A street lamp-oriented multi-source weather forecast-driven light storage scheduling method and system

Abstract

The present application relates to the field of photovoltaic energy storage management and intelligent power supply scheduling, in particular to a multi-source weather prediction driven photovoltaic storage scheduling method and system for street lamps, comprising the following steps: S1, obtaining hourly weather forecast data in a preset period from multiple heterogeneous weather data sources; S2, performing fusion processing on the hourly weather forecast data of each weather data source to generate integrated hourly power generation prediction and prediction confidence; S3, generating a scheduling instruction for the energy storage system based on the integrated hourly power generation prediction and prediction confidence, etc., and the scheduling instruction is dynamically adjusted according to the prediction confidence; S4, collecting actual operation data by executing the scheduling instruction; S5, forming a closed-loop adaptive scheduling by adaptively updating the fusion processing strategy based on the error between the actual operation data and the hourly weather forecast data. Through multi-source fusion, confidence evaluation and adaptive learning, the present application improves the prediction accuracy, maximizes the photovoltaic utilization rate and the energy storage life while ensuring the reliability of street lamp power supply.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic energy storage management and intelligent power supply scheduling, and to a photovoltaic energy storage scheduling method and system driven by multi-source weather forecasting for streetlights. Background Technology

[0002] With the deepening of smart city construction, urban streetlights, as an important public infrastructure, have expanded their functions from simple lighting to multiple fields such as information sensing, communication hubs, and security monitoring. Meanwhile, in response to the national "dual-carbon" strategy, an increasing number of streetlights are integrating photovoltaic power generation units and energy storage batteries, forming independent or grid-connected microgrid systems that integrate photovoltaic and energy storage. This system can generate and store solar energy during the day, and power the streetlights and their associated loads at night, effectively reducing dependence on traditional grid power and reducing carbon emissions.

[0003] However, the operational efficiency of such photovoltaic-storage street light systems is highly dependent on meteorological conditions, especially solar irradiance. Inaccurate weather forecasts can lead to serious errors in energy dispatch.

[0004] For example, if the power generation for the next day is overestimated, the system may over-discharge at night or reduce charging during off-peak hours, resulting in insufficient power the next day and affecting the normal illumination of streetlights; conversely, if the power generation is underestimated, cheap off-peak electricity may be wasted due to overcharging, or the energy storage batteries may be overcharged, affecting their lifespan.

[0005] Currently, common scheduling strategies are mainly based on historical average data or a single public weather forecast source, which has obvious drawbacks: the predictive ability of a single meteorological data source is limited in local areas and specific weather phenomena (such as short-term severe convective weather), and the performance of different data sources varies in different seasons and regions, making reliance on a single source risky; most systems adopt fixed scheduling rules, such as filling the system during off-peak hours regardless of the weather, which cannot adapt to the uncertainty of meteorological conditions, lacks an assessment and response mechanism for the reliability of forecast results, and the system cannot learn from historical forecast errors and cannot dynamically adjust the degree of trust in different data sources, resulting in persistent errors and the inability to optimize the scheduling strategy.

[0006] Therefore, there is an urgent need in this field for a method and system for scheduling solar-powered streetlights that can comprehensively utilize multi-source meteorological information, possess high-precision forecasting capabilities, and can adaptively learn and optimize, in order to overcome the aforementioned shortcomings of existing technologies.

[0007] In view of the above-mentioned problems, this technical solution designs a photovoltaic energy storage scheduling method and system for streetlights driven by multi-source weather forecasting. Summary of the Invention

[0008] To achieve the objectives of this invention, the following technical solution is adopted:

[0009] To achieve the above-mentioned objectives, this application provides a multi-source weather forecast-driven photovoltaic-storage scheduling method and system for streetlights, comprising the following:

[0010] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0011] The primary objective of this invention is to provide a photovoltaic-storage scheduling method driven by multi-source weather forecasting for streetlights. This method can significantly improve the accuracy of photovoltaic power generation forecasting by integrating forecast information from multiple heterogeneous weather data sources, and make intelligent energy storage scheduling decisions based on the confidence level of the forecast results.

[0012] Another objective of this invention is to provide a corresponding optical-storage scheduling system that can implement the above-mentioned methods and has the capabilities of online learning, adaptive optimization, and fault degradation operation, forming an efficient closed-loop control system.

[0013] The technical solution of this invention is as follows:

[0014] A multi-source weather forecast-driven photovoltaic-storage scheduling method for streetlights includes the following steps:

[0015] S1 obtains hourly weather forecast data for a preset time period from multiple heterogeneous meteorological data sources;

[0016] S2 integrates hourly weather forecast data from various meteorological data sources to generate integrated hourly power generation forecasts and corresponding forecast confidence levels.

[0017] S3 generates dispatch instructions for the energy storage system based on integrated hourly power generation forecasts and forecast confidence levels, combined with load forecasts and the current status of the energy storage system. The strategy for formulating dispatch instructions is dynamically adjusted according to the forecast confidence level.

[0018] S4 executes scheduling instructions and collects actual operational data;

[0019] Based on the error between actual operational data and hourly weather forecast data, S5 adaptively updates the fusion processing strategy to form a closed-loop adaptive scheduling.

[0020] Furthermore, in S1, during the electricity price trough time window within a preset time period, hourly weather forecast data for the next day is requested in parallel from multiple heterogeneous meteorological data sources.

[0021] Furthermore, the multiple heterogeneous meteorological data sources include at least two or more of the following: forecast data from the China Meteorological Administration Data Center, global numerical weather prediction products, and commercial converged weather platforms.

[0022] Furthermore, the fusion process in S2 includes the following steps:

[0023] S21 performs time-series alignment and feature extraction on hourly weather forecast data from various meteorological data sources, extracting hourly total horizontal irradiance and cloud cover parameters related to photovoltaic power generation.

[0024] S22 assigns initial weights to the hourly weather forecast data of each meteorological data source based on the historical forecast errors and short-term correction factors of each meteorological data source and performs weighted fusion to obtain integrated hourly power generation forecasts.

[0025] S23 calculates the forecast confidence level of the integrated hourly power generation forecast, which is calculated by weighting multiple factors including source consistency, historical accuracy scores of each source, short-term observation correction response capability, and seasonal / regional adaptability.

[0026] Furthermore, S3 includes:

[0027] Based on the integrated hourly power generation forecast and prediction confidence, combined with load forecast and the current state of the energy storage system, the sufficiency of photovoltaic power generation is calculated. The sufficiency is constrained by the available space for energy storage, grid connection limitations and minimum standby SOC (state of charge).

