CN116050653A - Closed-loop prediction-decision scheduling method for power system based on data driving - Google Patents
Closed-loop prediction-decision scheduling method for power system based on data driving Download PDFInfo
- Publication number
- CN116050653A CN116050653A CN202310160847.5A CN202310160847A CN116050653A CN 116050653 A CN116050653 A CN 116050653A CN 202310160847 A CN202310160847 A CN 202310160847A CN 116050653 A CN116050653 A CN 116050653A
- Authority
- CN
- China
- Prior art keywords
- renewable energy
- prediction
- decision
- unit combination
- economy
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 39
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 16
- 238000012545 processing Methods 0.000 claims abstract description 9
- 238000011156 evaluation Methods 0.000 claims abstract description 7
- 238000000605 extraction Methods 0.000 claims abstract description 7
- 239000013598 vector Substances 0.000 claims description 33
- 230000001186 cumulative effect Effects 0.000 claims description 3
- 238000005315 distribution function Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000005457 optimization Methods 0.000 description 12
- 230000008569 process Effects 0.000 description 5
- 238000010248 power generation Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000011144 upstream manufacturing Methods 0.000 description 3
- 238000012512 characterization method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005096 rolling process Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 230000036316 preload Effects 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 230000003442 weekly effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06312—Adjustment or analysis of established resource schedule, e.g. resource or task levelling, or dynamic rescheduling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Engineering & Computer Science (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Health & Medical Sciences (AREA)
- Educational Administration (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a closed-loop prediction-decision scheduling method of a power system based on data driving, which comprises the following steps: in the data processing stage, carrying out feature extraction by adopting standard regression coefficients to determine the most relevant feature class, and selecting a training sample for the next stage based on the Wasserstein distance; in the training stage, an ERM model taking the unit combination economy as a prediction evaluation index is constructed based on the determined characteristic category and a training sample, and a Lagrange decomposition algorithm is adopted for solving, so that a renewable energy predictor taking the unit combination economy as a guide is obtained; in the prediction-decision stage, a renewable energy source predictor is embedded into a traditional unit combination model to obtain a prediction-decision model capable of simultaneously carrying out renewable energy source prediction and unit combination decision. The method and the system can consider the influence of the predicted value on the economy of the unit combination in the prediction stage, and provide the method and the system which take the economy of the unit combination as the guiding predicted value so as to improve the economy of the unit combination.
Description
Technical Field
The invention relates to the field of power system optimization scheduling, in particular to a closed-loop prediction-decision scheduling method of a power system based on data driving.
Background
Network constrained combined power unit (NCUC) problems have been considered as one of the most important applications in power system operation and power market clearing. Specifically, NCUC is typically implemented by an Independent System Operator (ISO) to determine optimal crew combinations, power generation schedules, and backup plans with minimal system operating costs.
Typically, ISO performs NCUC in open loop prediction, and then optimizes the (O-PO) framework, as shown in (a) of fig. 1. In O-PO, the upstream prediction step generates precision-oriented predictions (e.g., about renewable energy sources and loads) based on traditional statistical metrics (e.g., average absolute error); then NCUC is implemented in a downstream optimization step, taking the prediction as input. However, statistically more accurate predictions do not necessarily enable more economical NCUC planning. In practice, NCUC economy refers to the actual system costs, including the start-up and shut-down costs of the generators involved in the pre-day NCUC planning and rescheduling problem, as well as the actual power generation costs of the generators corresponding to the final scheduling level of the rescheduling problem, which can be calculated after solving the pre-day NCUC (based on renewable energy predictions) and rescheduling (based on given NCUC decisions and renewable energy) problems.
To improve the NCUC economy of O-PO, one emerging technique is to generate a cost-oriented prediction by feeding back certain information of the downstream optimization (e.g., the cost incurred by the prediction) to the upstream prediction and measuring the prediction quality by the cost incurred, not just the statistical prediction accuracy. As shown in fig. 1 (b), this idea is called closed-loop prediction and optimization (C-PO). The C-PO technique performs cost-oriented prediction and optimization in a single step by a simple and efficient data-driven method, and enhances economy. However, considering that data driven approaches involve complex constraints that may lead to infeasibility, the improved C-PO predicts the target coefficients of Linear Programming (LP) and Mixed Integer Linear Programming (MILP) problems through an intelligent prediction then optimization (SPO) framework. The core of the SPO framework is the SPO penalty function, which measures the difference between the optimal target corresponding to the prediction and the actual implementation.
