Optimal dispatching method for photovoltaic power storage system in computer room based on load demand

Abstract

This invention discloses an optimized scheduling method for a data center photovoltaic (PV) energy storage system based on load demand, belonging to the field of energy dispatching technology. The method includes: collecting load data, operational constraint data, and available PV power from the data center PV energy storage system; generating a unified timestamp and constructing a unified state record written to a historical cache; performing feasibility correction on the proposed energy storage power; outputting and issuing energy storage power commands; and simultaneously generating execution feedback information and correction information based on the load data and available PV power; and forming candidate evaluation sample records and strategy replay records based on the execution feedback information and correction information, respectively, and writing them to the historical cache. This invention calculates the gap-margin factor within a rolling window for deterministic preference mapping screening, ensuring that the strategy satisfies SOC and power constraints and is consistently reproducible, reducing jumps and corrections, improving PV absorption, and suppressing fluctuations in levelized cost of electricity (LCOE).

Description

Technical Field

[0001] This invention relates to the field of energy dispatching technology, and in particular to an optimized dispatching method for a data center photovoltaic energy storage system based on load demand. Background Technology

[0002] In recent years, data center server rooms have gradually been equipped with photovoltaic (PV) and energy storage systems to form an integrated power supply system, with energy management evolving from traditional fixed-value charging and discharging to online optimization control oriented towards rolling windows. Related technologies are typically based on load monitoring, PV available power prediction, and SOC state estimation, combined with time-synchronous sampling, operational constraint modeling, and multi-objective optimization and closed-loop feedback mechanisms to achieve coordinated scheduling of electricity purchase, battery charging and discharging, and PV consumption, and are adapted to operational scenarios such as time-of-use pricing and demand management.

[0003] Existing data center rolling scheduling systems for photovoltaic-storage are mostly based on point prediction and fixed-weight multi-objective functions. They generally lack quantitative characterization of supply-demand gaps and operational margins under uncertainties in load and photovoltaic output. This makes prediction errors sensitively amplified in the constrained neighborhood of SOC boundaries and charging / discharging power limits. At the same time, strategy selection often relies on heuristic rules, making it difficult to form reproducible and consistent decision criteria. This leads to policy jumps and conservative / aggressive switching within the rolling time domain, making it difficult to stably balance curtailment suppression and electricity purchase cost control. Summary of the Invention

[0004] To address the problems mentioned in the background section, the present invention provides the following technical solution:

[0005] This invention provides an optimized scheduling method for a data center photovoltaic energy storage system based on load demand, which includes collecting load data, operating constraint data and available photovoltaic power of the data center photovoltaic energy storage system, generating a unified timestamp, and constructing a unified status record and writing it into a historical cache.

[0006] Based on historical cache, conservative boundary predictions of the upper and lower limits of load and photovoltaic power are made for load data and available photovoltaic power in the future rolling window, generating a conservative prediction sequence and calculating the gap-margin factor.

[0007] Based on the conservative prediction sequence and operational constraint data, a constrained multi-objective candidate search is performed within the future rolling window to generate a Pareto candidate policy set. Then, a deterministic preference mapping is performed using the gap-margin factor to filter the candidate subset.

[0008] Perform candidate fine-tuning within the candidate subset, determine the reference strategy, and output a limited energy storage power fine-tuning on the current energy storage power to obtain the energy storage power to be executed;

[0009] The feasibility of the proposed energy storage power is revised, and the energy storage power command is output and issued. At the same time, the execution feedback information and correction information are generated by combining load data and available photovoltaic power.

[0010] Based on the execution feedback information and correction information, candidate evaluation sample records and strategy replay records are generated and written to the historical cache.

[0011] As a preferred embodiment of the load demand-based optimized scheduling method for a data center photovoltaic energy storage system according to the present invention, the specific steps for constructing a unified state record and writing it into the historical cache are as follows:

[0012] The system acquires load data, available photovoltaic power, and operational constraint data within the same acquisition period and generates a unified timestamp. The load data includes rigid loads and adjustable loads, and the operational constraint data includes energy storage operating status, energy storage state of charge, energy storage power boundary, state of charge boundary, electricity price, and derating alarm.

[0013] The load data, available photovoltaic power, and operational constraint data are aligned according to a unified timestamp, and missing items are filled in. The aligned load data, available photovoltaic power, and operational constraint data are merged to generate a unified status record, which is then written to the historical cache in ascending order of the unified timestamp.

[0014] As a preferred embodiment of the load demand-based optimized scheduling method for a data center photovoltaic energy storage system according to the present invention, the following steps are taken: based on historical buffering, conservative boundary predictions are performed on the load data and available photovoltaic power within the future rolling window to predict the upper and lower bounds of the load and photovoltaic power, generating a conservative prediction sequence and calculating the gap-margin factor.

[0015] Extract the current unified timestamp based on the historical cache, and generate a historical sequence by tracing back the set historical length. Then, perform validity screening and missing data filling on the historical sequence to obtain a valid historical sequence.

[0016] Based on the effective historical sequence, load data historical samples for the corresponding time period are extracted according to the unified timestamp of the future rolling window, and the upper limit of rigid load and the upper limit of adjustable load are determined according to the preset quantile level, and the load conservative boundary is synthesized.

[0017] Based on valid historical sequence data, the historical samples of photovoltaic available power for the corresponding time period are extracted according to the unified timestamp of the future rolling window. The minimum value of the historical samples of photovoltaic available power is taken as the next photovoltaic conservative boundary. The reduction ratio is determined according to the reduction alarm, and the ratio limit is modified for the photovoltaic conservative boundary.

[0018] The conservative load boundary and the conservative photovoltaic boundary are aligned and combined according to the unified timestamp of the future rolling window to generate a conservative forecast sequence.

[0019] Based on the conservative prediction sequence, the power difference between the photovoltaic conservative boundary and the load conservative boundary is calculated for each time period. After conversion by time period, the difference is accumulated. The maximum positive value of the accumulated value is the maximum margin, and the minimum negative value of the accumulated value is recorded as the maximum gap. The gap-margin factor is then output.

[0020] As a preferred embodiment of the load demand-based optimized scheduling method for a data center photovoltaic energy storage system according to the present invention, the specific steps for generating the Pareto candidate strategy set are as follows:

[0021] Based on the conservative prediction sequence, energy storage operating status and energy storage power boundary, determine the conservative load boundary, photovoltaic conservative boundary and energy storage power feasible range within the future rolling window, and generate multiple sets of candidate energy storage power sequences within the feasible range of energy storage power.

[0022] Perform operational constraint verification and boundary correction on the candidate energy storage power sequence to obtain a set of feasible candidate strategies;

[0023] Based on the set of feasible candidate strategies, the changes in energy storage status of feasible candidate strategies are deduced over time periods and energy balance calculations are performed to obtain the purchased electricity and curtailed solar power corresponding to each feasible candidate strategy. The electricity purchase cost is then calculated based on the electricity price, forming a multi-objective evaluation result.

[0024] The dominance relationship is compared on the multi-objective evaluation results, the dominated strategies are eliminated and the non-dominated strategies are retained, and a Pareto candidate strategy set is generated.

[0025] As a preferred embodiment of the load demand-based optimized scheduling method for a data center photovoltaic energy storage system according to the present invention, the specific steps for obtaining the candidate subset are as follows:

[0026] The preference direction is determined based on the relationship between the maximum gap and the maximum margin in the gap-margin factor, and preference mapping rules are set according to the preference direction. When the gap takes priority, the electricity purchase cost is mapped to the first preference amount and the curtailed solar power is mapped to the second preference amount; when the margin takes priority, the curtailed solar power is mapped to the first preference amount and the electricity purchase cost is mapped to the second preference amount.

[0027] The preference mapping rule is used to map the multi-objective evaluation results of the Pareto candidate policy set to obtain the preference mapping results of each candidate policy.

[0028] The preference mapping results are deterministically sorted and quantitatively compared under preset filtering conditions to output candidate subsets.

[0029] As a preferred embodiment of the load demand-based optimized scheduling method for a data center photovoltaic energy storage system described in this invention, the following steps are taken: Candidate fine-tuning is performed within a candidate subset, a candidate strategy is deterministically selected as a reference strategy, and a limited-amplitude energy storage power fine-tuning is output based on the current energy storage power to obtain the energy storage power to be executed.

[0030] Based on the current unified timestamp and the current energy storage power, the absolute difference between the candidate energy storage power and the current energy storage power for the corresponding time period is read from the candidate subset one by one, and the candidate strategy with the smallest absolute difference is taken as the reference strategy.

