposition rate prediction model according to the occurrence frequency and amplitude change of the abnormal current drop mode, and triggering advance or delay adjustment of the photovoltaic cleaning period to obtain a corrected multi-time scale energy scheduling strategy;

[0013] a control execution module for controlling the charging and discharging power and timing of the energy storage unit based on the corrected multi-time scale energy scheduling strategy to achieve energy self-balancing of the off-grid system of the photovoltaic and energy storage in the hot savanna environment.

[0014] The technical effects and advantages of the off-grid system of the photovoltaic and energy storage suitable for the hot savanna terrain of the present application are as follows:

[0015] The present application accurately grasps the environmental microclimate characteristics, effectively overcomes the influence of dust deposition, prolongs the effective power generation time of the photovoltaic equipment, and reduces unnecessary maintenance costs. In complex terrain conditions, the present application can intelligently cope with the power generation efficiency fluctuations caused by shadow shading to ensure the maximum utilization of photovoltaic resources. Especially in high temperature environments, through the advanced thermal management mechanism, the thermal decay problem of the energy storage unit is effectively inhibited, the service life of the equipment is prolonged, and the energy conversion efficiency is improved. The multi-time scale energy scheduling strategy realized by the present application enhances the power supply reliability, enabling the off-grid system to calmly cope with the resource fluctuations and load changes brought by the alternation of dry and wet seasons. The real-time monitoring and dynamic adjustment mechanism enables the system to have adaptive ability and maintain stable operation in the face of unpredictable environmental changes. The present application not only improves the overall efficiency of the system, but also reduces the life cycle cost. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A schematic diagram of a light storage off-grid system suitable for tropical savanna terrain. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0018] The present application provides a light storage off-grid system suitable for tropical savanna terrain. The execution body of the system includes but is not limited to the following: a distributed energy management platform, a micro-grid control center, an intelligent photovoltaic maintenance system, a meteorological environment monitoring station, etc., which can be regarded as a general computing node of the present application, and the system control part includes but is not limited to: a cloud energy scheduling engine, a distributed microclimate analysis system, and an intelligent photovoltaic performance predictor.

[0019] Please refer to Figure 1 In an embodiment of the present application, a light storage off-grid system suitable for tropical savanna terrain includes:

[0020] A microclimate acquisition module is used to obtain microclimate time series data of the area where the photovoltaic array is located. The microclimate time series data includes key parameters such as wind speed and direction, atmospheric humidity, particulate matter concentration, temperature change, and solar radiation intensity, which are collected in real time through a distributed meteorological sensor network. The wind speed and direction data record the speed vector and direction change of air flow, the humidity data record the fluctuation of water content in the atmosphere in detail, the particulate matter concentration reflects the content distribution of dust and other particles in the air, and the temperature and solar radiation data record the change trend of environmental heat. These data provide comprehensive environmental parameters for subsequent analysis, ensuring the accurate adaptability and response capability of the system to the special environment of tropical savanna.

[0021] A deposition prediction module is used to construct a photovoltaic surface dust deposition rate prediction model according to the wind speed vector, humidity fluctuation, and particulate matter concentration in the microclimate time series data. The dust deposition rate prediction model includes a dry season acceleration factor and a wet season self-cleaning coefficient, which describes the accumulation law of dust on the photovoltaic surface under different climate conditions. The dry season acceleration factor represents the acceleration effect of dust accumulation under long-term rainless conditions, and the wet season self-cleaning coefficient quantifies the cleaning effect of rainfall on the photovoltaic surface. These indicators together constitute the core parameters of the dynamic prediction of dust deposition, providing a basis for performance prediction of the system under different seasonal conditions.

[0022] The terrain sheltering analysis module is configured to obtain terrain elevation data and photovoltaic array layout coordinates, calculate terrain shadow projection paths at different times, and generate a sheltering influence matrix containing dynamic evolution characteristics of shadow boundaries. The module first constructs a three-dimensional terrain model, and then calculates shadow projection based on solar trajectory to accurately simulate the sheltering effect of grassland undulating terrain on photovoltaic arrays, forming a time-space sheltering influence matrix to provide quantitative evaluation of terrain factors for power generation prediction.

[0023] The energy storage thermal management module is configured to collect multi-point temperature distribution data of an energy storage unit shell, calculate an internal temperature gradient field of the energy storage unit, and establish a thermal decay compensation model of energy storage efficiency according to the correlation between the temperature gradient field and the diurnal temperature difference. The module reveals the thermal characteristics of the energy storage unit in a hot environment through fine temperature monitoring and thermodynamic analysis, and constructs a temperature-efficiency mapping relationship to provide thermal environment adaptability strategies for energy storage management.

[0024] The power correction module is configured to combine the dust deposition rate prediction model, the sheltering influence matrix, and the thermal decay compensation model to construct a three-dimensional correction surface of photovoltaic available power. The three-dimensional correction surface has dimensions of time, spatial position, and environmental factors, and comprehensively describes the comprehensive influence of various environmental factors on photovoltaic power generation performance to provide an accurate power prediction basis for energy scheduling.

[0025] The strategy generation module is configured to generate multi-time scale energy scheduling strategies including hourly, daily, and weekly levels based on the three-dimensional correction surface and seasonal decomposition data of historical load demand. The multi-time scale energy scheduling strategies achieve peak clipping and valley filling through non-uniform segmentation of energy storage charging and discharging windows, and make forward-looking plans for demand fluctuations and resource fluctuations at different time scales to optimize system energy utilization efficiency.

[0026] The ripple monitoring module is configured to extract micro-fluctuation characteristics of photovoltaic array output by high-frequency sampling of inverter DC side current ripple, and identify abnormal current drop patterns associated with dust sheltering gradient in the micro-fluctuation characteristics. The module extracts dust distribution characteristics from subtle changes in current waveform through high-frequency signal analysis technology to realize real-time monitoring of the pollution state of photovoltaic surfaces and provide data support for cleaning and maintenance decisions.

[0027] The dynamic correction module is configured to dynamically correct the deposition rate parameter in the dust deposition rate prediction model according to the occurrence frequency and amplitude change of the abnormal current drop pattern, and trigger advance or delay adjustment of the photovoltaic cleaning period to obtain a corrected multi-time scale energy scheduling strategy. The module realizes adaptive adjustment of model parameters through comparison of real-time monitoring data and prediction models to improve prediction accuracy and optimize maintenance strategies.

[0028] The control execution module controls the charging and discharging power and timing of the energy storage units based on a modified multi-timescale energy dispatch strategy, achieving energy self-balancing of the off-grid photovoltaic-storage system in a tropical savanna environment. This module translates the strategy into specific control commands, adjusts the operating state of the energy storage system, achieves supply-demand balance, and ensures stable system operation.

[0029] The modules are connected via wired and / or wireless means to enable data transmission between them.

[0030] In this embodiment of the invention, the detailed implementation steps for constructing a prediction model for dust deposition rate on photovoltaic surfaces include:

[0031] The wind speed vector is decomposed into a tangential wind speed component parallel to the photovoltaic surface and a normal wind speed component perpendicular to the photovoltaic surface. Wind speed decomposition is a fundamental step in accurately evaluating the effectiveness of wind-driven cleaning, and the composite wind force is vectorized according to its direction of action. The decomposition process is based on photovoltaic installation angle and wind direction data, and uses vector projection calculation. The tangential wind speed component mainly affects the lateral movement of dust and the surface cleaning effect, while the normal wind speed component affects dust settling and stirring. The wind speed component calculation uses the triangular decomposition method, using the angle between the wind speed vector and the normal to the photovoltaic surface as the decomposition basis. This accurate decomposition method provides a directional data foundation for subsequent evaluation of wind-driven cleaning efficiency.

[0032] The ratio of the tangential wind speed component to a preset critical wind speed is calculated as the wind-driven cleaning efficiency factor. The cleaning efficiency factor is a key indicator for quantifying wind-driven cleaning capability, reflecting the effectiveness of actual wind speed in dust removal. The calculation process is based on fluid mechanics principles, considering the force characteristics of particles and surface adhesion. When the tangential wind speed exceeds the critical wind speed, the wind force is sufficient to overcome the adhesion between dust and the surface, resulting in effective cleaning; when the wind speed is below the critical value, the cleaning effect decreases significantly. The cleaning efficiency factor exhibits a non-linear relationship with the tangential wind speed, and the efficiency increase tends to level off after exceeding the critical value. This factor provides a dynamically adjustable parameter for dust accumulation prediction, enabling the model to accurately reflect the actual effect of wind-driven cleaning.

[0033] This method extracts the gradient rate of change during the rising phase of humidity fluctuation data, identifies periods where the rate of change exceeds a preset humidity threshold, and marks these periods as potential rainfall precursors. Humidity gradient analysis is an effective method for predicting rainfall events, identifying phases of rapid increase in atmospheric moisture through the first derivative characteristics of humidity time-series data. The analysis process first smooths the raw humidity data to reduce noise; then, it calculates the rate of change of humidity at adjacent time points, identifying periods where the rate of change exceeds the threshold; finally, through a comprehensive evaluation of duration and magnitude of change, it determines potential rainfall precursor periods. This rainfall prediction method based on dynamic humidity characteristics can detect rainfall signs earlier than simply relying on absolute humidity values, providing the system with early warning time.