[0028] The calculation of fitness includes:

[0029] Based on the integrated hourly power generation forecast, the estimated total power generation for the next day is calculated.

[0030] Estimate the total load demand for the next day based on the street light activation schedule and historical load curves;

[0031] The estimated net power generation for the next day is calculated by subtracting the total load demand for the next day from the estimated total power generation for the next day.

[0032] After calculating the difference between the maximum charging SOC limit and the current SOC, multiply it by the energy storage capacity to obtain the energy storage capacity.

[0033] The estimated net power generation for the next day is divided by the total load demand for the next day to obtain the power generation coverage rate.

[0034] The adequacy status is determined based on forecast confidence, power generation coverage, and available energy storage space.

[0035] The logic for determining the fitness status includes:

[0036] If the forecast confidence level is less than the forecast confidence threshold, it is judged as an unreliable state;

[0037] If the forecast confidence level is greater than or equal to the forecast confidence threshold

[0038] If the power generation coverage rate is greater than or equal to the power generation coverage rate threshold and the energy storage capacity is not less than the energy storage capacity threshold, then it is considered to be in a suitable state.

[0039] Otherwise, it is judged as an unsuitable state;

[0040] Based on the energy availability status, generate scheduling instructions for the energy storage system:

[0041] When the state is determined to be unsuitable, a first-class scheduling instruction is generated, which mainly uses mains power charging, and a higher target SOC is set.

[0042] When the condition is determined to be suitable, a second type of scheduling instruction is generated, which is mainly based on photovoltaic charging, and a lower target SOC is set.

[0043] When the system is determined to be in an unreliable state, a third type of scheduling instruction is generated, which primarily uses mains power for charging, and a medium target SOC is set.

[0044] Furthermore, the adaptive update in S5 adopts a sliding window strategy, saves the error sequence of the most recent N days, and calculates the mean square error and deviation trend of each meteorological data source based on the error sequence, and uses a smoothing factor to update the weight of each data source in the fusion process.

[0045] Furthermore, the method also includes:

[0046] When the central dispatch fails or the forecast confidence is extremely low, S6 switches to a local conservative control strategy. The local conservative control strategy ensures that the energy storage SOC is not lower than the preset minimum emergency value and prioritizes meeting the nighttime lighting load.

[0047] Furthermore, the scheduling instructions include the target end of SOC, the charging power curve for different time periods, the instruction validity period, and the allowed range for local adjustments.

[0048] A multi-source weather forecast-driven photovoltaic-storage scheduling system for streetlights, the system comprising:

[0049] The data acquisition module is used to acquire hourly weather forecast data for a preset time period from multiple heterogeneous meteorological data sources;

[0050] The multi-source fusion and learning module is used to fuse forecast data from various data sources to generate integrated hourly power generation forecasts and corresponding forecast confidence levels.

[0051] The central dispatch server is used to generate dispatch instructions for the energy storage system based on integrated hourly power generation forecasts and forecast confidence levels, combined with load forecasts and the current status of the energy storage system. The strategy for formulating dispatch instructions is dynamically adjusted according to the forecast confidence level.

[0052] The execution and monitoring module is used to execute scheduling instructions and collect actual operation data;

[0053] The adaptive learning module is used to adaptively update the processing strategy of the data fusion and prediction module based on the error between actual operating data and integrated hourly weather forecast data, forming a closed-loop adaptive scheduling.

[0054] Furthermore, the system also includes a local controller, which is configured to execute a preset conservative strategy when communication with the central dispatch server is interrupted or the confidence level of the received instruction is extremely low, and has a BMS interface, an MPPT controller interface and a charge and discharge control interface to execute local safety protection logic.

[0055] Operations and maintenance database: Used to store all historical forecast data, actual operation data, model parameters, historical weights and event logs, supporting data backtracking, auditing and manual intervention.

[0056] The system also includes a regional coordination module, which coordinates the charging and discharging behavior of multiple street light sites at the regional level, so that the total grid-connected power of the region does not exceed the transformer capacity or grid connection limit.

[0057] Compared with existing technologies, the present invention has the following significant advantages: by adopting multi-source heterogeneous meteorological data fusion technology, combined with historical error analysis and short-term observation correction, the accuracy of photovoltaic power generation prediction is significantly improved; at the same time, through dynamic weight allocation and confidence assessment, the system can automatically identify and reduce the impact of unreliable prediction sources, avoiding energy storage scheduling errors caused by meteorological prediction deviations.

[0058] By using a confidence-based dynamic adequacy determination method, which comprehensively considers multiple factors such as forecast reliability, load demand, and energy storage status, the charging and discharging strategies are automatically adjusted according to different confidence levels, thereby optimizing energy utilization while ensuring lighting reliability.

[0059] In the event of a communication interruption or extremely low prediction confidence, the system can seamlessly switch to a local conservative control mode, using preset safety policies to protect basic lighting functions and significantly improve the system's robustness.

[0060] By setting a reasonable SOC operating range to avoid overcharging and over-discharging, and by combining temperature monitoring and power limiting measures, the lifespan of energy storage batteries can be effectively extended. Attached Figure Description

[0061] Figure 1 This is an overall flowchart of a photovoltaic-storage scheduling method driven by multi-source weather forecasting for streetlights.

[0062] Figure 2 This is a detailed flowchart of the fusion process in step S2.

[0063] Figure 3 This is a schematic diagram of the logic determination of suitability in step S3.

[0064] Figure 4 This is a schematic diagram illustrating the generation of scheduling instructions in step S3.

[0065] Figure 5 This is an architectural block diagram of a multi-source weather forecast-driven photovoltaic energy storage scheduling system for streetlights.

[0066] Figure 6 It is a framework diagram of a computer system. Detailed Implementation

[0067] The implementation of the technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0068] The technical solution of this invention is as follows:

[0069] like Figure 1 As shown, a multi-source weather forecast-driven photovoltaic-storage scheduling method for streetlights includes the following steps:

[0070] S1: Obtain meteorological forecast data for a preset time period from multiple heterogeneous meteorological data sources;

[0071] During the electricity price off-peak hours within a preset time period each day (e.g., from 23:00 to 5:00 the next day), the system initiates data requests in parallel to multiple heterogeneous meteorological data sources through the data acquisition module to obtain hourly meteorological forecast data for the next day.