Recently, C-PO has also been applied to power system optimization scheduling problems, however, extending C-PO to MILP-based NCUC may be challenging, especially for large-scale power systems. Thus, the present invention proposes a feature driven C-PO framework with three modules to improve the economics of NCUC. First, the data processing module performs feature and scene selection to identify the correct feature type and training scene. Given the features and scenarios, the cost-oriented modeling and training module then forms an Empirical Risk Minimization (ERM) problem based on SPO losses, which is solved by a decomposition based on Lagrangian Relaxation (LR) to obtain a cost-oriented renewable energy prediction model. Finally, the closed-loop prediction and optimization module integrates the trained renewable energy prediction model and the NCUC formula into a feature-driven NCUC prescription model that jointly performs cost-oriented renewable energy power prediction and NCUC optimization.
Disclosure of Invention
The invention aims to provide a closed-loop prediction-decision scheduling method of a power system based on data driving, which can consider the influence of a predicted value on the economy of unit combination in a prediction stage, and provide a method for guiding the economy of unit combination to the predicted value so as to improve the economy of unit combination.
In order to achieve the above object, the present invention is realized by the following technical scheme:
a closed-loop prediction-decision scheduling method of a power system based on data driving comprises the following steps:
s1: in the data processing stage, carrying out feature extraction by adopting standard regression coefficients to determine the most relevant feature class, and selecting a training sample for the next stage based on the Wasserstein distance;
s2: in the training stage, an ERM model taking the unit combination economy as a prediction evaluation index is constructed based on the determined characteristic category and a training sample, and a Lagrange decomposition algorithm is adopted for solving, so that a renewable energy predictor taking the unit combination economy as a guide is obtained;
s3: in the prediction-decision stage, a renewable energy source predictor is embedded into a traditional unit combination model to obtain a prediction-decision model capable of simultaneously carrying out renewable energy source prediction and unit combination decision.
Optionally, step S1 performs feature extraction by using standard regression coefficients to determine the most relevant feature class, which specifically includes:
quantifying the importance of each feature type on a renewable energy power plant j by the following steps 1, 2:
step 1: for the t hour of the renewable energy power plant j, determining available renewable energy power vectors in its historyAnd a feature vector f t,1 ,...,f t,|N| And regression coefficient beta t,j,0 ,...,β t,j,|N| ;
Step 2: the average Standard Regression Coefficient (SRC) for each feature type n over all |τ| hours is calculated by
wherein ,σtj and σtn Respectively is and />J represents a renewable energy power plant set, N represents a characteristic type set, and T represents a scheduling period set;
will have the highest valueThe feature type n of (c) is selected as the relevant feature of the renewable energy power plant j.
Alternatively, regression coefficient beta t,j,0 ,...,β t,j,|N| Calculated by least squares method:
optionally, step S1 selects a training sample for the next stage based on the wasperstein distance, specifically including:
predicting power using renewable energy sourcesAnd the actual renewable energy source power +>Wasserstein distance W between h' Quantitatively evaluating the extreme value of the latest historical scene h';
w of T representative scenes h' The scene corresponding to the median value of (2) is determined as the training scene of the future scheduling day.
Alternatively, wasserstein distance W h' The calculation formula of (2) is as follows:
wherein , and />Respectively-> and />Is a cumulative distribution function of experience of->Representing the predicted output of renewable energy sources in the latest historical scene h +.>And the actual output of the renewable energy source under the latest historical scene h' is represented.