[0034] By statistically analyzing the time intervals between potential rainfall precursors and actual rainfall events, a mapping relationship between humidity gradients and rainfall lead time is established. This mapping relationship is a key component for improving the accuracy of rainfall forecasting, as it correlates humidity variation characteristics with actual rainfall timing. The establishment process is based on historical data analysis, collecting a large number of sample pairs of humidity gradients and rainfall time intervals, and constructing a functional relationship through regression analysis. The mapping model considers the variability of different seasons and meteorological conditions, adopting a piecewise function structure and setting corresponding prediction parameters for different gradient ranges. This mapping relationship enables the system to accurately predict rainfall time based on real-time humidity changes, providing a time benchmark for assessing the self-cleaning effect of photovoltaics.

[0035] Based on the diurnal fluctuation curve of particulate matter concentration, the peak occurrence time and duration of the fluctuation curve are extracted to calculate the effective settling time window for particulate matter. The effective settling time window is a temporal parameter for accurately estimating the dust accumulation rate, reflecting the main time period distribution of particulate matter settling. The extraction process first analyzes the diurnal variation pattern of particulate matter concentration, identifying the concentration peaks and their temporal patterns; then, combined with wind speed data, it assesses the settling conditions at different times; finally, it determines the effective deposition time window for particulate matter on the photovoltaic surface. In tropical savanna regions, settling peaks typically occur in the morning and evening, which are related to local wind direction changes and human activity patterns. This time window parameter gives the deposition prediction temporal accuracy, avoiding the averaging error caused by simple daily averages.

[0036] A dust deposition rate prediction model is generated by coupling the wind sweeping efficiency factor, mapping relationship, and effective settling time window. Coupled calculation is a key step in integrating the influence of multiple factors, unifying wind force, humidity, and particulate matter characteristics into the deposition kinetics model. The calculation process uses a weighted integral method, comprehensively considering the dynamic changes of each factor on the time axis to obtain the net deposition rate. The formula for the dust deposition rate prediction model is:

[0037] Rs=[∫(C(t)·(1-E(t))·T(t)·dt]·Ad·(1+Ds)·Wc;

[0038] Where Rs is the dust deposition rate, C(t) is the particulate matter concentration at time t, E(t) is the wind sweeping efficiency factor at time t, T(t) is the settling time window weight at time t, Ad is the dry season acceleration factor, Ds is the exponential function term of the number of consecutive rainless days, and Wc is the wet season self-cleaning coefficient.

[0039] The dry season acceleration factor, determined by an exponential function of the number of consecutive rainless days, reflects the accelerated deposition effect under prolonged drought conditions. The wet season self-cleaning coefficient, determined by the reciprocal of the product of rainfall intensity and rainfall interval, quantifies the washing effect of rainfall. This multi-factor coupled prediction model can adapt to the seasonal characteristics of tropical savanna climates and provides accurate predictions of dust impacts for photovoltaic performance evaluation.

[0040] In this embodiment of the invention, the detailed implementation steps for calculating the terrain shadow projection path at different times and generating an occlusion influence matrix containing the dynamic evolution characteristics of shadow boundaries include:

[0041] Based on topographic elevation data, a 3D topographic mesh model is constructed. The mesh nodes of the 3D topographic mesh model contain latitude and longitude coordinates and elevation values. The 3D topographic mesh model is the fundamental component of shadow simulation, transforming discrete elevation data into a continuous surface representation. The construction process employs either Triangular Irregular Network (TIN) or Digitized Grid Model (DEM) techniques, selecting an appropriate representation method based on data density and topographic features. For grassland terrain, DEM data at a resolution of 30m or higher is typically used to capture micro-topographic undulations. The model construction process includes steps such as data interpolation, outlier handling, and boundary smoothing to ensure the continuity and realism of the terrain surface. The completed 3D topographic mesh provides an accurate surface geometry basis for illumination simulation, reflecting the gentle slopes and local topographic variations unique to tropical grasslands.

[0042] Based on the geographical location of the photovoltaic array deployment coordinates, the time-varying functions of the solar altitude angle and azimuth angle throughout the year are calculated. Solar trajectory calculation is a necessary prerequisite for shadow projection, providing the spatiotemporal variation patterns of the light source position. The calculation process is based on an astronomical model, considering factors such as Earth's revolution, rotation, and axial tilt, to accurately calculate the apparent solar motion at a specific geographical location. For tropical savanna regions, the solar trajectory exhibits significant seasonal variations, with a larger solar altitude angle and longer daylight hours in summer, and vice versa in winter. The solar position calculation formula employs a standard astronomical algorithm, comprehensively considering parameters such as latitude, longitude, date, and time, outputting continuous functions of the solar altitude angle and azimuth angle, providing dynamic light source parameters for shadow projection.

[0043] In a 3D terrain mesh model, the shadow projection trajectory of each terrain protrusion node on the photovoltaic array plane is tracked using a time-varying function as a parameter. Shadow tracing is the core algorithm for shadow simulation, determining the shading effect of terrain on the photovoltaic array through ray projection technology. The tracing process first identifies all possible terrain protrusions that may cause shading, and then calculates the projection position of the light rays after being truncated by the terrain on the photovoltaic plane based on the geometric relationship between the sun, terrain, and photovoltaic array. The algorithm uses ray tracing, modeling sunlight as rays emanating from the sun's position, finding their intersection with the terrain model to obtain the shading point, and then projecting them onto the photovoltaic plane to obtain the shadow point. To improve computational efficiency, a multi-resolution shadow calculation strategy is implemented, using a low-resolution model for distant terrain and a high-resolution model for nearby terrain. This accurate shadow tracing method can capture the irregular shadow shapes generated by complex terrain and their changes over time.

[0044] The shadow projection trajectory is discretized into a set of shadow boundary points with time intervals on the order of minutes. Discretization transforms the continuous shadow process into a computer-processable discrete representation, facilitating subsequent analysis and applications. An adaptive sampling strategy is employed, increasing sampling density during periods of rapid shadow change and appropriately decreasing sampling frequency during periods of gradual change. For typical tropical savanna terrain, shadow changes are rapid at higher solar altitude angles, requiring a higher sampling rate; while shadow changes are relatively gradual in the early morning and evening, allowing for a lower sampling frequency. The discretized shadow boundary point set is typically represented by a coordinate sequence, recording the precise position of the shadow outline on the photovoltaic plane and its evolution over time, providing a discrete data foundation for the analysis of shadow dynamic characteristics.

[0045] The centroid offset distance and area change rate between adjacent shadow boundary point sets are calculated as dynamic evolution characteristics of the shadow boundary. These dynamic evolution characteristics are key indicators describing the spatiotemporal variation of shadows, reflecting the dynamic impact of terrain shading on photovoltaic systems. The calculation process first performs geometric analysis on the shadow boundary point set at each time step, extracting the centroid position and coverage area; then, it calculates the centroid displacement vector and area change percentage between adjacent time steps; finally, it constructs a velocity and acceleration model of shadow evolution based on the temporal changes of these parameters. The centroid offset reflects the overall movement trend of the shadow, while the area change rate reflects the expansion or contraction of the shadow range. These dynamic characteristics provide a quantitative basis for predicting shadow change trends and assessing the impact of shading.

[0046] Based on the dynamic evolution characteristics of shadow boundaries, an shading impact matrix is ​​constructed. This matrix is ​​the final output of terrain shading analysis, quantifying the degree of shadowing influence on each photovoltaic sub-unit at different times. The construction process first divides the photovoltaic array into sub-unit grids of appropriate granularity, typically corresponding to the physical partitions of the photovoltaic modules. Then, for each time moment, the proportion of each sub-unit covered by shadow is calculated, forming a coverage matrix. Finally, the matrices for each time moment are organized according to a time series to form a complete spatiotemporal shading impact matrix. The row index of the matrix represents the time series, the column index represents the photovoltaic array sub-unit number, and the matrix element value is the shading rate of the corresponding sub-unit, ranging from [0,1], where 0 represents no shading and 1 represents complete shading. This matrix comprehensively describes the spatiotemporal distribution impact of terrain shading on the photovoltaic system, providing accurate quantification of terrain factors for subsequent power prediction and scheduling optimization.

[0047] In this embodiment of the invention, the detailed implementation steps for establishing a thermal decay compensation model for energy storage efficiency based on the correlation between the temperature gradient field and the diurnal temperature range of the environment include:

[0048] Spatial interpolation is performed on the multi-point temperature distribution data of the energy storage unit's outer shell to generate a continuous temperature distribution field. Spatial interpolation is a key step in constructing a continuous temperature field from discrete measurement points, providing a complete temperature surface for thermal analysis. The interpolation process employs the radial basis function (RBF) method, which is well-adapted to irregular sampling point distributions and can preserve local features and gradient information of the temperature field. The implementation process first preprocesses the collected temperature data, removing outliers and performing calibration; then, appropriate kernel function parameters are determined, typically choosing a quadratic function as the radial basis function to balance smoothness and accuracy; finally, global interpolation calculations are performed to generate a high-resolution temperature distribution field. For energy storage units in typical tropical savanna environments, the temperature distribution usually exhibits a non-uniform characteristic of high temperature on the sunny side and low temperature on the shaded side, with temperature differences reaching 10-15℃. This temperature non-uniformity has a significant impact on energy storage efficiency and needs to be accurately captured.

[0049] The temperature gradient field is obtained by calculating the radial and axial temperature gradient vectors of a continuous temperature distribution field. Temperature gradient field calculation is a fundamental step in heat flow analysis, revealing the direction and intensity distribution of heat transfer. The calculation process is based on numerical differentiation methods, performing directional derivative calculations on the continuous temperature field. A central difference scheme is employed to ensure the accuracy and stability of the gradient calculation. The gradient calculation formula is as follows:

[0050]

[0051] in, Let T(x,y,z) be the temperature gradient vector, and let T(x,y,z) be the temperature value at point (x,y,z). and These are the partial derivatives of temperature in the three orthogonal directions.