[0072] The meteorological data sources must include at least hourly forecasts from the authoritative China Meteorological Administration (CMA) data center, supplemented by at least one other source, such as the Global Numerical Prediction Product (GFS) of the European Centre for Medium-Range Weather Forecasts (ECMWF) or commercial high-precision fusion weather platforms (such as Windy, Caiyun Weather, etc.).

[0073] This multi-source parallel acquisition mechanism ensures the timeliness and diversity of data, and ensures the accuracy of subsequent data fusion;

[0074] It should be noted that the data acquisition module (also known as the network module) obtains JSON or other structured packets via HTTP / HTTPS, performs packet integrity, timestamp and geographic matching verification, filters out expired or abnormal data, and obtains the verified hourly weather forecast data for the next day.

[0075] JSON is a lightweight data exchange format, an open standard file and data exchange format that is easy for humans to read and write, and also easy for machines to parse and generate.

[0076] S2: Based on the hourly weather forecast data from various meteorological data sources obtained in S1, the data are fused and processed to generate an integrated weather forecast and the corresponding forecast confidence level;

[0077] like Figure 2 As shown, the fusion processing steps are as follows:

[0078] S21: Data time series alignment and feature extraction: Due to the differences in update frequency and spatiotemporal resolution of different meteorological data sources, the original data (hourly meteorological forecast data from each meteorological data source) are first time series aligned to the same timestamp and geographic grid. Then, key feature parameters closely related to photovoltaic power generation are extracted from the aligned data, including hourly total horizontal irradiance and cloud cover.

[0079] Optionally, auxiliary parameters that affect the efficiency of photovoltaic panels, such as ambient temperature and humidity, can also be extracted as needed.

[0080] S22: Initial weighting and multi-source data fusion: Weighted fusion is performed based on historical forecast errors and short-term correction factors from various meteorological data sources;

[0081] Calculate preliminary performance indicators (long-term / medium-term / short-term) for each meteorological data source, statistically analyze the typical error trends of each source in different windows from the error assessment data (e.g., bias, trend of mean absolute error, whether there is systematic overestimation / underestimation, etc.), and output "long-term performance score" and "short-term performance score".

[0082] Generate initial fusion weights: Map the above performance scores to initial weight preferences. Sources with good long-term performance receive higher base weights, while sources with outstanding short-term performance receive additional short-term correction factors. These weights are referred to as "fusion candidate weights," and the availability of meteorological data sources is recorded. It should be noted that missing or delayed data will lower the upper limit of the weight for that source.

[0083] Weighted fusion is performed using candidate weights: hourly weather forecast data (e.g., total horizontal irradiance) from each meteorological data source are multiplied by their corresponding weights and summed to obtain the final integrated hourly power generation forecast. At the same time, cross-source differences (the distribution or gap between forecasts from each source) are recorded during the fusion process as input for subsequent confidence sub-items.

[0084] S23: Confidence Assessment: Confidence Calculation: Calculate the forecast confidence level of the integrated hourly power generation forecast. The forecast confidence level is a value between 0 and 1, comprehensively evaluated from four dimensions:

[0085] (1) Source consistency: The source consistency index uses historical meteorological observations and meteorological forecast data from various sources. It inputs hourly weather forecasts from all sources at the current time or within a preset fluctuation window, while using historical meteorological observations as a reference. For the hourly forecast to be evaluated, it calculates the degree of difference between forecasts from various sources.

[0086] If multiple forecasts are similar to each other and do not significantly conflict with historical observation trends, a high consistency score is given.

[0087] If the differences are large or contradictory and cannot be explained by historical observation bias, a low score will be assigned.

[0088] This measures the consistency of hourly predictions from different sources. The smaller the difference between sources, the higher the consistency. This is achieved by statistically analyzing the cross-source differences (e.g., mean absolute difference or variance) of the current hourly set of each source and mapping them to a 0-1 score, where the smaller the difference, the higher the score.

[0089] It should be noted that the source consistency index is mainly used to assess short-term consistency. If some sources are missing, the maximum score for that item is reduced based on the inherent differences of the remaining sources to reflect the decrease in information content. When consistency is high, more balanced integration can be allowed in the short term. When consistency is low, online learning will favor sources with high historical accuracy and reduce immediate trust in fluctuating sources.

[0090] (2) Historical accuracy score: Based on the past N days, i.e. based on historical actual power generation and meteorological forecast error of meteorological data source: input the historical hourly meteorological forecast and the corresponding historical actual power generation of each source (the sliding window can be set to 30 days), align and statistically analyze the typical error performance of the source within the sliding window by hour (e.g. deviation trend, whether it is systematically overestimated or underestimated, degree of error fluctuation), and then map the performance of "long-term stable and small error" to high score, and "long-term large error or obvious fluctuation" to low score. Evaluate for different hourly periods (e.g. daytime high-incidence period and non-power generation period) to support hourly scoring. Finally, map these statistics to the accuracy score in the range of 0-1, i.e. the better the historical performance, the higher the score.

[0091] It should be noted that the historical accuracy score is mainly used to assess long-term or medium-term consistency, and this score constitutes the base value of the weight (long-term baseline). When updating the weight in online learning, priority is given to retaining the share of sources with high long-term accuracy.

[0092] (3) Short-term observation correction response capability: assess the speed and accuracy of meteorological data source correction response to ultra-short-term (e.g., the next 2-6 hours) observation data; based on the short-term observation correction factor, through the difference record between short-term near-real-time observation and the corresponding time period forecast, as well as the historical record of error reduction after the application of short-term correction in the past, assess the deviation between the current short-term observation and the forecast from each source, and check whether the error has significantly decreased after the application of short-term correction to similar deviations in the past.

[0093] If a source can quickly and stably align its predictions with the actual observations after incorporating short-term observations, then that source has a high short-term response score. In the absence of short-term observations, the short-term response score is downgraded to low confidence or neutral.

[0094] The speed and effectiveness of the corrections made by the source or fusion algorithm after incorporating near real-time ground observations (e.g., radiometers, short-time satellites) are measured.

[0095] By comparing the difference between short-term observations and forecasts, the proportion of error reduction after short-term correction is evaluated. Those with strong responsiveness and significant error reduction score higher.

[0096] It should be noted that when there are effective short-term observations and a good correction history, the system can briefly increase the weight of the source or correction model on an hourly basis to quickly improve hourly predictions.

[0097] Seasonal / Regional Adaptability: Some meteorological data sources perform better in specific seasons (such as rainy season, dry season) or regions. This factor is used to adjust the confidence level under different seasons or regions.

