A method, system, medium and processor for evaluating power system peak shaving capacity of a domain large model

Abstract

The application provides a power system peak shaving capacity evaluation method of a domain large model, comprising the following steps: S1, obtaining a historical data set and calculating a corresponding index data set; S2, supplementing the index data set to the historical data set to form a peak shaving capacity evaluation case data set; S3, performing semanticization on the data set to obtain a complete semantic data set; S4, inputting the complete semantic data set into a large language model to search for similar information, and obtaining a richness-enhanced semantic data set; S5, decoding the richness-enhanced semantic data set, inputting the decoded data set into a domain model for domainization, and then performing semantic encoding to obtain a domain-enhanced semantic data set; S6, inputting the domain-enhanced semantic data set into the large language model for model reinforcement training; S7, then inputting system data to be evaluated to obtain an evaluation result; and S8, performing semantic decoding on the result. The problems of insufficient accuracy of case prediction results of multi-source information training data, great randomness, and poor professional matching degree are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of peak load regulating capacity evaluation, and in particular to a power system peak load regulating capacity evaluation method and system based on a domain large model, a medium and a processor. BACKGROUND

[0002] The peak load regulating capacity is the difference between the total maximum adjustable output and the total minimum technical output of the peak load regulating power plant in the system. Its main role is to maintain power balance and keep the system frequency stable, and to ensure that the power grid can meet the electricity demand during peak periods. For example, a Chinese patent application with the application number CN202211631509.7 discloses a peak load regulating capacity market capacity demand evaluation system and method. The system includes a login authentication module, a data entry module, an uncertainty calculation module, a peak load regulating capacity calculation module, and a data output module. The uncertainty calculation module calculates n load power curves, new energy curves, and external power curves according to historical curve related data. The peak load regulating capacity demand calculation module calculates the minimum output gear of the thermal power unit and the peak load regulating capacity demand of each gear to achieve the expected new energy curtailment rate based on the curves and the relevant parameters of the power system transmitted by the data entry module. The data output module outputs an evaluation report. The random time series production simulation technology considering the uncertainty of future load growth, new energy generation uncertainty, and external power uncertainty is used to realize the accurate evaluation function of the peak load regulating capacity market.

[0003] However, the existing power system peak load regulating capacity evaluation method cannot accurately predict and evaluate complex data cases containing multi-source information, and the prediction accuracy of cases containing multi-source information training data is insufficient. In addition, the existing large model prediction method has large randomness and poor professional fit in the evaluation of power system peak load regulating capacity. In addition, the existing peak load regulating capacity evaluation method cannot effectively solve the problem of inaccurate data evaluation in the case of lack of training cases.

[0004] Therefore, a power system peak load regulating capacity evaluation method, system, medium, and processor based on a domain large model are needed. SUMMARY

[0005] To solve the problems of large randomness and poor professional fit of single large model prediction evaluation results in the prior art, the present application provides a power system peak load regulating capacity evaluation method, system, medium, and processor based on a domain large model, which can solve the problems of large randomness and poor professional fit of single large model prediction evaluation results. The method uses a large model to search and screen cases to solve the problem of inaccurate prediction and evaluation in the case of lack of actual training cases. The specific technical solutions are as follows:

[0006] A power system peak load regulating capacity evaluation method based on a domain large model, comprising:

[0007] S1: Obtain a historical data set of a power system, calculate corresponding peak shaving capacity evaluation indexes according to the historical data set, and obtain a corresponding index data set;

[0008] S2: Supplement the index data set to the historical data set to jointly constitute a peak shaving capacity evaluation case data set;

[0009] S3: Perform semanticization on the obtained peak shaving capacity evaluation case data set and the data set of the system to be evaluated to obtain a complete semantic data set

[0010] S4: Input the complete semantic data set to a large language model to search for similar information and screen cases to expand the number of cases, and obtain a richness-enhanced semantic data set

[0011] S5: Perform semantic decoding on the richness-enhanced semantic data set , input the decoded data set to a domain model to field the data set, and then perform semantic encoding on the fielded data set to obtain a domain-enhanced semantic data set

[0012] S6: Input the semantic domain-enhanced semantic data set to a large language model for model reinforcement training;

[0013] S7: Input the semantic data of the system to be evaluated into the large language model after reinforcement to evaluate the peak shaving demand result, and obtain an evaluation result in the form of an evaluation.

[0014] S8: Perform semantic decoding on the evaluation result in the form of an evaluation to obtain the peak shaving capacity equivalent load curve data of the system to be evaluated as the result of the peak shaving capacity evaluation.