[0052] For cylindrical energy storage units, a cylindrical coordinate system is typically used to represent the gradient, including radial, circumferential, and axial components, which better reflects the geometry of the energy storage unit. The temperature gradient field visually displays the heat flow dynamics inside the energy storage unit. High gradient regions usually correspond to critical areas of heat accumulation or dissipation, and are key areas of focus for thermal management.

[0053] The region with the largest gradient magnitude in the temperature gradient field is extracted and marked as the heat accumulation core region. Identifying the heat accumulation core region is a crucial step in locating hotspots and determining the area of ​​highest thermal stress within the energy storage unit. The identification process is based on gradient magnitude analysis, calculating the Euclidean norm of the temperature gradient vector to find the connected regions with the highest gradient intensity. In practice, morphological processing and threshold segmentation techniques are employed to extract regions with gradient magnitudes exceeding a set threshold (typically 1.5 times the average). Morphological closing operations are then used to connect these local regions, forming a complete heat accumulation core region. In tropical savanna environments, the heat accumulation core region of an energy storage unit typically appears in the inner region of the directly exposed surface or near high-heat-generating components in the power conversion circuit. These regions are where thermal stress is concentrated and where efficiency degradation is most pronounced, decisively impacting overall performance.

[0054] The peak and trough temperatures of the core area accumulating heat are statistically analyzed over a 24-hour period, and the peak-to-trough temperature difference is calculated as the effective diurnal temperature range (DNR) of the energy storage unit. The effective DNR is a key indicator for evaluating the thermal cycling intensity of an energy storage unit, reflecting the actual temperature stress experienced by the unit. The statistical process is based on time-series temperature data analysis, recording the highest and lowest temperatures in the core area within a 24-hour cycle, and calculating the difference as the effective DNR. Data acquisition typically uses a time resolution of 1 hour or higher to ensure accurate capture of diurnal temperature variations. In tropical savanna regions, the diurnal temperature range is usually large, with the effective DNR of energy storage units reaching 20-30°C, significantly higher than in temperate regions. This intense temperature cycling has a significant impact on energy storage lifetime and efficiency. The effective DNR is directly related to the temperature sensitivity of electrochemical performance and is a fundamental parameter for constructing thermal degradation models.

[0055] A deviation value is established between the effective diurnal temperature range and the ambient diurnal temperature range. This deviation value characterizes the thermal inertia of the energy storage unit. Thermal inertia characteristic analysis is a crucial step in understanding the thermal response characteristics of energy storage, quantifying the response delay and damping effect of the energy storage unit to changes in ambient temperature. The analysis process compares the time series of temperature changes in the energy storage core area with those in the ambient temperature range, calculating the amplitude ratio and phase difference between the two. The thermal inertia characteristic is represented by a deviation function, with the formula:

[0056] θ(t) = Ts(t) - Te(t-τ);

[0057] Where θ(t) is the temperature deviation at time t, Ts(t) is the temperature of the energy storage core area, Te(t-τ) is the ambient temperature at the lag time τ, and τ is the thermal response delay time, which is determined through cross-correlation analysis.

[0058] The magnitude and pattern of the deviation directly reflect the thermal capacity and thermal resistance characteristics of the energy storage unit. Systems with high thermal inertia exhibit small-amplitude, large-lag temperature changes, which help alleviate temperature stress but prolong the heat dissipation process; low thermal inertia, on the other hand, does the opposite. Thermal inertia characteristics provide a time-dimensional basis for thermal management strategies, guiding the timing and intensity adjustment of the cooling system's start-up and shutdown.

[0059] Based on thermal inertia characteristics and the number of charge-discharge cycles of the energy storage unit, the thermal decay coefficient of energy storage efficiency is calculated. The thermal decay coefficient is a key parameter for quantifying the impact of temperature on energy storage efficiency, establishing a quantitative relationship between temperature change and performance degradation. The calculation process combines experimental data and theoretical models, comprehensively considering the cumulative effects of temperature and cyclic fatigue. The formula for calculating the thermal decay coefficient is:

[0060] η(T,N)=η0·[1-αT·(T-Tref)-βN·f(T)·N];

[0061] Where η(T,N) is the energy storage efficiency at temperature T and number of cycles N, η0 is the baseline efficiency under standard conditions, αT is the temperature coefficient, T is the actual temperature, Tref is the reference temperature, βN is the cycle decay coefficient, f(T) is the temperature acceleration factor, and N is the number of charge-discharge cycles.

[0062] The temperature acceleration factor f(T) is described using a modified Arrhenius equation, which describes the accelerating effect of temperature increase on cycle life. For energy storage systems in tropical savanna regions, cycle degradation under high-temperature conditions is typically 2-3 times faster than under standard conditions, making thermal management a critical factor in maintaining efficiency. The thermal degradation coefficient directly affects the energy conversion efficiency and lifespan of the energy storage system and is a core reference value for thermal management decisions.

[0063] Based on the thermal decay coefficient, a thermal decay compensation model is constructed. This model serves as a decision-making tool for optimizing energy storage operation strategies, proactively adjusting operating parameters to maintain efficiency by predicting temperature effects. The model's construction employs a combination of data-driven and physical modeling methods, integrating the thermal decay coefficient with temperature prediction algorithms to form an adaptive compensation mechanism. The model incorporates two core compensation strategies: reducing the upper limit of charging power during high-temperature periods to avoid additional heat generation and temperature increases; and extending the discharge preheating time during low-temperature periods to ensure the electrochemical system reaches its optimal operating temperature. The compensation strategy parameters are determined using a nonlinear optimization algorithm, with the objective function balancing efficiency improvement and energy availability. This intelligent compensation mechanism enables the energy storage system to adapt to the extreme temperature differences of tropical savanna environments, maintaining stable energy conversion efficiency and extending system lifespan, making it a key optimization method for energy storage systems in tropical environments.

[0064] In this embodiment of the invention, the detailed implementation steps for constructing a three-dimensional modified surface for usable photovoltaic power by combining a dust deposition rate prediction model, a shading influence matrix, and a thermal attenuation compensation model include:

[0065] The cumulative dust thickness sequence of each photovoltaic sub-unit was extracted from the dust deposition rate prediction model, and the corresponding light transmittance attenuation curve was calculated. Transmittance calculation is a crucial step in quantifying the impact of dust, transforming physical deposition into an effect on optical performance. The calculation process is based on a modified model of the Beer-Lambert law, considering the scattering and absorption characteristics of dust particles. Dust particles in tropical savanna regions typically contain a high amount of organic matter and mineral mixtures, and their optical properties differ significantly from desert dust. An empirical fitting formula was used for transmittance calculation:

[0066] T(d) = T0·e (-k·d) ;

[0067] Where T(d) is the transmittance when the dust thickness is d, T0 is the baseline transmittance of the clean surface (usually 1), k is the attenuation coefficient, and d is the cumulative dust thickness.

[0068] The attenuation coefficient k was determined through on-site calibration, taking into account the particle size distribution and composition characteristics of local dust. The transmittance attenuation curve directly reflects the attenuation effect of dust accumulation on incident light, which is the direct cause of power loss and provides an optical attenuation factor for subsequent power prediction.

[0069] The shading rate values ​​of each photovoltaic sub-unit at each time point are extracted from the shading impact matrix and converted into the effective light-receiving area ratio. Shading conversion is a key step in quantifying the impact of terrain, transforming shadow coverage into an effective area ratio. The conversion process considers the photovoltaic response characteristics under different shading conditions, especially the mechanism of the bypass diode. For typical photovoltaic modules, the impact mechanisms of complete shading and partial shading are different and need to be treated separately. The formula for calculating the effective light-receiving area ratio is:

[0070] A(s) = 1 - γ·s;

[0071] Where A(s) is the proportion of the effective light-receiving area when the occlusion rate is s, γ is the occlusion sensitivity coefficient, and s is the occlusion rate.

[0072] The shading sensitivity coefficient γ is typically greater than 1, reflecting the nonlinear impact of shading on photovoltaic performance. This nonlinearity stems from the electrical connection characteristics and hotspot effects within the photovoltaic module. The effective light-receiving area ratio directly affects the power generation capacity of photovoltaics and is a spatial modulation factor for power calculation.

[0073] The overall solar availability coefficient for each photovoltaic sub-unit is obtained by multiplying the light transmittance attenuation curve with the effective light-receiving area ratio. Calculating the availability coefficient is a core step in the comprehensive assessment of optical effects, unifying the attenuation effects from different dimensions into a single index. The calculation process employs an element-wise multiplication method, considering both the independent and combined effects of dust and shading. The formula for calculating the overall solar availability coefficient is:

[0074] C(t,i)=T(d(t,i))·A(s(t,i));

[0075] Where C(t,i) is the comprehensive illuminance availability coefficient of photovoltaic sub-unit i at time t, T(d(t,i)) is the corresponding transmittance, and A(s(t,i)) is the corresponding effective light-receiving area ratio.

[0076] The comprehensive coefficient ranges from [0,1], where 0 indicates completely unusable and 1 indicates completely usable. This coefficient comprehensively reflects the combined impact of environmental factors on photovoltaic power reception and is the fundamental modulation factor for calculating power generation.