[0031] Based on the relationship between the candidate energy storage power corresponding to the reference strategy and the current energy storage power, the fine-tuning direction is determined and the energy storage power fine-tuning amount is formed. After the energy storage power fine-tuning amount is subjected to amplitude limiting processing, it is superimposed with the current energy storage power to obtain the energy storage power to be executed.

[0032] As a preferred embodiment of the load demand-based optimized scheduling method for a data center photovoltaic energy storage system according to the present invention, the specific steps for performing feasibility correction on the proposed energy storage power and outputting the energy storage power command are as follows:

[0033] Based on the historical cache, the energy storage state of charge and state of charge boundary corresponding to the current unified timestamp are read, and the energy storage state of charge is estimated by combining the energy storage power to be executed.

[0034] When the calculated state of charge of the energy storage exceeds the state of charge boundary, the corrected energy storage power is determined as the energy storage power command; when the calculated state of charge of the energy storage does not exceed the state of charge boundary, the energy storage power to be executed is determined as the energy storage power command, and the energy storage power command is output and issued.

[0035] As a preferred embodiment of the load demand-based optimized scheduling method for a data center photovoltaic energy storage system according to the present invention, the specific steps for outputting execution feedback information and correction information are as follows:

[0036] Energy balance calculations are performed based on load data, available photovoltaic power, and energy storage power commands to obtain the purchased electricity and the amount of curtailed photovoltaic power.

[0037] The unified timestamp, energy storage power instruction, purchased electricity volume, curtailed solar power volume and electricity purchase cost are collected to form execution feedback information;

[0038] The difference between the energy storage power to be executed and the energy storage power command is calculated as the power difference value, and then aggregated with the current unified timestamp to form correction information.

[0039] As a preferred embodiment of the load demand-based optimized scheduling method for a data center photovoltaic energy storage system according to the present invention, the execution feedback information and correction information respectively form candidate evaluation sample records and strategy replay records. The specific steps are as follows.

[0040] Based on the execution feedback information, extract the unified timestamp of the execution cycle, the power value of the energy storage power instruction, the purchased electricity and the curtailed solar power, generate candidate evaluation sample records, and write them into the historical cache;

[0041] Based on the correction information, the power difference is extracted and associated with the reference policy to form a policy replay record, which is then written to the historical cache.

[0042] As a preferred embodiment of the load demand-based optimized scheduling method for a data center photovoltaic energy storage system described in this invention, the preset screening condition refers to screening candidate strategies based on the quantitative preference value of the top-ranked candidate strategy after the Pareto candidate strategy set has been deterministically sorted, and retaining candidate strategies with consistent quantitative preference values ​​to form a candidate subset.

[0043] The beneficial effects of this invention are as follows:

[0044] By incorporating quantifiable indicators into the strategy ranking process to mitigate supply and demand risks under uncertain disturbances, the reliance on empirical weighting and manual parameter tuning is reduced. This allows for clear criteria and stable output in candidate strategy selection. Within a rolling window, the gap-margin factor is calculated for deterministic preference mapping and screening, ensuring that strategies meet SOC and power constraints and are consistently reproducible. This reduces jumps and corrections, improves photovoltaic absorption, and suppresses fluctuations in the levelized cost of electricity. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart of an optimized scheduling method for a data center photovoltaic energy storage system based on load demand.

[0047] Figure 2 This is a flowchart for calculating the gap-margin factor.

[0048] Figure 3 A flowchart for generating a Pareto candidate policy set.

[0049] Figure 4 A flowchart for determining the reference strategy. Detailed Implementation

[0050] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0051] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0052] Reference Figures 1-4 This is one embodiment of the present invention, which provides an optimized scheduling method for a data center photovoltaic energy storage system based on load demand, including the following steps:

[0053] S1: Collect load data, operating constraint data, and available photovoltaic power of the photovoltaic energy storage system in the computer room, generate a unified timestamp, and construct a unified status record to write to the historical cache.

[0054] S1.1: Acquire load data, available photovoltaic power and operating constraint data within the same acquisition period, and generate a unified timestamp. The load data includes rigid load and adjustable load, and the operating constraint data includes energy storage working status, energy storage state of charge, energy storage power boundary, state of charge boundary, electricity price and derating alarm.

[0055] Set a fixed collection cycle duration (default 15 minutes, configurable), with the end time boundary of each collection cycle as the trigger time. The collection terminal and each data source device use the same time zone and perform network time synchronization and periodic time synchronization through the NTP protocol to ensure consistency across data times.

[0056] When the trigger time arrives, the trigger time is aligned forward to the nearest collection cycle time boundary no later than the trigger time to obtain a unified timestamp.

[0057] The acquisition time range is determined based on a unified timestamp, which is the duration of one collection cycle prior to the unified timestamp up to the unified timestamp.

[0058] Within the acquisition time frame, load data, available photovoltaic power, and operational constraint data are acquired separately.

[0059] Validity filtering is performed on each field sampling point. When the sampling point timestamp falls within the acquisition time range and the field value can be parsed as a number or mapped to a dictionary value, the field validity is determined. When the sampling point timestamp does not fall within the acquisition time range, or the field value cannot be parsed as a number and cannot be mapped to a dictionary value, it is determined as an invalid sampling point.

[0060] Power-related fields include rigid load, adjustable load, and available photovoltaic power. When a field value is not empty, not less than zero, and does not exceed the upper limit of the rated power of the corresponding equipment, it is determined to be a valid sampling point; otherwise, it is determined to be an invalid sampling point. The upper limit of the rated power of the equipment is determined according to the equipment nameplate.

[0061] The energy storage operating status field is considered a valid sampling point when the field value falls within the range of status codes defined in the status dictionary table; otherwise, it is considered an invalid sampling point. The status dictionary table is configured according to the energy storage device communication protocol and includes status name and status code fields, as well as charging status code, discharging status code, and standby status code.

[0062] For the energy storage state of charge field, a valid sampling point is determined when the field value meets the format requirements and is not less than zero and not greater than one. When the state of charge boundary is a valid sampling point within the same acquisition period, it should also be within the range of the upper and lower limits of the state of charge; otherwise, it is determined as an invalid sampling point.

[0063] Boundary fields include energy storage power boundaries and state of charge boundaries. A field is considered a valid sampling point if its value meets the boundary format requirements, its lower bound is not greater than its upper bound, and its value does not exceed the allowed configuration range; otherwise, it is considered an invalid sampling point.

[0064] The allowable configuration range is determined by the rated capacity of the energy storage converter, the allowable charge range of the battery, and the protection limit, and is configured through the power boundary parameter table and the state of charge boundary parameter table; the power boundary parameter table includes a lower power field and a higher power field, and the state of charge boundary parameter table includes a lower charge field and a higher charge field.

[0065] The electricity price field is considered a valid sampling point if the field value is not empty, meets the price format requirements, and is within the valid electricity price range; otherwise, it is considered an invalid sampling point.

[0066] The effective range of electricity price and the time-of-use electricity price table are configured according to the standards published by the electricity price platform and can be updated according to the standards. The time-of-use electricity price table includes the start time of the time period, the end time of the time period, and the electricity price field.

[0067] The derating alarm field is considered a valid sampling point when its value maps to the alarm level in the alarm dictionary table; otherwise, it is considered an invalid sampling point.

[0068] The alarm dictionary table is used to define the correspondence between alarm levels and derating ratios, and includes fields for alarm level and derating ratio. The derating ratio can be maintained and updated, and the derating ratio does not decrease as the alarm level increases. The derating ratio value range is set to zero to one.

[0069] Calculate the arithmetic mean of the effective sampling points for power-related fields, and output the representative values ​​of the rigid load sampling period, the adjustable load sampling period, and the photovoltaic available power sampling period.

[0070] For the data field of the operation constraint, select the valid sampling point whose time is closest to and no later than the unified timestamp within the acquisition time range, as the valid value at the end of the acquisition period, and use it as the representative value of the acquisition period of the corresponding field. If there is no valid sampling point for a certain field within the acquisition time range, the representative value of the field acquisition period is set to empty.

[0071] It should be noted that the energy storage operating state and the energy storage state of charge are collectively referred to as the energy storage state, and the energy storage power boundary and the state of charge boundary are collectively referred to as the energy storage boundary.

[0072] S1.2: Align the load data, available photovoltaic power and operating constraint data according to a unified timestamp, and perform fill-in processing on missing items. Merge the aligned load data, available photovoltaic power and operating constraint data to generate a unified status record, and write it into the historical cache in ascending order of the unified timestamp.

[0073] The representative values ​​of each field under the same unified timestamp are grouped into the same set. When the representative value of a field's collection period is empty or there are no valid sampling points within the acquisition time range, the field is marked as a missing item.