Optionally, step S2 builds an ERM model with unit combination economy as a predictive evaluation index based on the given feature class and the training sample, and specifically includes:
predicted output of a given renewable energy source in a scene s isAnd the actual output of renewable energy source in scene s is +.>SPO loss->The definition is as follows: /> wherein /> and />Are respectively-> and />The determined optimal target value;
by means ofAnd the NCUC model is used for forming the following ERM model and training a cost-oriented prediction model H;
in the above formula, subscript s represents a variable corresponding to scene s, and x represents an M-dimensional vector of a binary variable in the NCUC problem; y represents a vector of continuous variables; c and d represent cost vectors; A. b and F represent constant matrices; g is the right-hand vector of all constraints except renewable energy power limits; fy is less than or equal to H * f represents a renewable energy power limit constraint; lambda H 1 Is a 1-norm adjustment term used to prevent an H overfitting, where λ is the overshoot parameter.
Optionally, step S2 is solved by using a lagrangian decomposition algorithm to obtain a renewable energy predictor guided by unit combination economy, and specifically includes:
decomposing an ERM model original problem into |S| disjoint sub-problems by adopting a Lagrangian decomposition algorithm, wherein each sub-problem corresponds to a training scene, and solving the sub-problems in parallel;
and calculating a global optimal solution of the ERM model by adopting an iterative algorithm based on a secondary gradient to obtain the renewable energy predictor taking the unit combination economy as a guide.
Optionally, the prediction-decision model obtained in step S3, which can simultaneously perform renewable energy prediction and unit combination decision, is specifically as follows:
wherein x represents an M-dimensional vector of binary variables in the NCUC problem; y represents a vector of continuous variables; c and d represent cost vectors; A. b and F represent constant matrices; g is the right-hand vector of all constraints except renewable energy power limits; fy is less than or equal to H * f represents renewable energy power limit.
Compared with the prior art, the invention has the following advantages:
1) The invention provides a novel feature driven C-PO framework for improving NCUC economy. By utilizing the characterization data, NCUC structure (constraints and targets), and induced NCUC costs, C-PO can provide cost-oriented renewable energy power predictions to improve NCUC economics.
2) The customized Lagrangian decomposition-based method can be applied to effectively solve the C-PO model based on MILP.
Drawings
For a clearer description of the technical solutions of the present invention, the drawings that are needed in the description will be briefly introduced below, it being obvious that the drawings in the following description are one embodiment of the present invention, and that, without inventive effort, other drawings can be obtained by those skilled in the art from these drawings:
FIG. 1 is an illustration of a different prediction-post-optimization framework provided by the present invention;
FIG. 2 is a flow chart for implementing NCUC feature driven closed loop prediction and optimization in accordance with the present invention;
FIG. 3 is a C-PO implementation of rolling-based daily NCUC provided by the present invention.
Detailed Description
The following provides a further detailed description of the proposed solution of the invention with reference to the accompanying drawings and detailed description. The advantages and features of the present invention will become more apparent from the following description. It should be noted that the drawings are in a very simplified form and are all to a non-precise scale, merely for the purpose of facilitating and clearly aiding in the description of embodiments of the invention. For a better understanding of the invention with objects, features and advantages, refer to the drawings. It should be understood that the structures, proportions, sizes, etc. shown in the drawings are for illustration purposes only and should not be construed as limiting the invention to the extent that any modifications, changes in the proportions, or adjustments of the sizes of structures, proportions, or otherwise, used in the practice of the invention, are included in the spirit and scope of the invention which is otherwise, without departing from the spirit or essential characteristics thereof.
In the conventional power system scheduling, an open-loop prediction-decision is adopted, an upstream predictor firstly provides a renewable energy source predicted value for the power system, and a downstream Unit module (UC) is further optimized based on the predicted value. However, the open-loop prediction-decision scheduling predicts with statistical accuracy as a guide, and does not consider a specific optimization scenario downstream, so that the economy of the unit combination cannot be guaranteed. Aiming at the problem, the invention provides a closed-loop prediction-decision scheduling method of an electric power system based on data driving, which can consider the influence of a predicted value on the economy of unit combination in a prediction stage, and takes the economy of unit combination as a guiding predicted value so as to improve the economy of unit combination. The method comprises a data processing stage, a training stage and a prediction-decision stage in sequence. In the data processing stage, feature extraction is performed using standard regression coefficients, and training samples are selected for the next stage based on Wasserstein distance. In the training stage, an empirical risk minimization (Empirical risk minimization, ERM) model taking the unit combination economy as a prediction evaluation index is constructed based on given feature types and training samples, and a Lagrange decomposition algorithm is adopted for solving, so that the renewable energy predictor taking the unit combination economy as a guide is finally obtained. In the prediction-decision stage, a predictor is embedded into a traditional unit combination model to obtain a prediction-decision model capable of simultaneously carrying out renewable energy prediction and unit combination decision.