[0077] Based on the comprehensive solar availability factor and the rated photovoltaic power, the real-time theoretical output power of each photovoltaic sub-unit is calculated. Power calculation is a fundamental step in performance evaluation, transforming environmental impact factors into actual power generation predictions. The calculation process is based on the standard photovoltaic output model, adjusted for real-time irradiance and temperature coefficient. The formula for calculating the real-time theoretical output power is:

[0078]

[0079] Where P(t,i) is the theoretical output power of photovoltaic sub-unit i at time t, Pstc is the rated power under standard test conditions, C(t,i) is the comprehensive irradiance availability factor, I(t) is the real-time irradiance intensity, Istc is the standard irradiance intensity, α is the temperature coefficient, Tc(t) is the battery temperature, and Tstc is the standard test temperature.

[0080] This calculation model comprehensively considers environmental degradation, light intensity, and temperature effects, and can accurately predict the actual power generation performance of photovoltaics in tropical savanna environments, providing a reliable basis for energy planning.

[0081] An energy transfer link model is established to correlate photovoltaic (PV) power generation with energy storage efficiency, and charge / discharge efficiency correction coefficients are extracted from a thermal degradation compensation model. Establishing this link model is a crucial step in system integration, connecting independent subsystems into an energy flow network. The modeling process is based on energy flow graph analysis, tracing the entire energy transfer path and efficiency nodes from PV power generation to energy storage conversion. The model structure includes PV power generation units, power conversion units, energy storage charge / discharge units, and inverter output units, with each node having corresponding efficiency parameters and environmental response characteristics. The efficiency correction coefficients extracted from the thermal degradation compensation model are directly applied to the energy storage nodes to adjust their energy conversion efficiency under different temperature conditions. This link model expands system analysis from isolated components to overall performance evaluation, providing a unified framework for system-wide optimization.

[0082] By substituting the real-time theoretical output power into the energy transfer link model and incorporating charge / discharge efficiency correction coefficients, the actual usable power after passing through the energy storage stage is calculated. Usable power calculation is a crucial step in the final performance evaluation, converting theoretical power generation into practically usable energy. The calculation process is based on the energy balance principle, considering efficiency losses and temperature effects during the energy storage conversion process. The actual usable power calculation needs to distinguish between direct power supply and energy storage release paths, and consider their dynamic allocation at different times. For the energy storage path, the overall efficiency of the charge / discharge cycle is significantly affected by temperature; adjustments are made using efficiency correction coefficients to ensure prediction accuracy. This end-to-end power calculation method overcomes the limitations of traditional photovoltaic systems that only focus on the generation side, achieving a comprehensive evaluation from solar energy harvesting to end-user availability, providing a holistic perspective for system design and operational optimization.

[0083] A coordinate system was constructed with three dimensions: time, spatial location of the photovoltaic array, and comprehensive environmental impact factors. The comprehensive environmental impact factors were obtained through a weighted combination of dry season acceleration factors, wet season self-cleaning coefficients, and thermal inertia characteristics. Coordinate system construction is a fundamental step in 3D visualization, providing an intuitive framework for representing complex data. The construction process first defines three orthogonal axes: the time axis uses hours or days as units, the spatial location axis corresponds to the physical layout coordinates of the photovoltaic array, and the environmental impact factor axis represents a comprehensive representation of multiple environmental parameters. The comprehensive environmental impact factors were calculated through a weighted combination.

[0084] E = w1·Ad + w2·Wc + w3·θ;

[0085] Where E is the comprehensive environmental impact factor, Ad is the dry season acceleration factor, Wc is the wet season self-cleaning coefficient, θ is the thermal inertia characteristic parameter, and w1, w2 and w3 are weighting coefficients, and w1+w2+w3=1.

[0086] The weighting coefficients are dynamically adjusted based on the current season and system characteristics, reflecting the relative importance of different environmental factors under specific conditions. This three-dimensional coordinate system unifies time, space, and environmental variables into a single framework, providing a mathematical foundation for a comprehensive expression of the performance of complex systems.

[0087] The actual available power is mapped to grid nodes in the coordinate system, forming a discrete power distribution point set. Power mapping is a key step in data preparation, organizing the calculation results into an interpolable discrete point set. The mapping process employs a regular sampling strategy, defining grid points of appropriate density in three-dimensional space and calculating the power value corresponding to each grid point. The sampling density needs to balance computational complexity and interpolation accuracy, typically using hourly intervals on the time axis, component-level partitioning on the spatial axis, and uniform segmentation on the environmental factor axis. For grid points not directly calculated under specific environmental conditions, a weighted average of nearest neighbors is used for estimation. The resulting discrete point set contains performance data of the system under various conditions, providing raw data support for surface construction.

[0088] Cubic spline surface interpolation is performed on a discrete power distribution point set to generate a three-dimensional corrected surface. Surface interpolation is the final step in the continuous representation, transforming discrete data points into a smooth and continuous mathematical surface. The interpolation process employs cubic B-spline technology, which maintains the continuity and smoothness of the physical model while accurately fitting the sampled data. The interpolation algorithm constructs basis functions in three dimensions, then forms a three-dimensional interpolation function through tensor product to calculate the power value at any point. Natural boundary constraints are used to ensure that the surface has reasonable extensibility at the data range edges. The final generated three-dimensional corrected surface is a continuous function that can take any time point, spatial location, and environmental conditions as input and output the corresponding predicted available power value. This surface comprehensively describes the performance characteristics of the system in a tropical savanna environment and serves as the fundamental mathematical model for energy scheduling optimization.

[0089] In this embodiment of the invention, the detailed implementation steps for generating multi-timescale energy dispatch strategies, including hourly, daily, and weekly levels, based on the seasonal decomposition data of the three-dimensional modified surface and historical load demand include:

[0090] Seasonal decomposition is performed on historical load demand data to obtain trend, seasonal, and random components. Seasonal decomposition is a fundamental step in load analysis, breaking down the time series into components with different characteristics. The decomposition process employs the STL method, which effectively handles nonlinear trends and multi-period seasonality. The decomposition formula is:

[0091] Y(t) = T(t) + S(t) + R(t);

[0092] Where Y(t) is the original load sequence, T(t) is the trend component, representing the long-term trend; S(t) is the seasonal component, representing the periodic change pattern; and R(t) is the random component, representing irregular fluctuations.

[0093] The decomposition process first determines the appropriate seasonal cycle length (typically a natural cycle such as 24 hours, 7 days, or 365 days), and then extracts each component through iterative weighted regression. For tropical savanna regions, load typically exhibits significant seasonal differences, with distinct electricity consumption patterns between the dry and wet seasons. This decomposition method effectively captures these characteristics, providing a basis for time-of-use load forecasting.

[0094] The seasonal components are divided into two sub-cycles based on the dry and wet seasons, and the daily average and variance of load demand are calculated for each sub-cycle. Cycle division is a crucial step in adapting to seasonal differences and developing differentiated strategies for different climatic conditions. The division process determines the dry and wet season time ranges based on meteorological data, and then extracts the corresponding seasonal load component data. For each sub-cycle, statistical characteristics are calculated:

[0095]

[0096] Where μs is the daily average load of sub-period s, σs 2 Let be the load variance of sub-period s, Ns be the number of days contained in the sub-period, and L(t) be the load value at time t within the sub-period.

[0097] Statistical characteristics reflect the average load level and fluctuations in different seasons; the wet season typically has a higher average load and lower fluctuations, while the dry season is the opposite. This seasonal variation analysis provides time-segmented load forecasts for scheduling strategies and forms the basis for rational resource allocation.

[0098] Based on a 3D modified surface, hourly photovoltaic (PV) available power prediction sequences are extracted. Power prediction is a core step in resource assessment, transforming environmental models into expected available energy. The prediction process uses future weather forecasts and system states as inputs, calculating the corresponding power output through the 3D modified surface. The calculation employs the path integral method, considering the temporal evolution of environmental parameters to obtain a more accurate prediction sequence. The prediction results are organized as hourly time series, visually demonstrating the time-varying characteristics of future power generation capacity and providing resource-side data for energy balance planning. For tropical savanna regions, the dry season typically has higher solar radiation intensity and lower cloud cover variability, resulting in higher prediction accuracy; however, the wet season is affected by cloud cover and rainfall, increasing prediction uncertainty and requiring more frequent real-time corrections.

[0099] The calculation of the photovoltaic available power forecast sequence and the hourly power deficit distribution of load demand is performed. In the power deficit distribution, a positive deficit represents energy storage discharge demand, while a negative deficit represents energy storage charging capacity. Deficit calculation is a fundamental step in energy balance planning, quantifying the supply-demand difference and energy storage demand. The calculation process uses the direct interpolation method, subtracting available power from load demand at the corresponding time point to obtain the deficit sequence. The formula for calculating the deficit distribution is:

[0100] D(t) = L(t) - P(t);

[0101] Where D(t) is the power deficit at time t, L(t) is the load demand at the corresponding time, and P(t) is the available photovoltaic power at the corresponding time.

[0102] A positive deficit indicates insufficient energy, which needs to be replenished from energy storage; a negative deficit indicates excess energy, which can be added to energy storage reserves. The deficit distribution sequence intuitively reflects the energy imbalance state of the system in the time dimension, providing a basis for energy storage scheduling decisions and is a prerequisite for achieving peak shaving and valley filling.

[0103] Based on the power deficit distribution, the system divides the energy storage charging priority period and the discharging reserved period on a daily time scale. The charging priority period is selected from the period with the largest continuous negative deficit. Time period division is the core step of daily-scale scheduling, allocating time within a day to energy operations of different priorities. The division process uses a dynamic programming algorithm to find the time period segmentation scheme with the maximum energy gain. In specific implementation, firstly, the deficit sequence of 24 hours is accumulated and summed to find the continuous period with the largest cumulative negative deficit as the charging priority zone; then, based on the energy storage capacity constraint, the distribution and capacity allocation of the discharging reserved period are determined. The charging priority period usually corresponds to noon when sunlight is strongest, while the discharging reserved period corresponds to the peak electricity consumption in the evening and early morning. This time period division method based on energy balance enables the system to optimize energy utilization efficiency on a daily scale, reducing waste and deficits.