[0074] Missing items are filled in using a fixed filling order and fixed filling rules. The fixed filling order is: rigid load, adjustable load, available photovoltaic power, energy storage operating status, state of charge boundary, energy storage state of charge, energy storage power boundary, electricity price and derating alarm. The fixed filling rules are as follows:

[0075] When rigid loads are missing, the upper limit of the rated power of the corresponding load equipment is taken as the compensation value;

[0076] When adjustable load is missing, the upper limit of the rated power of the corresponding load equipment shall be used as the compensation value.

[0077] When the available photovoltaic power is missing, it is taken as zero. The filling is a conservative filling to avoid recording an overly optimistic record due to the lack of photovoltaic power.

[0078] When the energy storage working state is missing, the standby state code is used;

[0079] When the charged state boundary is missing, the allowable range recorded in the charged state boundary parameter table is taken as the filler value of the charged state boundary.

[0080] When the state of charge (SOC) of the energy storage is missing, the arithmetic mean of the lower and upper bounds of the SOC is taken as the filler value to avoid the filler value being fixedly biased towards the lower or upper bound, which would introduce an unfounded bias in the charging and discharging directions. When it is necessary to limit the discharge risk, the lower bound of the SOC is taken as the filler value, and when it is necessary to limit the charging risk, the upper bound of the SOC is taken as the filler value, and the filler SOC value is limited to the range of the SOC boundary.

[0081] When the energy storage power boundary is missing, the allowable range recorded in the power boundary parameter table is used as the filler value;

[0082] When the electricity price is missing, the electricity price corresponding to the time-of-use electricity price table at the same timestamp is used; when the time-of-use electricity price table is unavailable, the arithmetic mean of the lower limit and the upper limit of the effective electricity price range is used as the filler value.

[0083] When a derating alarm is missing, the alarm level corresponding to the maximum derating ratio in the alarm dictionary table is taken as the filler value to maintain the conservatism of photovoltaic-related records when alarm information is missing.

[0084] A unified status record is generated by merging fields in a fixed order, namely: unified timestamp, rigid load, adjustable load, available photovoltaic power, energy storage operating status, energy storage state of charge, energy storage power boundary, state of charge boundary, electricity price, and derating alarm.

[0085] Write the unified state record to the historical cache and perform time consistency processing:

[0086] When the history cache is empty, the unified state record is written to the history cache as the first record.

[0087] When the historical cache is not empty, if the unified timestamp is the same as the last record, it is overwritten and written to the last position; if the unified timestamp is later than the last record, it is appended and written to the last position.

[0088] If the unified timestamp is earlier than the last record, it is determined whether it falls within the preset out-of-order tolerance window. The out-of-order tolerance window is represented by the number of collection cycles, with two collection cycles by default, which can be configured.

[0089] If it is not earlier than the lower limit of the window, write to the corresponding timestamp position and maintain the unified timestamp increment order; otherwise, do not write and record out-of-order events to ensure that the unified state records in the historical cache are arranged in the unified timestamp increment order.

[0090] S2: Based on historical cache, conservative boundary predictions of the upper and lower limits of load and photovoltaic power are performed on the load data and available photovoltaic power in the future rolling window, generating a conservative prediction sequence and calculating the gap-margin factor.

[0091] S2.1: Extract the current unified timestamp based on the historical cache, and generate a historical sequence by tracing back the set historical length. Perform validity screening and missing data filling on the historical sequence to obtain a valid historical sequence.

[0092] Read the unified state record corresponding to the current moment from the historical cache and extract the unified timestamp.

[0093] Using a unified timestamp as the endpoint, the system traces back a set historical length, which corresponds to several collection cycle durations, and generates a historical timestamp list in ascending order of the unified timestamp. When a certain timestamp does not have a corresponding record, a unified status record with an empty field is generated to fill in the timestamp.

[0094] For example, the default historical length is selected to cover a seven-day historical collection period, and can be configured to fourteen or thirty days. The historical length setting is based on the following: at least one complete period must be covered to reflect intraday and intraweek patterns. When the number of valid historical samples is insufficient to support statistical stability, the historical length will be automatically extended to fourteen or thirty days.

[0095] Perform field validity screening on each historical sequence. The screening targets include rigid load, adjustable load, available photovoltaic power, energy storage operating status, energy storage state of charge, energy storage power boundary, state of charge boundary, electricity price and derating alarm. When a field is empty, the field is marked as a missing item.

[0096] Missing items are filled in using a fixed filling order and fixed filling rules, and the output is a complete and valid historical sequence for statistical purposes.

[0097] It should be noted that the determination of insufficient number of valid historical samples is as follows: if the number of valid samples that can be extracted from the same collection period number in the valid historical sequence used for statistics is less than the preset minimum number of samples, it is determined to be insufficient; the minimum number of samples is determined based on the requirements of statistical stability and confidence level, and the example value is 20 and can be configured; when the collection period duration is adjusted, the minimum number of samples remains unchanged according to the effective number of samples corresponding to the same collection period number, and the historical length is converted according to the collection period duration to keep the number of historical days covered unchanged.

[0098] S2.2: Based on the effective historical sequence, extract the historical load data samples for the corresponding time period according to the unified timestamp of the future rolling window, and determine the upper limit of rigid load and the upper limit of adjustable load according to the preset quantile level, and synthesize the conservative load boundary.

[0099] Starting from the unified timestamp at the end of the valid historical sequence, the data is extrapolated sequentially according to the collection cycle duration to generate multiple unified timestamps corresponding to the preset future rolling window coverage duration. These unified timestamps are then arranged in ascending order to obtain the unified timestamp sequence for the future rolling window.

[0100] For each unified timestamp in the future rolling window unified timestamp sequence, determine the collection cycle number obtained by counting segments according to the collection cycle duration from midnight of the day to the unified timestamp.

[0101] Extract rigid load field values ​​with the same acquisition cycle number from the valid historical sequence to generate a rigid load candidate set;

[0102] Extract adjustable load field values ​​with the same collection cycle number from the valid historical sequence to form an adjustable load candidate set.

[0103] When the number of samples in the rigid load candidate set is not zero, the rigid load candidate set is arranged in ascending order of numerical values ​​to generate an ordered sequence of rigid loads, and the position order of the ordered sequence of rigid loads is used as the sorting position number.

[0104] Based on the preset quantile level, determine the corresponding position of the quantile within the sorting position number range.

[0105] For example, a preset quantile level (example value range is 80% to 99%) indicates that the sample location is closer to the high end, thus obtaining a more conservative estimate of the upper bound of the load (less likely to underestimate the load); the lower the quantile level, the less conservative it is and the estimate is closer to the general level.

[0106] When the position corresponding to the quantile is not an integer, the quantile position number is determined by rounding up, and the quantile position number is limited to the range of the sample size; the value corresponding to the quantile position number in the ordered sequence of rigid loads is taken as the conservative boundary value of the rigid load with a unified timestamp.

[0107] When the number of samples in the adjustable load candidate set is not zero, the adjustable load candidate set is arranged in ascending order of values ​​to generate an ordered sequence of adjustable loads, and the position order of the ordered sequence of adjustable loads is used as the sorting position number.

[0108] The corresponding position of the quantile is determined within the sorting position number range based on the preset quantile level.

[0109] When the position corresponding to the quantile is not an integer, the quantile position number is determined by rounding up, and the quantile position number is limited to the range of the sample size. The value corresponding to the quantile position number in the ordered sequence of adjustable load is taken as the conservative boundary value of adjustable load with a unified timestamp.

[0110] When the number of samples in the candidate set of rigid loads is zero, the maximum value among all valid values ​​of the rigid load field in the valid historical sequence is selected as the conservative boundary value of the rigid load with a unified timestamp.

[0111] When the number of adjustable load candidate samples is zero, the maximum value among all valid values ​​of the adjustable load field in the valid historical sequence is selected as the conservative boundary value of adjustable load with a unified timestamp, which is used to avoid underestimating the load when there are insufficient samples.

[0112] The conservative boundary values ​​of rigid loads and adjustable loads corresponding to each unified timestamp are combined to form the conservative boundary value of the load, and the conservative boundary value of the load is collected and output in the order of the unified timestamp sequence of the future rolling window.

[0113] The load conservative boundary value corresponding to the unified timestamp is expressed as: ;

[0114] In the formula, To standardize timestamps, To unify timestamps The corresponding conservative load boundary value (unit: kilowatt). To unify timestamps The corresponding conservative boundary value for rigid load (unit: kilowatt). To unify timestamps The corresponding conservative boundary value for adjustable load (unit: kilowatt).

