CN117077960A - Day-ahead scheduling optimization method for regional comprehensive energy system - Google Patents
Day-ahead scheduling optimization method for regional comprehensive energy system Download PDFInfo
- Publication number
- CN117077960A CN117077960A CN202311073076.2A CN202311073076A CN117077960A CN 117077960 A CN117077960 A CN 117077960A CN 202311073076 A CN202311073076 A CN 202311073076A CN 117077960 A CN117077960 A CN 117077960A
- Authority
- CN
- China
- Prior art keywords
- model
- energy system
- scene
- regional
- adopting
- 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 82
- 238000005457 optimization Methods 0.000 title claims abstract description 53
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 27
- 230000005611 electricity Effects 0.000 claims abstract description 25
- 238000009826 distribution Methods 0.000 claims abstract description 22
- 230000015654 memory Effects 0.000 claims abstract description 21
- 238000003064 k means clustering Methods 0.000 claims abstract description 16
- 239000000203 mixture Substances 0.000 claims abstract description 14
- 230000005855 radiation Effects 0.000 claims abstract description 12
- 238000000342 Monte Carlo simulation Methods 0.000 claims abstract description 10
- 238000005070 sampling Methods 0.000 claims abstract description 10
- 230000007246 mechanism Effects 0.000 claims abstract description 9
- 238000013528 artificial neural network Methods 0.000 claims abstract description 8
- 230000009467 reduction Effects 0.000 claims abstract description 8
- 239000013598 vector Substances 0.000 claims description 41
- 230000006870 function Effects 0.000 claims description 20
- 238000010248 power generation Methods 0.000 claims description 19
- 239000007789 gas Substances 0.000 claims description 17
- 238000005338 heat storage Methods 0.000 claims description 16
- 238000013178 mathematical model Methods 0.000 claims description 15
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 12
- 239000002918 waste heat Substances 0.000 claims description 12
- 230000006835 compression Effects 0.000 claims description 10
- 238000007906 compression Methods 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 8
- 238000010521 absorption reaction Methods 0.000 claims description 7
- 229910052799 carbon Inorganic materials 0.000 claims description 7
- 230000008878 coupling Effects 0.000 claims description 7
- 238000010168 coupling process Methods 0.000 claims description 7
- 238000005859 coupling reaction Methods 0.000 claims description 7
- 239000003345 natural gas Substances 0.000 claims description 7
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 claims description 6
- 238000012423 maintenance Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 6
- 230000035772 mutation Effects 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000012614 Monte-Carlo sampling Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000005520 cutting process Methods 0.000 claims description 3
- 239000000446 fuel Substances 0.000 claims description 3
- 230000001537 neural effect Effects 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 230000001902 propagating effect Effects 0.000 claims 1
- 238000006243 chemical reaction Methods 0.000 description 6
- 230000003044 adaptive effect Effects 0.000 description 4
- 238000010276 construction Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 238000007599 discharging Methods 0.000 description 3
- 238000013499 data model Methods 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 238000005265 energy consumption Methods 0.000 description 2
- 238000010438 heat treatment Methods 0.000 description 2
- 230000001172 regenerating effect Effects 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 241000196324 Embryophyta Species 0.000 description 1
- 241000208125 Nicotiana Species 0.000 description 1
- 235000002637 Nicotiana tabacum Nutrition 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 101150067055 minC gene Proteins 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
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
- G06Q10/06—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
- G06Q10/06315—Needs-based resource requirements planning or analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
The invention provides a regional comprehensive energy system day-ahead scheduling optimization method taking the variable working condition characteristics and wind-light prediction errors of equipment into consideration, and a dynamic energy hub model of the regional comprehensive energy system is constructed by combining a mechanism model and an XGboost algorithm; based on regional historical climate data, a long-short-period memory neural network is adopted to predict the wind speed, the ambient temperature and the total solar radiation intensity in the future, so that a model prediction error is obtained; estimating probability density distribution of a prediction error through a Gaussian mixture model, generating a scene by combining a Monte Carlo simulation sampling method, and realizing scene reduction by adopting a K-means clustering method to obtain typical scene probability; and (3) taking the minimum daily operation cost as an optimization target, setting an electricity, heat and cold balance constraint, a load range constraint and a power grid electricity purchase quantity compromise scheme, and solving by adopting a self-adaptive differential algorithm based on a success history to obtain a daily scheduling optimization scheme of the regional comprehensive energy system. The method provided by the invention can effectively improve the economic efficiency of the daily scheduling of the regional comprehensive energy system.
Description
Technical Field
The invention belongs to the technical field of regional comprehensive energy system optimization operation, and particularly relates to a regional comprehensive energy system day-ahead scheduling optimization method considering equipment variable working condition characteristics and wind-light prediction errors.
Background
Along with the conversion of energy production and consumption modes in China, the regional integrated energy system (regional integrated energy system, RIES) has the characteristics of promoting the large-scale access of renewable energy sources and realizing the cascade high-efficiency utilization of the energy sources, and becomes one of the important development directions of the energy industry technology in China. RIES aims at high-efficiency clean utilization of energy, aims at renewable energy consumption, and meets the energy consumption requirements of different quality energy sources such as electricity, heat and cold in a region by coupling various energy situations such as solar energy, wind energy, natural gas, electric energy and the like, so that the RIES is one of important ways for realizing a 'double-carbon' target.
Scientific scheduling of different energy supply links of RIES is an important guarantee of economic and efficient operation, and construction of an equipment model is a precondition of system scheduling optimization. At present, the research of RIES modeling by Chinese scholars mainly depends on an Energy Hub (EH) to describe the coupling relation between different energy supply systems, and the efficiency of each energy supply device in the EH model is usually set to be a constant, however, the research proves that the actual efficiency of the energy conversion device deviates from a design value when the energy conversion device operates under an off-design working condition. The actual efficiency of energy conversion devices such as gas turbines, heat pumps, waste heat boilers is dynamically variable as load and other boundary conditions fluctuate. In response to this problem, some researchers have improved EH models to dynamic energy hub (dynamic energy hub, DEH) models that take into account plant variable-regime characteristics using a two-term fit or piecewise linear fit approach. However, RIES has a complex structure, numerous devices and mutual characteristics determine the complexity of the system modeling process, and the DEH model based on the linear fitting relation still has the problems of inaccurate description and incapability of dynamic update according to the device performance.
Meanwhile, the characteristic that the output of renewable energy sources such as wind and light is difficult to accurately predict is still another challenge facing RIES day-ahead scheduling. The RIES operation optimization method considering the uncertainty of the renewable energy output at present mainly comprises a robust optimization method and a random optimization method. And only the influence of the boundary value of the uncertainty parameter range on the optimization result is considered in the robust optimization process, so that a conservative optimization result is obtained, and the economical efficiency is poor. The random optimization method based on scene generation considers various possible scenes in which random variables possibly appear, and can better balance the robustness and economy of the system in operation. However, the energy supply in the random optimization scheduling scheme based on scene generation in the existing research must meet the load demand of each typical scene, and mainly depends on the improvement of the power grid purchase quantity to cope with the influence of uncertainty, thereby causing the increase of the total operation cost of the scheduling in the future.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a daily scheduling optimization method of an area comprehensive energy system, which considers the variable working condition characteristics of equipment and wind-light prediction errors. And a DEH accurate model considering the variable working condition characteristics of the equipment is constructed by adopting a hybrid driving method based on the mechanism and the data, and a power grid electricity purchase quantity compromise scheme is set, so that the economy of random optimization of RIES day-ahead scheduling is further improved.