[0015] Further, in step S1, the obtaining of the historical data set of the power system, the calculation of the corresponding peak shaving capacity evaluation indexes according to the historical data set, and the obtaining of the corresponding index data set include the following steps:

[0016] S11: Obtain a load data set, a conventional output data set, and a wind-solar output data set of a power system, and the formulas of the data sets are as follows:

[0017]

[0018] In the above formula, E1(x load ) is the load data set, x load represents variables of all types of power loads; n is the total number of time scales of data division; x tem is the ti Instantaneous temperature data; is the t i Instantaneous light intensity data; is the t i Instantaneous wind speed data; is the t i Instantaneous load active demand data;

[0019] E2(x gen,nom ) is a set of conventional output data; x gen,nom is a variable representing conventional output data; is the t i Instantaneous conventional unit cost data; represents the value of conventional active output at the t i Instantaneous conventional active output value;

[0020] E3(x gen,new ) is a set of wind-solar output data; x gen,new is a variable representing wind-solar output data; is the t i Instantaneous wind-solar active output data; is the t i Instantaneous photovoltaic active output data; is the t i Instantaneous wind power active output data;

[0021] S12: Calculate the time sequence characteristic index of the load, obtain the daily maximum peak-valley difference ΔP load,week and the peak-valley difference rate ρ load,week of the weekly maximum peak-valley day, form the peak-valley difference data set E4(ΔP load,week ) and the peak-valley difference rate data set E5(ρ load,week ), and the calculation formula is as follows:

[0022]

[0023] In the above formula, ΔP load,week represents the weekly statistical daily maximum peak-valley difference; P peak,week , P valley,week represent the peak value and the valley value of the weekly maximum peak-valley day; ρ load,week represents the peak-valley difference rate of the weekly maximum peak-valley day; is the conventional active output value from the 1st hour to the 24th hour of the day with the maximum peak-valley difference in a week, and the day with the maximum peak-valley difference is obtained by screening the day with the maximum peak-valley difference value;

[0024] S13: Calculate the anti-peaking characteristic index of wind-solar output, obtain the wind-solar anti-peaking rate, form the wind-solar anti-peaking rate data set E6(r repeak ), and the calculation formula is as follows:

[0025]

[0026] In the above formula, r repeak For wind and solar inverse peak rate; m count For the number of times the wind and solar power peaks are reversed; d total The total number of days counted;

[0027] S14: Calculate the equivalent load curve of peak-shaving capacity, and form the peak-shaving capacity equivalent load curve dataset E7[P] equ-load (d day,t The equivalent load curve of the peak-shaving capacity reflects the peak-shaving capacity of the power system; the calculation formula is as follows:

[0028] P equ-load (d day,t ) = P load (d day,t )-P newenergy (d day,t ),d day ∈[1,7],t∈[0,23];

[0029] In the above formula, P equ-load (d day,t P represents the equivalent load power of wind and solar power output on a certain day at time t. load (d day,t P represents the load power at time t on a certain day; newenergy (d day,t ) represents the wind and solar power output at time t on a certain day.

[0030] Furthermore, in S2, the step of supplementing the indicator dataset to the historical dataset to jointly constitute the peak-shaving capacity assessment case dataset refers to adding the peak-valley difference dataset E4(ΔP) to the historical dataset. load,week ), Peak-to-valley difference dataset E5(ρ load,week ), Wind and solar inverse peak rate dataset E6(r repeak ) and peak-shaving capacity equivalent load curve dataset E7[P equ-load (d day,t )] and load dataset E1(x load ), conventional output dataset E2(x gen,nom ) and the solar power output dataset E3(x gen,new These combinations together constitute the peak-shaving capacity assessment case dataset.

[0031] Furthermore, in step S3, the semantic formula is as follows:

[0032]

[0033] In the above formula, This represents the k-th dataset in the peak-shaving capacity assessment case dataset at time t, where k∈[1,7]. The seven datasets are the load datasets E1(x) and E2(x)(x). load ), conventional output data set E2(x gen,nom ), Wind and solar power output data set E3(x gen,new ), Peak-to-valley difference dataset E4(ΔP) load,week ), Peak-to-valley difference dataset E5(ρ load,week ), Wind and solar inverse peak rate dataset E6(r repeak ) and peak-shaving capacity equivalent load curve data E7[P equ-load (d day,t The time scale and the total number of moments are determined based on the size of the dataset; This is the semantic statement corresponding to the k-th dataset at time t; Encode() is a semantic encoder that compiles data by writing simple data type names and data correspondences into a simple sentence.

[0034] Furthermore, in step S4, the case screening includes the following steps:

[0035] S41: Semantic information must contain the existence of E1(x) load E2(x) gen,nom ) and E3(x gen,new The corresponding data, where E1(x) load ) must contain Data, E2(x) gen,nom ) must contain Data, E3(x) gen,new ) must contain At least one of the three data points;

[0036] S42: The total number of moments T contained in the semantic information is greater than or equal to 12.

[0037] Furthermore, in step S5, the semantic decoding process is the inverse process of semantic encoding, as shown in the following expression:

[0038]

[0039] In the above formula, Decode() is a semantic decoder. The units of data generated by Encode() and recognized by Decode() are unified by the large language model, and the corresponding unit conversion is also performed by the large model.