[0115] S2.3: Based on valid historical sequence data, extract historical samples of photovoltaic available power for the corresponding time period according to the unified timestamp of the future rolling window, take the minimum value of the historical samples of photovoltaic available power to obtain the next photovoltaic conservative boundary, and determine the reduction ratio according to the reduction alarm, and perform proportional limit correction on the photovoltaic conservative boundary.

[0116] The reduced rate alarm field value is read from the unified status record at the end of the valid historical sequence and used as the reduced rate alarm value in the future rolling window; when the reduced rate alarm field value is missing, the alarm level corresponding to the maximum reduced rate ratio in the alarm dictionary table is taken as the reduced rate alarm value.

[0117] The derating alarm level is mapped to the derating ratio based on the alarm dictionary table. The derating ratio is configured in the alarm level lookup table and can be maintained and updated. The derating ratio does not decrease as the alarm level increases. The derating ratio value is set to the range of zero to one.

[0118] For each unified timestamp in the unified timestamp sequence of the future rolling window, the photovoltaic available power field value with the same collection period number is extracted from the valid historical sequence according to the collection period number to form a photovoltaic available power candidate set.

[0119] When the number of samples in the candidate set of available photovoltaic power is not zero, the minimum value in the candidate set of available photovoltaic power is selected as the initial value of the photovoltaic conservative boundary with a unified timestamp to avoid overestimating the available photovoltaic power; when the number of samples is zero, the initial value of the photovoltaic conservative boundary is set to zero.

[0120] The initial value of the photovoltaic conservative boundary is adjusted according to the derating ratio to obtain the photovoltaic conservative boundary value, and the photovoltaic conservative boundary value is guaranteed to be no less than zero; the photovoltaic conservative boundary sequence is collected and output in the order of the unified timestamp sequence of the future rolling window.

[0121] The photovoltaic conservative boundary value corresponding to the unified timestamp is expressed as: ;

[0122] In the formula, To unify timestamps Corresponding conservative boundary for photovoltaic power (unit: kilowatts). To unify timestamps The corresponding candidate set of available photovoltaic power (unit: kilowatts). The reduction ratio obtained by mapping the reduction alarm (dimensionless, value range from 0 to 1). This represents the minimum value (in kilowatts) in the candidate set of available photovoltaic power.

[0123] S2.4: Align and combine the conservative load boundary and the conservative photovoltaic boundary according to the unified timestamp of the future rolling window to generate a conservative prediction sequence.

[0124] Using the unified timestamp sequence of future rolling windows as a benchmark, the conservative boundary values ​​of load and photovoltaic corresponding to the same unified timestamp are merged into a prediction boundary record.

[0125] All prediction boundary records are aggregated in ascending order of a uniform timestamp to obtain a conservative prediction sequence.

[0126] S2.5: Based on the conservative prediction sequence, calculate the power difference between the photovoltaic conservative boundary and the load conservative boundary for each time period, and accumulate them after conversion by time period. The maximum positive value of the accumulated value is the maximum margin, and the minimum negative value of the accumulated value is recorded as the maximum gap. Output the gap-margin factor.

[0127] Obtain the data collection period duration and convert it into a time length in hours.

[0128] For example, when the data collection period is 15 minutes, the hourly unit is 0.25 hours.

[0129] The cumulative impact of energy storage is initialized to zero, and the maximum margin and maximum gap are initialized to zero.

[0130] Traverse each prediction boundary record in the conservative prediction sequence in ascending order of unified timestamp, read the load conservative boundary value and photovoltaic conservative boundary value corresponding to the unified timestamp one by one, calculate the difference between the photovoltaic conservative boundary value and the load conservative boundary value, and use the difference as the power difference.

[0131] The power difference is converted into the energy change over a time period based on the hourly unit time length, and the energy change is accumulated into the cumulative impact of energy storage in the previous time period to obtain the cumulative impact of energy storage in the current time period.

[0132] The cumulative impact of energy storage is expressed as follows: ; ;

[0133] In the formula, For the first Cumulative impact of energy storage over a period of time (unit: kilowatt-hours). For the first Conservative boundary values ​​for photovoltaic power over a given time period (unit: kilowatts). For the first Conservative load boundary values ​​for a given time period (unit: kilowatts). The hourly unit duration value (unit: hour) is obtained by converting the data collection period duration. This refers to the sequence number of time periods within the future scrolling window, arranged in ascending order according to a unified timestamp. This is an initial value based on the starting point of the future rolling window, indicating that the cumulative impact of stored energy at the starting point of the window is initialized to zero.

[0134] During the traversal, the maximum margin and maximum deficit are updated based on the cumulative impact of the stored energy in the current time period:

[0135] When the cumulative impact of energy storage is positive, it is used as a margin candidate value;

[0136] When the margin candidate value is greater than the current maximum margin, update the maximum margin to the margin candidate value.

[0137] When the cumulative impact of energy storage is negative, the positive amplitude corresponding to the negative value will be used as a candidate value for the gap.

[0138] When the gap candidate value is greater than the current maximum gap, update the maximum gap to the gap candidate value.

[0139] After completing the traversal of the conservative forecast sequence, the maximum gap and maximum margin are output, and the maximum gap and maximum margin are combined to form the gap-margin factor.

[0140] The gap-margin factor is expressed as: ; ; ;

[0141] In the formula, This represents the maximum margin within the future rolling window (unit: kilowatt-hours). The maximum gap within the future rolling window (unit: kilowatt-hours). For gap-margin factor, For future scrolling window, the sequence number of all time periods The corresponding target quantity is taken as the maximum value.

[0142] It should be noted that the maximum gap is used to characterize the maximum insufficient energy scale of the load conservative boundary relative to the photovoltaic conservative boundary within the future rolling window, while the maximum margin is used to characterize the maximum surplus energy scale of the photovoltaic conservative boundary relative to the load conservative boundary within the future rolling window.

[0143] S3: Based on the conservative prediction sequence and the running constraint data, perform a constrained multi-objective candidate search within the future rolling window to generate a Pareto candidate policy set, and use the gap-margin factor to perform deterministic preference mapping screening to obtain a candidate subset.

[0144] S3.1: Based on the conservative prediction sequence, energy storage operating status and energy storage power boundary, determine the conservative boundary of load, photovoltaic conservative boundary and feasible range of energy storage power within the future rolling window, and generate multiple sets of candidate energy storage power sequences within the feasible range of energy storage power.

[0145] Read the unified state record corresponding to the current moment from the historical cache, extract the energy storage working state, energy storage state of charge, energy storage power boundary, state of charge boundary and electricity price under the current unified timestamp, and read the unified timestamp sequence of the future rolling window and the corresponding conservative prediction sequence.

[0146] The feasible range of energy storage power is determined based on the energy storage power boundary under the current unified timestamp, and the state limitation correction of the feasible range of energy storage power is combined with the energy storage operating state. When the energy storage operating state restricts the charging and discharging direction, only the power range corresponding to the allowed direction is retained. When the energy storage operating state does not allow charging and discharging, the feasible range is corrected to zero power.

[0147] The lower and upper bounds of the modified feasible range of energy storage power are used as the two candidate points, and zero power is used as the center candidate point. When zero power does not fall within the feasible range of energy storage power, the center candidate point is the energy storage power boundary value closest to zero power. Between the two candidate points and the center candidate point, the number of candidate points is the total number of points in the energy storage power candidate point set, including the two candidate points, the center candidate point, and the middle candidate point. The distance between adjacent candidate points is not less than the minimum power command resolution allowed by the energy storage device. After deduplication, all candidate points are arranged in ascending order of power value to form the energy storage power candidate point set.

[0148] Based on the electricity price field within the future rolling window, time periods with consecutively identical electricity prices are merged into electricity price segments, forming a list of electricity price segments.

[0149] When the electricity price field is missing for a certain time period within the future rolling window, the electricity price in the time-of-use electricity price table under the corresponding unified timestamp will be used to fill the gap first. If the time-of-use electricity price table is unavailable, the electricity price in the unified status record at the end of the valid historical sequence will be used to fill the gap.

[0150] For each electricity price segment, select a candidate point from the set of candidate energy storage power points and assign it to all time periods within the segment.

[0151] The candidate point selection combinations for all electricity price segments are traversed to form multiple candidate energy storage power sequences, which are then aggregated into a set of candidate energy storage power sequences.

[0152] When the number of candidate sequences exceeds a preset upper limit, candidate sequences that do not exceed the upper limit are truncated according to the generation order.

[0153] It should be noted that the number of candidate points and the minimum resolution of the power command are jointly determined by the energy storage converter nameplate capability and the equipment communication protocol; the pre-set upper limit of the number is based on the time available for calculation within the acquisition cycle and the calculation time for evaluating a single candidate sequence, with an example value of 200.