As shown in fig. 2, the method for closed-loop prediction-decision scheduling of the power system based on data driving provided by the invention specifically comprises the following steps:
step S1: in the data processing stage, feature extraction is performed using standard regression coefficients, and training samples are selected for the next stage based on Wasserstein distance.
The predictive model will be trained from historical data. Specifically, the data of a single historical scenario h includes NCUC costs that perfectly correspond to the actual renewable energy power(i.e. use->Solving a conventional crew combining model) and a feature vector f containing |n| feature types. Potential feature types may include daily preload predictions and renewable energy predictions that may be provided to the ISO during the daily phase. The data processing module performs feature selection and training scenario selection to identify relevant feature types and representative historical scenarios for each renewable energy power plant j.
1) Feature selection: in general, various feature types (e.g., area loading and raw predictions of renewable energy sources) may be employed in practice. If too few feature types are used to train the predictive model, the model may perform poorly in both training and actual prediction processes (i.e., under-fitting); on the other hand, if training is performed using too many feature types, the model may contain many parameters and become too complex to be easily implemented in the actual prediction process (i.e., over-fitting). Thus, feature selection is utilized to determine an appropriate set of relevant feature types for the predictive model, such that an accurate and easy-to-implement predictive model may be obtained while avoiding potential under-fit and over-fit problems.
The importance of each feature type on the renewable energy power plant j is quantified in particular by the following steps:
step 1: for the t hour of the renewable energy power plant j, a renewable energy power vector is available in its historyAnd a feature vector f t,1 ,...,f t,|N| And regression coefficient beta t,j,0 ,...,β t,j,|N| . Regression coefficients can be calculated by least squares:
Step 2: the average Standard Regression Coefficient (SRC) for each feature type n over all |τ| hours is calculated by wherein σtj and σtn Respectively-> and />Standard deviation of (2).
Wherein J represents a renewable energy power plant set, N represents a characteristic type set, and T represents a scheduling period set;
the larger the value of (c) is, the higher the importance of the feature type n on the renewable energy power plant j is. Finally, have the highest value->Is selected as the relevant feature of the renewable energy power plant j.
Among various feature selection methods, SRC is widely considered as a simple and effective method of identifying the relationship between feature types. In the present invention, the correlation of individual feature types based on a large number of historical scenarios to renewable energy output power is quantified using its interpretability, ensuring that the identified feature types are reasonable and interpretable.
2) Selection of training scenes: after selecting the relevant features, NT representative historical scenes in the latest NH history day should be identified to effectively train a predictive model for future scheduling days.
Using the Wasserstein distance W between renewable energy source predicted power and actual renewable energy source power h' To quantitatively evaluate the extremum of the latest historical scene h'. and />Respectively-> and />Is a cumulative distribution function of experience of->Representing the predicted output of renewable energy sources in the latest historical scene h +.>And the actual output of the renewable energy source under the latest historical scene h' is represented. Then W of T representative scenes h' The scene corresponding to the median value of (2) is determined as the training scene of the future scheduling day.
Step S2: in the training stage, an ERM model taking the unit combination economy as a prediction evaluation index is constructed based on given feature types and training samples, and a Lagrange decomposition algorithm is adopted for solving, so that the renewable energy predictor taking the unit combination economy as a guide is finally obtained.
The core idea of the closed-loop framework is to train a cost-oriented renewable energy power prediction model that is suitable for NCUC. The cost-oriented ERM problem based on SPO losses is constructed for training a cost-oriented renewable energy prediction model from the appropriate feature types and representative scenarios identified in the data processing module.
1) Cost-oriented ERM problem modeling: SPO losses are first defined to quantify the system operating cost losses due to imperfect renewable energy prediction information.