[0040] Furthermore, in step S5, the prediction training principle of the domain model is as follows:

[0041]

[0042] In the above formula, and is the training input data of the lth and the l+1th layer; Q(l) and Q(l+1) are the adaptive training weight coefficients of the lth and the l+1th layer respectively; b(l) and b(l+1) are the adaptive training bias coefficients of the lth and the l+1th layer respectively; is the input peak shaving capacity equivalent load curve training data, that is, the training output data; Softmax() is an activation function.

[0043] A power system peak shaving capacity evaluation system of a domain large model, applied to the domain large model power system peak shaving capacity evaluation method described above, comprising:

[0044] An index calculation module for obtaining a historical data set of a power system, calculating corresponding peak shaving capacity evaluation indexes according to the historical data set, and obtaining a corresponding index data set;

[0045] A construction module for supplementing the index data set to the historical data set to jointly constitute a peak shaving capacity evaluation case data set;

[0046] A semantic module for obtaining the peak shaving capacity evaluation case data set and the data set of the system to be evaluated to perform semanticization, obtaining a complete semantic data set

[0047] An expansion module for inputting the complete semantic data set to a large language model to search for similar information and select cases to expand the number of cases, and obtain a richness-enhanced semantic data set

[0048] A domain module for inputting the richness-enhanced semantic data set to perform semantic decoding, inputting the decoded data set to a domain model to field the data set, and performing semantic encoding on the fielded data set to obtain a domain-enhanced semantic data set

[0049] A training module for inputting the semantic domain-enhanced semantic data set to the large language model to perform model reinforcement training;

[0050] An evaluation module for inputting the semantic system to be evaluated into the reinforced large language model to evaluate the peak shaving demand result, and obtaining a review evaluation result;

[0051] A decoding module for performing semantic decoding on the review evaluation result to obtain the peak shaving capacity equivalent load curve data of the system to be evaluated as the result of the peak shaving capacity evaluation.

[0052] A computer-readable storage medium comprising a stored program, wherein the computer-readable storage medium controls a device in which the computer-readable storage medium is located to execute the power system peak regulation capacity evaluation method of the domain large model described above when the program is run.

[0053] A processor for running a program, wherein the power system peak regulation capacity evaluation method of the domain large model described above is executed when the program is run.

[0054] Compared with the prior art, the beneficial effects of the present application are:

[0055] 1. A power system peak regulation capacity evaluation method of a domain large model is proposed to solve the problem of insufficient accuracy of case prediction results of multi-source information training data; the method of applying a language large model combined with a domain model is used to solve the problem of large randomness of prediction results of a single large model and poor professional fit; and the method of using a large model to search and screen cases is used to solve the problem of inaccurate prediction and evaluation in the case of lack of actual training cases. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the specific embodiments or the prior art, the drawings needed in the specific embodiments or the prior art description will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, each element or part is not necessarily drawn according to the actual scale.

[0057] Figure 1 The first power system peak regulation capacity evaluation method of a domain large model is a flowchart.

[0058] Figure 2 The second power system peak regulation capacity evaluation method of a domain large model is a flowchart. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0060] It should be understood that when used in the present specification and the appended claims, the terms "comprise" and "include" indicate the presence of described features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or sets thereof.

[0061] It is also to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. As used in this specification and the appended claims, the singular forms "a," "an" and "the" include plural referents unless the context clearly dictates otherwise.

[0062] It is further to be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items, and that the term "at least one of' encompasses each possible combination of one or more of the associated listed items.

[0063] Embodiment one

[0064] Using a large language model for the enrichment and prediction evaluation tasks of the training data set can quickly make predictions of high-dimensional data and effectively deal with the lack of training data, while dealing with high complexity of training data cases can simplify the difficulty of data processing, save the step of manual parameter tuning, use domain models to assist large model training, and improve the professional fit of large model training data and the accuracy of prediction.

[0065] As shown in Figure 1 , Figure 2 is a flowchart of a power system peak shaving capacity evaluation method of a domain large model, including:

[0066] S1: Obtain a historical data set of a power system, calculate corresponding peak shaving capacity evaluation indexes according to the historical data set, and obtain a corresponding index data set.

[0067] Further, in step S1, the historical data set of the power system is obtained, the corresponding peak shaving capacity evaluation indexes are calculated according to the historical data set, and the corresponding index data set is obtained, including the following steps:

[0068] S11: Obtain a load data set, a conventional output data set and a wind-solar output data set of a power system, and each data set formula is as follows:

[0069]

[0070] In the above formula, E1(x load ) is the load data set, x load represents the variables of all types of power load; n is the total number of time scales of data division; x tem is the temperature data at the t i time; is the light intensity data at the t i time; is the wind speed data at the t i time; is the wind speed data at the t iActive load demand data at each time;

[0071] E2(x gen,nom ) is a set of conventional output data; x gen,nom is a variable representing conventional output data; is the conventional unit cost data at the t i time; represents the conventional active power output value at the t i time;

[0072] E3(x gen,new ) is a set of wind-solar output data; x gen,new is a variable representing wind-solar output data; is the wind-solar active power output data at the t i time; is the photovoltaic active power output data at the t i time; is the wind power active power output data at the t i time.