[0104] On a weekly timescale, the energy storage capacity reservation ratio for each day of the week is adjusted based on the predicted weekly shading trend using the shading impact matrix. Weekly adjustment is a crucial step in long-term planning, addressing cross-day energy allocation and adaptation to abnormal weather. The adjustment process is based on shading trend analysis, predicting the changing patterns of resource availability over the next week. The trend coefficient is calculated using linear regression; a positive value indicates gradually worsening shading, requiring advance reserves; a negative value indicates gradually lessening shading, allowing for appropriate reduction in reserves. The adjustment of the capacity reservation ratio is proportional to the trend coefficient, ensuring stable power supply under resource fluctuations. This proactive weekly adjustment enables the system to cope with prolonged rainy weather or seasonal transitions common in tropical savanna regions, improving the system's environmental adaptability and reliability.

[0105] By coupling hourly power deficit distribution, daily charging / discharging period division, and weekly capacity reservation ratio at multiple scales, a multi-timescale energy dispatch strategy is generated. Multi-scale coupling is the final step in dispatch integration, unifying planning across different time scales into a single coordination framework. The coupling process employs a hierarchical constraint propagation mechanism, where decisions at higher time scales serve as constraints at lower time scales, ensuring that local optimization conforms to the global objective. In implementation, firstly, the upper limit of daily available energy storage capacity is determined based on the weekly capacity reservation ratio; then, under this constraint, the specific allocation of daily charging / discharging periods is optimized; finally, it is refined into hourly real-time power control commands, forming a complete dispatch strategy. This multi-timescale coordination mechanism enables the system to simultaneously cope with short-term fluctuations and long-term changes, achieving full-spectrum energy management from real-time response to seasonal adaptation—a core strategy for achieving energy self-balancing in tropical savanna environments.

[0106] In this embodiment of the invention, the detailed implementation steps for extracting the micro-fluctuation characteristics of the photovoltaic array output by high-frequency sampling of the inverter's DC-side current ripple and identifying the abnormal current drop mode associated with the dust shading gradient in the micro-fluctuation characteristics include:

[0107] The inverter's DC-side current is sampled at a kilohertz frequency to obtain a high-frequency current sampling sequence. High-frequency sampling is a fundamental step in signal feature extraction, acquiring raw data containing information about microscopic fluctuations. A high-precision data acquisition system is used during the sampling process, typically employing a 16-bit or higher resolution analog-to-digital converter to ensure the capture of minute current changes. The sampling frequency is set in the 1-10kHz range, which is sufficient to capture the main current harmonic components and transient characteristics of the photovoltaic system. For high-capacity systems, a distributed acquisition architecture is used, with synchronous sampling at multiple key nodes, followed by timestamp alignment and merging. The sampled data undergoes preprocessing to remove noise and interference, forming a high-quality current time-domain signal sequence, providing accurate raw data for subsequent analysis.

[0108] Wavelet decomposition is performed on the high-frequency current sampling sequence to separate the high-frequency ripple component and the low-frequency fundamental component. Wavelet decomposition is a key step in signal processing, decomposing the composite signal into components with different frequency characteristics. The decomposition process employs multi-resolution discrete wavelet transform (DWT) with the Daubechies wavelet basis function, which has good time-frequency localization characteristics. The number of decomposition levels is typically set to 5-7 levels, which can effectively separate the kHz-level ripple component and the fundamental operating frequency component. The high-frequency detail coefficients obtained after decomposition correspond to the ripple characteristics, while the low-frequency approximation coefficients correspond to the fundamental characteristics. The high-frequency ripple component contains minute fluctuations caused by factors such as dust distribution, local shading, and component mismatch, and is a key signal source for dust detection; the low-frequency fundamental component reflects the overall power generation level and large-scale changes, and is used to determine the system's operating state.

[0109] The envelope of the high-frequency ripple component is calculated, and local minima are extracted. Envelope analysis is a crucial step in ripple feature enhancement, highlighting the changing trend of ripple intensity. The calculation process employs the Hilbert transform method, extracting the instantaneous amplitude of the ripple by constructing an analytic signal to form the envelope. The calculation formula is as follows:

[0110] env(t) = |x(t) + j·H[x(t)]|;

[0111] Where env(t) is the envelope value at time t, x(t) is the high-frequency ripple signal, H[x(t)] is the Hilbert transform of x(t), and |·| represents the modulus of the complex number.

[0112] The envelope smoothly connects the peak points of the ripple, visually demonstrating the time-varying characteristics of the ripple intensity. Local minima are identified through first-derivative zero-point detection; these points typically correspond to sudden drops in ripple intensity and serve as indicators of dust occlusion or local shadows. The distribution pattern and depth features of the minima contain spatial information about dust distribution, forming the foundational data for constructing the ripple fingerprint.

[0113] By statistically analyzing the occurrence intervals and amplitude distributions of local minima, a ripple fingerprint of the current ripple is constructed. This ripple fingerprint construction is the core step in feature extraction, transforming the time-domain signal into a recognizable feature vector. The construction process first performs statistical analysis on the minima sequence, calculating the time interval and amplitude distributions between adjacent points; then, it extracts key statistical features of these distributions, such as mean, variance, skewness, and kurtosis; finally, these features are combined into a high-dimensional feature vector, forming the characteristic fingerprint of the ripple fluctuation. Different types of dust distribution patterns produce different ripple fingerprints. Uniform dust layers typically exhibit low-frequency, low-amplitude fluctuations, while dust gradients produce high-frequency, large-amplitude, irregular fluctuations. The ripple fingerprint is a high-dimensional feature representation for dust state identification, providing a feature basis for subsequent pattern matching.

[0114] A mapping database was established between wave fingerprints and the local shading states of photovoltaic arrays. This database contains wave fingerprint feature vectors corresponding to different shading gradients. Database construction forms the knowledge foundation for pattern recognition, associating known patterns with feature representations. The construction process combines experimental data and physical models, collecting wave fingerprint samples under various known shading conditions and establishing a feature-state lookup table. The database employs a hierarchical index structure, organized by shading type, degree, and distribution characteristics, facilitating rapid retrieval and matching. For tropical savanna environments, the database pays special attention to specific patterns such as dust gradients and humidity stratification, which frequently occur in environments with alternating dry and wet seasons and exhibit specific wave fingerprint characteristics. The mapping database not only stores feature vectors but also includes corresponding physical interpretations and severity ratings, providing a comprehensive reference for subsequent decision-making.

[0115] The real-time acquired fluctuation fingerprints are matched with feature vectors in the mapping database for similarity. Similarity matching is the core operation of pattern recognition, comparing unknown samples with known patterns. The matching process employs a multi-metric fusion similarity calculation method, comprehensively considering multiple metrics such as Euclidean distance, cosine similarity, and Mahalanobis distance. The fusion similarity calculation formula is as follows:

[0116] S(F1,F2)=

[0117] w1·(1-d1(F1,F2))+w2·cos(F1,F2)+w3·(1-dm(F1,F2));

[0118] Where S(F1,F2) is the fusion similarity of feature vectors F1 and F2, d1 is the normalized Euclidean distance, cos is the cosine similarity, dm is the normalized Mahalanobis distance, and w1, w2 and w3 are weight coefficients, satisfying w1+w2+w3=1.

[0119] The similarity value ranges from [0,1], with higher values ​​indicating a higher degree of matching. The matching process employs a k-nearest neighbor search strategy to find the k patterns in the database most similar to the current fluctuating fingerprint, and then determines the most likely occlusion state and degree through weighted voting. This multi-method fusion matching strategy improves the robustness of recognition and can adapt to changing patterns in complex environments.

[0120] When the matching similarity exceeds a preset similarity threshold and the corresponding occlusion gradient exhibits a monotonically increasing trend, it is identified as an abnormal current decline pattern. The judgment criterion is the decision rule for anomaly identification, transforming the matching result into a clear state judgment. The judgment process comprehensively considers two dimensions: static similarity and dynamic trend. Static similarity reflects the degree of matching in the current state, and a threshold of 0.75-0.85 is typically set to ensure recognition reliability. The dynamic trend is judged by the gradient direction within a continuous time window; a monotonically increasing trend indicates that dust accumulation is intensifying and requires attention. When both conditions are met simultaneously, the system identifies it as an abnormal current decline pattern, triggering subsequent prediction model correction and maintenance decisions. This judgment method, combining static features and dynamic trends, greatly improves the accuracy of anomaly detection and early warning capabilities, creating conditions for proactive maintenance.

[0121] In this embodiment of the invention, the detailed implementation steps for dynamically correcting the deposition rate parameters in the dust deposition rate prediction model based on the frequency and amplitude changes of the abnormal current drop pattern include:

[0122] The cumulative frequency of abnormal current decline patterns within a 24-hour sliding time window was statistically analyzed. Frequency statistics are a fundamental step in anomaly assessment, quantifying the temporal distribution density of anomalous events. The statistical process employs a sliding window technique, maintaining a fixed-length time window and updating its content over time to continuously monitor data from the most recent 24 hours. Frequency calculation uses an event counting method, recording the total number of times the anomalous pattern was identified within the window, serving as a preliminary indicator of anomaly severity. In tropical savanna environments, frequency data typically exhibits clear diurnal and seasonal variations, highly correlated with environmental conditions. High frequencies usually indicate that dust accumulation has entered an accelerated phase or reached a critical threshold, requiring systemic intervention; low frequencies indicate that the dust impact is still within a controllable range. Frequency data provides temporal anomaly characteristics for subsequent assessments.

