A method for augmenting and rapidly generating training data for unit combination optimization problems

Abstract

This invention discloses a method for augmenting and rapidly generating training data for unit combination optimization problems, relating to the field of power system technology. The method includes: acquiring historical high-temporal-resolution power system unit combination input data, including a set of scheduling periods, denoted as the first-day-ahead dataset, and calculating the system net load; resampling the set of scheduling periods into subsets according to time-series resampling rules; extracting corresponding renewable energy output data and load forecast data from the first-day-ahead dataset by time period to obtain the second-day-ahead dataset; proportionally adjusting and expanding the renewable energy output data and load forecast data in the second-day-ahead dataset according to data augmentation rules to form the third-day-ahead dataset; constructing a unit combination model with the same sample size as the third-day-ahead dataset and solving it. This invention solves the problem of difficult training data collection when applying data-driven methods to high-temporal-resolution unit combination optimization problems and day-ahead clearing in the electricity spot market.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and more specifically to a method for augmenting and rapidly generating training data for unit combination optimization problems. Background Technology

[0002] The Safety-Constrained Unit Combination (SCUC) problem is a typical non-convex, mixed-integer optimization problem, often used for day-ahead clearing in the electricity spot market. The worst-case computational complexity of the SCUC problem is exponentially related to the number of integer variables. High time resolution (e.g., 15-minute, 10-minute, 5-minute) SCUC problems multiply the number of integer variables and constraints, sacrificing computational complexity for a safer and more economical power system dispatching scheme compared to the traditional 1-hour time resolution. Therefore, efficiently solving high-time-resolution SCUC problems has become a critical technical challenge that urgently needs to be addressed.

[0003] Data-driven methods can quickly identify redundant line constraints and unit integer variables in high-time-resolution SCUC problems, making them an effective measure to improve computational efficiency. However, the application of data-driven methods faces a circular dependency problem: the very difficulty in solving high-time-resolution SCUC problems necessitates data-driven methods to improve efficiency, but a sufficient amount of high-quality training data requires solving a large number of high-time-resolution SCUC instances beforehand. Therefore, how to efficiently generate diverse training data has become a key bottleneck restricting the application of data-driven methods in high-time-resolution SCUC problems. Summary of the Invention

[0004] To address the aforementioned shortcomings in existing technologies, this invention provides a method for augmenting and rapidly generating training data for unit combination optimization problems, which solves the problem of difficult training data collection when applying data-driven methods to high time-resolution unit combination optimization problems and day-ahead clearing in the electricity spot market.

[0005] To achieve the aforementioned objectives, the present invention employs the following technical solution: a method for augmenting and rapidly generating training data for unit combination optimization problems, comprising the following steps:

[0006] S1: Obtain historical high time resolution power system unit combination input data, including the scheduling period set, and denote it as the first day-ahead dataset. Calculate the system net load in the first day-ahead dataset.

[0007] S2: Based on the system net load concentrated in the data of the first day and the preset time-series resampling rules, the set of scheduling periods with high time resolution is resampled into a subset of scheduling periods with a set time resolution;

[0008] S3: Based on the resampled scheduling period subset, extract the corresponding high time resolution new energy output data and load forecast data from the first day's dataset by time period to form the set time resolution data, and obtain the second day's dataset;

[0009] S4: Based on the preset data augmentation rules, the renewable energy output data and load forecast data of the second-day dataset with a set time resolution are proportionally adjusted and expanded to form the third-day dataset.

[0010] S5: Construct a unit combination model with a set time resolution consistent with the number of samples in the dataset three days prior and solve it. Record the optimization solution results and their analysis results as new data to complete the training data augmentation and rapid generation of the unit combination optimization problem.