Predicted output of a given renewable energy source in a scene s isAnd the actual output of renewable energy source in scene s is +.>SPO loss->Defined as the followingWherein-> and />Are respectively-> and />And determining an optimal target value.
By usingAnd an NCUC model, an ERM model can be formed that is used to train a cost-oriented predictive model H.
In the above formula, subscript S represents a variable corresponding to scene S, S represents a set formed by scene S, and x represents an M-dimensional vector of a binary variable in the NCUC problem; y represents a vector of continuous variables; c and d represent cost vectors; A. b and F represent constant matrices; g is the right-hand vector of all constraints except renewable energy power limits; fy is less than or equal to H * f represents a renewable energy power limit constraint.
λ||H|| 1 Is a 1-norm adjustment term used to prevent an H overfitting, where λ is the overshoot parameter. The adjustment term λ H 1 Can be equivalently converted into a linear formula, and the above formula is used as the MILP problem of the unit combination. Solving this is essentially a training process that derives a trained cost-oriented predictive model H * 。
Slave machineFrom the point of view of learning, ERM can be regarded as a linear regression problem from the feature data f s Cost-oriented renewable energy power predictive search optimal linear mapping H to NCUC * 。
2) Cost-oriented ERM problem solving: LR-based decomposition is applied to effectively solve ERM, training a cost-oriented renewable energy prediction model H for NCUC * 。
(1) LR decomposition: LR decomposition can double some coupling constraints and decompose the original problem into |s| disjoint sub-problems, which can be solved in parallel. Subsequently, the original ERM is decomposed into |s| sub-questions, as shown in the following formula, where μ s,e Is H s The lagrangian multiplier for element e, S' refers to the index of scene set S.
Each sub-question corresponds to a training scenario. For this reason, the present patent can process |s| scene problems in parallel as in the above equation, instead of directly solving ERM.
(2) Iterative algorithm based on secondary gradient: to obtain a globally optimal solution for the original ERM, u should be determined s,e Is a suitable value for (a). Therefore, the sub-gradient algorithm is employed to iteratively update u by s,e Until the optimal gap is small enough. Alpha c Is a scalar step where ω is a constant scalar selected between 0 and 2.
The detailed procedure for the secondary gradient based method is described in table 1 below.
TABLE 1
Step S3: in the prediction-decision stage, a renewable energy source predictor is embedded into a traditional unit combination model to obtain a prediction-decision model capable of simultaneously carrying out renewable energy source prediction and unit combination decision.
By means of a trained cost-oriented renewable energy power prediction model H * Embedded into the original NCUC model to formulate a feature driven NCUC prescription model.
Where x represents the M-dimensional vector of binary variables in the NCUC problem; y represents a vector of continuous variables; c and d represent cost vectors; A. b and F represent constant matrices; g is the right-hand vector of all constraints except renewable energy power limits; fy is less than or equal to H * f represents a renewable energy power limit constraint.
Figure 3 shows a rolling based C-PO embodiment of the daily NCUC problem. Specifically, to balance the computational burden and quality of the NCUC prescription model, retraining was performed every 7 days, updated weekly.
With first H * For example, the H * Corresponding to 7 future scheduling days d 1,1 ,...,d 1,7 And NH recent history day h 1,1 ,...,h 1,NH . At d 1,1 Before at h 1,1 ,...,h 1,NH NT training scenarios were selected over the day to create ERM questions. In solving the ERM problem to obtain a first training H * Thereafter, a first NCUC prescription model may be constructed to solve for the d < th ] 1,1 ,...,d 1,7 Peripheral NCUC task. Over time, the above process is repeated to train a second H * And form the secondNCUC prescription model, which corresponds to day 7 NCUC task d daily 2,1 ,...,d 2,7 . NT is uniformly set to 2 to obtain the proper size of ERM.