[0154] S3.2: Perform operational constraint verification and boundary correction on the candidate energy storage power sequence to obtain a set of feasible candidate strategies.

[0155] Read each candidate energy storage power sequence from the candidate energy storage power sequence set, and perform operational constraint verification time by time segment according to the unified timestamp of the future rolling window:

[0156] When the candidate energy storage power for a time period falls within the feasible range of energy storage power, the power for that time period remains unchanged.

[0157] When the candidate energy storage power for a time period is less than the lower power limit, the power for that time period is adjusted to the lower power limit; when the candidate energy storage power for a time period is greater than the upper power limit, the power for that time period is adjusted to the upper power limit.

[0158] The candidate energy storage power sequence after time period verification and amplitude limiting correction is output as a feasible candidate strategy and associated with the generation order.

[0159] The set of feasible candidate strategies is subjected to duplicate elimination. For feasible candidate strategies with the same energy storage power value in all time periods within the future rolling window, only the strategy generated earlier is retained, and the rest are eliminated. The retained strategies are then aggregated to form a set of feasible candidate strategies.

[0160] S3.3: Based on the set of feasible candidate strategies, the changes in energy storage status of feasible candidate strategies are deduced over time periods and energy balance calculations are performed to obtain the purchased electricity and curtailed solar power corresponding to each feasible candidate strategy. The purchased electricity is then charged according to the electricity price to obtain the electricity purchase cost, thus forming a multi-objective evaluation result.

[0161] For each feasible candidate strategy, read the energy storage charge state, charge state boundary and energy storage working state corresponding to the current unified timestamp from the historical cache; read the collection cycle duration and convert it into hourly unit duration value.

[0162] Read the available energy capacity, charging efficiency, discharging efficiency, and positive and negative directions of energy storage power from the energy storage device parameter configuration table.

[0163] The convention for positive and negative directions of energy storage power is used to define the correspondence between the energy storage power symbol and the charging and discharging direction.

[0164] In this embodiment, positive energy storage power indicates the discharge direction, negative energy storage power indicates the charging direction, and zero energy storage power indicates the standby direction. The conservative prediction sequence is traversed in ascending order according to the unified timestamp of the future rolling window. The conservative boundary values ​​of the load and photovoltaics corresponding to each time period are read, and the candidate energy storage power corresponding to each time period is also read.

[0165] Within each time period, the energy storage operating status is updated according to the direction of the candidate energy storage power value: when the candidate energy storage power is positive, it is updated to the discharge state; when the candidate energy storage power is negative, it is updated to the charging state; and when the candidate energy storage power is zero, it is updated to the standby state.

[0166] Based on candidate energy storage power, hourly unit time value, available energy storage capacity, charging efficiency, and discharging efficiency, the energy storage state of charge (SOC) for the next time period is extrapolated, and the extrapolated values ​​are verified to be consistent with the SOC boundary.

[0167] When the extrapolated value falls within the state of charge boundary range, the strategy execution power corresponding to the time period is taken as the candidate energy storage power.

[0168] When the projected value is less than the state of charge limit, the energy storage state of charge in the next time period is corrected to the state of charge limit, and the strategy execution power corresponding to the time period is corrected to the corrected power that meets the state of charge limit.

[0169] When the projected value is greater than the upper limit of the state of charge, the energy storage state of charge for the next time period is corrected to the upper limit of the state of charge, and the strategy execution power corresponding to the time period is corrected to the corrected power that meets the upper limit of the state of charge.

[0170] Based on the conservative boundary value of load, the conservative boundary value of photovoltaic power and the power executed by the strategy, the purchased power and the curtailed power are calculated: when the remaining conservative boundary value of load after combining the conservative boundary value of photovoltaic power and the power executed by the strategy is positive, the positive remaining value is determined as the purchased power; otherwise, the purchased power is zero.

[0171] When the conservative boundary value of photovoltaic power is positive after the conservative boundary value of the comprehensive load and the power executed by the strategy, the positive surplus is determined as the curtailed power; otherwise, the curtailed power is zero.

[0172] Convert the purchased power and the curtailed power into purchased electricity and curtailed electricity respectively in hourly units and accumulate them, then output the accumulated value of purchased electricity and curtailed electricity corresponding to the feasible candidate strategies.

[0173] It should be noted that the energy storage equipment parameter configuration table is determined by the equipment nameplate parameters, energy storage equipment communication protocol parameters, and operation and maintenance configuration.

[0174] S3.4: Perform a dominance comparison on the multi-objective evaluation results, eliminate dominated strategies and retain non-dominated strategies, and generate a Pareto candidate strategy set.

[0175] For each feasible candidate strategy, the electricity price sequence is read in ascending order according to the unified timestamp of the future rolling window. The purchased electricity volume is multiplied by the electricity price for each time period and accumulated to obtain the electricity purchase cost.

[0176] The cost of electricity is expressed as follows: ;

[0177] In the formula, Candidate strategies Electricity purchase cost in the future rolling window (unit: yuan). Candidate strategies In the Electricity purchased during a specific time period (unit: kilowatt-hours). For the first Electricity price for each time period (unit: yuan / kWh) The number of time periods (dimensionless integer) within the future scrolling window.

[0178] The electricity purchase cost and the cumulative value of curtailed solar power are used together as the multi-objective evaluation result and associated with feasible candidate strategies. The field order of the multi-objective evaluation result is fixed as electricity purchase cost and cumulative value of curtailed solar power.

[0179] Perform a dominance relationship comparison on each feasible candidate strategy. Read the cumulative values ​​of electricity purchase cost and curtailed solar power corresponding to the feasible candidate strategies to be compared.

[0180] The feasible candidate strategy is searched in the set of feasible candidate strategies, excluding the feasible candidate strategy to be compared. If there is a feasible candidate strategy that simultaneously satisfies the following conditions: the electricity purchase cost is not higher than that of the feasible candidate strategy to be compared and the cumulative value of curtailed solar power is not higher than that of the feasible candidate strategy to be compared, and the electricity purchase cost is lower than that of the feasible candidate strategy to be compared or the cumulative value of curtailed solar power is lower than that of the feasible candidate strategy to be compared, then the feasible candidate strategy to be compared is determined as the dominated strategy and eliminated. The strategies that are not eliminated are then gathered to form the Pareto candidate strategy set.

[0181] S3.5: Determine the preference direction based on the relationship between the maximum gap and the maximum margin in the gap-margin factor, and set preference mapping rules based on the preference direction. When the gap takes priority, map the electricity purchase cost to the first preference amount and the curtailed solar power to the second preference amount; when the margin takes priority, map the curtailed solar power to the first preference amount and the electricity purchase cost to the second preference amount.

[0182] Based on the maximum gap and maximum margin in the gap-margin factor: when the maximum gap is not less than the maximum margin, the preference direction is determined to be gap priority; when the maximum gap is less than the maximum margin, the preference direction is determined to be margin priority.

[0183] The deterministic preference mapping is represented as: ;

[0184] In the formula, This represents the preference direction, used to determine the order (enumerated value) of preference quantities.

[0185] Based on the preference direction, solidify the preference mapping rules:

[0186] When the preference direction is gap priority, the first preference quantity is fixed as the electricity purchase cost, and the second preference quantity is fixed as the cumulative value of the curtailed solar power.

[0187] When the preference direction is margin priority, the first preference quantity is fixed as the cumulative value of abandoned solar power, and the second preference quantity is fixed as the cost of purchasing electricity.

[0188] By fixing the generation order of the Pareto candidate policy set to a stable decision key, the order is determined using the stable decision key when the first preference and the second preference are simultaneously the same.

[0189] S3.6: Use preference mapping rules to map the multi-objective evaluation results of the Pareto candidate policy set to obtain the preference mapping results of each candidate policy.

[0190] Read the multi-objective evaluation results associated with each Pareto candidate strategy from the Pareto candidate strategy set, and extract the cumulative values ​​of electricity purchase cost and curtailed solar power from the multi-objective evaluation results.

[0191] Generate preference mapping results based on the fixed preference mapping rules:

[0192] When the preference direction is gap priority, the first preference measure is the electricity purchase cost, and the second preference measure is the cumulative value of the curtailed solar power.

[0193] When the preference direction is margin priority, the first preference measure is the cumulative value of the curtailed solar power, and the second preference measure is the cost of purchasing electricity.

[0194] The first preference quantity, the second preference quantity, and the stable decision key are associated one-to-one with the corresponding Pareto candidate policies, and the results are aggregated to form a set of candidate policy preference mapping results.