[0011] Furthermore, the preset timing resampling rules in S2 include:

[0012] (1) Based on a 1-hour time resolution, the set of scheduling periods is decomposed into S non-overlapping subsets of scheduling periods that are associated with the scheduling time resolution, where the first subset is... The first sample A subset of scheduling periods Includes the following scheduling periods:

[0013]

[0014] in, The resampling interval is... , For scheduling time resolution, For the set of scheduling periods, For subset notation, For any symbol, Set of scheduling periods The scheduling time periods are represented by their indices, which indicate the sets of scheduling time periods. The The, the The, the The, the One, all the way up to the first One scheduling period;

[0015] (2) Calculate the maximum and minimum net load of the system per hour, and form 4 scheduling time period subsets in a certain order. The specific steps include the following:

[0016] a1: Set of scheduling periods Grouped hourly, the first The scheduling period included in the hour for:

[0017]

[0018] in, Set of scheduling periods The Each scheduling period, index The range of values ​​is based on the first The hour decided, for ;

[0019] a2: Statistics The first sample The time periods in which the maximum and minimum system net load values ​​occur across all scheduling periods within an hour;

[0020] a3: Based on the time periods of the system's maximum and minimum net load values, four scheduling time period subsets are formed in the following order: maximum-minimum, minimum-maximum, maximum-maximum-minimum, and minimum-minimum-maximum.

[0021]

[0022]

[0023]

[0024]

[0025] in, For the A subset of the maximum-minimum value scheduling periods from each sample. For the The minimum-maximum value scheduling period subset in each sample. For the A subset of the maximum-maximum-minimum scheduling periods from each sample. For the A subset of the minimum-minimum-maximum scheduling periods from a sample. and The first The first sample The index of the time period containing the maximum and minimum system net load values ​​across all scheduling periods within an hour. and They represent the first Hour and the Hour;

[0026] Obtained through the aforementioned time-series resampling rules A subset of scheduling time periods.

[0027] Furthermore, step S3 includes the following sub-steps:

[0028] S31: For each scheduling period subset, extract the new energy output data and load forecast data of the corresponding period in the dataset of the first day to form 1-hour resolution data;

[0029] The renewable energy output data includes the day-ahead predicted output data of renewable energy power plants, obtained from the data from the [missing information]. Based on the day-ahead predicted power output data of a sample of high temporal resolution renewable energy power stations, according to the first... A subset of scheduling periods Data slices extracted from the included scheduling periods:

[0030]

[0031] in, For the A sample of new energy power stations In the The time period of each scheduling period subset The recent forecast has been effective. For the A sample of new energy power stations In the scheduling period set Time period The following day's forecast showed that the force would be exerted. A collection of new energy power stations. For the A subset of scheduling periods;

[0032] The load forecast data includes day-ahead load forecast data, which is obtained by extracting the original day-ahead load forecast data according to the... Each subset of scheduling periods contains slices of data extracted from the scheduling periods:

[0033]

[0034] in, Indicates the first Sample nodes In the The time period of each scheduling period subset The load forecast values ​​for the day before, Indicates the first Sample nodes In the scheduling period set Next period The load forecast values ​​for the day before, For the set of load nodes;

[0035] S32: Each day, each scheduling period subset constitutes an independent sample, which together form the dataset of the second day prior.

[0036] Furthermore, step S4 includes the following sub-steps:

[0037] S41: Construct a set of proportional adjustment factors The formula is:

[0038]

[0039] in, To preset the adjustment factor limit, To adjust the step size, the expansion factor is: ;

[0040] S42: For the sample 1-hour time resolution renewable energy output data and load forecast data in the second-day-ahead dataset, which are expanded from the scheduling period subset of the first-day-ahead dataset, scale and expand them according to the proportional adjustment factor:

[0041]

[0042]

[0043] in, and They represent the first The first sample New energy power stations in the sample expanded from a subset of scheduling time periods During the period The predicted output and nodes of the day During the period The load forecast values ​​for the day before, and These respectively represent, based on the first... The new sample after scaling by a proportional adjustment factor Indicates the first One proportional adjustment factor;

[0044] S43: Subset of each scheduling period every day Each scaling factor, after being scaled and expanded, constitutes an independent sample, which together form the dataset from the third day prior.

[0045] Furthermore, the objective function for constructing a time-resolution unit combination model in S5, which has the same sample size as the dataset three days prior, is:

[0046]

[0047] Among them, in the first The first sample The first scheduling period subset is... Under the sample scaled by a proportional adjustment factor Total operating costs, To participate in the optimization of the unit set, For the A subset of scheduling periods , and The units Time period The binary variables represent the running status, whether a power-on event has occurred, and whether a shutdown event has occurred. , and The units Time period The no-load cost, start-up cost, and downtime cost, For the unit Time period Below the output level The relevant variable cost function for power generation.