[0098] Based on neighborhood similarity, by inputting hourly power generation observations of nearby street light stations and local historical meteorological observations, a set of stations that are geographically close and have similar climates or installed capacities is found. The historical and recent similarities between the target station's predicted power generation and the actual power generation in the neighborhood are compared.

[0099] If a neighboring region performs stably under the same weather conditions and is consistent with the current forecast, then the neighboring region similarity score is high.

[0100] If there is a consistent deviation between neighboring observations and current multi-source forecasts (e.g., multi-source forecasts are generally overestimated but neighboring observations are underestimated), the confidence level should be lowered and a possible regional error should be indicated.

[0101] Finally, based on historical seasonal / regional error statistics, a weighted score is mapped to the final score.

[0102] It should be noted that high neighborhood similarity supports increasing the weight of sources with good neighborhood performance or performing neighborhood correction on the fusion results; conversely, it prompts downweighting or triggers a more conservative strategy.

[0103] Finally, the scores of each sub-item are mapped to a unified 0-1 range according to the preset or adaptive configuration. Based on the system configuration (or the sub-item weights adjusted by online learning), the sub-items are weighted according to priority to synthesize hourly confidence scores. At the same time, the synthesized confidence scores are smoothed over time (fused with the confidence scores of the previous period by a smoothing factor).

[0104] By weighting and combining the scores from these four dimensions, a quantitative confidence score is finally output, which intuitively reflects the credibility of this prediction.

[0105] It should be noted that the confidence level outputs both hourly values ​​for hourly discharge or grid connection decisions and daytime or all-day aggregated values ​​for adequacy determination. The hourly confidence level can be weighted to obtain the comprehensive confidence level for daytime or critical windows, and the weights can be adjusted according to the hourly importance (e.g., the weight of peak daytime power generation periods is greater).

[0106] In addition, each confidence level output should be accompanied by the scores of each sub-item and an explanation of the main influencing factors, such as: the main reason is that the historical accuracy has decreased due to the recent overestimation of source A, which facilitates operation and maintenance interpretation and manual intervention.

[0107] S3: Based on the integrated hourly power generation forecast and the forecast confidence level, combined with the load forecast and the current state of the energy storage system, a dispatch instruction for the energy storage system is generated, wherein the strategy for formulating the dispatch instruction is dynamically adjusted according to the forecast confidence level.

[0108] Specifically, load forecasting combines the next day's load forecast of the street light site itself, that is, forecasting based on factors such as season, weekday or holiday; the current state of the energy storage system refers to the current state of charge (SOC) of the energy storage battery.

[0109] Among them, the sufficiency of photovoltaic power generation is calculated based on the integrated hourly power generation forecast and prediction confidence level. The sufficiency level is used to quantify the suitability of relying on photovoltaic power generation to meet daytime load and supplement nighttime lighting needs the next day. Its calculation not only considers the total predicted power generation, but also the acceptable space for energy storage: whether the current SOC is too low and needs to be charged first; the expected grid connection limit for the day: the local grid's restrictions on reverse power transmission (surplus power to the grid); and the minimum reserve SOC constraint: the minimum amount of electricity that must be reserved to cope with emergencies.

[0110] Specifically, the adequacy is determined as follows:

[0111] The system integrates hourly power generation forecasts, forecast confidence levels, load forecasts (based on the next day's lighting load curve (kW) and additional loads based on historical data), and the current status of the energy storage system (current state of charge (SOC) (%), maximum charging and discharging power (kW), and state of health (SOH) (%)) into the suitability judgment logic to calculate the expected net power generation, the space that can be accommodated by energy storage, and the power generation coverage rate.

[0112] like Figure 3 As shown, the suitability judgment logic is as follows:

[0113] The total expected power generation for the next day is calculated by integrating the hourly power generation forecasts and accumulating them hourly to obtain the total expected power generation for the whole day.

[0114] Estimate the total load demand for the next day. Based on the street light lighting schedule and historical load curves, estimate the electricity consumption for the next day, with the main demand at night and minimal or standby power during the day.

[0115] The current state of an energy storage system is represented by the available space for energy storage.

[0116] Power generation coverage ratio is the percentage of the expected net power generation for the next day relative to the total load demand for the next day.

[0117] That is, the estimated net power generation for the next day = the estimated total power generation for the next day - the total load demand for the next day;

[0118] Energy storage capacity = Energy storage capacity * (Maximum charging SOC limit - Current SOC);

[0119] Power generation coverage = Projected net power generation for the next day / Total load demand for the next day;

[0120] Based on the above factors and combined with grid connection and electricity price information, three categories of judgments are derived: "suitable", "unsuitable", and "unreliable".

[0121] The specific determination process is as follows:

[0122] If the forecast confidence level is less than the forecast confidence level threshold (usually 0.6 by default), it is judged as "unreliable", that is, a conservative strategy is adopted;

[0123] If the forecast confidence level is greater than or equal to the forecast confidence level threshold (usually 0.6 by default), then the next step is to make a judgment based on the power generation coverage and energy storage capacity.

[0124] Power generation coverage determination: If the power generation coverage is greater than or equal to the power generation coverage threshold (usually 70% by default), it is determined as "suitable". This means that the grid charging target can be reduced during off-peak hours, the charging target can be set to a lower SOC, and direct photovoltaic power supply and battery charging can be prioritized during the day.

[0125] If the power generation coverage rate is less than the power generation coverage rate threshold (usually 70% by default): it is judged as "unsuitable". Even if the forecast is reliable, if the power generation is insufficient to cover the load, the grid charging target should be increased during off-peak hours to ensure nighttime lighting.

[0126] The threshold for determining the energy storage capacity and energy storage acceptance space is:

[0127] If the energy storage capacity is less than a certain percentage (e.g., 20%) of the expected peak daytime power generation, it is considered that the energy storage capacity is insufficient. When the energy storage capacity is insufficient (i.e., the current SOC of the battery is close to the upper limit) and the expected daytime power generation is high, the off-peak charging target is reduced to avoid overcharging, and the priority is adjusted during the daytime to first use direct photovoltaic power and then slowly replenish the battery.

[0128] It should be noted that, regardless of the judgment result, the minimum emergency SOC (default 25%) is retained as a strong constraint and must not be overridden by scheduling.

[0129] If there are dynamic electricity prices or next-day electricity prices that are expected to generate revenue during peak hours, strategies can be adjusted while ensuring lighting reliability. For example, grid connection and discharge can be carried out during high-price periods to generate revenue, but the determination of suitability should still prioritize lighting reliability.