[0123] Calculate the mean current drop amplitude corresponding to the abnormal current drop pattern. Amplitude analysis is a key step in severity assessment, quantifying the intensity of the abnormal event. The calculation process first extracts the absolute value or relative percentage of the current drop in each abnormal event, and then calculates the arithmetic mean of the amplitudes of all events within the sliding window to obtain the average impact intensity. The formula for calculating the mean amplitude is:

[0124]

[0125] Where A is the average current drop, n is the number of abnormal events within the window, Ib is the baseline current value before the abnormality, and Ia is the lowest current value during the abnormality process.

[0126] The mean amplitude directly reflects the actual impact of dust obstruction on power generation performance. A larger value indicates a thicker or more unevenly distributed dust layer, resulting in a more severe degrade of system performance. Amplitude data provides anomalies in the intensity dimension for subsequent evaluation.

[0127] A dust accumulation assessment index is established based on the average of the cumulative frequency and the magnitude of current decrease. The construction of the assessment index is a crucial step in integrating multi-dimensional features, fusing frequency and magnitude information into a single evaluation indicator. The construction process employs a weighted product model, considering the combined influence of event frequency and severity. The formula for calculating the dust accumulation assessment index is as follows:

[0128]

[0129] Where I is the evaluation index, f is the cumulative frequency, fmax is the historical maximum frequency, A is the mean amplitude, Amax is the historical maximum amplitude, and α and β are weighting indices, usually set between 0.5 and 0.7, to adjust the relative importance of frequency and amplitude.

[0130] The evaluation index ranges from [0,1], with higher values ​​indicating more severe dust accumulation. This fusion index fully utilizes information from both time and intensity dimensions, providing a comprehensive assessment of the dust status and offering a quantitative basis for model correction.

[0131] A deviation analysis was performed between the dust accumulation assessment index and the predicted values ​​of the dust deposition rate prediction model to obtain the deviation. Deviation analysis is a core step in model evaluation, quantifying the degree of difference between predictions and actual observations. The analysis process is based on a comparison of the assessment index and the prediction model, calculating the relative deviation between the two on the same time scale. The deviation calculation formula is:

[0132]

[0133] Where δ is the relative deviation value, I is the actual evaluation index, and Ip is the cumulative index predicted by the model.

[0134] Bias values ​​can be positive or negative. Positive values ​​indicate that actual accumulation is faster than predicted, meaning the model underestimates the deposition rate; negative values ​​indicate that actual accumulation is slower than predicted, meaning the model overestimates the deposition rate. Bias analysis directly reflects the time-varying characteristics of model accuracy, providing directional guidance for parameter adjustment.

[0135] When the absolute value of the deviation exceeds a preset deviation threshold, the correction increment for the deposition rate parameter is calculated, and this correction increment is proportional to the deviation value. The calculation of the correction increment is a crucial step in parameter adjustment, determining the direction and magnitude of the model correction. The calculation process employs a proportional control strategy, ensuring the correction magnitude is proportional to the deviation size, thus guaranteeing the appropriateness and stability of the adjustment. The increment calculation formula is:

[0136] ΔR = γ·R·δ;

[0137] Where ΔR is the correction increment of the deposition rate parameter, γ is the correction coefficient (usually set between 0.1 and 0.3), R is the current deposition rate parameter value, and δ is the relative deviation.

[0138] The correction increment directly reflects the amount of parameter change that needs adjustment. Positive deviations result in positive increments, increasing the deposition rate; negative deviations result in negative increments, decreasing the deposition rate. The increment calculation uses proportional control rather than a fixed step size, ensuring the appropriateness of the adjustment ratio and avoiding oscillation problems caused by over-correction.

[0139] The correction increments are superimposed on the current deposition rate parameters of the dust deposition rate prediction model to complete dynamic parameter correction. The cumulative dust thickness sequence for each photovoltaic sub-unit is then recalculated to update the light transmittance attenuation curve, and subsequently, the 3D correction surface is updated. Parameter updating is the final step in model optimization, applying the correction results to the prediction model. The update process uses an incremental parameter approach, directly superimposing the calculated correction increments onto the current parameter values ​​to form new parameter settings. The update formula is:

[0140] Rnew = R + ΔR;

[0141] Where Rnew is the updated deposition rate parameter, R is the current parameter value, and ΔR is the correction increment.

[0142] After the parameters are updated, the system immediately recalculates the cumulative dust thickness sequence and extrapolates the future dust accumulation trend based on the new deposition rate parameters. Then, it updates the light transmittance attenuation curve to reflect the changes in the impact of dust on optical performance. Finally, it updates the three-dimensional correction surface and adjusts the power prediction model. This end-to-end update mechanism ensures that changes to a single parameter can be propagated to the entire system model, maintaining the consistency and accuracy of all parts of the model.

[0143] The system regenerates the predicted photovoltaic (PV) available power based on the updated 3D corrected surface and, combined with the adjustment time points of the PV cleaning cycle, corrects the charging and discharging period division and capacity reservation ratio in the multi-timescale energy dispatch strategy, resulting in the corrected multi-timescale energy dispatch strategy. Strategy correction is the final step in the system response, transforming model updates into operational adjustments. The correction process, based on the updated power prediction, reassesses future energy availability and adjusts the energy dispatch plan. For the PV cleaning cycle, the system adaptively adjusts the cleaning time points according to the changing trend of dust accumulation rate, accelerating or delaying the maintenance plan. For the energy dispatch strategy, the system re-optimizes the charging and discharging period division and capacity reservation ratio based on the corrected power curve to ensure energy supply and demand balance. This closed-loop feedback dispatch optimization mechanism enables the system to continuously learn and adapt to environmental changes, maintaining optimal operating conditions. It is the core technological guarantee for achieving reliable off-grid power supply in tropical savanna environments.

[0144] In this embodiment of the invention, the ratio of the tangential wind speed component to a preset critical wind speed is calculated as the wind-driven sweeping efficiency factor, and the method further includes:

[0145] Acquiring the tilt angle data of photovoltaic (PV) modules is a prerequisite for wind impact assessment, as it determines the spatial attitude of the PV surface. This acquisition is typically achieved through recording installation parameters or on-site measurements, recording the tilt angle of the PV modules relative to the horizontal plane. For fixed installation systems, the tilt angle is a static parameter; for tracking systems, dynamic tilt angle changes over different time periods need to be recorded. In tropical savanna regions, PV modules are usually installed at a smaller tilt angle (10-15°) to optimize the annual power generation distribution. Tilt angle data directly affects the effective direction and intensity of wind action and is a fundamental parameter for assessing wind sweeping effects.

[0146] The effective component of wind speed on the photovoltaic surface is calculated based on the angle between the surface tilt angle and the wind speed vector. This calculation of the effective component is a refinement step in wind decomposition, considering the three-dimensional spatial relationship between wind direction and the surface normal. The calculation process is based on the principle of vector projection, projecting the wind speed vector onto the tangent plane of the photovoltaic surface to obtain the effective cleaning component. The calculation needs to consider the relative relationship between the wind direction and the photovoltaic orientation; cleaning efficiency is highest when the orientations are aligned and lowest when they are perpendicular. This three-dimensional wind decomposition method more accurately reflects the actual wind field's effect on the photovoltaic surface than simple two-dimensional decomposition, providing a more precise physical basis for subsequent cleaning efficiency evaluation.

[0147] A relationship function between the effective action component and the particle size was established, characterizing the critical wind speed required to clear particles of different sizes. Establishing this function is the core step in determining the critical wind speed, quantifying the force characteristics of particles of different sizes. The process is based on fluid mechanics principles and experimental data, analyzing the minimum velocity required for wind to overcome particle adhesion. For spherical particles, the relationship between the critical wind speed and particle size is approximately:

[0148]

[0149] Where Vc(d) is the critical wind speed of particles with a diameter of d, k is the correction factor, ρp is the particle density, g is the gravitational acceleration, Cf is the friction coefficient, ρa is the air density, and Cd is the drag coefficient.

[0150] This relationship function reflects the impact of particle size on cleaning difficulty. Small particles, dominated by surface forces, are relatively more difficult to remove by wind; while large particles are mainly affected by gravity and require higher wind speeds to overcome. This particle size-dependent critical wind speed model allows the evaluation of cleaning efficiency to take into account the influence of dust composition, improving the physical accuracy of the model.

[0151] Particle size distribution analysis is performed on particulate matter concentration data to obtain the proportion of particulate matter in each size range. Distribution analysis is a crucial step in dust characteristic assessment, revealing the size composition of environmental particulate matter. The analysis process is based on air quality monitoring data, using light scattering or gravity sedimentation methods to measure particulate matter concentrations across different size ranges. Analysis typically divides particulate matter into multiple size intervals (e.g., PM1.0, PM2.5, PM10), calculating the percentage of particulate matter in each interval within the total concentration. Particulate matter in tropical savanna regions often exhibits a bimodal distribution, with the finer-sized peak corresponding to long-distance transported minerals and combustion products, and the coarser-sized peak corresponding to local dust and biogenic particles. Particle size distribution data directly affects the weighted calculation of critical wind speeds and is a key input for considering differences in dust characteristics.

[0152] Based on the proportion of particulate matter in each particle size range and its corresponding critical wind speed, a weighted average critical wind speed is calculated as the preset critical wind speed. Weighted averaging is the final step in the comprehensive evaluation, integrating the influence of different particle sizes proportionally. The calculation process uses a weighted averaging method, with the critical wind speed for each particle size weighted according to its corresponding particulate matter proportion. The formula for calculating the weighted average critical wind speed is:

[0153] Vcrit=Σ(P i ·Vc(d i ));

[0154] Where Vcrit is the weighted average critical wind speed, and P i Vc(d) represents the proportion of particles in particle size range i. i ) represents the critical wind speed for the corresponding interval.