[0048] The present invention also employs the following technical solution: a training data augmentation and rapid generation device for unit combination optimization problems, comprising:

[0049] Memory, used to store computer programs;

[0050] A processor is used to execute the computer program to implement the training data augmentation and rapid generation method for the above-described unit combination optimization problem.

[0051] The beneficial effects of this invention are as follows: This invention proposes a method for enhancing and rapidly generating training data for unit combination optimization problems. Based on time-series resampling and proportional expansion adjustment methods, it enhances and rapidly generates historical high-temporal-resolution unit combination data, achieving rapid acquisition of training data. This method significantly reduces the difficulty of training data collection and provides effective data support for the application of data-driven methods in large-scale high-temporal-resolution unit combination optimization problems. Attached Figure Description

[0052] Figure 1 A flowchart of a training data augmentation and rapid generation method for a unit combination optimization problem. Detailed Implementation

[0053] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0054] like Figure 1 As shown, a method for augmenting and rapidly generating training data for a unit combination optimization problem includes the following steps:

[0055] S1: Obtain historical high time resolution power system unit combination input data, including the scheduling period set, and denote it as the first day-ahead dataset. Calculate the system net load in the first day-ahead dataset.

[0056] S2: Based on the system net load concentrated in the data of the first day and the preset time-series resampling rules, the set of scheduling periods with high time resolution is resampled into a subset of scheduling periods with a set time resolution;

[0057] S3: Based on the resampled scheduling period subset, extract the corresponding high time resolution new energy output data and load forecast data from the first day's dataset by time period to form the set time resolution data, and obtain the second day's dataset;

[0058] S4: Based on the preset data augmentation rules, the renewable energy output data and load forecast data of the second-day dataset with a set time resolution are proportionally adjusted and expanded to form the third-day dataset.

[0059] S5: Construct a unit combination model with a set time resolution consistent with the number of samples in the dataset three days prior and solve it. Record the optimization solution results and their analysis results as new data to complete the training data augmentation and rapid generation of the unit combination optimization problem.

[0060] In step S1, historical high-temporal-resolution power system unit combination input data, including the scheduling period set, is obtained. The input data for each day constitutes an independent sample, collectively forming the first day-ahead dataset. The high-temporal-resolution power system unit combination input data for each sample in the first day-ahead dataset includes:

[0061] Set of scheduling periods: All scheduling periods within the next 24 hours, using... The specific number of time slots depends on the scheduling time resolution. High time resolution generally refers to 15 minutes, 10 minutes, or 5 minutes, in which case the number of scheduling time slots... The values ​​are 96, 144, and 288 respectively. The following explanation uses a 15-minute scheduling time resolution as an example.

[0062] Power grid topology data: the connection relationships between power network nodes and transmission lines, and the output power transfer distribution factor matrix of each type of generating unit and node load to each transmission line;

[0063] Cost data involved in optimizing the generating unit: start-up cost, shutdown cost, no-load cost, and variable generation cost function for each time period;

[0064] The following operational characteristic data of the unit are used to optimize the unit's minimum / maximum technical output, unit's ramp-up capability, and unit's minimum continuous start-up and shutdown time.

[0065] Load day forecast data: Load day forecast data for each node, with the same time resolution as the scheduling time resolution;

[0066] Forecasted output data for new energy power plants day-ahead; the time resolution of the forecasted output data for new energy power plants day-ahead is consistent with the dispatch time resolution.

[0067] The formula for calculating the net system load for each sample in the first day's dataset is as follows:

[0068]

[0069] in, Represents the set of scheduling periods. Indicates the first Sample time period The system net load value, Indicates the first Sample nodes During the period The load forecast values ​​for the day before, Indicates the first A sample of new energy power stations During the period The recent forecast has been effective. This refers to a collection of new energy power stations, including photovoltaic power stations and wind farms. Represents the set of load nodes;

[0070] The preset timing resampling rules in S2 include:

[0071] (1) Based on a 1-hour time resolution, the set of scheduling periods is decomposed into S non-overlapping subsets of scheduling periods that are associated with the scheduling time resolution, where the first subset is... The first sample A subset of scheduling periods Includes the following scheduling periods:

[0072]

[0073] in, The resampling interval is... , The scheduling time resolution can be 15 minutes, 10 minutes, or 5 minutes, in which case the number of scheduling time period subsets is 4, 6, or 12, respectively. For the set of scheduling periods, For subset notation, For any symbol, Set of scheduling periods The scheduling time periods are represented by their indices, which indicate the sets of scheduling time periods. The The, the The, the The, the One, all the way up to the first One scheduling period;

[0074] (2) Calculate the maximum and minimum net load of the system per hour, and form 4 scheduling time period subsets in a certain order. The specific steps include the following:

[0075] a1: Set of scheduling periods Grouped hourly, the first The scheduling period included in the hour for:

[0076]

[0077] in, Set of scheduling periods The Each scheduling period, index The range of values ​​is based on the first The hour decided, for ;

[0078] a2: Statistics The first sample The time periods in which the maximum and minimum system net load values ​​occur across all scheduling periods within an hour;

[0079] a3: Based on the time periods of the system's maximum and minimum net load values, four scheduling time period subsets are formed in the following order: maximum-minimum, minimum-maximum, maximum-maximum-minimum, and minimum-minimum-maximum.

[0080]

[0081]

[0082]

[0083]

[0084] in, For the first A subset of the maximum-minimum value scheduling periods from each sample. For the first The minimum-maximum value scheduling period subset in each sample. For the first A subset of the maximum-maximum-minimum scheduling periods from each sample. For the first A subset of the minimum-minimum-maximum scheduling periods from a sample. and The first The first sample The index of the time period containing the maximum and minimum system net load values ​​across all scheduling periods within an hour. and They represent the first Hour and the Hour;

[0085] Obtained through the aforementioned time-series resampling rules A subset of scheduling time periods.

[0086] According to the first S3 Resampling of individual samples A subset of scheduling periods High-temporal-resolution renewable energy output and load forecast data are extracted from the dataset from the first day and divided into time periods to form 1-hour time-resolution data:

[0087] S3 includes the following sub-steps:

[0088] S31: For each scheduling period subset, extract the new energy output data and load forecast data of the corresponding period in the dataset of the first day to form 1-hour resolution data;

[0089] The renewable energy output data includes the day-ahead predicted output data of renewable energy power plants, obtained from the data from the [missing information]. Based on the day-ahead predicted power output data of a sample of high temporal resolution renewable energy power stations, according to the first... A subset of scheduling periods Data slices extracted from the included scheduling periods:

[0090]

[0091] in, For the first A sample of new energy power stations In the The time period of each scheduling period subset The recent forecast of output, because Including representative time periods for each hour of the following day, it is therefore converted to a 1-hour time resolution. For the first A sample of new energy power stations In the scheduling period set Time period The current day's forecast output is high temporal resolution data, the first A subset of scheduling periods and scheduling time set Time period Both indicate the same scheduling period. It is a collection of new energy power stations, including photovoltaic power stations and wind farms. For the first A subset of scheduling periods;

[0092] The load forecast data includes day-ahead load forecast data, which is obtained by extracting the original day-ahead load forecast data according to the... Each subset of scheduling periods contains slices of data extracted from the scheduling periods:

[0093]

[0094] in, Indicates the first Sample nodes In the The time period of each scheduling period subset The load forecast values ​​for the day before, Indicates the first Sample nodes In the scheduling period set Next period The load forecast values ​​for the day before, For the set of load nodes;

[0095] S32: Each day, each scheduling period subset constitutes an independent sample, which together form the second day's dataset. Samples from the same day inherit the same grid topology data, cost data of the units participating in optimization, and operating characteristic data of the units participating in optimization, retaining only the corresponding scheduling period subset, the 1-hour time resolution load forecast data, and the new energy power plant forecast output data.