[0130] The final determination result is: suitable, unsuitable, or unreliable, used to determine whether to use mains power for charging during off-peak hours, the target SOC for charging, and whether to retain the daytime photovoltaic priority strategy.

[0131] Within the preset time period trough window, the charging target SOC and power curves are generated by comprehensively considering the charging suitability and constraints (i.e., SOC upper and lower limits, lifespan strategy, etc.). At the same time, the daytime discharge suggestion for the next day is generated. The central dispatch server sends the dispatch instruction to the local controller, waits for confirmation, and then enters the execution and monitoring. The actual power generation and consumption data are uploaded the next day and compared with the prediction error to update the fusion weight and threshold.

[0132] like Figure 4 As shown, the scheduling instruction decision is generated:

[0133] Based on the adequacy determination results, generate differentiated scheduling strategies;

[0134] If the battery is deemed "unsuitable" for charging, the system will generate a first-type dispatch instruction that prioritizes mains power charging, setting a high target SOC to ensure that the battery is charged to a sufficient level during off-peak electricity price periods to cope with potential insufficient photovoltaic power generation the following day.

[0135] If the deemed adequate charging level is "suitable," the system will generate a second type of dispatch instruction primarily for photovoltaic charging, setting a lower target SOC. This approach only provides minimal battery charging during off-peak hours, reserving ample energy storage capacity to accommodate abundant photovoltaic power during the day, maximizing the utilization of clean energy while reducing grid charging costs.

[0136] If the assessment of the suitability is "unreliable", the system will generate a third type of dispatch instruction that prioritizes mains power charging, setting a medium target SOC to avoid insufficient power supply the next day due to unreliable photovoltaic forecasts. In this case, the photovoltaic system will be downgraded to a backup, and only a small amount of supplementary charging will be allowed when the actual photovoltaic power generation is monitored to be sufficient in real time.

[0137] It should be noted that a higher target SOC is typically 90% of the target value, a lower target SOC is typically 40% of the target value, and a medium target SOC is typically 60% of the target value.

[0138] S4: Execute the scheduling instructions and collect actual operating data: Send the scheduling instructions generated in S3 to the local controllers of each street light. The scheduling instructions specifically include: the target end SOC, the charging power curves for different time periods (to achieve smooth charging and avoid impacting the power grid), the instruction validity period, and the permission for the local controller to make adjustments within a certain range.

[0139] The local controller executes scheduling commands to control the operating status of the mains charger, MPPT controller, and DC-DC converter. After execution, the system collects data such as actual photovoltaic power generation, battery SOC changes, and operation logs, and uploads them to the central database for subsequent error backtracking and analysis.

[0140] S5: Closed-loop adaptive weight update: Based on the actual operational data collected in S4, the system initiates a learning mechanism and adopts a sliding window strategy (e.g., saving the hourly error sequence for the most recent 30 days) to calculate the hourly weather forecast data error (e.g., mean square error, MSE) and bias trend for each meteorological data source within the window period. Then, a smoothing factor (e.g., 0.1) is used to update the fusion weights of each data source.

[0141] The smoothing factor serves to ensure that the weights change smoothly, preventing drastic fluctuations due to abnormal weather on a single day, thus enhancing the stability of the system and forming a closed-loop feedback mechanism to ensure that the system can continuously optimize its predictive capabilities.

[0142] S6: Fault Degradation and Local Conservative Strategy: When the central control server is unavailable, communication is interrupted, or the calculated confidence level is extremely low (below a more stringent second threshold, such as 0.3), the local controller will automatically switch to a pre-set conservative strategy. The local conservative strategy is primarily used to ensure the system's most basic security and functionality.

[0143] Conservative strategies include: maintaining the SOC at or above the preset minimum emergency SOC (e.g., 30%) to ensure that it can support emergency lighting for at least several nights; prohibiting discharge to the grid during communication interruptions to prevent safety hazards such as islanding effects; and limiting the maximum discharge power to protect battery life.

[0144] At the same time, by supporting multiple communication links, such as primary cellular or backup NB-IoT or Ethernet, the reliability of command issuance and telemetry can be improved.

[0145] It should be noted that the thresholds involved in the judgment logic (such as the forecast confidence threshold, the power generation coverage threshold, etc.) are default values. The specific values ​​specified in the text are only one case, and those skilled in the art can set them according to actual applications.

[0146] like Figure 5 As shown, this invention provides a multi-source weather forecast-driven solar energy storage scheduling system for streetlights that implements the above-described method. The system includes:

[0147] Central Dispatch Server: The "brain" of the system, responsible for executing the core algorithms of S2, S3, and S6, generating dispatch instructions. It is mainly used to generate dispatch instructions for the energy storage system based on integrated hourly power generation forecasts and forecast confidence levels, combined with load forecasts and the current status of the energy storage system. The strategy for formulating dispatch instructions is dynamically adjusted according to the forecast confidence level.

[0148] Data acquisition module: responsible for communicating with external meteorological data sources and executing the data acquisition task of S1, that is, acquiring hourly meteorological forecast data for a preset period from multiple heterogeneous meteorological data sources;

[0149] Multi-source fusion and learning module: Usually integrated into the central dispatch server in software form, it includes feature extraction, data fusion, forecast confidence calculation and online learning units. It is mainly used to fuse forecast data from various data sources to generate integrated hourly power generation forecasts and corresponding forecast confidence scores.

[0150] The adaptive learning module is used to adaptively update the processing strategy of the data fusion and prediction module based on the error between actual operating data and hourly weather forecast data, forming a closed-loop adaptive scheduling.

[0151] Execution and monitoring module: includes specific power electronic devices, such as bidirectional DC-DC converters, AC chargers, photovoltaic panels, energy storage batteries, etc., used to execute the scheduling instructions and collect actual operating data;

[0152] Local controller: An embedded device deployed in each street light, responsible for receiving and executing central commands, and implementing conservative strategies in abnormal situations. It typically integrates a BMS (Battery Management System) interface, an MPPT (Maximum Power Point Tracking) controller interface, and a DC-DC converter / AC charger control interface;

[0153] Operations and maintenance database: Used to store all historical forecast data, actual operation data, model parameters, historical weights and event logs, supporting data backtracking, auditing and manual intervention.

[0154] Optional, regional dispatch coordination module: used to coordinate the charging and discharging behavior of multiple streetlights at the regional level to ensure that the total power does not exceed the grid connection limit of the upstream transformer or power grid.