The technical scheme adopted for solving the technical problems is as follows:
a regional comprehensive energy system day-ahead scheduling optimization method taking the variable working condition characteristics and wind-light prediction errors of equipment into consideration combines a mechanism model and an XGboost algorithm to construct a regional comprehensive energy system dynamic energy hub model; based on regional historical climate data, a long-short-period memory neural network is adopted to predict the wind speed, the ambient temperature and the total solar radiation intensity in the future, so that a model prediction error is obtained; estimating probability density distribution of a prediction error through a Gaussian mixture model, generating an initial scene by combining a Monte Carlo simulation sampling method, and realizing scene reduction by adopting a K-means clustering method to obtain typical scene probability; and (3) taking the minimum daily operation cost as an optimization target, setting an electricity, heat and cold balance constraint, a load range constraint and a power grid electricity purchasing quantity compromise scheme, and solving by adopting a self-adaptive differential algorithm based on a success history to obtain a daily scheduling optimization strategy of the regional comprehensive energy system. The method specifically comprises the following steps:
step 1, combining an XGboost algorithm and a mechanism model to construct a dynamic energy hub model of an area comprehensive energy system; the regional comprehensive energy system mainly comprises energy supply equipment including photovoltaic power generation, wind power generation, batteries, a heat storage tank, a gas turbine, an electric boiler, a waste heat boiler, a compression refrigerator and an absorption refrigerator;
Step 2, based on regional historical climate data, predicting wind speed, ambient temperature and total solar radiation intensity in a region to be researched by adopting a long-short-period memory neural network to obtain a prediction error data set;
step 3, generating a typical scene based on a scene generation method: estimating probability distribution of a prediction error data set by adopting a Gaussian mixture model, generating a scene by combining a Monte Carlo simulation sampling method, and realizing scene reduction by combining a K-means clustering method to obtain probability of each typical scene;
and 4, setting a heat, electricity and cold balance constraint, a load range constraint and power grid purchase quantity compromise scheme of the regional comprehensive energy system by taking the minimum daily operation cost as an optimization target, adopting a self-adaptive differential algorithm based on success history, and carrying out optimization solution on an objective function to obtain a RIES day-ahead dispatching optimization strategy.
Further, the step 1 specifically includes the following steps:
step 101, constructing mathematical models of main energy supply equipment of an area comprehensive energy system, wherein the mathematical models comprise mathematical models of photovoltaic power generation, wind power generation, batteries, heat storage tanks, gas turbines, electric boilers, waste heat boilers, compression type refrigerators and absorption type refrigerator output; wherein,
The photovoltaic power generation model is as follows:
wherein P is PV Is the output power of the photovoltaic system, kW; η (eta) PV，STC The efficiency of the photovoltaic module under standard test conditions is shown; a is that PV The area of the photovoltaic array at the peak value of the array power is 1m 2 ；G g Is the total radiation intensity of solar energy, kW/m 2 The method comprises the steps of carrying out a first treatment on the surface of the Mu is the temperature coefficient of output power, and is-0.0064/DEG C; t (T) a Is ambient temperature, deg.c; t (T) STC Is the temperature under standard test conditions, 25 ℃; t (T) NOC Is the temperature of a standard working unit, 45 ℃.
The wind power generation model is as follows:
in the formula, v i ，v r And v 0 The cut-in speed, the rated speed and the cut-out speed of the wind characteristic curve are respectively 3m/s,10m/s and 20m/s; v is wind speed, m/s; η (eta) m 、η e The mechanical efficiency and the power generation efficiency of the wind generating set are respectively 0.80 and 0.95; ρ is the air density taken to be 1.225kg/m 3 ；A WT Is the rotor area, m 2 ；C p Is a dynamic coefficient, 0.4 is taken; p (P) WT,r Is rated power of the wind driven generator, kW.
Battery charging process data model:
SOC bat (t)＝SOC bat (t-Δt)[1-δ sd (t)]+η B P in Δt (3)
battery discharge process data model:
SOC bat (t)＝SOC bat (t-Δt)[1-δ sd (t)]-η B P out Δt (4)
in SOC bat (t) is the remaining capacity of the battery at the end of the t period, kW.h; SOC (State of Charge) bat (t- Δt) is the remaining capacity of the battery at the end of the t- Δt time period, kW.h; t represents the time step length when calculating parameters, h; delta sd Representing the period of delta t of the battery Taking the self-discharge rate to be 0.002/h; η (eta) B Representing the charge and discharge efficiency of the battery pack, and taking 0.95; p (P) in Representing charging power, kW; p (P) out Represents discharge power, kW; Δt represents charge and discharge time, and the system operation is simulated in a time-by-time calculation mode, and 1h is taken.
Mathematical model of heat storage process of heat storage tank:
HSS(t)＝HSS(t-Δt)[1-δ HS (t)]+η HS Q in Δt (5)
mathematical model of heat storage tank exothermic process:
HSS(t)＝HSS(t-Δt)[1-δ HS (t)]-η HS Q out Δt (6)
wherein HSS (t) is the residual heat of the heat storage box at the end of the t period, kW.h; HSS (t- Δt) is the remaining heat of the heat storage tank at the end of the t- Δt period, kW.h; delta HS Representing the heat loss rate of the heat stored in the heat storage box in the delta t time period; η (eta) HS Representing energy conversion efficiency in the process of storing heat energy and discharging heat energy of the heat storage device; q (Q) in Representing the input thermal power of the heat storage device, kW; q (Q) out The output thermal power of the heat storage device is indicated, kW.
Gas turbine power generation P GT,e The mathematical model of (kW) is as follows:
wherein eta is GT,e The power generation efficiency of the gas turbine; v (V) g,GT For the natural gas consumption rate of the gas turbine, m 3 /h；q g For natural gas calorific value, 38931kJ/Nm is taken 3 。
Thermal power P for waste heat utilization of gas turbine GT,h The (kW) mathematical model can be described as follows:
wherein eta is GT,h Is the heating efficiency of the gas turbine.
Output heat power P of waste heat boiler WB The (kW) mathematical model can be described as:
P WB ＝η WB P GT,h (9)
Wherein eta is WB Is the efficiency of the waste heat boiler.
Electric boiler output thermal power P EB,h The (kW) mathematical model can be described as:
P EB,h ＝η EB P EB,e (10)
wherein eta is EB Efficiency as an electric boiler; p (P) EB,e For inputting the electric power of the electric boiler, kW.
Refrigerating power P of absorption refrigerator AR,c The (kW) mathematical model is as follows:
P AR,c ＝COP AR P AR,h (11)
wherein P is AR,h The input thermal power of the absorption refrigerator is represented by kW; COP of AR The refrigerating coefficient of the absorption refrigerator is expressed and is the ratio of the output cold quantity and the input heat quantity of the absorption refrigerator.
Refrigerating power P of compression refrigerator CR,c The (kW) mathematical model can be described as follows:
P CR,c ＝COP CR P CR,e (12)
wherein P is CR,e The electric power input by the compression refrigerator is kW; COP of CR The refrigerating coefficient of the compression type refrigerator is the ratio of the output cold quantity and the power consumption of the compression type refrigerator.
And 102, based on historical operation data, performing variable working condition characteristic modeling on each device in the step 101 by adopting an XGBoost algorithm to obtain the mapping relation among the device load rate, the ambient temperature, the local atmospheric pressure and the device efficiency, thereby completing the construction of the dynamic hub model. Wherein, the XGBoost model is input as the equipment load rate l GT Ambient temperature T a,GT Local atmospheric pressure p GT The output is the equipment efficiency eta；
Step 103, constructing a system energy hub dynamic model, and constructing a regional comprehensive energy system energy hub dynamic model by combining the mathematical modeling of the equipment in step 101, the equipment efficiency under the multi-boundary condition and the coupling relation before the equipment, which are obtained in step 102, wherein the formula is as follows:
Wherein η is the efficiency of the device; l is the load factor of the equipment and is defined as the ratio of the actual output power of the equipment to the rated capacity; p (P) i,r Rated power for i devices; p (P) G The power of the system purchased from the power grid; p (P) E,out 、P H,out 、P C,out The electric, heat and cold outputs of the functional devices are respectively.
Further, in step 102, the xGBoost algorithm is adopted to model the variable working condition characteristics of each device, and the method specifically comprises the following steps:
step A, setting regression values of XGBoost models as follows:
wherein,representing the efficiency η, x of each device obtained from the XGBoost model i Representing model input vectors, including device load rate l GT Ambient temperature T a,GT Local atmospheric pressure p GT T is the number of decision trees, f k The corresponding structure is q k The leaf weight is omega k Is the kth independent tree of (2);
step B, setting an objective function of the XGBoost model as follows:
wherein l is a minutely convex loss function for representing regression valuesAnd true value y i Error between n represents the number of samples, Ω (f) represents the minimum regularization term, +.>T k And ω represents the number of leaf nodes in the tree and the weight value of the leaf, γ represents the penalty coefficient, λ is the regular term coefficient, and C represents a constant;
through second-order taylor expansion, a loss function of the XGBoost regression model is constructed and expressed as follows:
Wherein,representing the first partial derivative and the second partial derivative of the loss function, respectively, I j ＝{i|q(X i ) =j) } represents the sample set of leaf j.
The step C, XGBoost model iterates as follows:
wherein,representing regression values of the ith group of samples after t iterations; />Representing the initial value of the i-th set of samples.
The above steps are repeated until the model loss function converges or the end condition of training is reached.
And (3) inputting the efficiency of different load rates, ambient temperature and equipment efficiency under the atmospheric pressure obtained by the XGBoost model into a formula (13) to obtain the RIES energy hub dynamic model.
Further, the step 2 specifically includes the following steps:
step 201, respectively inputting wind speed, ambient temperature and historical climate data set X of total solar radiation intensity into an input layer of an LSTM model t And then propagates forward, entering the hidden layer.
Step 202, hidden layer according to X t H at the last moment t-1 And c t-1 Calculating to obtain H at the current moment t And c t The method comprises the steps of carrying out a first treatment on the surface of the Wherein H is t-1 For the output of the memory block at time t-1, c t And c t-1 The unit state vectors are respectively at the time t and the time t-1;
step 203, utilize O t And c t Calculating to obtain H t And will hide H of the layer t And c t To the neural unit at the next moment; wherein,
O t ＝sigmoid(W O X t +W O H t-1 +B O ) (18)
c t ＝F t c t-1 +I t tanh(W c X t +W c H t-1 +B c ) (19)
F t ＝sigmoid(W F X t +W F H t-1 +B F ) (21)
I t ＝sigmoid(W I X t +W I H t-1 +B I ) (22)
H t ＝O t tanh(c t ) (23)
t is the moment of time F t Is amnestic door, I t For input gate, O t Is an output door; sigmoid and hyperbolic tangent function tanh are two activation functions; w (W) F 、W I 、W O 、W c The weights of the forget gate, the input gate, the output gate and the unit state vector are respectively; b (B) F 、B I 、B O 、B c Bias terms of forget gate, input gate, output gate and cell state vector, respectively.
Step 204, substituting the output value and the true value of the model into the loss functionUpdating the parameter weight value by adopting a gradient back propagation method; wherein H is the length of the segmentation window, and m is the dimension of the training set data; p (P) i For model inversion value, y i Is the true value of the model;
step 205, repeating the above steps until the neural network model converges or reaches the training termination condition.
Further, the step 3 specifically includes the following steps:
step 301, estimating probability density distribution of the climate data prediction error obtained in the step 2 by adopting a Gaussian mixture model; the probability distribution form of the multi-element Gaussian mixture model is set as follows:wherein X is a historical climate data vector, x= [ X ] 1 ,x 2 ,…,x n ] T The method comprises the steps of carrying out a first treatment on the surface of the M is the number of Gaussian mixture model submodels omega m Is the weight of the submodel and ω m≥0 ，/>φ(X|θ m ) Is a multi-dimensional single gaussian probability density function of the m-th sub-model, Wherein mu m Sum sigma m Respectively representing the mean and covariance matrix of the kth sub-model;
step 302, generating an initial scene set with a certain scale based on the obtained wind and light prediction error probability distribution model and by combining a Monte Carlo simulation sampling method;
and 303, cutting the generated scene set by combining a K-means clustering method to obtain the occurrence probability of each typical scene.
Further, in step 303, the generated scene set is cut in combination with the K-means clustering method, which includes the following steps:
step a, for an initial scene set s= [ S ] generated by a monte carlo sampling method 1 ,…,S q ,…,S Q ]Setting an initial cluster K=3, and randomly generating a cluster center u k Obtaining a cluster center matrix U= [ U ] 1 ,u 2 ,u 3 ]；
Step B, calculating each scene S q With each cluster center u k Euclidean distance between the two scenes S according to the principle of minimum Euclidean distance q Dividing into the nearest clusters;
step C, calculating a sample mean value in a new cluster and taking the sample mean value as a new cluster center, and then turning to step B until the cluster center is not changed any more, ending the K-means clustering process, and calculating a DBI value at the moment, whereinK is the cluster number, i, j=1, 2, …, K, and i+.j, s i ，s j Representing the average distance from each point to the center point within the class of clusters i and j, respectively, also called cluster diameter, d ij Representative is the inter-class center point distance of cluster i and cluster j.
And D, repeating the steps A-C, and sequentially calculating the clustering condition when the clustering number K is 4 to 15. And then selecting the corresponding clustering result with the minimum DBI value as the optimal clustering scene to be reserved.
Further, the step 4 specifically includes the following steps:
step 401, considering the carbon emission of the system, setting a scheduling optimization target, and setting the daily total running cost C of the system total The lowest, namely:
minC total ＝C F +C M +C G +C CO2,tot (24)
wherein C is F 、C M 、C G 、C CO2，tot Respectively fuel cost, equipment maintenance cost, electricity purchasing cost and CO 2 And (5) discharging cost.
Wherein C is gas Is the price of natural gas, τ i Is i equipment maintenance coefficient, C B Is the electricity price of the power grid, C CO2 Is CO 2 Discharge trade price, E CO2 Is system CO 2 Discharge amount.
Step 402, setting a power grid purchase quantity compromise scheme, wherein the purchase quantity at each moment is determined by the following formula:
wherein,the power purchasing power at the time t in the day-ahead power grid dispatching scheme is obtained; k (K) best K=1, 2, K, for a typical scene number best ；p k Probability of occurrence for the kth typical scene; p (P) tG,k Is the actual power purchase of the power grid at the time t in the kth typical scene.
Step 403, setting energy balance constraint, wherein the demand of energy and supply are balanced at each moment in the scheduling period, including electric, thermal and cold energy balance. And setting a load range constraint, wherein the load ratio of each device is in the load range.
And 404, solving the model by adopting an adaptive differential algorithm based on the success history, and determining a day-ahead scheduling strategy of the regional comprehensive energy system.
Further, step 404, using an adaptive differential algorithm based on a success history to solve the model, determines a day-ahead scheduling policy for the regional integrated energy system, comprising the steps of:
step A, initializing data; setting population scale U, decision variable dimension D and maximum evolution algebra G max The initial population is represented as a set of real parameter vectors x i ＝(x 1 ,x 2 ,···,x D )，i＝1,2,···,U。
Step B, mutation; in each generation, current-to-pbest/1 mutation strategy is adopted from an existing population vector x i,G Generates a variation vector v i,G As shown in formula (30).
v i,G ＝x i,G +F i ·(x pbest,G -x i,G )+F i ·(x r1,G -x r2,G ) (30)
Wherein F is i Representing the scale factor, being the forward control parameter of the differential vector, x pbest,G Is the best individual in the G generation population, and the index r 1 ，r 2 Are mutually exclusive integers randomly selected from a population.
Step C, crossing; the variance vector v is obtained by a binomial crossing method i,G And the target vector x i,G Cross-mixing is shown in formula (31).
When the randomly generated value in the [0,1 ] range is less than or equal to the crossing rate CR i At the time, u j,i,G Assigning as element v in a variant vector j,i,G 。j rand Is one in [1, D]Decision variable indexes generated randomly in the range.
Step D, generating all test vectors u i,G After that, by comparing the test vectors u i,G And a target vector x i,G Is selected to generate the next generation population.
And E, calculating population fitness. Judging whether the evolution reaches the maximum evolution algebra G max If yes, stopping evolution and outputting; if not, go to step B and continue the iteration.
Wherein, i individual scale factors CR in the population i And crossing rate F i The selection of (c) follows the following principle:
m of U items stored in historic storage of SHADE algorithm CR ，M F . First, M is CR,i 、M F,i (i=1.,), the content of U) was initialized to 0.5.
In each generation, differential evolution control parameters F and CR are adaptively generated based on historical parameters, test vectors are generated, selection is applied, and historical memory data (M CR ，M F ). This process is repeated until some termination criteria is reached.
In each generation, first from [1, U]An index r is randomly selected i Then, according to formulas (33) and (34), control parameter CR used by the i individual is generated i And F i 。
Wherein, randn i (μ,σ 2 )，randc i (μ,σ 2 ) Respectively represent the mean value is mu and the variance is sigma 2 Randomly generated values in the normal distribution and the cauchy distribution. If an in [0,1 ] is generated]CR outside of range i The value is replaced by the limit value (0 or 1) closest to the generated value. If F is generated i >1, then F i Taking 1, if F i At < 0, regenerating using formula (34) until an effective F is produced i Values.
CR used by successful individuals in the equation i And F i The value is recorded at Z CR And Z F At the end of the generation, the content of the memory is updated as follows:
the index z (1. Ltoreq.z. Ltoreq.U) determines the location in memory to be updated. At the beginning of the search, z is initialized to 1. K increases when a new element is inserted into memory. If z>U, z is set to 1. In the G generation, the z-th element in memory is updated. Notably, in formulas (35) and (36), when all individuals in the G generation are unable to generate test vectors better than the parent, i.e.The memory is not updated.
Weighted arithmetic mean in equation (36), mean WA (Z CR ) Calculated according to equation (37).
Δf z ＝|f(u z,G )-f(x z,G )| (39)
Weighted average mean in equation (40) WL (Z F ) Calculated using the following formula, and mean WA (Z CR ) As such, an improved amount is used to affect the adaptivity of the parameters.
The invention has the beneficial effects that:
(1) The invention provides a regional comprehensive energy system day-ahead scheduling optimization method taking the variable working condition characteristics and wind-light prediction errors of equipment into consideration, a regional comprehensive energy system dynamic energy hub model is built by combining a mechanism model and an XGboost algorithm, the variable working condition operation characteristics of RIES equipment are considered, the actual efficiency of the built equipment is a mapping relation with load, ambient temperature and atmospheric pressure, and the problems that a DEH model based on a linear fitting relation is inaccurate and cannot be updated dynamically according to the performance of the equipment are solved; based on regional historical climate data, a long-short-period memory neural network is adopted to predict the wind speed, the ambient temperature and the total solar radiation intensity in the future to obtain a model prediction error, and the climate data is adopted to replace the traditional prediction error based on wind and light output, so that the problem that the prediction error is large due to lack of enough equipment output historical operation data, and the fitting probability distribution is inaccurate is avoided; estimating probability density distribution of a prediction error through a Gaussian mixture model, generating a scene by combining a Monte Carlo simulation sampling method, and realizing scene reduction by adopting a K-means clustering method to obtain typical scene probability; the daily operation cost is used as an optimization target, an electricity, heat and cold balance constraint, a load range constraint and a power grid electricity purchasing quantity compromise scheme are set, a self-adaptive differential algorithm based on a success history is adopted for solving, a regional comprehensive energy system daily scheduling optimization scheme is obtained, and the compromise power grid electricity purchasing scheme is adopted, so that daily scheduling cost can be effectively reduced, and balance between system operation uncertainty and economy is considered.
(2) The future scheduling optimization method of the domain comprehensive energy system provided by the invention takes into account the variability of the actual operation condition of RIES energy supply equipment and the complexity of the influence of multiple factors on the equipment output, and builds the mapping relation between the equipment efficiency and the load and the environmental boundary according to the actual operation condition of the equipment, thereby improving the DEH model and improving the accuracy of the model; meanwhile, by adopting a random optimization method based on scene generation, an optimization target for seeking carbon emission reduction and operation economy is reasonably set, a compromise power grid electricity purchasing scheme is provided, and the economy of daily scheduling of the regional comprehensive energy system is improved.
Drawings
FIG. 1 is a general flow chart of a dispatching optimization method of an area comprehensive energy system;
FIG. 2 is a flow chart of the dynamic energy hub model construction of the present invention;
FIG. 3 is a flow chart of climate data prediction in accordance with the present invention;
FIG. 4 is a flow chart of scene generation and curtailment according to the present invention;
FIG. 5 is a flow chart of a day-ahead scheduling optimization strategy for solving RIES according to the present invention;
fig. 6 is a schematic diagram of an apparatus of an area integrated energy system according to an embodiment of the present invention.
Detailed Description
For a clearer understanding of technical features, objects and effects of the present invention, a detailed description of embodiments of the present invention will be made with reference to the accompanying drawings.
FIG. 1 is a general flow chart of a day-ahead dispatch optimization method for an area integrated energy system. The energy efficiency day-ahead scheduling optimization of the regional comprehensive energy system comprises the following steps:
step 1, combining an XGboost algorithm and a mechanism model to construct a dynamic energy hub model of an area comprehensive energy system; the regional comprehensive energy system mainly comprises energy supply equipment, such as photovoltaic power generation, wind power generation, batteries, a heat storage tank, a gas turbine, an electric boiler, a waste heat boiler, a compression refrigerator and an absorption refrigerator;
the regional comprehensive energy system is formed by coupling a plurality of energy systems such as electric power, heating power, natural gas and the like, and renewable energy sources such as solar energy, wind energy and the like are connected to supply the energy systems such as electricity, heat, cold and the like in the region. Therefore, the energy supply equipment of the regional comprehensive energy system comprises photovoltaic power generation, wind power generation, batteries, heat storage tanks, gas turbines, electric boilers, waste heat boilers, compression type refrigerators and absorption type refrigerators. The research on the day-ahead scheduling of regional comprehensive energy is essentially the planned distribution of the output of the energy supply equipment.
The regional comprehensive energy system has more devices and strong coupling relevance, and the normal operation of each energy supply device in the variable working condition and deviating from the design working condition is the normal operation. The actual efficiency of energy conversion equipment including gas turbines, heat pumps and waste heat boilers is dynamically changed along with load and other boundary condition fluctuation, and is not a constant value. According to the invention, the XGboost algorithm and the mechanism model are combined, the variability of the operation condition and the complexity of the influence of multiple factors on the equipment output are fully considered, the mapping relation between the equipment efficiency and the load and the environment boundary is constructed according to the actual operation condition of the equipment, the dynamic energy hub model of the regional comprehensive energy system is constructed, and the accuracy of the model is improved.
Fig. 2 is a flow chart of the dynamic energy hub model construction according to the present invention, and the step 1 includes:
step 101, constructing mathematical models of main energy supply equipment of an area comprehensive energy system, wherein the mathematical models comprise mathematical models of photovoltaic power generation, wind power generation, batteries, heat storage tanks, gas turbines, electric boilers, waste heat boilers, compression type refrigerators and absorption type refrigerator output;
102, based on historical operation data, modeling the variable working condition characteristics of each device in the step 101 by adopting an XGBoost algorithm to obtain the mapping relation among the device load rate, the ambient temperature, the local atmospheric pressure and the device efficiency, wherein the XGBoost model is input into the device load rate l GT Ambient temperature T a,GT Local atmospheric pressure p GT The output is the equipment efficiency eta;
step 103, constructing a system energy hub dynamic model, and constructing an area comprehensive energy system energy hub dynamic model by combining the equipment mathematical modeling in the step 101, the equipment efficiency under the multi-boundary condition and the coupling relation before equipment, which are obtained in the step 102; the following formula is shown:
wherein η is the efficiency of the device; l is the load factor of the equipment and is defined as the ratio of the actual output power of the equipment to the rated capacity; p (P) i,r Rated power for i devices; p (P) G The power of the system purchased from the power grid; p (P) E,out 、P H,out 、P C,out The electric, heat and cold outputs of the functional devices are respectively.
In step 102, the XGBoost algorithm is adopted to model the variable working condition characteristics of each device, and the method specifically comprises the following steps:
step A, setting regression values of XGBoost models as follows:
wherein,representing the efficiency η, x of each device obtained from the XGBoost model i Representing model input vectors, including device load rate l GT Ambient temperature T a,GT Local atmospheric pressure p GT T is the number of decision trees, f k The corresponding structure is q k The leaf weight is omega k Is the kth independent tree of (2);
step B, setting a loss function of the XGBoost model as follows:
/>
wherein l is a minutely convex loss function for representing regression valuesAnd true value y i Error between n represents the number of samples, Ω (f) represents the minimum regularization term, +.>T k And ω represents the number of leaf nodes in the tree and the weight value of the leaf, γ represents the penalty coefficient, λ is the regular term coefficient, and C represents a constant;
through second-order taylor expansion, a loss function of the XGBoost regression model is constructed and expressed as follows:
wherein,representing the first partial derivative and the second partial derivative of the loss function, respectively, I j ＝{i|q(X i ) =j) } represents the sample set of leaf j.
The step C, XGBoost model iterates as follows:
wherein,representing regression values of the ith group of samples after t iterations; />Representing the initial value of the i-th set of samples.
The above steps are repeated until the model loss function converges or the end condition of training is reached.
And (3) inputting the efficiency of different load rates, ambient temperature and equipment efficiency under the atmospheric pressure obtained by the XGBoost model into a formula (1), thus obtaining the RIES energy hub dynamic model.
Step 2, based on regional historical climate data, predicting wind speed, ambient temperature and total solar radiation intensity in a region to be researched by adopting a long-short-period memory neural network to obtain a prediction error data set;
the invention adopts long-term memory neural network wind speed, ambient temperature, solar energy total radiation intensity and other climate parameters to predict, and replaces a prediction error model of wind power and photovoltaic output. The scale of the prediction data can influence the accuracy of the prediction model to a certain extent, and the accuracy of probability distribution estimation can be influenced due to the fact that the regional comprehensive energy system actually put into operation is less, and the wind power and photovoltaic output data which can be obtained are limited due to the confidentiality of the data, and the data scale is reduced. However, the historical climate data of the region to be studied can be obtained according to the geographic position of the region to be studied. Therefore, in consideration of the practical application feasibility and operability, the invention adopts the historical climate data to predict so as to obtain a prediction error.
FIG. 3 is a flow chart of the climate data prediction according to the present invention, wherein the step 2 comprises:
step 201, respectively inputting wind speed, ambient temperature and historical climate data set X of total solar radiation intensity into an input layer of an LSTM model t And then propagates forward, entering the hidden layer.
Step 202, hidden layer according to X t H at the last moment t-1 And c t-1 Calculating to obtain H at the current moment t And c t The method comprises the steps of carrying out a first treatment on the surface of the Wherein H is t-1 For the output of the memory block at time t-1, c t Is a cell state vector.
Step 203, utilize O t And c t Calculating to obtain H t And will hide H of the layer t And c t To the neural unit at the next moment; wherein,
O t ＝sigmoid(W O X t +W O H t-1 +B O ) (6)
c t ＝F t c t-1 +I t tanh(W c X t +W c H t-1 +B c ) (7)
F t ＝sigmoid(W F X t +W F H t-1 +B F ) (9)
I t ＝sigmoid(W I X t +W I H t-1 +B I ) (10)
H t ＝O t tanh(c t ) (11)
t is the moment of time F t Is amnestic door, I t For input gate, O t Is an output door; sigmoid and hyperbolic tangent function tanh are two activation functions; w (W) F 、W I 、W O 、W c The weights of the forget gate, the input gate, the output gate and the unit state vector are respectively; b (B) F 、B I 、B O 、B c Bias terms of forget gate, input gate, output gate and cell state vector, respectively.
Step 204, substituting the output value and the true value of the model into the loss functionUpdating the parameter weight value by adopting a gradient back propagation method; wherein H is the length of the segmentation window, and m is the dimension of the training set data; p (P) i For model inversion value, y i For the true value of the model
Step 205, repeating the above steps until the neural network model converges or reaches the training termination condition.
Step 3, generating a typical scene based on a scene generation method: estimating probability distribution of a prediction error data set by adopting a Gaussian mixture model, generating a scene by combining a Monte Carlo simulation sampling method, and realizing scene reduction by combining a K-means clustering method to obtain probability of each typical scene;
reasonable determination of the number of curtailed scenes is a key to random optimized scheduling based on scene generation. The invention adopts DBI as the standard for evaluating the clustering number. And when the DBI is minimum, the corresponding clustering result is reserved as the optimal clustering scene.
Fig. 4 is a scene generation and clipping flowchart, and the step 3 includes:
step 301, estimating probability density distribution of the climate data prediction error obtained in the step 2 by adopting a Gaussian mixture model; the probability distribution form of the multi-element Gaussian mixture model is set as follows:wherein X is a historical climate data vector, x= [ X ] 1 ,x 2 ,…,x n ] T The method comprises the steps of carrying out a first treatment on the surface of the M is the number of Gaussian mixture model submodels omega m Is the weight of the submodel and ω m≥0 ，/>φ(X|θ m ) Is a multi-dimensional single gaussian probability density function of the m-th sub-model,wherein mu m Sum sigma m Respectively representing the mean and covariance matrix of the kth sub-model;
Step 302, generating an initial scene set with a certain scale based on the obtained wind and light prediction error probability distribution model and by combining a Monte Carlo simulation sampling method;
and 303, cutting the generated scene set by combining a K-means clustering method to obtain the occurrence probability of each typical scene.
In step 303, the method for reducing the generated scene set by combining the K-means clustering method comprises the following steps:
step a, for an initial scene set s= [ S ] generated by a monte carlo sampling method 1 ,…,S q ,…,S Q ]Setting an initial cluster K=3, and randomly generating a cluster center u k Obtaining a cluster center matrix U= [ U ] 1 ,u 2 ,u 3 ]；
Step B, calculating each scene S q With each cluster center u k Euclidean distance between the two scenes S according to the principle of minimum Euclidean distance q Dividing into the nearest clusters;
step C, calculating a sample mean value in a new cluster and taking the sample mean value as a new cluster center, and then turning to step B until the cluster center is not changed any more, ending the K-means clustering process, and calculating a DBI value at the moment, whereinK is the cluster number, i, j=1, 2, …, K, and i+.j, s i ，s j Representing the average distance from each point to the center point within the class of clusters i and j, respectively, also called cluster diameter, d ij Representative is the inter-class center point distance of cluster i and cluster j.
And D, repeating the steps A-C, and sequentially calculating the clustering condition when the clustering number K is 4 to 15. And then selecting the corresponding clustering result with the minimum DBI value as the optimal clustering scene to be reserved.
And 4, setting a heat, electricity and cold balance constraint, a load range constraint and power grid purchase quantity compromise scheme of the regional comprehensive energy system by taking the minimum daily operation cost as an optimization target, adopting a self-adaptive differential algorithm based on success history, and carrying out optimization solution on an objective function to obtain a RIES day-ahead dispatching optimization strategy.
Generally, the dispatching optimization of the regional comprehensive energy system takes the minimum daily operation cost as an optimization target, but the daily operation cost only considers CO 2 Is not limited by the cost of (a). Today, the carbon emission costs along with the two carbon targets are also considered. Meanwhile, energy supply in a random optimization scheduling scheme based on scene generation must meet the load demand of each typical scene, and the influence of uncertainty is mainly dealt with by improving the power grid purchase quantity, so that the total operation cost of scheduling in the future is increased. In the invention, a compromise electricity purchasing scheme is provided, so that the day-ahead dispatching cost can be effectively reduced, and the balance between the uncertainty of system operation and economy is realized.
FIG. 5 is a flowchart of a day-ahead schedule optimization strategy for solving RIES, said step 4 comprising:
step 401, considering the carbon emission of the system, setting a scheduling optimization target, and setting the daily total running cost C of the system total The lowest, namely:
wherein C is F 、C M 、C G 、C CO2，tot Respectively fuel cost, equipment maintenance cost, electricity purchasing cost and CO 2 And (5) discharging cost.
Wherein C is gas Is the price of natural gas, τ i Is i equipment maintenance coefficient, C B Is the electricity price of the power grid, C CO2 Is CO 2 Discharge trade price, E CO2 Is system CO 2 Discharge amount.
Step 402, setting a power grid purchase quantity compromise scheme, wherein the purchase quantity at each moment is determined by the following formula:
wherein,the power purchasing power at the time t in the day-ahead power grid dispatching scheme is obtained; k (K) best K=1, 2, K, for a typical scene number best ；p k Probability of occurrence for the kth typical scene; p (P) tG,k Is the actual power purchase of the power grid at the time t in the kth typical scene.
Step 403, setting energy balance constraint, wherein the demand of energy and supply are balanced at each moment in the scheduling period, including electric, thermal and cold energy balance. And setting a load range constraint, wherein the load ratio of each device is in the load range.
And 404, solving the model by adopting an adaptive differential algorithm based on the success history, and determining a day-ahead scheduling strategy of the regional comprehensive energy system.
Further, step 404, using an adaptive differential algorithm based on a success history to solve the model, determines a day-ahead scheduling policy for the regional integrated energy system, comprising the steps of:
step A, initializing data; setting population scale U, decision variable dimension D and maximum evolution algebra G max The initial population is represented as a set of real parameter vectors x i ＝(x 1 ,x 2 ,···,x D )，i＝1,2,···,U。
Step B, mutation; in each generation, current-to-pbest/1 mutation strategy is adopted from an existing population vector x i,G Generates a variation vector v i,G As shown in formula (26).
v i,G ＝x i,G +F i ·(x pbest,G -x i,G )+F i ·(x r1,G -x r2,G ) (18)
Wherein F is i Representing the scale factor, being the forward control parameter of the differential vector, x pbest,G Is the best individual in the G generation population, and the index r 1 ，r 2 Are mutually exclusive integers randomly selected from a population.
Step C, crossing; the variance vector v is obtained by a binomial crossing method i,G And the target vector x i,G Cross-mixing, as shown in formula (27)Shown.
When the randomly generated value in the [0,1 ] range is less than or equal to the crossing rate CR i At the time, u j,i,G Assigning as element v in a variant vector j,i,G 。j rand Is one in [1, D]Decision variable indexes generated randomly in the range.
Step D, generating all test vectors u i,G After that, by comparing the test vectors u i,G And a target vector x i,G Is selected to generate the next generation population.
And E, calculating population fitness. Judging whether the evolution reaches the maximum evolution algebra G max If yes, stopping evolution and outputting; if not, go to step B and continue the iteration.
Wherein, i individual scale factors CR in the population i And crossing rate F i The selection of (c) follows the following principle:
m of U items stored in historic storage of SHADE algorithm CR ，M F . First, M is CR,i 、M F,i (i=1.,), the content of U) was initialized to 0.5.
In each generation, differential evolution control parameters F and CR are adaptively generated based on historical parameters, test vectors are generated, selection is applied, and historical memory data (M CR ，M F ). This process is repeated until some termination criteria is reached.
In each generation, first from [1, U]An index r is randomly selected i Then, according to the formulas (21) and (22), the control parameter CR used by the i individual is generated i And F i 。
Wherein, randn i (μ,σ 2 )，randc i (μ,σ 2 ) Respectively represent the mean value is mu and the variance is sigma 2 Randomly generated values in the normal distribution and the cauchy distribution. If an in [0,1 ] is generated]CR outside of range i The value is replaced by the limit value (0 or 1) closest to the generated value. If F is generated i >1, then F i Taking 1, if F i At < 0, regenerating by applying formula (23) until an effective F is produced i Values.
CR used by successful individuals in the equation i And F i The value is recorded at Z CR And Z F At the end of the generation, the content of the memory is updated as follows:
the index z (1. Ltoreq.z. Ltoreq.U) determines the location in memory to be updated. At the beginning of the search, z is initialized to 1. K increases when a new element is inserted into memory. If z>U, z is set to 1. In the G generation, the z-th element in memory is updated. Notably, in formulas (23) and (24), when all individuals in the G generation are unable to generate test vectors better than the parent, i.e.The memory is not updated.
Weighted arithmetic mean in equation (24), mean WA (Z CR ) Calculated according to equation (25).
Δf z ＝|f(u z,G )-f(x z,G )| (27)
Weighted average mean in equation (24) WL (Z F ) Calculated using the following formula, and mean WA (Z CR ) As such, an improved amount is used to affect the adaptivity of the parameters.
The technical proposal of the invention is implemented in a certain area of the tobacco station city of Shandong province, and the area occupies 12.8km 2 The comprehensive energy system equipment configuration is shown in fig. 6, and the data for training the climate prediction model is selected from time-by-time historical climate data of the local 2016-2020 obtained from a Meteonarm 8 database, and the total data is 43800 groups, wherein 80% of the data are used for training the model, and 20% of the data are used for testing. 10000 climates and scene sets are generated by adopting a Monte Carlo sampling method based on climate data prediction errors, the scenes are cut by adopting a K-means clustering method, when the clustering number is 9, the DBI value is minimum, and the optimal typical scene number K is the smallest best Selected as 9. Probability p of 9 scene correspondences k 14.14%,11.03%,10.59%,17.84%,15.77%,6.26%,9.89%,5.06% and 9.42%, respectively. By adopting a random optimization scheme of compromise purchase of electricity, the actual total running cost of the system is reduced by 3.20%. The result shows that the future scheduling scheme of the regional comprehensive energy system can effectively improve the economical efficiency of the system operation.
The above description is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above examples, and all technical solutions belonging to the concept of the present invention belong to the protection scope of the present invention. It should be noted that modifications and adaptations to the present invention may occur to one skilled in the art without departing from the principles of the present invention and are intended to be within the scope of the present invention.
Claims (8)
1. A day-ahead scheduling optimization method for an area comprehensive energy system is characterized in that a dynamic energy hub model of the area comprehensive energy system is built by combining a mechanism model and an XGboost algorithm; based on regional historical climate data, predicting wind speed, ambient temperature and total solar radiation intensity by adopting a long-short-period memory neural network to obtain a model prediction error; estimating probability density distribution of a prediction error through a Gaussian mixture model, generating a scene by combining a Monte Carlo simulation sampling method, and realizing scene reduction by adopting a K-means clustering method to obtain typical scene probability; and (3) taking the minimum daily operation cost as an optimization target, setting an electricity, heat and cold balance constraint, a load range constraint and a power grid electricity purchase quantity compromise scheme, and solving by adopting a self-adaptive differential algorithm based on a success history to obtain a daily scheduling optimization scheme of the regional comprehensive energy system.
2. The regional integrated energy system day-ahead scheduling optimization method of claim 1, comprising the steps of:
step 1, combining an XGboost algorithm and a mechanism model to construct a dynamic energy hub model of an area comprehensive energy system; the regional comprehensive energy system mainly comprises energy supply equipment including photovoltaic power generation, wind power generation, batteries, a heat storage tank, a gas turbine, an electric boiler, a waste heat boiler, a compression refrigerator and an absorption refrigerator;
step 2, based on regional historical climate data, predicting wind speed, ambient temperature and total solar radiation intensity in a region to be researched by adopting a long-short-period memory neural network to obtain a prediction error data set;
step 3, generating a typical scene based on a scene generation method: estimating probability distribution of a prediction error data set by adopting a Gaussian mixture model, generating an initial scene set by combining a Monte Carlo simulation sampling method, and realizing scene reduction by combining a K-means clustering method to obtain probability of each typical scene;
and 4, setting an electricity, heat and cold balance constraint, a load range constraint and a power grid electricity purchasing quantity compromise scheme by taking the minimum daily operation cost as an optimization target, adopting a self-adaptive differential algorithm based on success history, and carrying out optimization solution on an objective function to obtain a daily scheduling optimization scheme of the regional comprehensive energy system.
3. The method for optimizing daily schedule of regional integrated energy system according to claim 2, wherein the step 1 specifically comprises the steps of:
step 101, constructing mathematical models of main energy supply equipment of an area comprehensive energy system, wherein the mathematical models comprise mathematical models of photovoltaic power generation, wind power generation, batteries, heat storage tanks, gas turbines, electric boilers, waste heat boilers, compression type refrigerators and absorption type refrigerator output;
102, based on historical operation data, performing variable working condition characteristic modeling on each device in the step 101 by adopting an XGBoost algorithm to obtain a mapping relation among the device load rate, the ambient temperature, the local atmospheric pressure and the device efficiency;
step 103, constructing a system energy hub dynamic model, and constructing a regional comprehensive energy system energy hub dynamic model by combining the mathematical modeling of the equipment in step 101, the equipment efficiency under the multi-boundary condition and the coupling relation before the equipment, which are obtained in step 102, wherein the formula is as follows:
wherein η is the efficiency of the device; l is the load factor of the equipment and is defined as the ratio of the actual output power of the equipment to the rated capacity; p (P) i,r Rated power for i devices; p (P) G The power of the system purchased from the power grid; p (P) E,out 、P H,out 、P C,out The electric, heat and cold outputs of the functional devices are respectively.
4. The method for optimizing daily schedule of regional integrated energy system according to claim 2, wherein the step 2 specifically comprises the steps of:
step 201, respectively inputting wind speed, ambient temperature and historical climate data set X of total solar radiation intensity into an input layer of an LSTM model t Then propagating forward, inputting a hidden layer;
step 202, hidden layer according to X t Output H of the last time t-1 And cell state vector c t-1 Calculating to obtain H at the current moment t And c t ；
Step 203, utilize O t And c t Calculating to obtain H t And will hide H of the layer t And c t To the neural unit at the next moment; wherein,
O t ＝sigmoid(W O X t +W O H t-1 +B O ) (2)
c t ＝F t c t-1 +I t tanh(W c X t +W c H t-1 +B c ) (3)
B t ＝I t c t +F t B t-1 (5)
F t ＝sigmoid(W F X t +W F H t-1 +B F ) (6)
I t ＝sigmoid(W I X t +W I H t-1 +B I ) (7)
H t ＝O t tanh(c t ) (8)
t is the moment of time F t Is amnestic door, I t For input gate, O t Is an output door; sigmoid and hyperbolic tangent functionThe number tanh is two activation functions; w (W) F 、W I 、W O 、W c The weights of the forget gate, the input gate, the output gate and the unit state vector are respectively; b (B) F 、B I 、B O 、B c Bias items of a forget gate, an input gate, an output gate and a unit state vector respectively;
step 204, substituting the output value and the true value of the model into the loss functionUpdating the parameter weight value by adopting a gradient back propagation method; wherein H is the length of the segmentation window, and m is the dimension of the training set data; p (P) i For model inversion value, y i Is the true value of the model;
step 205, repeating the above steps until the neural network model converges or reaches the training termination condition.
5. The method for optimizing daily schedule of regional integrated energy system according to claim 2, wherein the step 3 specifically comprises the steps of:
step 301, estimating probability density distribution conditions of the climate data prediction errors obtained in the step 2 by adopting a Gaussian mixture model;
step 302, generating an initial scene set with a certain scale based on the obtained wind and light prediction error probability distribution model and by combining a Monte Carlo simulation sampling method;
and 303, cutting the generated initial scene set by combining a K-means clustering method to obtain the occurrence probability of each typical scene.
6. The method for optimizing day-ahead scheduling of an area integrated energy system according to claim 5, wherein said step 303 comprises:
step a, for an initial scene set s= [ S ] generated by a monte carlo sampling method 1 ,…,S q ,…,S Q ]Setting an initial cluster K=3, and randomly generating a cluster center u k Obtaining a cluster center matrix U= [ solution ]u 1 ,u 2 ,u 3 ]；
Step B, calculating each scene S q With its cluster center u k Euclidean distance between the two scenes S according to the principle of minimum Euclidean distance q Dividing into the nearest clusters;
step C, calculating a sample mean value in a new cluster and taking the sample mean value as a new cluster center, and then turning to step B until the cluster center is not changed any more, ending the K-means clustering process, and calculating a DBI value at the moment, wherein K is the cluster number, i, j=1, 2, …, K, and i+.j, s i ，s j Representing the average distance from each point to the center point within the class of clusters i and j, respectively, also called cluster diameter, d ij Representative is the inter-class center point distance of cluster i and cluster j;
step D, repeating the steps A-C, and sequentially calculating the clustering condition when the clustering number K is 4 to 15; and then selecting the corresponding clustering result with the minimum DBI value as the optimal clustering scene to be reserved.
7. The method for optimizing daily schedule of regional integrated energy system according to claim 2, wherein the step 4 specifically comprises the steps of:
step 401, considering the carbon emission of the system, setting a scheduling optimization target, and setting the daily total running cost C of the system total The lowest, namely:
wherein C is F 、C M 、C G 、C CO2，tot Respectively fuel cost, equipment maintenance cost, electricity purchasing cost and CO 2 The discharge cost;
wherein C is gas Is the price of natural gas, τ i Is i equipment maintenance coefficient, C B Is the electricity price of the power grid, C CO2 Is CO 2 Discharge trade price, E CO2 Is system CO 2 Discharge amount;
step 402, setting a power grid purchase quantity compromise scheme, wherein the purchase quantity at each moment is determined by the following formula:
wherein,the power purchasing power at the time t in the day-ahead power grid dispatching scheme is obtained; k (K) best K=2, 3, K, for a typical scene number best ；p k Probability of occurrence for the kth typical scene; p (P) tG,k The actual power purchasing power of the power grid at the time t in the kth typical scene;
step 403, setting energy balance constraint, wherein the balance between the demand of energy and the supply at each moment in the scheduling period is maintained, including electric, thermal and cold energy balance; setting a load range constraint, wherein the load rate of each device is in the load range;
and 404, adopting a self-adaptive differential algorithm based on success history to solve the model to determine a day-ahead scheduling strategy of the regional comprehensive energy system.
8. The method for day-ahead schedule optimization of a regional integrated energy system of claim 7, wherein said step 404 comprises:
step A, initializing data; setting population scale U, decision variable dimension D and maximum evolution algebra G max The initial population is represented as a set of real parameter vectors x i ＝(x 1 ,x 2 ,···,x D )，i＝1,2,···,U；
Step B, mutation; in each generation, current-to-pbest/1 mutation strategy is adopted from an existing population vector x i,G Generates a variation vector v i,G As shown in formula (15);
v i,G ＝x i,G +F i ·(x pbest,G -x i,G )+F i ·(x r1,G -x r2,G ) (15)
wherein F is i Representing the scale factor, being the forward control parameter of the differential vector, x pbest,G Is the best individual in the G generation population, and the index r 1 ，r 2 Is a mutually exclusive integer randomly selected from a population;
step C, crossing; the variance vector v is obtained by a binomial crossing method i,G And the target vector x i,G Cross-mixing as shown in formula (16);
when the randomly generated value in the [0,1 ] range is less than or equal to the crossing rate CR i At the time, u j,i,G Assigning as element v in a variant vector j,i,G ；j rand Is one in [1, D]Decision variable indexes generated randomly in the range;
step D, generating all the testsTest vector u i,G After that, by comparing the test vectors u i,G And a target vector x i,G Selecting and generating a next generation population;
step E, calculating population fitness: judging whether the evolution reaches the maximum evolution algebra G max If yes, stopping evolution and outputting; if not, go to step B and continue the iteration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311073076.2A CN117077960A (en) | 2023-08-24 | 2023-08-24 | Day-ahead scheduling optimization method for regional comprehensive energy system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311073076.2A CN117077960A (en) | 2023-08-24 | 2023-08-24 | Day-ahead scheduling optimization method for regional comprehensive energy system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117077960A true CN117077960A (en) | 2023-11-17 |
Family
ID=88709378
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311073076.2A Pending CN117077960A (en) | 2023-08-24 | 2023-08-24 | Day-ahead scheduling optimization method for regional comprehensive energy system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117077960A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117353359A (en) * | 2023-12-05 | 2024-01-05 | 国网浙江省电力有限公司宁波供电公司 | Battery combined energy storage and power supply method and system |
| CN117713176A (en) * | 2024-02-06 | 2024-03-15 | 内蒙古科技大学 | Source network charge storage low-carbon operation method and device, electronic equipment and storage medium |
-
2023
- 2023-08-24 CN CN202311073076.2A patent/CN117077960A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117353359A (en) * | 2023-12-05 | 2024-01-05 | 国网浙江省电力有限公司宁波供电公司 | Battery combined energy storage and power supply method and system |
| CN117353359B (en) * | 2023-12-05 | 2024-04-12 | 国网浙江省电力有限公司宁波供电公司 | Battery combined energy storage and power supply method and system |
| CN117713176A (en) * | 2024-02-06 | 2024-03-15 | 内蒙古科技大学 | Source network charge storage low-carbon operation method and device, electronic equipment and storage medium |
| CN117713176B (en) * | 2024-02-06 | 2024-05-03 | 内蒙古科技大学 | Source network charge storage low-carbon operation method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Short-term wind power forecasting based on support vector machine with improved dragonfly algorithm | |
| Pu et al. | Optimal sizing for an integrated energy system considering degradation and seasonal hydrogen storage | |
| CN108206543B (en) | Energy router based on energy cascade utilization and operation optimization method thereof | |
| CN117077960A (en) | Day-ahead scheduling optimization method for regional comprehensive energy system | |
| CN110620402B (en) | Distributed planning operation joint optimization method and system for electricity-gas hybrid system | |
| Qi et al. | Optimal configuration of concentrating solar power in multienergy power systems with an improved variational autoencoder | |
| CN111555355B (en) | Scheduling strategy and optimization method for water-light-storage combined power generation | |
| CN111754037B (en) | Long-term load hybrid prediction method for regional terminal integrated energy supply system | |
| CN104598985B (en) | Methods of electric load forecasting | |
| CN111598289A (en) | Distributed optimization method of integrated energy system considering LSTM photovoltaic output prediction | |
| Wang et al. | Two-stage stochastic planning for integrated energy systems accounting for carbon trading price uncertainty | |
| Sodsong et al. | Short-term solar PV forecasting using gated recurrent unit with a cascade model | |
| CN113592133A (en) | Energy hub optimal configuration method and system | |
| Toosi et al. | Machine learning for performance prediction in smart buildings: Photovoltaic self-consumption and life cycle cost optimization | |
| CN112072643A (en) | Light-storage system online scheduling method based on depth certainty gradient strategy | |
| Dong et al. | Day-ahead prediction of wind power based on conditional generative adversarial network | |
| Mujeeb et al. | Deep learning based carbon emissions forecasting and renewable energy's impact quantification | |
| CN112580897A (en) | Multi-energy power system optimal scheduling method based on parrot algorithm | |
| CN115409645A (en) | Comprehensive energy system energy management method based on improved deep reinforcement learning | |
| CN112734451B (en) | Green house multi-energy system based on non-cooperative game and optimization method | |
| CN115021331A (en) | Clean energy-based ground water wind-solar integrated compensation method based on contribution coefficient method | |
| Fang et al. | Energy scheduling and decision learning of combined cooling, heating and power microgrid based on deep deterministic policy gradient | |
| Wu et al. | Study on optimal dispatching strategy of regional energy microgrid | |
| CN113095534A (en) | Wind power prediction method combining ARIMA and improved Elman neural network | |
| Khaligh Fard et al. | Electricity Supply Model of Conventional Residential Buildings in Tehran with Priority on Renewable Energy Using Adaptive Fuzzy-neural Inference System |
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 |