[0155] The weighted average comprehensively reflects the overall sweeping characteristics of local dust composition and serves as a benchmark parameter for evaluating the efficiency of wind-driven sweeping. This parameter dynamically adjusts with changing environmental conditions, typically being higher during the dry season and lower during the wet season, reflecting the impact of seasonal variations in dust composition.

[0156] The ratio of the real-time tangential wind speed component to the weighted average critical wind speed is calculated. When the ratio is greater than 1, the wind-driven cleaning efficiency factor increases linearly; when the ratio is less than 1, the wind-driven cleaning efficiency factor decreases exponentially. Efficiency factor calculation is the final step in quantifying cleaning capacity, establishing a non-linear relationship between wind speed and cleaning efficiency. The calculation uses a piecewise function model to reflect the efficiency variation characteristics across different wind speed ranges. The efficiency factor calculation formula is:

[0157]

[0158] Where E(v) is the wind sweeping efficiency factor, v is the tangential wind speed component, Vcrit is the weighted average critical wind speed, and α, β and γ are fitting parameters, which are usually determined by experimental data.

[0159] This piecewise nonlinear model accurately reflects the physical characteristics of wind scavenging. When the wind speed exceeds a critical value, the scavenging capacity increases linearly with the wind speed; when the wind speed is below the critical value, the scavenging capacity decreases sharply, exhibiting an exponential decay trend. The wind scavenging efficiency factor provides a dynamically adjustable parameter for calculating the deposition rate, enabling the model to respond to real-time wind condition changes and improving prediction accuracy.

[0160] In this embodiment of the invention, the detailed implementation steps for adjusting the energy storage capacity reservation ratio for each day of the week based on the predicted trend of shading changes within the week using the shading impact matrix on a weekly time scale include:

[0161] Shading rate data for seven consecutive days is extracted from the shading impact matrix to construct a weekly shading rate change curve. Data extraction is a fundamental step in trend analysis, obtaining weekly-scale shading time-series data. The extraction process selects a seven-day time segment from the shading impact matrix, calculates the average shading rate or cumulative shading amount for each day, forming a daily-scale shading sequence. Calculation methods typically use daily averages or weighted averages, considering the actual impact of shading on power generation. For each photovoltaic sub-unit, the shading curve can be extracted individually, and then area-weighted to obtain the comprehensive shading curve for the entire system. This weekly-scale shading data intuitively reflects the changes in the impact of terrain shadows on the system over the next week, providing fundamental data for trend analysis.

[0162] Curve fitting was performed on the weekly occlusion rate change curve to obtain the trend coefficient of the occlusion rate change. Trend fitting is a key step in extracting change characteristics, transforming discrete data into a continuous function expression. The fitting process uses a linear regression method, with time as the independent variable and occlusion rate as the dependent variable, to calculate the best-fit line. The trend coefficient is the slope of the fitted line, directly reflecting the rate and direction of change of the occlusion rate. The formula for calculating the trend coefficient is:

[0163]

[0164] Where k is the trend coefficient, n is the number of data points (usually 7), and t _ Let r be the time sequence number of day i. _ This represents the corresponding occlusion rate value.

[0165] The sign of the trend coefficient reflects the direction of change; a positive value indicates that shading is gradually worsening, while a negative value indicates that shading is gradually lessening. The absolute value of the coefficient reflects the rate of change; the larger the value, the more drastic the change. Trend analysis provides predictive information on future resource availability for scheduling strategies and is a key basis for forward-looking decision-making.

[0166] When the trend coefficient is positive, it indicates that shading is gradually worsening, and the proportion of reserved energy storage capacity should be increased in the later part of the week. Adjusting the reservation strategy is a key measure to address resource reduction, mitigating future shortages by storing energy in advance. The adjustment process is based on trend prediction. When an worsening shading trend is detected, the system proactively increases energy reserves in the early part of the week to prepare for resource shortages in the later part. In practice, this typically involves increasing the priority of energy storage charging and increasing the proportion of reserved energy storage capacity 3-4 days before the week begins, ensuring stable power supply even in the later days when shading worsens. This proactive energy management strategy allows the system to respond to foreseeable resource changes in advance, avoiding the risk of power shortages caused by reactive responses.

[0167] When the trend coefficient is negative, it indicates that shading is gradually decreasing, and the energy storage capacity reservation ratio is reduced in the later part of the week. Relaxing the reservation is a reasonable strategy to cope with increased resources and avoid excessive storage leading to energy waste. The adjustment process is based on trend prediction. When a decreasing shading trend is detected, the system appropriately reduces the energy storage requirements in the early part of the week, releasing more energy for immediate power supply. In practice, the priority of energy storage charging is typically reduced 3-4 days in the early part of the week, decreasing the energy storage capacity reservation ratio and improving the immediate utilization rate of energy. This flexible energy management strategy enables the system to maximize energy utilization efficiency while ensuring power supply security and avoiding unnecessary energy storage losses.

[0168] The ratio of the absolute value of the trend coefficient to a preset trend threshold is used as the adjustment amplitude coefficient for the capacity reservation ratio. Amplitude determination is a crucial step in adjustment quantification, translating trend strength into specific operational parameters. The calculation process is based on the normalization of the trend coefficient, determining the relative strength of the adjustment by comparing it with a preset threshold. The formula for calculating the amplitude coefficient is:

[0169]

[0170] Where α is the adjustment amplitude coefficient, k is the trend coefficient, kth is the preset trend threshold, and αmax is the maximum adjustment amplitude limit (usually set to 0.3-0.5).

[0171] The amplitude coefficient ranges from [0, αmax], with a larger value indicating a larger adjustment amplitude. This dynamic adjustment mechanism based on trend strength enables the system to take corresponding response measures according to the severity of the change, avoiding overreaction while ensuring sufficient adaptability.

[0172] Based on the adjustment range coefficient, the baseline capacity reservation ratio set at the beginning of the week is adjusted incrementally or incrementally to generate differentiated energy storage capacity reservation ratios for each day of the week, ensuring continuous system power supply under changing shading trends. Reservation ratio generation is the final step in strategy implementation, translating adjustment decisions into specific operational parameters. The generation process employs a gradual adjustment method, calculating the daily change in the reservation ratio based on the number of days and the trend direction. The reservation ratio adjustment formula is:

[0173]

[0174] Where R(d) is the energy storage capacity reservation ratio on day d, R0 is the baseline reservation ratio (usually set to 0.2-0.3), α is the adjustment range coefficient, k is the trend coefficient, D is the total number of days (usually 7), and sign() is the sign function, which returns the positive or negative sign of the parameter.

[0175] This invention achieves optimized regulation of the entire process of a photovoltaic-storage system in the unique environment of tropical savanna through microclimate acquisition, sedimentation prediction, topographic shading analysis, energy storage thermal management, power correction, multi-timescale strategy generation, ripple monitoring, and dynamic correction. It is environmentally adaptable and highly efficient, capable of in-depth analysis and performance evaluation under complex climate and topographic conditions, effectively addressing dust deposition and topographic shading issues, and providing a complete energy self-balancing solution.

[0176] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0177] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0178] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A photovoltaic energy storage off-grid system suitable for tropical savanna terrain, characterized in that, include: The microclimate acquisition module is used to acquire time-series microclimate data of the area where the photovoltaic array is located; The deposition prediction module is used to construct a prediction model for the dust deposition rate on the photovoltaic surface based on the wind speed vector, humidity fluctuation and particulate matter concentration in the microclimate time series data. The terrain shading analysis module is used to acquire terrain elevation data and photovoltaic array deployment coordinates, calculate the terrain shadow projection path at different times, and generate an shading influence matrix that includes the dynamic evolution characteristics of shadow boundaries. The energy storage thermal management module is used to collect multi-point temperature distribution data of the outer shell of the energy storage unit, calculate the internal temperature gradient field of the energy storage unit, and establish a thermal decay compensation model for energy storage efficiency based on the correlation between the temperature gradient field and the ambient day-night temperature difference. The power correction module is used to combine the dust deposition rate prediction model, the shading influence matrix and the thermal attenuation compensation model to construct a three-dimensional correction surface for the photovoltaic usable power. The strategy generation module is used to generate multi-timescale energy scheduling strategies based on the three-dimensional modified surface and the seasonal decomposition data of historical load demand. The ripple monitoring module is used to extract the micro-fluctuation characteristics of the photovoltaic array output by high-frequency sampling of the DC side current ripple of the inverter, and to identify the abnormal current drop mode associated with the dust obstruction gradient in the micro-fluctuation characteristics. The dynamic correction module is used to dynamically correct the deposition rate parameters in the dust deposition rate prediction model based on the frequency and amplitude changes of the abnormal current drop mode, and trigger the adjustment of the photovoltaic cleaning cycle to advance or delay, so as to obtain the corrected multi-timescale energy dispatch strategy. The control execution module is used to control the charging and discharging power and timing of the energy storage unit based on the modified multi-timescale energy scheduling strategy.