[0155] As an embodiment 1 of the present invention: Street light single-station scheduling based on the method of the present invention

[0156] Reference Figure 1 This embodiment uses a single photovoltaic-storage street light site as an example to explain in detail the execution process of this method. The street light is equipped with a 200W photovoltaic panel, a 1kWh lithium energy storage battery, and a 5G micro base station load.

[0157] S1 obtains hourly weather forecast data for a preset time period from multiple heterogeneous meteorological data sources;

[0158] At 23:00 on a summer evening (the start of a pre-defined trough time window), the central dispatch server, through its data acquisition module, simultaneously initiated API calls to three meteorological data sources: the China Meteorological Administration (CMA), the public data interface of the European Centre for Medium-Range Weather Forecasts (ECMWF), and a commercial weather service provider. The requested data was the hourly weather forecast for the following day (24 hours) based on the latitude and longitude coordinates of the streetlights, with key fields including GHI (Total Horizontal Irradiance) and cloud cover.

[0159] S2 integrates hourly weather forecast data from various meteorological data sources to generate integrated hourly power generation forecasts and corresponding forecast confidence levels.

[0160] After receiving JSON responses from the three data sources, the central dispatch server first parses and aligns the timestamps.

[0161] For example, unify all data to the Beijing time zone and use the hour as the timestamp. Then extract the total horizontal irradiance value (unit: W / m²) and cloud cover percentage (%) for each data source and each hour. Then organize the above data into a structured data table and store it in a temporary database.

[0162] Assume that the current fusion weights of the three data sources (CMA, ECMWF, and commercial source) are 0.5, 0.3, and 0.2, respectively (these weights were learned through S6 over a period of time).

[0163] For the time "tomorrow at 12:00", the GHI values ​​predicted by the three data sources are 800 W / m², 750 W / m², and 820 W / m², respectively. Therefore, the ensemble predicted GHI value = 800 × 0.5 + 750 × 0.3 + 820 × 0.2 = 789 W / m².

[0164] The same calculations were performed for all 24 hours to obtain the integrated hourly power generation prediction curve.

[0165] Forecast confidence calculation:

[0166] Source consistency: Calculate the standard deviation of the three GHI values, assuming it is 35 W / m². According to the preset mapping relationship, this item scores 0.7.

[0167] Historical accuracy score: The RMSE of three data sources at 12:00 over the past 30 days was queried, and the overall score for this item was calculated to be 0.8.

[0168] Short-term observation correction response capability: The deviation between the ultra-short-term forecast and the actual observation was checked from 6:00 to 10:00 a.m. on the same day. It was found that the deviation of each source was small, and the score for this item was 0.9.

[0169] Seasonal / Regional Adaptability: It is currently summer. According to historical statistics, commercial sources perform better in predicting convective weather in summer. Therefore, this item has a weighted score of 0.85.

[0170] Finally, the four scores are weighted and averaged. Assuming the weights of the four dimensions are 0.3, 0.4, 0.2, and 0.1 respectively, the final prediction confidence is: 0.7×0.3 + 0.8×0.4 + 0.9×0.2 + 0.85×0.1 = 0.785.

[0171] S3 Based on the integrated hourly power generation forecast and the forecast confidence level, combined with the load forecast and the current state of the energy storage system, a scheduling instruction for the energy storage system is generated, wherein the strategy for formulating the scheduling instruction is dynamically adjusted according to the forecast confidence level; S4 Execute the scheduling instruction and collect actual operating data; S5 Based on the error between the actual operating data and the hourly weather forecast data, the strategy of the fusion processing is adaptively updated to form a closed-loop adaptive scheduling.

[0172] Integrated hourly power generation forecasts indicate a total power generation of approximately 2.5 kWh for the following day, with a confidence level of 0.785. The projected load for 5G micro base stations and lighting for the following day is 1.8 kWh. The current battery SOC is 40%.

[0173] Calculation of energy storage suitability: Based on a maximum battery capacity of 1kWh, the current capacity is 0.6kWh. Assuming no grid connection restrictions and a minimum standby SOC of 20%. Calculations show that the expected net power generation the following day (2.5 - 1.8 = 0.7kWh) is greater than the energy storage capacity (0.6kWh), and the forecast confidence level is greater than the forecast confidence level threshold. Therefore, the "energy storage suitability" is determined to be high (suitable).

[0174] Based on a high suitability and a confidence level (0.785) higher than the preset threshold (0.6), the system is deemed "suitable." At this point, a "photovoltaic charging-centric" strategy is generated, i.e., the second type of scheduling instruction: The target end SOC is set at 40% (a relatively low value, while reserving 60% capacity for daytime photovoltaic power). A charging power curve is generated: during off-peak hours (23:00-5:00), charging at 200W for 3 hours can charge from 40% to 50%. The instruction is valid until 5:00 the next day and allows the local controller to fine-tune within ±5% of the SOC based on real-time conditions.

[0175] If the charging suitability is determined to be "unsuitable" or "unreliable", the corresponding dynamic charging adjustment control will be carried out in accordance with the above-mentioned scheduling instructions, which will not be described in detail here.

[0176] S6 adaptive update uses a sliding window strategy to save the error sequence of the most recent N days, and calculates the mean square error and deviation trend of each meteorological data source based on the error sequence. It then uses a smoothing factor to update the weight of each data source in the fusion process.

[0177] The following day, the system collected actual meteorological observation data (such as from local radiometers) and calculated actual power generation. It found that the ECMWF source was most closely predicted during the midday period, while the CMA source was slightly overestimated.

[0178] Error calculation is then performed: that is, within a 30-day sliding window, the new mean square error of each source is calculated.

[0179] Based on the new error, the new theoretical weights are calculated to be (CMA: 0.45, ECMWF: 0.35, Commercial Source: 0.20); then updated using a smoothing factor α = 0.1.

[0180] New weight _CMA = 0.9 × 0.5 + 0.1 × 0.45 = 0.495

[0181] New weight _ECMWF = 0.9 × 0.3 + 0.1 × 0.35 = 0.305

[0182] New weight_Business = 0.9 × 0.2 + 0.1 × 0.20 = 0.200

[0183] The weights were updated smoothly, and the weights of ECMWF were slightly increased.

[0184] Local conservative strategy;

[0185] Suppose that one day, the central dispatch server is temporarily paralyzed due to a network attack and is unable to issue commands. After the command validity period expires and no new commands are received, the local controller automatically triggers a conservative strategy: stopping all grid-connected discharge activities.

[0186] If the current SOC is 35%, which is higher than the minimum emergency SOC (20%), then power the streetlights as originally planned.