[0195] S3.7: Perform deterministic sorting on the preference mapping results and output candidate subsets based on preset filtering conditions.

[0196] Read the first preference quantity, second preference quantity, stable decision key and associated Pareto candidate policy one by one from the candidate policy preference mapping result set to form a Pareto candidate policy list with sorting key.

[0197] Perform deterministic sorting on the Pareto candidate policy list with sorting keys: sort in ascending order by the first preference value; if the first preference values ​​are the same, sort in ascending order by the second preference value; if the first preference value and the second preference value are the same, sort in ascending order by the stable decision key to obtain a deterministic sorted list.

[0198] Deterministically sorted lists apply preset filtering criteria:

[0199] Read the first and second preference values ​​corresponding to the first strategy in the deterministic ranking list and use them as the filtering criteria;

[0200] Before the screening and comparison, the first and second preference values ​​of each strategy in the deterministic ranking list, as well as the screening benchmark, are quantified according to a uniform measurement precision: the electricity purchase cost is quantified to a preset decimal place in monetary units (e.g., to 0.01 yuan), and the cumulative value of curtailed solar power is quantified to a preset decimal place in electricity units (e.g., to 0.01 kWh). The quantified first and second preference values ​​are then used as the first preference value comparison value and the second preference value comparison value, respectively.

[0201] Read the first preference value and the second preference value of each strategy in the deterministic ranking list one by one, and compare them with the first preference value and the second preference value corresponding to the screening benchmark. If any comparison value is inconsistent with the comparison value corresponding to the screening benchmark, the strategy is eliminated. If both the first preference value and the second preference value are consistent with the comparison value corresponding to the screening benchmark, the strategy is retained. The retained strategies are gathered in the order of the deterministic ranking list to form a candidate subset.

[0202] It should be noted that when the number of candidate subsets exceeds the preset size limit (three in the example), the deterministically sorted list is retained from front to back up to the size limit.

[0203] S4: Perform candidate fine-tuning within the candidate subset, determine the reference strategy, and output a limited energy storage power fine-tuning on the current energy storage power to obtain the energy storage power to be executed.

[0204] S4.1: Based on the current unified timestamp and the current energy storage power, read the absolute difference between the candidate energy storage power and the current energy storage power for the corresponding time period from the candidate subset one by one, and take the candidate strategy with the smallest absolute difference as the reference strategy.

[0205] Read the unified state record corresponding to the current moment from the historical cache and extract the current unified timestamp.

[0206] The energy storage power is read from the historical cache as the current energy storage power; if the historical cache does not contain the energy storage power command value from the previous acquisition cycle, the current energy storage power is zero.

[0207] Read each candidate strategy in the candidate subset and locate the candidate energy storage power of the candidate strategy in the time period corresponding to the current unified timestamp.

[0208] For each candidate subset, calculate the absolute difference between the candidate energy storage power and the current energy storage power, and select the candidate strategy with the smallest absolute difference as the reference strategy.

[0209] When multiple candidate strategies have the same absolute difference, the candidate strategy with the highest ranking among the candidate subsets is selected as the reference strategy.

[0210] S4.2: Based on the relationship between the candidate energy storage power corresponding to the reference strategy and the current energy storage power, determine the fine-tuning direction and form the energy storage power fine-tuning amount. After performing amplitude limiting processing on the energy storage power fine-tuning amount, it is superimposed with the current energy storage power to obtain the energy storage power to be executed.

[0211] Read the candidate energy storage power of the reference strategy for the time period corresponding to the current unified timestamp, and read the current energy storage power.

[0212] The fine-tuning direction is determined based on the relationship between the candidate energy storage power and the current energy storage power: when the candidate energy storage power is greater than the current energy storage power, the fine-tuning direction is the direction of increasing power; when the candidate energy storage power is less than the current energy storage power, the fine-tuning direction is the direction of decreasing power; when the candidate energy storage power is equal to the current energy storage power, the fine-tuning direction is the direction of zero change.

[0213] Calculate the difference between the candidate energy storage power and the current energy storage power, and use the difference as a candidate value for the fine-tuning of the energy storage power.

[0214] Based on the set of candidate energy storage power points, and sorted in ascending order of value; calculate the spacing between adjacent candidate points, and take the minimum value of the adjacent spacing as the energy storage power fine-tuning limit value;

[0215] When the number of energy storage power candidate points is insufficient (e.g., less than two candidate points), adjacent spacing cannot be formed between candidate points. The energy storage power fine-tuning limit value is the minimum of the difference between the current energy storage power and the lower limit of the power, and the difference between the current energy storage power and the upper limit of the power. The energy storage power fine-tuning limit value is consistent with the energy storage power unit.

[0216] Amplitude limiting processing is applied to the candidate values ​​of energy storage power fine-tuning: when the absolute value of the candidate value of energy storage power fine-tuning is greater than the amplitude limit value of energy storage power fine-tuning, the energy storage power fine-tuning is taken as the value whose absolute value is equal to the amplitude limit value of energy storage power fine-tuning and whose direction is consistent with the fine-tuning direction.

[0217] When the absolute value of the candidate value for energy storage power fine-tuning is not greater than the energy storage power fine-tuning limit, the energy storage power fine-tuning amount shall be taken as the candidate value for energy storage power fine-tuning.

[0218] The initial value of the energy storage power to be executed is obtained by superimposing the fine-tuning amount of the energy storage power with the current energy storage power.

[0219] Read the lower and upper bounds of the feasible range of energy storage power, and apply boundary limiting to the initial value of the energy storage power to be executed:

[0220] When the initial value of the energy storage power to be executed is less than the lower power limit, the energy storage power to be executed shall be the lower power limit.

[0221] When the initial value of the energy storage power to be executed is greater than the upper limit of the power, the energy storage power to be executed shall be the upper limit of the power.

[0222] When the initial value of the energy storage power to be executed falls within the feasible range of energy storage power, the initial value of the energy storage power to be executed shall be taken as the energy storage power to be executed.

[0223] S5: Perform feasibility correction on the proposed energy storage power, output and issue energy storage power command, and generate execution feedback information and correction information by combining load data and available photovoltaic power.

[0224] S5.1: Based on the historical cache, read the energy storage state of charge and state of charge boundary corresponding to the current unified timestamp, and calculate the estimated value of energy storage state of charge in combination with the energy storage power to be executed.

[0225] Read the unified state record corresponding to the current moment from the historical cache, and extract the current unified timestamp, energy storage charge state and charge state boundary.

[0226] The estimated state of charge (SBC) of the energy storage for the next data collection cycle is calculated based on the current SBC, using the following calculation method:

[0227] In the direction of discharge, the energy storage power to be executed forms the change in external output energy within the time period corresponding to the hourly unit duration value. The change in external output energy is determined by the magnitude of the energy storage power to be executed and the hourly unit duration value.

[0228] Discharge efficiency is used to characterize the proportion of external output energy that can be generated after the energy on the battery side decreases. The ratio of the change in external output energy to the change in battery side energy is consistent with the discharge efficiency. The change in battery side energy is derived from the relationship between the change in external output energy and discharge efficiency. The change in battery side energy is greater than or equal to the change in external output energy.

[0229] The change in energy on the battery side is converted into a change in state of charge based on the available energy capacity of the energy storage. The calculated value of the energy storage state of charge is the current energy storage state of charge minus the corresponding change in state of charge.

[0230] In the charging direction, the energy storage power to be executed will form the change in external input energy within the time period corresponding to the hourly unit duration value. The change in external input energy is determined by the magnitude of the energy storage power to be executed and the hourly unit duration value.

[0231] Charging efficiency is used to characterize the proportion of external input energy converted into an increase in battery energy. The ratio of battery energy change to external input energy change is consistent with charging efficiency. Battery energy change is determined by the relationship between external input energy change and charging efficiency. Battery energy change is less than or equal to external input energy change.

[0232] The change in energy on the battery side is converted into a change in state of charge based on the available energy capacity of the energy storage. The estimated value of the energy storage state of charge is the current energy storage state of charge plus the corresponding change in state of charge.

[0233] In the standby direction, the calculated state of charge of the energy storage is consistent with the current state of charge of the energy storage.

[0234] The estimated state of charge of the energy storage for the next data acquisition cycle is expressed as follows: ;

[0235] In the formula, The calculated state of charge (SBC) value for the next data collection cycle is dimensionless. This represents the current state of charge of the energy storage (dimensionless, value range 0 to 1). The hourly unit duration value (unit: hour) is obtained by converting the data collection period duration. Available energy capacity for energy storage (unit: kilowatt-hours). Charging efficiency (dimensionless, 0 to 1). Discharge efficiency (dimensionless, 0 to 1). The equivalent power amplitude on the charging side (unit: kilowatts). The equivalent power amplitude on the discharge side (unit: kilowatt). The proposed energy storage capacity (unit: kilowatt, as agreed) For the discharge direction, (In the direction of charging).