[0096] S4 includes the following sub-steps:

[0097] S41: Construct a set of proportional adjustment factors The formula is:

[0098]

[0099] in, To preset the adjustment factor limit, To adjust the step size, the expansion factor is: ;

[0100] S42: For the sample 1-hour time resolution renewable energy output data and load forecast data in the second-day-ahead dataset, which are expanded from the scheduling period subset of the first-day-ahead dataset, scale and expand them according to the proportional adjustment factor:

[0101]

[0102]

[0103] in, and They represent the first The first sample New energy power stations in the sample expanded from a subset of scheduling time periods During the period The predicted output and nodes of the day During the period The load forecast values ​​for the day before, and These respectively represent, based on the first... The new sample after scaling by a proportional adjustment factor Indicates the first One proportional adjustment factor;

[0104] S43: Subset of each scheduling period every day Each proportional adjustment factor, after scaling and expansion, constitutes an independent sample, collectively forming the third-day day-ahead dataset. Samples from the same day and the same scheduling period subset inherit the same grid topology data, cost data of participating units, operating characteristic data of participating units, and scheduling period subset. Only the day-ahead load forecast data and the day-ahead power output forecast data of renewable energy power plants with the corresponding proportional adjustment factor scaling and expansion are retained.

[0105] In S5, the dataset is in the third day. The first sample The first scheduling period subset is... Before constructing the unit combination optimization model corresponding to the sample scaled by the proportional adjustment factor, it is necessary to use the power system unit combination input data according to the first... Individual samples of high temporal resolution power system unit combination input data and dispatch time resolution Convert and match with a 1-hour time resolution:

[0106] Cost data for the units involved in optimization:

[0107]

[0108]

[0109]

[0110]

[0111] Among them, the The first sample A subset of scheduling periods and scheduling time set Time period Both indicate the same scheduling period. , , and The first The first sample The first scheduling period subset is... The sample unit after scaling by a proportional adjustment factor Time period No-load costs, start-up costs, shutdown costs and related costs of the unit During the period The power generation variable cost function related to the power output level. , , and The first A set of sample scheduling periods Time period Lower unit No-load costs, start-up costs, shutdown costs and related costs of the unit During the period The power generation variable cost function related to the output level, the conversion ratio and the dispatch time resolution Related, for ;

[0112] Operating characteristic data of the unit involved in optimization:

[0113]

[0114]

[0115]

[0116]

[0117]

[0118]

[0119] in, , , , , and The first The first sample The first scheduling period subset is... The sample unit after scaling by a proportional adjustment factor Maximum technical output, minimum technical output, maximum uphill rate, maximum downhill rate, minimum continuous uptime, and minimum continuous downtime parameters. , , , , and The first A set of sample scheduling periods Lower unit Maximum technical output, minimum technical output, maximum uphill speed, maximum downhill speed, minimum continuous uptime, and minimum continuous downtime parameters;

[0120] The objective function for constructing the time-resolution unit combination model in S5, which has the same sample size as the dataset three days prior, is:

[0121]

[0122] Among them, in the first The first sample The first scheduling period subset is... Under the sample scaled by a proportional adjustment factor Total operating costs, To participate in the optimization of the unit set, For the first A subset of scheduling periods , and The units Time period The binary variables represent the running status, whether a power-on event has occurred, and whether a shutdown event has occurred. , and The units Time period The no-load cost, start-up cost, and downtime cost, For the unit Time period Below the output level The relevant variable cost function for power generation.

[0123] The constraints of the above unit combination optimization model include upper and lower limits of unit output, unit start-up and shutdown status constraints, minimum continuous start-up and shutdown constraints, unit ramping constraints, power system supply and demand balance constraints, power system positive and negative reserve constraints, and line power flow safety constraints.

[0124] The upper and lower limits of the unit output are constrained as follows:

[0125]

[0126] in, and The first The first sample The first scheduling period subset is... The sample unit after scaling by a proportional adjustment factor Minimum and maximum technical output;

[0127] The unit start-up and shutdown status constraints are:

[0128]

[0129] in, Indicates the first The first sample The first scheduling period subset is... The sample unit after scaling by a proportional adjustment factor Time period The running status under;

[0130] The minimum continuous start-stop constraint for the unit is:

[0131]

[0132]

[0133] in, and They represent the first The first sample The first scheduling period subset is... The sample unit after scaling by a proportional adjustment factor Time period Did a power-on event or a shutdown event occur? and They represent the generating units. The minimum continuous start-up time parameter and the minimum continuous stop-down time parameter;

[0134] The unit's ramp-up constraint is:

[0135]