[0073] S12: Calculate the time sequence characteristic index of the load, which includes the peak-valley difference and the peak-valley difference rate. The peak-valley difference and the peak-valley difference rate of the day with the maximum daily peak-valley difference in each week are used to measure the maximum daily peak-valley difference in each week and the proportion of the daily peak-valley difference to the peak value of the load, to obtain the daily maximum peak-valley difference ΔP load,week and the peak-valley difference rate ρ load,week of the day with the maximum peak-valley difference in each week, forming a peak-valley difference data set E4(ΔP load,week ) and a peak-valley difference rate data set E5(ρ load,week ), and the calculation formula is as follows:

[0074]

[0075] In the above formula, ΔP load,week represents the daily maximum peak-valley difference calculated by week; P peak,week , P valley,week represent the peak value and the valley value of the day with the maximum peak-valley difference in each week; ρ load,week represents the peak-valley difference rate of the day with the maximum peak-valley difference in each week; is the conventional active power output value from the 1st hour to the 24th hour of the day with the maximum peak-valley difference in each week, which is obtained by screening the day with the maximum peak-valley difference.

[0076] S13: Calculate the anti-peaking characteristic index of wind-solar output, to obtain the wind-solar anti-peaking rate, forming a wind-solar anti-peaking rate data set E6(r repeak ), and the calculation formula is as follows:

[0077]

[0078] In the above formula, r repeak For wind and solar inverse peak rate; m count For the number of times the wind and solar power peaks are reversed; d total The total number of days counted; number of times wind and solar peak shifts (m) count The calculation requires first determining the conventional power output peak shaving rate and the wind and solar power output peak shaving rate; the expression for the conventional power output peak shaving rate is:

[0079]

[0080] Where, r load (d day ) represents the normal peak output rate over a 24-hour period; P load,max (d day P represents the maximum daily load power for that load; load,min (d day () represents the minimum daily power output for that load. This represents the normal active power output from the 1st to the 24th hour of the day; the expression for the wind and solar power output peak regulation rate is:

[0081]

[0082] Where, r ne (d day P represents the peak load regulation rate of solar and wind power output over a 24-hour period. ne,max (d day P represents the maximum power output of the solar power system in a single day. ne,min (d day This represents the minimum power output of the solar power system in a single day. and The data represent the active power output of photovoltaic power and the active power output of wind power from the first hour to the 24th hour of the day, respectively.

[0083] After calculating the normal power output peak shaving rate and the wind and solar power output peak shaving rate for each day, if the wind and solar power output peak shaving rate is greater than the normal power output peak shaving rate, then that day is an anti-peak shaving day, i.e., r. equ-load (d day )>r load (d day If m is the number of times the wind and solar peaks are reversed, then... count Add one to the value.

[0084] S14: Calculate the equivalent load curve of peak-shaving capacity, and form the peak-shaving capacity equivalent load curve dataset E7[P] equ-load (d day,t); the peak regulation capacity equivalent load curve reflects the peak regulation capacity of the power system as a result. Among all the output curves, the curve with the maximum value of the wind-solar anti-peak regulation rate index value calculated by S13 is selected as the time sequence process curve with the most obvious anti-peak regulation characteristic, and the curve with the maximum value of the daily maximum peak-valley difference index value calculated by S12 is selected as the daily output curve with the maximum peak-valley difference. The combination of the time sequence process curve with the most obvious anti-peak regulation characteristic and the daily output curve with the maximum peak-valley difference is the equivalent load time sequence process curve, which is the peak regulation capacity equivalent load curve of the power grid. The calculation formula of the peak regulation capacity equivalent load of seven days with one hour as a period is as follows:

[0085] P equ-load (d day,t )=P load (d day,t )-P newenergy (d day,t ),d day ∈[1,7],t∈[0,23];

[0086] In the above formula, P equ-load (d day,t ) is the equivalent load power of the wind-solar output of a day at t o'clock; P load (d day,t ) is the load power at t o'clock of a day; and P newenergy (d day,t ) is the wind-solar output power at t o'clock of a day.

[0087] The calculation of the conventional output data E2(x repeak ) and the wind-solar output data E3(x load,week ) corresponding to the time sequence process curve with the most obvious anti-peak regulation characteristic, i.e., the maximum r gen,nom , and the daily output curve with the maximum peak-valley difference, i.e., the maximum ΔP gen,new , can obtain the equivalent load time sequence process curve composed of P equ-load (d day,t ) data.

[0088] S2: The index data set is supplemented to the historical data set to jointly constitute the peak regulation capacity evaluation case data set.