[0236] The calculated state of charge (SBC) value of the energy storage is compared with the SBC boundary: if the calculated SBC value falls between the SBC boundary and the upper SBC boundary, it is determined that the energy storage power to be executed will not cause the energy storage SBC to exceed the SBC boundary; if the calculated SBC value is less than the SBC boundary or greater than the upper SBC boundary, it is determined that the energy storage power to be executed will cause the energy storage SBC to exceed the SBC boundary.

[0237] S5.2: When the calculated state of charge of the energy storage exceeds the state of charge boundary, the corrected energy storage power is determined as the energy storage power command; when the calculated state of charge of the energy storage does not exceed the state of charge boundary, the energy storage power to be executed is determined as the energy storage power command, and the energy storage power command is output and issued.

[0238] When it is determined that the proposed energy storage power will cause the energy storage state of charge to exceed the state of charge boundary, the energy storage power is determined based on the comparison between the calculated energy storage state of charge and the state of charge boundary:

[0239] When the calculated state of charge of energy storage is less than the limit of the state of charge, the corrected energy storage power shall be taken as the target limit of the state of charge.

[0240] When the calculated state of charge of the energy storage is greater than the upper limit of the state of charge, the corrected energy storage power is taken as the upper limit of the state of charge.

[0241] Based on the current energy storage state of charge, correction target, hourly unit duration value, available energy storage capacity, charging efficiency, and discharging efficiency, the corrected energy storage power that meets the correction target is calculated, and the corrected energy storage power is used as the energy storage power command power value.

[0242] When it is determined that the energy storage power to be executed will not cause the energy storage state of charge to exceed the state of charge boundary, the energy storage power to be executed is directly used as the energy storage power command power value.

[0243] The energy storage power command value is associated with the current unified timestamp, and the energy storage power command is output and sent for execution.

[0244] S5.3 performs energy balance calculations based on load data, available photovoltaic power, and energy storage power commands to obtain the purchased electricity and the curtailed photovoltaic power.

[0245] When the next acquisition cycle is triggered, the acquisition cycle duration is read and converted into an hourly unit duration value. The unified status record of the execution cycle associated with the energy storage power command is read from the historical cache. The rigid load, adjustable load and photovoltaic available power of the execution cycle are read, and the power value of the energy storage power command is read.

[0246] The rigid load and the adjustable load are combined to obtain the load power of the execution cycle.

[0247] The impact of energy storage power command values ​​on energy balance based on the positive and negative directions of energy storage power:

[0248] When the energy storage power command value is positive, it indicates the direction of discharge. A positive value of the energy storage power command value is used as the power supply on the load side to participate in the balancing process.

[0249] When the energy storage power command power value is negative, it indicates the charging direction. The positive amplitude corresponding to the negative value of the energy storage power command power value is used as the newly added electricity consumption to participate in the balance.

[0250] When the energy storage power command value is zero, it indicates the standby direction and does not change the energy balance relationship.

[0251] The required power and supplied power for the execution cycle are calculated based on the energy balance relationship. The load power and the newly added electricity consumption for the execution cycle are summed to obtain the required power for the execution cycle.

[0252] The available photovoltaic power during the execution cycle is summed with the power supplied to the load side to obtain the power supplied during the execution cycle.

[0253] When the demand power of the execution cycle is higher than the supply power of the execution cycle, the difference is determined as the power purchased in the execution cycle; when the demand power of the execution cycle is not higher than the supply power of the execution cycle, the power purchased in the execution cycle is zero.

[0254] When the power supplied during the execution cycle is higher than the power required during the execution cycle, the difference is determined as the curtailed power during the execution cycle; when the power supplied during the execution cycle is not higher than the power required during the execution cycle, the curtailed power during the execution cycle is zero.

[0255] The purchased electricity volume is obtained by multiplying the purchased electricity volume by the hourly unit time value during the execution period, and the curtailed photovoltaic power volume is obtained by multiplying the curtailed photovoltaic power volume by the hourly unit time value during the execution period.

[0256] S5.4: Collect unified timestamps, energy storage power instructions, purchased electricity, curtailed solar power, and electricity purchase costs to form execution feedback information.

[0257] When the next collection cycle is triggered, the electricity price field is read from the unified status record of the execution cycle from the historical cache; when the electricity price field is missing, the electricity price corresponding to the unified timestamp of the execution cycle is retrieved from the time-of-use electricity price table.

[0258] When the time-of-use electricity meter is unavailable, the arithmetic mean of the lower and upper bounds of the effective electricity price range shall be taken as the electricity price.

[0259] The cost of purchasing electricity for the execution period is calculated based on the amount of electricity purchased and the electricity price for the execution period.

[0260] The execution cycle unified timestamp, energy storage power command power value, purchased electricity, curtailed solar power and electricity purchase cost are collected into execution effect records, and the execution effect records are compiled into execution feedback information.

[0261] S5.5: Calculate the difference between the energy storage power to be executed and the energy storage power command as the power difference value, and aggregate it with the current unified timestamp to form correction information.

[0262] Calculate the difference between the energy storage power to be executed and the energy storage power command value, use the difference as the power difference value, establish a correlation between the current unified timestamp and the power difference value, form correction information and output it.

[0263] S6: Based on the execution feedback information and correction information, candidate evaluation sample records and strategy replay records are formed respectively.

[0264] S6.1: Based on the execution feedback information and correction information, candidate evaluation sample records and strategy replay records are generated and written to the historical cache.

[0265] The execution feedback information is used to read the unified timestamp of the execution cycle, the power value corresponding to the energy storage power instruction, the purchased electricity and the curtailed solar power.

[0266] The execution cycle unified timestamp, the power value corresponding to the energy storage power instruction, the purchased electricity and the curtailed solar power are collected into candidate evaluation sample records in a fixed field order.

[0267] Write candidate evaluation sample records to the candidate evaluation sample record area of ​​the historical cache: when there are candidate evaluation sample records with the same execution period and the same timestamp in the historical cache, perform overwrite writing;

[0268] An append write is performed when there are no candidate evaluation sample records with the same execution period and unified timestamp in the historical cache.

[0269] Perform time consistency processing on the candidate evaluation sample recording area:

[0270] When the unified timestamp of the execution cycle is later than the unified timestamp recorded at the end of the candidate evaluation sample record area, it is appended;

[0271] When the unified timestamp of the execution cycle equals the unified timestamp of the last record in the candidate evaluation sample record area, it is overwritten.

[0272] When the unified timestamp of the execution cycle is earlier than the unified timestamp recorded at the end of the candidate evaluation sample recording area, it is determined whether it falls into the out-of-order tolerance window. If it falls into the window, it is written according to the unified timestamp position and the unified timestamp increment order is maintained. If it does not fall into the window, it is not written and the out-of-order event is recorded.

[0273] S6.2: Extract the power difference based on the correction information and associate it with the reference policy to form a policy replay record and write it to the historical cache.

[0274] Read the current unified timestamp and power difference from the correction information.

[0275] Based on the reference strategy, the candidate strategy is read to generate the sequence number, and the unified timestamp corresponding to the starting point of the future scrolling window is read from the historical cache.

[0276] The reference strategy identifier is determined by combining the unified timestamp of the starting point of the future scrolling window with the candidate strategy generation sequence number, so that the reference strategy identifier is globally unique within the same future scrolling window.

[0277] When the number of candidate sequences exceeds the preset upper limit and is truncated according to the generation order, only the candidate strategies that are first in the generation order are retained and the remaining candidate strategies are deleted. The generation sequence number of the retained candidate strategies remains unchanged and is not reordered, so that the reference strategy identifier remains unchanged and can be stably traced.

[0278] The current unified timestamp, reference policy identifier, and power difference are grouped into a policy playback record in a fixed field order, namely the current unified timestamp, reference policy identifier, and power difference.

[0279] For policy replay records in the historical cache, time consistency processing is performed: when the current unified timestamp is later than the unified timestamp of the last record, an append write is performed;

[0280] When the current unified timestamp is equal to the unified timestamp of the last record, an overwrite write is performed; when the current unified timestamp is earlier than the unified timestamp of the last record, it is determined whether it falls within the out-of-order tolerance window. If it falls within the out-of-order tolerance window, it is written according to the unified timestamp position and the unified timestamp increment order is maintained. If it does not fall within the out-of-order tolerance window, no write is performed and the out-of-order event is recorded.