[0187] If the SOC is below 20%, lighting power will be limited or the micro base station load will be shut down to prioritize basic lighting.

[0188] As an embodiment 2 of the present invention: Regional cooperative scheduling based on the system of the present invention

[0189] This embodiment describes a park-level system containing hundreds of streetlights.

[0190] The system architecture includes a central dispatch server that acquires data from multiple meteorological data sources via a data acquisition module. A multi-source fusion and learning module runs within the server, generating a unified, integrated hourly power generation forecast and forecast confidence level for the entire industrial park.

[0191] Based on predictions, the central dispatch server initially generates individual dispatch instructions for each street light. Subsequently, the regional coordination module begins operation. By summarizing the planned charging power of all street lights, it is found that if all operations are carried out according to plan, the peak power at the beginning of the off-peak period will reach 500kW, exceeding the 400kW safety limit of the park's transformers.

[0192] At this point, the regional coordination module activates the optimization algorithm to "shave peaks and fill valleys" in the charging sequence without significantly affecting the final target SOC of each street light.

[0193] For example, the charging start time for some streetlights was delayed by one hour, and the charging power for another group was reduced from 5kW to 4kW. After optimization, the total charging power curve was smoothed, and the peak value was controlled within 380kW.

[0194] The optimized instruction set is distributed to the local controllers of each street light. Each local controller monitors the battery status through its BMS interface, manages the photovoltaic input through the MPPT controller interface, and performs precise charge and discharge control through the DC-DC converter / mains charger control interface, according to the instructions.

[0195] All operational data ultimately flows back to the operations and maintenance database, where it is used by the multi-source fusion and learning module to update weights and provide administrators with visual reports and audit trails. The entire system forms a perfect closed loop from prediction, collaborative decision-making, distributed execution to centralized learning.

[0196] It should be noted that the performance of this invention under boundary conditions is as follows: (1) When facing severe convective weather or sudden extreme weather: the forecast confidence drops rapidly, and this system prioritizes switching to a conservative strategy and retains more backup SOCs to cope with possible photovoltaic power generation shortages.

[0197] (2) When a single data source fails or the network is abnormal: the system will automatically downgrade to the remaining sources for fusion and appropriately reduce the confidence level. If all sources are unavailable, the system will directly enter the local conservative strategy.

[0198] (3) When under multi-site grid constraints: The regional coordinated dispatch module prioritizes ensuring the SOC of nighttime lighting-related sites based on constraints such as transformer capacity, and restricts discharge or postpones charging of non-critical sites when necessary.

[0199] The parameters for embodiments of the present invention are as follows:

[0200] Off-peak charging window: The default is from 23:00 to 5:00 the next day, but it can also be set to any off-peak electricity price period according to actual implementation requirements or conditions;

[0201] Forecast confidence threshold: 0.6 (default);

[0202] Power generation coverage threshold: 70% (default);

[0203] Minimum emergency SOC: 25%, Maximum charging SOC (primarily for reducing battery cycle time): 95%

[0204] Retrospective window: 30 days; Short-term correction window: 7 days

[0205] Smoothing factor (weight update): 0.8 (old weights) + 0.2 (new estimated weights);

[0206] The above parameters should be adjusted by the operations and maintenance or system administrator as configurable items. It should be noted that the above values ​​are for implementation reference only, and the specific values ​​may be adjusted according to the on-site conditions and policies.

[0207] Among them, the adequacy judgment threshold, confidence threshold, and conservative strategy trigger threshold all adopt a closed-loop adjustment mechanism rather than fixed values; the system uses historical backtracking as a basis to statistically analyze the performance of the judgment strategy under different weather scenarios on a weekly or monthly basis, such as the actual power supply achievement rate after adequacy judgment and the number of times the backup SOC is insufficient due to misjudgment.

[0208] If too many misjudgments lead to a decline in lighting quality, the conservatism threshold will be gradually increased, for example, the confidence threshold will be increased from 0.6 to 0.65.

[0209] If the system runs stably for a long time and the economic benefits are good, the conservative threshold can be gradually reduced to improve economic efficiency. The threshold adjustment strategy records the threshold change log and can be manually rolled back.

[0210] like Figure 6 The diagram illustrates the structure of a computer server according to an embodiment of this application. This server can vary significantly due to differences in configuration or performance, and may include one or more processors and one or more memories. The one or more memories store at least one computer program, which is loaded and executed by the one or more processors to implement the methods provided in the various method embodiments described above. Of course, the server may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The server may also include other components for implementing device functions, which will not be elaborated upon here.

[0211] To ensure the feasibility of the system, the key data interfaces and fields are given below:

[0212] The data acquisition module should connect to various sources and support the following fields: source identifier, timestamp and time zone, geographic coordinates (longitude, latitude, altitude / grid offset), hourly irradiance forecast (corresponding hour array), cloud cover / precipitation probability / temperature sequence, forecast validity period and model running time, confidence level / source metadata (e.g., number of model set members).

[0213] 2) Local operation reporting interface; for example, the local controller-central dispatch server should at least include:

[0214] Equipment identification (e.g., street light group ID, controller ID, etc.), time-series telemetry: hourly power generation, charging / discharging power, SOC, BMS temperature, operating mode, alarm / event log (including event time, level and equipment screenshot), local observation: if there is a ground radiometer or local weather sensor, upload the observation values.

[0215] Operation confirmation: The local controller confirms the receipt and execution of instructions issued by the central dispatch server.

[0216] 3) The format of the scheduling command issued by the central dispatch server includes: command ID and version, target execution time window (start and end time), target SOC or power curve (hourly granularity or finer), priority (urgent, high, normal), adjustable parameters (e.g., the range that allows the local controller to adjust autonomously for certain anomalies), instruction validity period and rollback conditions, and signature / authentication information (to ensure the credibility of the instruction source).

[0217] 4) Historical database storage elements: raw meteorological data and fusion results (retaining source identifiers), comparison of forecasts and actual power generation (hourly), dispatch decision records (key input factors and output instructions), and history of weights and model versions (for auditing purposes).

[0218] As an implementation of the present invention, the deployment process includes the following steps:

[0219] Site survey: Obtain information on street light installed capacity, energy storage parameters, communication conditions, and mains power contract terms;

[0220] Initial parameter calibration: Import historical meteorological data and historical power generation observations to perform initial multi-source weight estimation and forecast confidence baseline setting;

[0221] Trial run phase (generally recommended to be at least 30 days): Run in safe mode (e.g., first use suggested instructions and then manually confirm), collect actual errors for model calibration;

[0222] Online optimization: Enable online weight learning and gradually transition threshold adjustment from manual control to automatic closed-loop;

[0223] Operations and maintenance training: Explaining the fault ticket process, manual intervention methods for thresholds, and daily anomaly analysis to operations and maintenance personnel;

[0224] Version management: All algorithms and decision rules should be version controlled and change records should be kept. Important changes should be subject to regression testing.