The present invention employs an IEEE-24 node system to evaluate the effectiveness of NCUC economic improvement of C-PO and LR-based decomposition in solving the ERM problem of C-PO. The actual data of belgium ISO is scaled appropriately to build renewable energy, load, and characteristic data for case analysis. Belgium ISO operates a 2-zone grid, including 11 provinces (load bus), 11 solar power plants (SPFs), 4 land wind farms (WPFs), and 1 marine WPF. Based on the predictions for each SPF and WPF, a prediction synthesis for each regional SPF and WP is also calculated. Thus, the present invention has 22 types of features, including hour tags, belgium load predictions, and raw predictions (i.e., precision-oriented predictions) for 16 individual power plants and 4 regional power plant collections, which are readily available during the day-ahead stage. ISO typically performs NCUC by solving a traditional crew combining model using raw renewable energy predictions (i.e., O-PO).
In the dataset, 2018 and 2019 data were used for feature selection and 2020 data were used to evaluate the C-PO framework. The feature selection step identifies the original renewable energy prediction as the most relevant feature type f. Thereby, the prediction model H * Described is adjusting the original forecast f of a single renewable energy power plant to derive a cost-oriented renewable energy power forecast H * f has no unit weight.
In 2020, O-PO and C-PO were compared by 366 daily NCUC runs. Load C ls Penalty cost of 10 is set to 5 $/MWh。
Cost comparison of C-PO and O-PO: table 2 below compares the average annual cost of C-PO to O-PO. Both the NCUC SUSD cost and the system expected cost of C-PO are higher than O-PO (22052 $ against 21919 $ against 268090 $ against 285530), which means that C-PO is more prone to adopting a more conservative NCUC strategy, i.e., putting into more conventional gensets. That is, C-PO uses more regular gensets to support the load than O-PO, and some operating state gensets may not be fully or economically used, resulting in higher power generation costs (273530 dollars versus 269760 dollars). However, C-PO is superior to O-PO in rescheduling SUSD cost (US $ 13828 versus US $ 23911) and actual system cost (US $ 309410 versus US $ 315590). This is because the conservative NCUC strategy of the C-PO has enough flexibility to be used for scheduling adjustments, avoiding the non-rotating back-up of large numbers of generators that are not combined in the NCUC, thus improving the economics of the NCUC.
Prediction error comparison of C-PO and O-PO: table 3 below compares the prediction errors of C-PO and O-PO. In addition to the Mean Absolute Error (MAE) and the RMSE, the mean over-prediction error (MOPE) and the mean under-prediction error (MUPE) are also given, with the following formulas:
analysis of two selective schedule days: the results of two selective schedule days for C-PO and O-PO are given in Table 4 below, 22 in 6 and 29 in 2020. Specifically, 22 days 6 in 2020, C-PO achieved better NCUC economy (from the MAE perspective) with better prediction accuracy; and at 29 months 12 and 2020, C-PO achieved better NCUC economy with poorer prediction accuracy. Table 4 below further shows that C-PO with custom predictions yields NCUC strategies with lower SUSD costs ($21000 versus $22500) compared to O-PO. In addition, the NCUC strategy in C-PO may reduce SUSD costs ($10500 against $28500) and power generation costs ($ 269270 against $ 271080) in scheduling issues. This is mainly because the C-PO mobilizes less standby than the O-PO in the rescheduling problem. Thus, the C-PO has a customized cost-oriented prediction of renewable energy, enabling higher prediction accuracy and lower actual system cost (300770 dollars versus 322080 dollars).
TABLE 2
TABLE 3 Table 3
TABLE 4 Table 4
Based on the above analysis of the 24-node system, the following conclusions can be drawn:
(1) Statistically more accurate predictions of renewable energy power in O-PO do not necessarily lead to lower real system costs;
(2) In terms of NCUC economy, the proposed C-PO is superior to the benchmark O-PO;
(3) The excellent performance stems from the idea of custom cost-oriented renewable energy power prediction for NCUC, which is achieved through a closed-loop structure in C-PO.
Compared with the traditional open-loop power system scheduling method, the method has the beneficial effects that:
1) The invention provides a novel feature driven C-PO framework for improving NCUC economy. By utilizing the characterization data, NCUC structure (constraints and targets), and induced NCUC costs, C-PO can provide cost-oriented renewable energy power predictions to improve NCUC economics.
2) The customized Lagrangian decomposition-based method can be applied to effectively solve the C-PO model based on MILP.