2. The system according to claim 1, characterized in that, The step of constructing a prediction model for the dust deposition rate on photovoltaic surfaces based on the wind speed vector, humidity fluctuations, and particulate matter concentration in the microclimate time-series data includes: The wind speed vector is decomposed into a tangential wind speed component parallel to the photovoltaic surface and a normal wind speed component perpendicular to the photovoltaic surface. The ratio of the tangential wind speed component to the preset critical wind speed is calculated and used as the wind-driven sweeping efficiency factor. Extract the gradient change rate of the humidity rise phase from the humidity fluctuation data, identify the time periods when the humidity gradient change rate is greater than a preset humidity threshold, and mark them as potential rainfall precursor periods; The time interval between the potential rainfall precursor periods and the actual rainfall events is statistically analyzed to establish a mapping relationship between humidity gradient and rainfall lead time. Based on the intraday fluctuation curve of particulate matter concentration, the peak occurrence time and duration of the fluctuation curve are extracted, and the effective settling time window of particulate matter is calculated. The dust deposition rate prediction model is generated by coupling the wind sweeping efficiency factor, the mapping relationship, and the effective settling time window.

3. The system according to claim 1, characterized in that, The calculation of terrain shadow projection paths at different times generates an occlusion influence matrix that includes the dynamic evolution features of shadow boundaries, including: Based on the terrain elevation data, a three-dimensional terrain mesh model is constructed; Based on the geographical location of the photovoltaic array deployment coordinates, calculate the time-varying functions of the solar altitude angle and azimuth angle throughout the year; In the three-dimensional terrain mesh model, the time-varying function is used as a parameter to track the shadow projection trajectory of each terrain protrusion node on the photovoltaic array plane; The shadow projection trajectory is discretized into a set of shadow boundary points with time intervals on the order of minutes; Calculate the centroid offset distance and area change rate between adjacent shadow boundary point sets as dynamic evolution characteristics of the shadow boundary; Based on the dynamic evolution characteristics of the shadow boundary, the occlusion influence matrix is ​​constructed.

4. The system according to claim 1, characterized in that, The step of establishing a thermal decay compensation model for energy storage efficiency based on the correlation between the temperature gradient field and the diurnal temperature range of the environment includes: Spatial interpolation is performed on the multi-point temperature distribution data of the energy storage unit shell to generate a continuous temperature distribution field of the energy storage unit shell; Calculate the temperature gradient vectors of the continuous temperature distribution field in the radial and axial directions to obtain the temperature gradient field; Extract the region with the largest gradient magnitude in the temperature gradient field and mark it as the heat accumulation core region; The peak and valley temperatures of the heat accumulation core area are statistically analyzed over a 24-hour period, and the peak-valley temperature difference is calculated as the effective diurnal temperature difference of the energy storage unit. Establish the deviation value between the effective diurnal temperature difference and the ambient diurnal temperature difference, and the deviation value characterizes the thermal inertia characteristics of the energy storage unit; Based on the thermal inertia characteristics and the number of charge-discharge cycles of the energy storage unit, the thermal decay coefficient of the energy storage efficiency is calculated; based on the thermal decay coefficient, the thermal decay compensation model is constructed.

5. The system according to claim 1, characterized in that, The method of constructing a three-dimensional modified surface for photovoltaic usable power by combining the dust deposition rate prediction model, the shading influence matrix, and the thermal attenuation compensation model includes: The cumulative dust thickness sequence of each photovoltaic sub-unit is extracted from the dust deposition rate prediction model, and the corresponding light transmittance attenuation curve is calculated. The shading rate values ​​of each photovoltaic sub-unit at each time moment are extracted from the shading influence matrix and converted into the effective light-receiving area ratio; The light transmittance attenuation curve is multiplied by the effective light-receiving area ratio to obtain the comprehensive light availability coefficient of each photovoltaic sub-unit. Based on the comprehensive solar availability factor and the photovoltaic rated power, calculate the real-time theoretical output power of each photovoltaic sub-unit; A power transfer link model between photovoltaic power generation and energy storage efficiency is established, and the charge and discharge efficiency correction coefficient is extracted from the thermal attenuation compensation model. Substitute the real-time theoretical output power into the energy transfer link model, and combine it with the charging and discharging efficiency correction coefficient to calculate the actual usable power after passing through the energy storage stage; Construct a coordinate system with three dimensions: time, spatial location of the photovoltaic array, and comprehensive environmental influencing factors; The actual available power is mapped to the grid nodes of the coordinate system to form a discrete power distribution point set; The discrete power distribution point set is subjected to cubic spline surface interpolation to generate the three-dimensional modified surface.

6. The system according to claim 1, characterized in that, The generation of multi-timescale energy scheduling strategies based on the seasonal decomposition data of the three-dimensional modified surface and historical load demand includes: The historical load demand data is seasonally decomposed to obtain trend components, seasonal components, and random components. The seasonal components are divided into two sub-cycles according to the dry season and the wet season, and the daily average and variance of load demand in the two sub-cycles are calculated respectively. Based on the three-dimensional modified surface, an hourly photovoltaic available power prediction sequence is extracted; Calculate the predicted sequence of available photovoltaic power and the hourly power deficit distribution of the load demand; Based on the power deficit distribution, the energy storage charging priority period and the discharge reserved period are divided on a daily time scale. On a weekly timescale, the energy storage capacity reservation ratio for each day of the week is adjusted based on the weekly shading change trend predicted by the shading impact matrix. The hourly power deficit distribution, the daily charging and discharging time period division, and the weekly capacity reservation ratio are coupled at multiple scales to generate the multi-time-scale energy scheduling strategy.

7. The system according to claim 1, characterized in that, The process of extracting micro-fluctuation characteristics of the photovoltaic array output through high-frequency sampling of the inverter's DC-side current ripple, and identifying abnormal current drop patterns associated with dust obstruction gradients within these micro-fluctuation characteristics, includes: The inverter DC-side current is sampled at a high frequency of kilohertz to obtain a high-frequency current sampling sequence. Wavelet decomposition is performed on the high-frequency current sampling sequence to separate the high-frequency ripple component and the low-frequency fundamental component; Calculate the envelope of the high-frequency ripple component and extract the local minimum points of the envelope; By statistically analyzing the occurrence intervals and amplitude distribution of the local minimum points, a fluctuation fingerprint of the current ripple is constructed. A mapping database is established between the fluctuation fingerprint and the local shading state of the photovoltaic array. The mapping database contains fluctuation fingerprint feature vectors corresponding to different shading gradients. The real-time collected fluctuation fingerprint is matched with the feature vector in the mapping database for similarity matching; When the matching similarity is greater than the preset similarity threshold and the corresponding occlusion gradient shows a monotonically increasing trend, it is identified as the abnormal current decrease mode.

8. The system according to claim 1, characterized in that, The step of dynamically correcting the deposition rate parameters in the dust deposition rate prediction model based on the frequency and amplitude changes of the abnormal current drop pattern includes: Statistically count the cumulative frequency of occurrence of the abnormal current drop mode within the sliding time window; Calculate the average value of the current drop amplitude corresponding to the abnormal current drop mode; A dust accumulation assessment index is established based on the average of the cumulative frequency of occurrence and the current decrease magnitude. The deviation is obtained by performing a deviation analysis between the dust accumulation assessment index and the predicted value of the dust deposition rate prediction model; When the absolute value of the deviation is greater than the preset deviation threshold, the correction increment of the deposition rate parameter is calculated. The correction increment is superimposed on the current deposition rate parameter of the dust deposition rate prediction model to complete the dynamic correction of the parameters; and the cumulative dust thickness sequence of each photovoltaic sub-unit is recalculated to update the light transmittance attenuation curve, and then the three-dimensional correction surface is updated. Based on the updated three-dimensional modified surface, the predicted value of photovoltaic available power is regenerated, and combined with the adjustment time node of the photovoltaic cleaning cycle, the division of charging and discharging periods and the capacity reservation ratio in the multi-timescale energy dispatch strategy are corrected to obtain the corrected multi-timescale energy dispatch strategy.

9. The system according to claim 2, characterized in that, The calculation of the ratio of the tangential wind speed component to the preset critical wind speed, as the wind-driven sweeping efficiency factor, further includes: Obtain photovoltaic module surface tilt angle data; The effective component of wind speed on the photovoltaic surface is calculated based on the angle between the surface tilt angle and the wind speed vector. Establish a relationship function between the effective action component and the particle size, wherein the relationship function characterizes the critical wind speed at which particles of different sizes are swept away; The particle size distribution of the particulate matter concentration data is analyzed to obtain the proportion of particulate matter in each particle size range; Based on the proportion of particulate matter in each particle size range and the corresponding critical wind speed, a weighted average critical wind speed is calculated and used as the preset critical wind speed. The ratio of the real-time tangential wind speed component to the weighted average critical wind speed is calculated. When the ratio is greater than 1, the wind sweeping efficiency factor increases linearly, and when the ratio is less than 1, the wind sweeping efficiency factor decreases exponentially.

10. The system according to claim 5, characterized in that, The adjustment of the energy storage capacity reservation ratio for each day of the week based on the predicted shading change trend of the shading impact matrix at the weekly time scale includes: Extract the occlusion rate data for seven consecutive days from the occlusion impact matrix and construct the weekly occlusion rate change curve; Curve fitting is performed on the intra-week occlusion rate change curve to obtain the trend coefficient of the occlusion rate change; When the trend coefficient is positive, the energy storage capacity reservation ratio is increased in the later part of the week. When the trend coefficient is negative, the energy storage capacity reservation ratio is reduced in the later part of the week. The ratio of the absolute value of the trend coefficient to the preset trend threshold is calculated and used as the adjustment range coefficient of the capacity reservation ratio. Based on the adjustment range coefficient, the baseline capacity reservation ratio set at the beginning of the week is adjusted incrementally or incrementally to generate the differentiated energy storage capacity reservation ratio for each day of the week.

Images