[0136] in, and They represent the first The first sample The first scheduling period subset is... The sample unit after scaling by a proportional adjustment factor Maximum downhill and maximum uphill speeds Indicates the unit During the period The output level is below;

[0137] The power system supply and demand balance constraints are:

[0138]

[0139] in, and They represent the first The first sample The first scheduling period subset is... New energy power stations in the sample after scaling by a proportional adjustment factor Time period The following day's predicted output and nodes Time period The following are the day-ahead load forecasts. It is a collection of new energy power stations, including photovoltaic power stations and wind farms. It is a set of load nodes;

[0140] The positive and negative reserve constraints of the power system are:

[0141]

[0142]

[0143] in, Indicates the system's positive reserve rate. This indicates the system's negative reserve rate.

[0144] The power flow safety constraints of the line are:

[0145]

[0146] in, For the line Steady-state power flow limit value, , and The units New energy power stations and nodes For the line The power transfer distribution factor, A collection of transmission lines;

[0147] Finally, a commercial solver was used to solve the unit combination optimization model, yielding the optimal start-up and shutdown schedule matrix for each unit. The optimized solution results are included.

[0148] Based on the optimization results of the unit combination optimization model, the 1st unit can be further analyzed and determined. The first sample The first scheduling period subset is... The set of units that must be started and stopped and the set of units with flexibility that experience start-up and shutdown events, scaled by a proportional adjustment factor:

[0149]

[0150]

[0151]

[0152] in, To participate in the optimization of the unit set, This represents the set of units that must be started, when the optimal start-up and shutdown plan matrix for the units is... medium-sized units When all rows are 1, the unit That is, the set of elements that must be turned on. Represents the set of units that must be shut down, when the optimal start-up and shutdown plan matrix for the units is... medium-sized units When all rows are 0, the unit That is, the set of units that must be shut down. This indicates a flexible set of generating units, when the units... When a unit belongs to neither the set of units that must be started nor the set of units that must be stopped, it is considered an element of the flexible unit set. This represents the difference operation.

[0153] The aforementioned mandatory start / stop and flexible unit sets, along with data on power flow and unit output for each time period and line, together constitute a complete analysis result.

[0154] Example 2: A training data augmentation and rapid generation device for a unit combination optimization problem, comprising:

[0155] Memory, used to store computer programs;

[0156] A processor is used to execute the computer program to implement the training data augmentation and rapid generation method for the above-described unit combination optimization problem.

[0157] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of the invention.

Claims

1. A method for augmenting and rapidly generating training data for a unit combination optimization problem, characterized in that, Includes the following steps: S1: Obtain historical high time resolution power system unit combination input data, including the scheduling period set, and denote it as the first day-ahead dataset. Calculate the system net load in the first day-ahead dataset. S2: Based on the system net load concentrated in the data of the first day and the preset time-series resampling rules, the set of scheduling periods with high time resolution is resampled into a subset of scheduling periods with a set time resolution; S3: Based on the resampled scheduling period subset, extract the corresponding high time resolution new energy output data and load forecast data from the first day's dataset by time period to form the set time resolution data, and obtain the second day's dataset; S4: Based on the preset data augmentation rules, the renewable energy output data and load forecast data of the second-day dataset with a set time resolution are proportionally adjusted and expanded to form the third-day dataset. S5: Construct a unit combination model with a set time resolution consistent with the number of samples in the dataset three days prior and solve it. Record the optimization solution results and their analysis results as new data to complete the training data augmentation and rapid generation of the unit combination optimization problem.