While the present invention has been described in detail through the foregoing description of the preferred embodiment, it should be understood that the foregoing description is not to be considered as limiting the invention. Many modifications and substitutions of the present invention will become apparent to those of ordinary skill in the art upon reading the foregoing. Accordingly, the scope of the invention should be limited only by the attached claims.
Claims (8)
1. The closed-loop prediction-decision scheduling method of the power system based on data driving is characterized by comprising the following steps of:
s1: in the data processing stage, carrying out feature extraction by adopting standard regression coefficients to determine the most relevant feature class, and selecting a training sample for the next stage based on the Wasserstein distance;
s2: in the training stage, an ERM model taking the unit combination economy as a prediction evaluation index is constructed based on the determined characteristic category and a training sample, and a Lagrange decomposition algorithm is adopted for solving, so that a renewable energy predictor taking the unit combination economy as a guide is obtained;
s3: in the prediction-decision stage, a renewable energy source predictor is embedded into a traditional unit combination model to obtain a prediction-decision model capable of simultaneously carrying out renewable energy source prediction and unit combination decision.
2. The method for closed-loop prediction-decision scheduling of a data-driven power system according to claim 1, wherein step S1 uses standard regression coefficients for feature extraction to determine the most relevant feature class, and specifically comprises:
quantifying the importance of each feature type on a renewable energy power plant j by the following steps 1, 2:
step 1: for the t hour of the renewable energy power plant j, determining available renewable energy power vectors in its historyAnd a feature vector f t,1 ,...,f t,|N| And regression coefficient beta t,j,0 ,...,β t,j,|N| ;
Step 2: the average Standard Regression Coefficient (SRC) for each feature type n over all |τ| hours is calculated by
wherein ,σtj and σtn Respectively is and />J represents a renewable energy power plant set, N represents a characteristic type set, and T represents a scheduling period set;
4. the method for closed-loop prediction-decision scheduling of a power system based on data driving according to claim 1, wherein step S1 selects training samples for the next stage based on the waserstein distance, specifically comprising:
predicting power using renewable energy sourcesAnd the actual renewable energy source power +>Wasserstein distance W between h' Quantitatively evaluating the extreme value of the latest historical scene h';
w of T representative scenes h' The scene corresponding to the median value of (2) is determined as the training scene of the future scheduling day.
5. The data-driven power system closed-loop prediction-decision scheduling method as claimed in claim 4, wherein Wasserstein distance W h' The calculation formula of (2) is as follows:
6. The method for closed-loop prediction-decision scheduling of a power system based on data driving as claimed in claim 1, wherein step S2 builds an ERM model using unit combination economy as a prediction evaluation index based on a given feature class and training samples, and specifically comprises:
predicted output of a given renewable energy source in a scene s isAnd the actual output of renewable energy source in scene s is +.>SPO loss->The definition is as follows: /> wherein /> and />Are respectively-> and />The determined optimal target value;
by means ofAnd NCUC model, forming ERM model for training cost-oriented predictive model H as follows:
in the above formula, subscript s represents a variable corresponding to scene s, and x represents an M-dimensional vector of a binary variable in the NCUC problem; y represents a vector of continuous variables; c and d represent cost vectors; A. b and F represent constant matrices; g is the right-hand vector of all constraints except renewable energy power limits; fy is less than or equal to H * f represents a renewable energy power limit constraint; lambda H 1 Is a 1-norm adjustment term used to prevent an H overfitting, where λ is the overshoot parameter.
7. The method for closed-loop prediction-decision scheduling of a power system based on data driving as claimed in claim 1, wherein the step S2 is solved by using a lagrangian decomposition algorithm to obtain a renewable energy predictor guided by unit combination economy, and specifically comprises:
decomposing an ERM model original problem into |S| disjoint sub-problems by adopting a Lagrangian decomposition algorithm, wherein each sub-problem corresponds to a training scene, and solving the sub-problems in parallel;
and calculating a global optimal solution of the ERM model by adopting an iterative algorithm based on a secondary gradient to obtain the renewable energy predictor taking the unit combination economy as a guide.