[0281] In summary, this invention incorporates supply and demand risks under uncertain disturbances into the strategy ranking process using quantifiable indicators, reducing reliance on empirical weighting and manual parameter tuning. This allows for clear criteria and stable output in candidate strategy selection. Furthermore, by calculating the gap-margin factor within a rolling window for deterministic preference mapping and screening, the invention ensures that strategies meet SOC and power constraints and are consistently reproducible, reducing jumps and corrections, improving photovoltaic grid integration, and suppressing fluctuations in the levelized cost of electricity.

[0282] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for optimizing scheduling of a photovoltaic power storage system in a machine room based on load demand, characterized in that: include, Collect load data, operating constraint data, and available photovoltaic power of the photovoltaic energy storage system in the computer room, generate a unified timestamp, and construct a unified status record to write to the historical cache; Based on historical cache, conservative boundary predictions of the upper and lower limits of load and photovoltaic power are made for load data and available photovoltaic power in the future rolling window, generating a conservative prediction sequence and calculating the gap-margin factor. Based on the conservative prediction sequence and operational constraint data, a constrained multi-objective candidate search is performed within the future rolling window to generate a Pareto candidate policy set. Then, a deterministic preference mapping is performed using the gap-margin factor to filter the candidate subset. Perform candidate fine-tuning within the candidate subset, determine the reference strategy, and output a limited energy storage power fine-tuning on the current energy storage power to obtain the energy storage power to be executed; The feasibility of the proposed energy storage power is revised, and the energy storage power command is output and issued. At the same time, the execution feedback information and correction information are generated by combining load data and available photovoltaic power. Based on the execution feedback information and correction information, candidate evaluation sample records and strategy replay records are generated and written to the historical cache.

2. The optimized scheduling method for a data center photovoltaic energy storage system based on load demand as described in claim 1, characterized in that: The specific steps for constructing a unified state record and writing it into the historical cache are as follows. The system acquires load data, available photovoltaic power, and operational constraint data within the same acquisition period and generates a unified timestamp. The load data includes rigid loads and adjustable loads, and the operational constraint data includes energy storage operating status, energy storage state of charge, energy storage power boundary, state of charge boundary, electricity price, and derating alarm. The load data, available photovoltaic power, and operational constraint data are aligned according to a unified timestamp, and missing items are filled in. The aligned load data, available photovoltaic power, and operational constraint data are merged to generate a unified status record, which is then written to the historical cache in ascending order of the unified timestamp.

3. The optimized scheduling method for a data center photovoltaic energy storage system based on load demand as described in claim 1, characterized in that: The specific steps for generating the conservative prediction sequence and calculating the gap-margin factor are as follows: Extract the current unified timestamp based on the historical cache, and generate a historical sequence by tracing back the set historical length. Then, perform validity screening and missing data filling on the historical sequence to obtain a valid historical sequence. Based on the effective historical sequence, load data historical samples for the corresponding time period are extracted according to the unified timestamp of the future rolling window, and the upper limit of rigid load and the upper limit of adjustable load are determined according to the preset quantile level, and the load conservative boundary is synthesized. Based on valid historical sequence data, the historical samples of photovoltaic available power for the corresponding time period are extracted according to the unified timestamp of the future rolling window. The minimum value of the historical samples of photovoltaic available power is taken as the next photovoltaic conservative boundary. The reduction ratio is determined according to the reduction alarm, and the ratio limit is modified for the photovoltaic conservative boundary. The conservative load boundary and the conservative photovoltaic boundary are aligned and combined according to the unified timestamp of the future rolling window to generate a conservative forecast sequence. Based on the conservative prediction sequence, the power difference between the photovoltaic conservative boundary and the load conservative boundary is calculated for each time period. After conversion by time period, the difference is accumulated. The maximum positive value of the accumulated value is the maximum margin, and the minimum negative value of the accumulated value is recorded as the maximum gap. The gap-margin factor is then output.

4. The optimized scheduling method for a data center photovoltaic energy storage system based on load demand as described in claim 1, characterized in that: The specific steps for generating the Pareto candidate policy set are as follows: Based on the conservative prediction sequence, energy storage operating status and energy storage power boundary, determine the conservative load boundary, photovoltaic conservative boundary and energy storage power feasible range within the future rolling window, and generate multiple sets of candidate energy storage power sequences within the feasible range of energy storage power. Perform operational constraint verification and boundary correction on the candidate energy storage power sequence to obtain a set of feasible candidate strategies; Based on the set of feasible candidate strategies, the changes in energy storage status of feasible candidate strategies are deduced over time periods and energy balance calculations are performed to obtain the purchased electricity and curtailed solar power corresponding to each feasible candidate strategy. The electricity purchase cost is then calculated based on the electricity price, forming a multi-objective evaluation result. The dominance relationship is compared on the multi-objective evaluation results, the dominated strategies are eliminated and the non-dominated strategies are retained, and a Pareto candidate strategy set is generated.

5. The optimized scheduling method for a data center photovoltaic energy storage system based on load demand as described in claim 1, characterized in that: The specific steps for obtaining the candidate subset are as follows: The preference direction is determined based on the relationship between the maximum gap and the maximum margin in the gap-margin factor, and the preference mapping rule is set according to the preference direction. When the gap takes priority, the electricity purchase cost is mapped to the first preference amount and the curtailed photovoltaic power is mapped to the second preference amount. When margin is prioritized, the curtailed solar power is mapped to the first preferred quantity and the electricity purchase cost is mapped to the second preferred quantity; The preference mapping rule is used to map the multi-objective evaluation results of the Pareto candidate policy set to obtain the preference mapping results of each candidate policy. The preference mapping results are deterministically sorted and quantitatively compared under preset filtering conditions to output candidate subsets.

6. The optimized scheduling method for a data center photovoltaic energy storage system based on load demand as described in claim 1, characterized in that: The specific steps to obtain the energy storage power to be executed are as follows: Based on the current unified timestamp and the current energy storage power, the absolute difference between the candidate energy storage power and the current energy storage power for the corresponding time period is read from the candidate subset one by one, and the candidate strategy with the smallest absolute difference is taken as the reference strategy. Based on the relationship between the candidate energy storage power corresponding to the reference strategy and the current energy storage power, the fine-tuning direction is determined and the energy storage power fine-tuning amount is formed. After the energy storage power fine-tuning amount is subjected to amplitude limiting processing, it is superimposed with the current energy storage power to obtain the energy storage power to be executed.

7. The optimized scheduling method for a data center photovoltaic energy storage system based on load demand as described in claim 1, characterized in that: The specific steps for performing feasibility correction on the proposed energy storage power and outputting the energy storage power command are as follows. Based on the historical cache, the energy storage state of charge and state of charge boundary corresponding to the current unified timestamp are read, and the energy storage state of charge is estimated by combining the energy storage power to be executed. When the calculated state of charge of the energy storage exceeds the state of charge boundary, the corrected energy storage power is determined as the energy storage power command; when the calculated state of charge of the energy storage does not exceed the state of charge boundary, the energy storage power to be executed is determined as the energy storage power command, and the energy storage power command is output and issued.

8. The optimized scheduling method for a data center photovoltaic energy storage system based on load demand as described in claim 1, characterized in that: The specific steps for outputting execution feedback information and correction information are as follows. Energy balance calculations are performed based on load data, available photovoltaic power, and energy storage power commands to obtain the purchased electricity and the amount of curtailed photovoltaic power. The unified timestamp, energy storage power instruction, purchased electricity volume, curtailed solar power volume and electricity purchase cost are collected to form execution feedback information; The difference between the energy storage power to be executed and the energy storage power command is calculated as the power difference value, and then aggregated with the current unified timestamp to form correction information.

9. The optimized scheduling method for a data center photovoltaic energy storage system based on load demand as described in claim 1, characterized in that: The execution feedback information and correction information are respectively used to form candidate evaluation sample records and strategy replay records. The specific steps are as follows. Based on the execution feedback information, extract the unified timestamp of the execution cycle, the power value of the energy storage power instruction, the purchased electricity and the curtailed solar power, generate candidate evaluation sample records, and write them into the historical cache; Based on the correction information, the power difference is extracted and associated with the reference policy to form a policy replay record, which is then written to the historical cache.

10. The optimized scheduling method for a data center photovoltaic energy storage system based on load demand as described in claim 5, characterized in that: The preset screening criteria refer to the selection of candidate strategies based on the quantitative preference value of the top-ranked candidate strategy after the Pareto candidate strategy set is deterministically sorted, and the selection of candidate strategies with consistent quantitative preference values ​​is based on the quantitative preference value of the top-ranked candidate strategy.

Images