[0089] Further, in S2, the supplement of the index data set to the historical data set to jointly constitute the peak regulation capacity evaluation case data set refers to the supplement of the peak-valley difference data set E4(ΔP load,week ), the peak-valley difference rate data set E5(ρ load,week ), the wind-solar anti-peak regulation rate data set E6(r repeak ), and the peak regulation capacity equivalent load curve data set E7[P equ-load (d day,t )] to the load data set E1(xload ), conventional power output dataset E2(x gen,nom ), and wind-solar power output dataset E3(x gen,new ) are combined to form the peak shaving capacity evaluation case dataset.

[0090] The dataset of the system to be evaluated is the input data for peak shaving capacity evaluation, and the dataset types include load dataset E1(x load ), conventional power output dataset E2(x gen,nom ), and wind-solar power output dataset E3(x gen,new ).

[0091] S3: The dataset of the obtained peak shaving capacity evaluation case and the dataset of the system to be evaluated are semantized to obtain a complete semantized dataset

[0092] Further, in step S3, the semantization formula is as follows:

[0093]

[0094] In the above formula, represents the kth dataset in the peak shaving capacity evaluation case dataset at time t, k ∈ [1, 7], and the seven datasets are respectively load dataset E1(x load ), conventional power output dataset E2(x gen,nom ), wind-solar power output dataset E3(x gen,new ), peak-valley difference dataset E4(ΔP load,week ), peak-valley difference rate dataset E5(ρ load,week ), wind-solar anti-peak shaving rate dataset E6(r repeak ), and peak shaving capacity equivalent load curve data E7[P equ-load (d day,t )], the time scale and the total number of time points are determined according to the size of the dataset; is the semantized statement corresponding to the kth dataset at time t; Encode() is a semantization encoder that compiles simple data type names and data into a simple sentence in the form of a simple sentence.

[0095] Taking the load data as an example, it can be semantized as “the 2nd time illumination intensity is x solar,2 lux, the 2nd time wind speed is x wind,2 m / s, the 2nd time temperature is x tem,2 ℃, and the 2nd time load active demand is x load,2 kW”; the semantized data is input into a large language model; and the entire case dataset After semanticization, the complete semantic dataset for this case is obtained. Where T is the total number of time points.

[0096] S4: Complete semantic dataset The data is input into a large language model for similarity information search and case filtering to expand the number of cases, resulting in a richer semantically enhanced dataset.

[0097] This step leverages the advantages of large language models trained on massive amounts of data to improve the accuracy of training data for power system peak-shaving capacity assessment.

[0098] Furthermore, in step S4, the case screening includes the following steps:

[0099] S41: Semantic information must contain the existence of E1(x) load E2(x) gen,nom ) and E3(x gen,new The corresponding data, where E1(x) load ) must contain Data, E2(x) gen,nom ) must contain Data, E3(x) gen,new ) must contain At least one of the three data points;

[0100] S42: The total number of moments T contained in the semantic information is greater than or equal to 12.

[0101] S5: Enhancing the richness of semantic datasets Semantic decoding is performed, the decoded dataset is input into the domain model for domain-specific dataset localization, and then the localized dataset is semantically encoded to obtain the domain-enhanced semantic dataset.

[0102] Furthermore, in step S5, the semantic decoding process is the inverse process of semantic encoding, as shown in the following expression:

[0103]

[0104] In the above formula, Decode() is a semantic decoder. The units of the data generated by Encode() and recognized by Decode() are unified by the large language model, and the corresponding unit conversion is also performed by the large model. This enables the expression "the cost of conventional units at time 3 is x" to be translated into Chinese. cost,3 10,000 yuan, the normal active power output at the third moment is "kW" converted to At the same time, datasets of the same type but from different times are integrated.

[0105] Further, in step S5, the data set is fielded, that is, the deep learning prediction method is used to predict the peak shaving capacity equivalent load curve data E7[P equ-load (d day,t )] of the new case; the decoded data set of the original case semantics is input into the field model for training, and the prediction training principle formula of the field model is:

[0106]

[0107] In the above formula, and are the training input data of the lth and (l+1)th layers; Q(l) and Q(l+1) are the adaptive training weight coefficients of the lth and (l+1)th layers, respectively; b(l) and b(l+1) are the adaptive training bias coefficients of the lth and (l+1)th layers, respectively; is the input peak shaving capacity equivalent load curve training data, that is, the training output data; Softmax() is an activation function.

[0108] After the field model training is completed, the new case data set in the richness enhanced semantic data set after semantic decoding is input, and the peak shaving capacity equivalent load curve data E7[P equ-load (d day,t )] of each new case is predicted;

[0109] After the equivalent load curve data is predicted, the non-semantic field enhanced semantic data set is obtained by integration. Thereafter, the semantic formula in step S3 is used to perform semanticization to obtain the semantic field enhanced semantic data set

[0110] S6: Input the semantic field enhanced semantic data set into the large language model for model reinforcement training.