2. The method for training data augmentation and rapid generation of the unit combination optimization problem according to claim 1, characterized in that, The preset timing resampling rules in S2 include: (1) Based on a 1-hour time resolution, the set of scheduling periods is decomposed into S non-overlapping subsets of scheduling periods that are associated with the scheduling time resolution, where the first subset is... The first sample A subset of scheduling periods Includes the following scheduling periods: in, The resampling interval is... , For scheduling time resolution, For the set of scheduling periods, For subset notation, For any symbol, Set of scheduling periods The scheduling time periods are represented by their indices, which indicate the sets of scheduling time periods. The The, the The, the The, the One, all the way up to the first One scheduling period; (2) Calculate the maximum and minimum net load of the system per hour, and form 4 scheduling time period subsets in a certain order. The specific steps include the following: a1: Set of scheduling periods Grouped hourly, the first The scheduling period included in the hour for: in, Set of scheduling periods The Each scheduling period, index The range of values ​​is based on the first The hour decided, for ; a2: Statistics The first sample The time periods in which the maximum and minimum system net load values ​​occur across all scheduling periods within an hour; a3: Based on the time periods of the system's maximum and minimum net load values, four scheduling time period subsets are formed in the following order: maximum-minimum, minimum-maximum, maximum-maximum-minimum, and minimum-minimum-maximum. in, For the A subset of the maximum-minimum value scheduling periods from each sample. For the The minimum-maximum value scheduling period subset in each sample. For the A subset of the maximum-maximum-minimum scheduling periods from each sample. For the A subset of the minimum-minimum-maximum scheduling periods from a sample. and The first The first sample The index of the time period containing the maximum and minimum system net load values ​​across all scheduling periods within an hour. and They represent the first Hour and the Hour; Obtained through the aforementioned time-series resampling rules A subset of scheduling time periods.

3. The method for training data augmentation and rapid generation of the unit combination optimization problem according to claim 2, characterized in that, S3 includes the following sub-steps: S31: For each scheduling period subset, extract the new energy output data and load forecast data of the corresponding period in the dataset of the first day to form 1-hour resolution data; The renewable energy output data includes the day-ahead predicted output data of renewable energy power plants, obtained from the data from the [missing information]. Based on the day-ahead predicted power output data of a sample of high temporal resolution renewable energy power stations, according to the first... A subset of scheduling periods Data slices extracted from the included scheduling periods: in, For the A sample of new energy power stations In the The time period of each scheduling period subset The recent forecast has been effective. For the A sample of new energy power stations In the scheduling period set Time period The following day's forecast showed that the force would be exerted. A collection of new energy power stations. For the A subset of scheduling periods; The load forecast data includes day-ahead load forecast data, which is obtained by extracting the original day-ahead load forecast data according to the... Each subset of scheduling periods contains slices of data extracted from the scheduling periods: in, Indicates the first Sample nodes In the The time period of each scheduling period subset The load forecast values ​​for the day before, Indicates the first Sample nodes In the scheduling period set Next period The load forecast values ​​for the day before, For the set of load nodes; S32: Each day, each scheduling period subset constitutes an independent sample, which together form the dataset of the second day prior.

4. The method for training data augmentation and rapid generation of the unit combination optimization problem according to claim 3, characterized in that, S4 includes the following sub-steps: S41: Construct a set of proportional adjustment factors The formula is: in, To preset the adjustment factor limit, To adjust the step size, the expansion factor is: ; S42: For the sample 1-hour time resolution renewable energy output data and load forecast data in the second-day-ahead dataset, which are expanded from the scheduling period subset of the first-day-ahead dataset, scale and expand them according to the proportional adjustment factor: in, and They represent the first The first sample New energy power stations in the sample expanded from a subset of scheduling time periods During the period The predicted output and nodes of the day During the period The load forecast values ​​for the day before, and These respectively represent, based on the first... The new sample after scaling by a proportional adjustment factor Indicates the first One proportional adjustment factor; S43: Subset of each scheduling period every day Each scaling factor, after being scaled and expanded, constitutes an independent sample, which together form the dataset from the third day prior.

5. The method for training data augmentation and rapid generation of the unit combination optimization problem according to claim 4, characterized in that, The objective function for constructing the time-resolution unit combination model in S5, which has the same sample size as the dataset three days prior, is: Among them, in the first The first sample The first scheduling period subset is... Under the sample scaled by a proportional adjustment factor Total operating costs, To participate in the optimization of the unit set, For the A subset of scheduling periods , and The units Time period The binary variables represent the running status, whether a power-on event has occurred, and whether a shutdown event has occurred. , and The units Time period The no-load cost, start-up cost, and downtime cost, For the unit Time period Below the output level The relevant variable cost function for power generation.

6. A device for augmenting and rapidly generating training data for a unit combination optimization problem, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the training data augmentation and rapid generation method for the unit combination optimization problem as described in any one of claims 1-5.

Images