[0225] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0226] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0227] The above description is merely one or more embodiments of this specification and is not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included in the claims of this specification.

Claims

1. A photovoltaic-storage scheduling method driven by multi-source weather forecasting for streetlights, characterized in that, The method includes the following steps: S1 obtains hourly weather forecast data for a preset time period from multiple heterogeneous meteorological data sources; in S1, during the electricity price trough time window within the preset time period, hourly weather forecast data for the next day is requested in parallel from multiple heterogeneous meteorological data sources. S2 fuses hourly weather forecast data from various meteorological data sources to generate integrated hourly power generation forecasts and corresponding forecast confidence levels; the fusion process in S2 includes the following steps: S21 performs time-series alignment and feature extraction on hourly weather forecast data from each of the meteorological data sources, and extracts hourly total horizontal irradiance and cloud cover parameters related to photovoltaic power generation. S22 assigns initial weights to the hourly weather forecast data of each meteorological data source based on the historical forecast errors and short-term correction factors of each meteorological data source and performs weighted fusion to obtain integrated hourly power generation forecasts. S23 calculates the forecast confidence level of the integrated hourly power generation forecast, wherein the forecast confidence level is calculated by weighting multiple factors among source consistency, historical accuracy scores of each source, short-term observation correction response capability, and seasonal / regional adaptability; S3 generates a scheduling instruction for the energy storage system based on the integrated hourly power generation forecast and the forecast confidence level, combined with the load forecast and the current state of the energy storage system. The scheduling instruction formulation strategy is dynamically adjusted according to the forecast confidence level. S3 includes: calculating the sufficiency of photovoltaic power generation based on the integrated hourly power generation forecast and the forecast confidence level, combined with the load forecast and the current state of the energy storage system. The sufficiency is constrained by the energy storage's acceptable space, grid connection limitations, and minimum standby SOC. The calculation of the fitness includes: Based on the integrated hourly power generation forecast, the estimated total power generation for the next day is calculated. Estimate the total load demand for the next day based on the street light activation schedule and historical load curves; The estimated net power generation for the next day is calculated by subtracting the total load demand for the next day from the estimated total power generation for the next day. After calculating the difference between the maximum charging SOC limit and the current SOC, multiply it by the energy storage capacity to obtain the energy storage capacity. The estimated net power generation for the next day is divided by the total load demand for the next day to obtain the power generation coverage rate. Based on the forecast confidence level, the power generation coverage rate, and the energy storage capacity, the adequacy status is determined. The logic for determining the suitability status includes: If the prediction confidence level is less than the prediction confidence threshold, it is determined to be an unreliable state; If the prediction confidence level is greater than or equal to the prediction confidence threshold... If the power generation coverage rate is greater than or equal to the power generation coverage rate threshold and the energy storage capacity is not less than the energy storage capacity threshold, then it is determined to be a suitable state. Otherwise, it is judged as an unsuitable state; Based on the aforementioned energy availability status, a scheduling instruction for the energy storage system is generated: When the state is determined to be unsuitable, a first-class scheduling instruction is generated, which mainly uses mains power charging, and a higher target SOC is set. When the condition is determined to be suitable, a second type of scheduling instruction is generated, which is mainly based on photovoltaic charging, and a lower target SOC is set. When the system is determined to be in an unreliable state, a third type of dispatch instruction is generated, which mainly uses mains power for charging, and a medium target SOC is set. S4 executes the scheduling instructions and collects actual operating data; S5 adaptively updates the fusion processing strategy based on the error between the actual operating data and the hourly weather forecast data, forming a closed-loop adaptive scheduling; S6 When the central dispatch fails or the forecast confidence is extremely low, switch to the local conservative control strategy. The local conservative control strategy ensures that the energy storage SOC is not lower than the preset minimum emergency value and prioritizes meeting the nighttime lighting load. The scheduling instructions include the target end of SOC, the charging power curve for different time periods, the instruction validity period, and the allowed range for local adjustments.

2. The method according to claim 1, characterized in that, The multiple heterogeneous meteorological data sources include at least two or more of the following: forecast data from the China Meteorological Administration Data Center, global numerical weather prediction products, and commercial converged weather platforms.

3. The method according to claim 1, characterized in that, The adaptive update in S5 adopts a sliding window strategy, saves the error sequence of the most recent N days, calculates the mean square error and deviation trend of each meteorological data source based on the error sequence, and uses a smoothing factor to update the weight of each data source in the fusion process.

4. A multi-source weather forecast-driven photovoltaic-storage scheduling system for streetlights, based on the multi-source weather forecast-driven photovoltaic-storage scheduling method for streetlights as described in any one of claims 1-3, characterized in that, The system includes: The data acquisition module is used to acquire hourly weather forecast data for a preset time period from multiple heterogeneous meteorological data sources; The multi-source fusion and learning module is used to fuse forecast data from various data sources to generate integrated hourly power generation forecasts and corresponding forecast confidence levels. The central dispatch server is used to generate dispatch instructions for the energy storage system based on the integrated hourly power generation forecast and forecast confidence level, combined with load forecast and the current status of the energy storage system, wherein the strategy for formulating the dispatch instructions is dynamically adjusted according to the forecast confidence level. The execution and monitoring module is used to execute the scheduling instructions and collect actual operation data; An adaptive learning module is used to adaptively update the processing strategy of the data fusion and prediction module based on the error between the actual operating data and the integrated hourly weather forecast data, forming a closed-loop adaptive scheduling.

5. The system according to claim 4, characterized in that, The system also includes a local controller, which is configured to execute a preset conservative strategy when communication with the central scheduling server is interrupted or the confidence level of the received instruction is extremely low, and has a BMS interface, an MPPT controller interface and a charge / discharge control interface to execute local safety protection logic. Operations and maintenance database: used to store all historical forecast data, actual operation data, model parameters, historical weights and event logs, supporting data backtracking, auditing and manual intervention; The system also includes a regional coordination module, which coordinates the charging and discharging behavior of multiple street light sites at the regional level, so that the total grid-connected power of the region does not exceed the transformer capacity or grid connection limit.

Images