[0111] S7: Input the semantic evaluation system data into the large language model after reinforcement to evaluate the peak shaving demand result, and obtain the evaluation result of the review; the semantic peak-valley difference data set W4(ΔP load,week ), the semantic peak-valley difference rate data set W5(ρ load,week ), the semantic wind-light anti-peak shaving rate data set W6(r repeak ), and the semantic peak shaving capacity equivalent load curve data W7[P equ-load (d day,t )] are obtained as the evaluation result output by the large language model.

[0112] S8: using the semantic decoding formula in step S5, the semantic decoding of the evaluation result of the review is performed, and the equivalent load curve data of the peak shaving capacity of the system to be evaluated is obtained as the evaluation result of the peak shaving capacity.

[0113] The present scheme has the following advantages and effects relative to the prior art:

[0114] (1) For the case of insufficient accuracy of case prediction results containing multi-source information training data, the domain large model method can effectively capture the relationship between multiple factors, and can use the semantic encoding decoder to convert effective time series data into semantic information for training.

[0115] (2) For the case that the evaluation result of the power system peak shaving capacity by the single large model prediction method is random and has poor professional fit, the method of using domain model to assist large model training data is used to process the training data from the large model, improve the effectiveness and professional fit of the training data set; at the same time, the domain model can adapt to cases containing various data, solving the problem that some cases cannot directly solve the peak shaving capacity evaluation index.

[0116] (3) For the case of lacking actual training case data set, the method of searching and screening training case data by large model effectively extracts prior information and improves the richness of the model training data set.

[0117] Embodiment two

[0118] A power system peak shaving capacity evaluation system of a domain large model, applied to the power system peak shaving capacity evaluation method of the domain large model described above, comprising:

[0119] An index calculation module for obtaining a historical data set of a power system, calculating corresponding peak shaving capacity evaluation indexes according to the historical data set, and obtaining a corresponding index data set;

[0120] A construction module for supplementing the index data set to the historical data set to jointly constitute a peak shaving capacity evaluation case data set;

[0121] A semantic module for performing semanticization on the obtained peak shaving capacity evaluation case data set and the data set of the system to be evaluated to obtain a complete semantic data set

[0122] An expansion module for expanding the complete semantic data set to a large language model to search for similar information and select cases to expand the number of cases, and obtain a richness-enhanced semantic data set

[0123] A domain module for inputting the richness-enhanced semantic data set S1: obtaining a historical data set of a power system, calculating corresponding peak shaving capacity evaluation indexes according to the historical data set, and obtaining corresponding index data set; S2: supplementing the index data set to the historical data set to jointly constitute a peak shaving capacity evaluation case data set; S3: performing semantic coding on the obtained peak shaving capacity evaluation case data set and a data set of a system to be evaluated to obtain a complete semantic data set

[0124] a training module for inputting the semantic domain-enhanced semantic data set into a large language model for model reinforcement training;

[0125] an evaluation module for inputting the semantic system data to be evaluated into the large language model after reinforcement for peak regulation demand result evaluation, to obtain a semantic evaluation result;

[0126] a decoding module for performing semantic decoding on the semantic evaluation result to obtain a peak shaving capacity equivalent load curve data of the system to be evaluated as a peak shaving capacity evaluation result.

[0127] Embodiment three

[0128] A computer-readable storage medium comprising a stored program, wherein the computer-readable storage medium controls the device where the computer-readable storage medium is located to execute the power system peak shaving capacity evaluation method of the domain large model described above when the program is running.

[0129] Embodiment four

[0130] A processor for running a program, wherein the power system peak shaving capacity evaluation method of the domain large model described above is executed when the program is running.

[0131] The application provides a power system peak shaving capacity evaluation method of a domain large model, comprising: S1: obtaining a historical data set of a power system, calculating corresponding peak shaving capacity evaluation indexes according to the historical data set, and obtaining corresponding index data set; S2: supplementing the index data set to the historical data set to jointly constitute a peak shaving capacity evaluation case data set; S3: performing semantic coding on the obtained peak shaving capacity evaluation case data set and a data set of a system to be evaluated to obtain a complete semantic data set S4: inputting the complete semantic data set into a large language model for similar information searching and case screening to expand the number of cases, to obtain a richness-enhanced semantic data set S5: inputting the richness-enhanced semantic data set into a large language model for semantic decoding, inputting the decoded data set into a domain model for data set domainization, and then performing semantic coding on the domainized data set to obtain a domain-enhanced semantic data set S6: inputting the semantic domain-enhanced semantic data set Input into a large language model for model reinforcement training; S7: input the semanticized to-be-evaluated system data into the large language model after reinforcement for peak regulation demand result evaluation, and obtain the evaluation result of the evaluation; S8: semantic decoding is performed on the evaluation result of the evaluation, and the equivalent load curve data of the peak regulation capacity of the to-be-evaluated system is obtained as the result of the peak regulation capacity evaluation. The problem of insufficient accuracy of case prediction results of multi-source information training data is solved; the method of applying a large language model combined with a domain model is used to solve the problem of large randomness of a single large model prediction evaluation result and poor professional matching degree; the method of using a large model to search and screen cases is used to solve the problem of inaccurate prediction and evaluation in the case of lack of actual training cases.

[0132] Those skilled in the art can appreciate that the units of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both, and in order to clearly illustrate the interchangeability of hardware and software, the components of the examples have been described in the above description in general terms. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered as beyond the scope of the present application.

[0133] In the embodiments provided by the present application, it should be understood that the division of units is only a logical functional division, and actual implementation can have another division manner, for example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc.

[0134] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0135] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0136] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the specification of the present application.

Claims

1. A method for evaluating the peak-shaving capacity of a power system using a large-scale domain model, characterized in that, include: S1: Obtain the historical dataset of the power system, calculate the corresponding peak-shaving capacity assessment index based on the historical dataset, and obtain the corresponding index dataset; S2: Supplement the indicator dataset to the historical dataset to form a case dataset for peak-shaving capacity assessment; S3: Semantize the obtained datasets of peak-shaving capacity assessment cases and the datasets of the systems to be evaluated to obtain a complete semantic dataset. S4: Complete semantic dataset The data is input into a large language model for similarity information search and case filtering to expand the number of cases, resulting in a richer semantically enhanced dataset. S5: Enhancing the richness of semantic datasets Semantic decoding is performed, the decoded dataset is input into the domain model for domain-specific dataset localization, and then the localized dataset is semantically encoded to obtain the domain-enhanced semantic dataset. S6: Enhance semantic domains with semantic datasets The input is fed into a large language model for model reinforcement training; S7: Input semantically encoded data of the system to be evaluated into the enhanced large language model to evaluate the peak-shaving demand results and obtain the evaluation results; S8: Semantically decode the evaluation results to obtain the peak-shaving capacity equivalent load curve data of the system to be evaluated as the result of the peak-shaving capacity evaluation.

2. The power system peak-shaving capacity assessment method based on a large-scale domain model according to claim 1, characterized in that, In step S1, obtaining the historical data set of the power system, calculating the corresponding peak-shaving capacity assessment indicators based on the historical data set, and obtaining the corresponding indicator dataset includes the following steps: S11: Obtain the load dataset, conventional power output dataset, and wind and solar power output dataset of the power system. The formulas for each dataset are as follows: In the above formula, E1(x load ) represents the load dataset, x load Variables representing all types of electrical load; n is the total number of time scales for data partitioning; x tem It is the tth i Temperature data at any given time; It is the tth i Real-time light intensity data; It is the tth i Real-time wind speed data; It is the tth i Real-time load active power demand data; E2(x gen,nom ) is the set of regular output data; x gen,nom It is a variable representing regular output data; It is the tth i Cost data for conventional generating units at any given time; Indicates the t-th i The normal active power output value at any given time; E3(x gen,new ) is a set of wind and solar power output data; x gen,new These are variables representing wind and solar power output data; For the tth i The glory of each moment is reflected in the data on contributions and efforts; For the tth i Real-time photovoltaic active power output data; For the tth i Real-time wind power active power output data; S12: Calculate the time-series characteristics of the load to obtain the daily maximum peak-to-valley difference ΔP calculated on a weekly basis. load,week The peak-valley difference rate ρ of the maximum peak-valley difference per day of the week load,week This forms the peak-valley difference dataset E4(ΔP) load,week ) and peak-valley difference dataset E5(ρ load,week The calculation formula is as follows: In the above formula, ΔP load,week P represents the daily maximum peak-to-valley difference, calculated on a weekly basis. peak,week P valley,week The peak and trough values ​​represent the days with the largest peak-to-trough difference in a week; ρ load,week Peak-to-valley difference rate representing the day with the largest peak-to-valley difference in a week; The normal active power output values ​​from the 1st to the 24th hour of the day with the largest peak-to-valley difference in a week are obtained by selecting the day with the largest peak-to-valley difference. S13: Calculate the anti-peak shaving characteristic index of wind and solar power output, obtain the wind and solar anti-peak shaving rate, and form the wind and solar anti-peak shaving rate dataset E6(r). repeak The calculation formula is as follows: In the above formula, r repeak For wind and solar inverse peak rate; m count For the number of times the wind and solar power peaks are reversed; d total The total number of days counted; S14: Calculate the equivalent load curve of peak-shaving capacity, and form the peak-shaving capacity equivalent load curve dataset E7[P] equ-load (d day,t The equivalent load curve of the peak-shaving capacity reflects the peak-shaving capacity of the power system; the calculation formula is as follows: P equ-load (d day,t )=P load (d day,t )-P newenergy (d day,t ),d day ∈[1,7],t∈[0,23]; In the above formula, P equ-load (d day,t P represents the equivalent load power of wind and solar power output on a certain day at time t. load (d day,t P represents the load power at time t on a certain day; newenergy (d day,t ) represents the wind and solar power output at time t on a certain day.

3. The power system peak-shaving capacity assessment method for a large-scale domain model according to claim 2, characterized in that, In S2, the phrase "supplementing the indicator dataset to the historical dataset to jointly constitute the peak-shaving capacity assessment case dataset" refers to adding the peak-valley difference dataset E4(ΔP) to the historical dataset. load,week ), Peak-to-valley difference dataset E5(ρ load,week ), Wind and solar inverse peak rate dataset E6(r repeak ) and peak-shaving capacity equivalent load curve dataset E7[P equ-load (d day,t )] and load dataset E1(x load ), conventional output dataset E2(x gen ,nom) and the wind and solar power output dataset E3(x gen,new These combinations together constitute the peak-shaving capacity assessment case dataset.

4. The power system peak-shaving capacity assessment method for a large-scale domain model according to claim 2, characterized in that, In step S3, the semantic formula is as follows: In the above formula, This represents the k-th dataset in the peak-shaving capacity assessment case dataset at time t, where k∈[1,7]. The seven datasets are the load datasets E1(x) and E2(x)(x). load ), conventional output data set E2(x gen,nom ), Wind and solar power output data set E3(x gen,new ), Peak-to-valley difference dataset E4(ΔP) load,week ), Peak-to-valley difference dataset E5(ρ load,week ), Wind and solar inverse peak rate dataset E6(r repeak ) and peak-shaving capacity equivalent load curve data E7[P equ-load (d day,t The time scale and the total number of moments are determined based on the size of the dataset; This is the semantic statement corresponding to the k-th dataset at time t; Encode() is a semantic encoder that compiles data by writing simple data type names and data correspondences into a simple sentence.

5. The power system peak-shaving capacity assessment method for a large-scale domain model according to claim 2, characterized in that, In step S4, the case screening includes the following steps: S41: Semantic information must contain the existence of E1(x) load E2(x) gen,nom ) and E3(x gen,new The corresponding data, where E1(x) load ) must contain Data, E2(x) gen,nom ) must contain Data, E3(x) gen,new ) must contain At least one of the three data points; S42: The total number of moments T contained in the semantic information is greater than or equal to 12.

6. The power system peak-shaving capacity assessment method based on a large-scale domain model according to claim 4, characterized in that, In step S5, the semantic decoding process is the inverse process of semantic encoding, and the expression is as follows: In the above formula, Decode() is a semantic decoder. The units of data generated by Encode() and recognized by Decode() are unified by the large language model, and the corresponding unit conversion is also performed by the large model.

7. The power system peak-shaving capacity assessment method for a large-scale domain model according to claim 4, characterized in that, In step S5, the prediction training principle of the domain model is as follows: In the above formula, and The input data for training the l-th and l+1-th layers are given; Q(l) and Q(l+1) are the adaptive training weight coefficients for the l-th and l+1-th layers, respectively. b(l) and b(l+1) are the adaptive training bias coefficients of the l-th and l+1-th layers, respectively; The input is the peak-shaving capacity equivalent load curve training data, i.e., the training output data; Softmax() is the activation function.

8. A power system peak-shaving capacity assessment system based on a large-scale domain model, characterized in that, A method for assessing the peak-shaving capacity of a power system applied to a large-scale model according to any one of claims 1 to 7, comprising: The indicator calculation module is used to obtain historical data sets of the power system, calculate the corresponding peak-shaving capacity assessment indicators based on the historical data sets, and obtain the corresponding indicator datasets. The building module is used to supplement the indicator dataset to the historical dataset, together forming the peak-shaving capacity assessment case dataset; The semantic module is used to semantically represent the datasets of the obtained peak-shaving capacity assessment cases and the datasets of the systems to be evaluated, resulting in a complete semantically represented dataset. The extension module is used to add the complete semantic dataset. The data is input into a large language model for similarity information search and case filtering to expand the number of cases, resulting in a richer semantically enhanced dataset. Domain module, which is used to enrich semantically enhanced datasets. Semantic decoding is performed, the decoded dataset is input into the domain model for domain-specific dataset localization, and then the localized dataset is semantically encoded to obtain the domain-enhanced semantic dataset. The training module is used to augment semantic domain datasets. The input is fed into a large language model for model reinforcement training; The evaluation module is used to input semantically encoded data of the system to be evaluated into the enhanced large language model to evaluate the peak demand results and obtain the evaluation results. The decoding module is used to semantically decode the evaluation results to obtain the peak-shaving capacity equivalent load curve data of the system to be evaluated as the result of the peak-shaving capacity evaluation.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device containing the computer-readable storage medium to perform the power system peak-shaving capacity assessment method of any one of claims 1 to 7 for a large-scale model.

10. A processor, characterized in that, The processor is used to run a program, wherein the program executes the power system peak-shaving capacity assessment method of any one of claims 1 to 7 when it runs.

Images