8. The method for closed-loop prediction-decision scheduling of a power system based on data driving as claimed in claim 7, wherein the prediction-decision model capable of simultaneously performing the combination decision of the renewable energy prediction and the unit obtained in step S3 is specifically as follows:
s.t.Ax+By≤g
Fy≤H * f
x=[x 1 ,...,x M ]∈{0,1} M
wherein x represents an M-dimensional vector of binary variables in the NCUC problem; y represents a vector of continuous variables; c and d represent cost vectors; A. b and F represent constant matrices; g is the right-hand vector of all constraints except renewable energy power limits; fy is less than or equal to H * f represents the power limit of renewable energy sources, H * Representing the renewable energy predictor.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310160847.5A CN116050653A (en) | 2023-02-23 | 2023-02-23 | Closed-loop prediction-decision scheduling method for power system based on data driving |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310160847.5A CN116050653A (en) | 2023-02-23 | 2023-02-23 | Closed-loop prediction-decision scheduling method for power system based on data driving |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116050653A true CN116050653A (en) | 2023-05-02 |
Family
ID=86120130
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310160847.5A Pending CN116050653A (en) | 2023-02-23 | 2023-02-23 | Closed-loop prediction-decision scheduling method for power system based on data driving |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116050653A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117674123A (en) * | 2023-12-07 | 2024-03-08 | 上海交通大学 | Wind power short-term power prediction method considering operation cost of power system |
-
2023
- 2023-02-23 CN CN202310160847.5A patent/CN116050653A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117674123A (en) * | 2023-12-07 | 2024-03-08 | 上海交通大学 | Wind power short-term power prediction method considering operation cost of power system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Marín et al. | Prediction interval methodology based on fuzzy numbers and its extension to fuzzy systems and neural networks | |
Alamaniotis et al. | Evolutionary multiobjective optimization of kernel-based very-short-term load forecasting | |
EP1907996A1 (en) | Load prediction based on-line and off-line training of neural networks | |
Von Krannichfeldt et al. | Online ensemble approach for probabilistic wind power forecasting | |
Liu et al. | Modelling, analysis and improvement of an integrated chance-constrained model for level of repair analysis and spare parts supply control | |
Alamaniotis et al. | Monthly load forecasting using kernel based gaussian process regression | |
CN112085285A (en) | Bus load prediction method and device, computer equipment and storage medium | |
Wang et al. | Approximating multi-purpose AC optimal power flow with reinforcement trained artificial neural network | |
CN116050653A (en) | Closed-loop prediction-decision scheduling method for power system based on data driving | |
Lee et al. | Generative model-based hybrid forecasting model for renewable electricity supply using long short-term memory networks: A case study of South Korea's energy transition policy | |
Huo et al. | Integrating learning and explicit model predictive control for unit commitment in microgrids | |
Akhtar et al. | Deep learning methods utilization in electric power systems | |
Eren et al. | A comprehensive review on deep learning approaches for short-term load forecasting | |
Sen | Time series prediction based on improved deep learning | |
Hosseini et al. | Data-driven multi-resolution probabilistic energy and reserve bidding of wind power | |
Chen et al. | Regional wind-photovoltaic combined power generation forecasting based on a novel multi-task learning framework and TPA-LSTM | |
CN117494906A (en) | Natural gas daily load prediction method based on multivariate time series | |
Phan et al. | Application of a new Transformer-based model and XGBoost to improve one-day-ahead solar power forecasts | |
Hertel et al. | Evaluation of transformer architectures for electrical load time-series forecasting | |
Cao et al. | Simultaneous prediction for multiple key performance indicators in semiconductor wafer fabrication | |
Wu et al. | Physics-informed ai surrogates for day-ahead wind power probabilistic forecasting with incomplete data for smart grid in smart cities | |
Emamian et al. | Solar power forecasting with LSTM network ensemble | |
Fu et al. | Masked multi-step probabilistic forecasting for short-to-mid-term electricity demand | |
Rosato et al. | A Combined Deep Learning Approach for Time Series Prediction in Energy Environments | |
Suhartono et al. | Hybrid Machine Learning for Forecasting and Monitoring Air Pollution in Surabaya |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |