CN111008790A - Hydropower station group power generation electric scheduling rule extraction method - Google Patents

Hydropower station group power generation electric scheduling rule extraction method Download PDF

Info

Publication number
CN111008790A
CN111008790A CN201911335948.1A CN201911335948A CN111008790A CN 111008790 A CN111008790 A CN 111008790A CN 201911335948 A CN201911335948 A CN 201911335948A CN 111008790 A CN111008790 A CN 111008790A
Authority
CN
China
Prior art keywords
community
gravity
scheduling
power generation
hydropower station
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
Application number
CN201911335948.1A
Other languages
Chinese (zh)
Inventor
莫莉
易敏
汪涛
谌沁
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huazhong University of Science and Technology
Original Assignee
Huazhong University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Huazhong University of Science and Technology filed Critical Huazhong University of Science and Technology
Priority to CN201911335948.1A priority Critical patent/CN111008790A/en
Publication of CN111008790A publication Critical patent/CN111008790A/en
Pending legal-status Critical Current

Links

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
    • G06Q10/06—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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • 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
    • G06Q10/067—Enterprise or organisation modelling
    • 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
    • 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
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
    • Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
    • 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

Abstract

The invention belongs to the field of hydropower station energy optimization, and particularly discloses a hydropower station group power generation regulation extraction method, which comprises the steps of establishing a target function with the maximum annual power generation amount of a hydropower station group as a target and a constraint condition thereof, and solving to obtain an optimal solution of output flow of each time period of each reservoir; establishing a generalized regression network by taking the input flow and the reservoir water level corresponding to each time interval of each reservoir as input and the output flow as output; and optimizing a generalized regression network smoothing factor to obtain a scheduling function by adopting a particle swarm method based on the known optimal solution of the input flow, the reservoir water level and the output flow corresponding to each time interval of each reservoir and the training error minimum objective function. According to the invention, deep learning and cascade hydropower station power generation scheduling are combined, the generalized regression network corresponding to the scheduling function is established, the cascade hydropower station power generation scheduling rule based on the community gravity particle swarm and the generalized regression network is extracted, the defect that the basic traditional particle swarm algorithm is easy to fall into local optimization is overcome, and the reliability is high.

Description

Hydropower station group power generation electric scheduling rule extraction method
Technical Field
The invention belongs to the field of hydropower station energy optimization, and particularly relates to a hydropower station group power generation scheduling rule extraction method.
Background
The optimal scheduling of the cascade hydropower station is important for the maximization of the basin benefit, however, due to the uncertainty of the runoff, particularly the poor flood season runoff forecasting precision, the scheduling scheme made on the basis of the inflow certainty assumption is difficult to popularize and apply in the actual production environment. The scheduling rule is an important mode for scheduling the water guide reservoir, and the method outputs decision variables such as flow out of the reservoir, water level at the end of a time period and the like by inputting main decision factors such as forecast runoff and water level and has strong operability. Generally, a scheduling rule is extracted in a hidden random scheduling mode, and a basic idea is to extract the scheduling rule by taking a deterministic scheduling process as a scheme set, so that the theory is complete and the implementation is convenient.
In the traditional research method, machine learning algorithms such as a support vector machine, a random forest, an RBF neural network and the like are introduced into scheduling rule extraction, most of training sets are directly divided into a test set and a training set, then the training sets participate in training, the test set is used for testing, and the importance of a verification set is ignored, so that the model has weak adaptability to new data, and overfitting is easily caused; secondly, the runoff sequence input by the model has time characteristics with different scales, and the existing research does not give full consideration; finally, as the neural network model generally has a large number of parameters to be debugged, the reliability of manual selection according to experience is questioned.
Disclosure of Invention
The invention provides a hydropower station group power generation dispatching rule extraction method, which is used for solving the technical problem that the reliability of an extracted dispatching rule is poor due to overfitting caused by neglecting a verification set and complicated parameter adjustment in neural network training in the conventional hydropower station group power generation dispatching rule extraction method.
The technical scheme for solving the technical problems is as follows: a hydropower station group power generation scheduling rule extraction method comprises the following steps:
establishing an objective function with the maximum annual energy production of the hydropower station group as a target and a constraint condition thereof, and solving the objective function to obtain an optimal solution of the output flow of each reservoir in each period;
establishing a generalized regression network as a dispatching function model by taking the input flow and the reservoir water level corresponding to each period of each reservoir as input and output flows as output;
and optimizing a smoothing factor of the generalized regression network by adopting a swarm optimization method based on the known input flow and reservoir water level corresponding to each time interval of each reservoir and the output flow optimal solution and taking the minimum training error as a target to obtain a scheduling function and finish the extraction of the scheduling rule.
The invention has the beneficial effects that: the invention is based on the current artificial intelligence hotspot technology, combines deep learning with cascade hydropower station power generation scheduling, establishes a generalized regression network corresponding to a scheduling function, determines a hydropower station scheduling function by combining an optimal solution of output flow determined by a target function and a constraint condition based on a community gravity-center particle swarm and by fast iterative optimization of generalized regression network parameters, extracts a cascade hydropower station power generation scheduling rule based on the community gravity-center particle swarm and a generalized regression network (CPSO-GRNN), overcomes the defect that the basic traditional particle swarm algorithm is easy to fall into local optimization, has high reliability, accelerates the optimization and training of generalized regression network parameters, can still maintain higher precision under the condition of insufficient sample sets, and can provide decision support for the extraction of the large and medium cascade hydropower station power generation scheduling rule.
On the basis of the technical scheme, the invention can be further improved as follows.
Further, the solving of the objective function specifically includes: and solving the objective function by adopting a POA algorithm.
The invention has the further beneficial effects that: the medium-and-long-term power generation scheduling model of the hydropower station group has the characteristics of high dimension, nonlinearity and the like, and because the traditional dynamic programming is easy to fall into dimension disaster when solving the problems, a heuristic optimization algorithm has certain randomness and cannot ensure convergence to a global optimal solution. Therefore, the POA algorithm between the two is selected for model calculation, the multi-stage decision problem is converted into a plurality of two-stage optimization problems through the POA algorithm, time and space complexity is lower compared with DP, and the POA algorithm can be quickly and accurately converged to a global optimal solution.
Further, the method further comprises:
and evaluating the scheduling function by adopting a cross verification method so as to adjust the scheduling function.
Further, the method further comprises:
and evaluating the fitting degree of the deterministic optimal scheduling corresponding to the optimal solution of the output flow by the scheduling function so as to adjust the scheduling rule, wherein the fitting degree comprises the fitting degree of the output flow, the calculation speed and the generalization performance.
The invention has the further beneficial effects that: after the scheduling function is obtained, various evaluations are further performed on the scheduling function, so that the reliability of the scheduling function is further improved.
Further, optimizing a smoothing factor of the generalized regression network by using a particle swarm optimization method based on the center of gravity of the community specifically comprises:
s1, randomly generating values of initial smoothing factors, and constructing a plurality of smoothing factor communities;
s2, obtaining an optimal smoothing factor of each community cut to the current iteration based on the training error minimum objective function;
and S3, determining the center of gravity of the community based on the optimal training error value corresponding to the current iteration of each community, and taking the center of gravity of the community as the global optimal position of each community for the next iteration until the training error calculated based on the center of gravity of the community reaches a preset value, wherein the center of gravity of the community is the smoothing factor of the generalized regression network.
The invention has the further beneficial effects that: the particle swarm optimization (CPSO) based on the center of gravity of the community is obtained by improving the Particle Swarm Optimization (PSO) by introducing concepts of community and community center of gravity. The original particle swarm algorithm utilizes the positions of 'individual optimum' and 'global optimum' to update the speed and the position of the original particle swarm algorithm by a certain mechanism, so that a better solution is obtained. And the CPSO algorithm divides a population into a plurality of communities, each community is an independent particle swarm, the number of particles among the communities and optimization parameters are relatively independent, when each iteration is finished, each community elects a global optimal position of the community and shares the global optimal position with other communities, then the gravity center position of the community is calculated and is used as the global optimal position of the whole population, and each community uses the value to perform speed updating in the next iteration. Based on the method, the optimal position of each community contributes to the global optimal position of the community, so that the phenomenon that all communities evolve towards the optimal value of a certain community to lose diversity is avoided, and the reliability and the robustness of the scheduling function extraction are improved.
Further, the S4 includes:
calculating the sum of all community training errors corresponding to the current iteration, and calculating the gravity coefficient of each community by adopting the sum based on the universal gravity distance and the gravity value negative correlation characteristic;
and calculating the gravity center of the community based on the gravity coefficient of each community, replacing the optimal position of the community of a speed updating formula in the particle swarm algorithm by the gravity center of the community, and respectively iterating for the next time until the training error obtained by calculation based on the gravity center of the community reaches a preset value.
Further, the attraction coefficient of each community is expressed as:
Figure BDA0002330907330000041
wherein r isijGravitation coefficient for the jth community of the ith iteration, SiIs the sum, n is the number of colonies, CijThe annual energy production error of the jth community of the ith iteration is calculated;
the center of gravity of the community is expressed as:
Figure BDA0002330907330000042
wherein p is·Is center of gravity of the community, PijAnd (4) the optimal smoothing factor of the jth community for the ith iteration.
Further, the optimizing the smoothing factor of the generalized regression network specifically includes: and (4) computing the particle swarm algorithm based on the center of gravity of the community by adopting a parallel computing method to complete the optimization of the smoothing factor.
The invention has the further beneficial effects that: the method provides a distributed parallel hyper-parameter optimization framework to complete the optimization process, each community completes the training process on an independent server node, and the result after each iteration is shared, so that the parallelism is realized, and the hyper-parameter optimization process is accelerated.
The invention also provides a hydropower station group power generation dispatching rule extracted by adopting any one of the methods for extracting the hydropower station group power generation dispatching rule.
The invention has the beneficial effects that: the hydropower station group power generation dispatching rule extraction method is obtained by combining deep learning and cascade hydropower station power generation dispatching based on the current artificial intelligence hotspot technology, establishing a generalized regression network corresponding to a dispatching function, determining an output flow optimal solution by combining an objective function and constraint conditions based on a community gravity-center particle swarm, quickly and iteratively optimizing generalized regression network parameters, determining a hydropower station dispatching function, and extracting a cascade hydropower station power generation dispatching rule based on the community gravity-center particle swarm and the generalized regression network (CPSO-GRNN). Therefore, the scheduling rule has high reliability and strong robustness.
The invention also provides a storage medium, wherein the storage medium stores instructions, and when the instructions are read by a computer, the instructions cause the computer to execute any one of the above methods for extracting the power generation dispatching rule of the hydropower station group.
Drawings
Fig. 1 is a flow chart of a method for extracting a hydropower station group power generation scheduling rule according to an embodiment of the present invention;
fig. 2 is a process diagram of a rich water year optimal scheduling process, an open water year optimal scheduling process, and a dry water year optimal scheduling process according to an embodiment of the present invention;
FIG. 3 is a flowchart of particle swarm optimization based on the center of gravity of a community according to an embodiment of the present invention;
FIG. 4 is a comparison graph of the effect of CPSO and PSO provided by the embodiment of the present invention under different test functions;
FIG. 5 is a comparison graph of the POA and CPSO-GRNN simulation scheduling process comparison for the rich water year, the POA and CPSO-GRNN simulation scheduling process comparison for the open water year, and the POA and CPSO-GRNN simulation scheduling process comparison for the dry water year.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Example one
A method 100 for extracting power generation regulation rules of a hydropower station group, as shown in fig. 1, includes:
step 110, establishing an objective function with the maximum annual energy production of the hydropower station group as a target and constraint conditions thereof, and solving the objective function to obtain an optimal solution of the output flow of each reservoir in each period;
step 120, establishing a generalized regression network as a dispatching function model by taking the input flow and the reservoir water level corresponding to each period of each reservoir as input and output flows as output;
and step 130, optimizing a smoothing factor of the generalized regression network by adopting a community gravity center-based particle swarm optimization method based on the known optimal solution of the input flow, the reservoir water level and the output flow corresponding to each time interval of each reservoir and aiming at the minimum training error to obtain a scheduling function, and finishing the extraction of the scheduling rule.
It should be noted that, in step 110, a model is first established, in which the maximum annual power generation amount of the cascade hydropower station group is a target, and actual radial data of the qingjiang cascade hydrobelock power station in 1951 and 2007 for a total of 57 years is subjected to simulation scheduling, so that the average output at a scheduling time period meets the requirement of ensuring the output condition to measure the reliability of the model. And when the guarantee output constraint is not satisfied, adding a penalty term to constrain the target function.
The objective function is:
Figure BDA0002330907330000061
in the formula, E is annual generated energy of the cascade power station; n and T respectively represent the number of the cascade power stations and the dispatching time period; pitRepresenting the output of the ith power station in the t period; pibThe guaranteed output of the ith power station is shown, β is a penalty coefficient, and the output calculation formula is that P is 9.81 QZ.
The constraint conditions for optimizing the scheduling model are specifically as follows:
1) water level restraint:
Figure BDA0002330907330000071
2) force restraint:
Figure BDA0002330907330000072
3) and (3) flow restriction:
Figure BDA0002330907330000073
4) and (3) water balance constraint: vit+1＝Vit+(Iit-Qit) Δ t; 5) water level amplitude variation restraint:
Figure BDA0002330907330000074
in the formula (I), the compound is shown in the specification,Z it、Zitand
Figure BDA0002330907330000078
respectively representing the minimum value, the average value and the maximum value of the water level of the power station i in the t-th time period;P it、Pitand
Figure BDA0002330907330000075
representing the minimum value, the maximum value and the minimum value of output of the power station i in the time period t;Q it、Qitand
Figure BDA0002330907330000076
respectively representing the lower ex-warehouse limit, the upper ex-warehouse limit and the upper ex-warehouse limit of the power station in the time period t; vitFor the storage capacity of the ith station during the t-th period, Vit+1The storage capacity in the t +1 time period; i isitAnd QitRespectively the warehousing flow and the discharging flow of the ith power station in the t-th time period; delta ZitWater level variation, delta, of the ith station during the t-th periodZ itAnd
Figure BDA0002330907330000077
respectively representing the lower limit and the upper limit of the water level amplitude.
The method comprises the steps of 120, firstly, determining input factors of a scheduling function, selecting face time period warehousing flow and reservoir water level, selecting output factors, selecting lower discharge flow of the face time period, considering that the total conditions of abundance, average and withered water (water abundance year, average water year and dry water year) of annual water quantity and the influence of flood and non-flood periods on incoming water, but the influence of the total conditions of abundance, average and withered water of the annual water quantity on scheduling decisions is large, meanwhile, aiming at the incoming water and the water level of the same magnitude in the year, the decision is different in different periods, namely, the response of flood and non-flood periods to the incoming water is different, therefore, considering the influence of the annual water factor and the scheduling period factor on the scheduling decisions, taking the annual water frequency as an index of the abundance of the incoming water, using a ten-day time period number as a time mark, establishing a scheduling rule extraction model considering warehousing runoff, the water level, the annual water index and the scheduling period index, namely, establishing the scheduling function expressed as O, L, lambda 6324, in an equation, expressing the warehousing runoff factor and the flow factor and the face of the existing warehouse entry time period, namely, and the increase of the flow factor of the method, and the existing scheduling factors of the annual water.
The result of the optimized scheduling model is the optimal water level value at the end of each time interval year by year in the scheduling period, and a long series of sample sets are generated through water power calculation (namely, through a water balance formula and a water power calculation equation). Fig. 2 shows typical scheduling results under different water conditions of rich (upper panel, 1998), flat (middle panel, 1999), and dry (lower panel, 2005).
The method is based on the current artificial intelligence hotspot technology, deep learning is combined with cascade hydropower station power generation scheduling, a generalized regression network corresponding to a scheduling function is established, the output flow optimal solution determined by combining a target function and a constraint condition based on a community gravity-center particle swarm is used for fast iterative optimization of generalized regression network parameters to determine the hydropower station scheduling function, a cascade hydropower station power generation scheduling rule based on the community gravity-center particle swarm and a generalized regression network (CPSO-GRNN) is extracted, the defect that a basic traditional particle swarm algorithm is easy to fall into local optimization is overcome, the reliability is high, the generalized regression network parameter optimization and training are accelerated, higher precision can be kept under the condition of sample set deficiency, and decision support can be provided for extracting the large and medium cascade hydropower station power generation scheduling rule.
Preferably, the solving objective function is specifically: and calculating by using a POA algorithm.
And performing simulation optimization scheduling on the historical long series runoff data by using a Progressive Optimization Algorithm (POA) to obtain a sample. The medium-and-long-term power generation scheduling model of the cascade hydropower station has the characteristics of high dimension, nonlinearity and the like, and because the problem is easy to fall into a dimension disaster when the traditional dynamic programming is used for solving the problem, a heuristic optimization algorithm has certain randomness and can not be guaranteed to converge to a global optimal solution. Therefore, the POA algorithm between the two is selected for model calculation, the multi-stage decision problem is converted into a plurality of two-stage optimization problems through the POA algorithm, time and space complexity is lower compared with DP, and the POA algorithm can converge on a global optimal solution.
Specifically, the warehousing flow of the history long series 1951-2007 of the Qingjiang cascade water distribution puerperium power station is used as input, the scheduling period takes a water conservation year as a unit, the monthly mean 376m is used as the initial and final water levels of the scheduling period, the water level constraint interval is [350, 400], and the solving step of the POA algorithm is as follows:
1) in a feasible region (preset according to a constraint condition) of the independent variable water level, generating a group of solutions by using a dynamic programming algorithm under the constraint condition, namely determining the water level of each time interval of each reservoir as an initial track (a scheduling period comprises a plurality of time intervals, and each reservoir corresponds to a track formed by the water levels of each time interval);
2) fixing the water level Z of each reservoirtAnd Zt+2(t is t-th period), and based on the objective function, the water level Z is optimized (two-stage method)t+1Then at water level Zt+1Restarting two stages for starting point (Z)tAnd Zt+2) Optimizing until the whole T period (scheduling period) is traversed;
3) and (4) taking the track of the iteration as the initial track of the next iteration (repeating the step 2)) until the iterated track (water level) is not changed, and obtaining the optimal solution. It should be noted that the optimal solution is the water level of each reservoir in each period. And further calculating according to the water level to obtain the output flow.
Preferably, the method further comprises:
evaluating a scheduling function by adopting a cross verification method so as to adjust the scheduling function;
alternatively, the first and second electrodes may be,
and evaluating the fitting degree of the deterministic optimization scheduling corresponding to the scheduling function and the output flow optimal solution to adjust the scheduling rule, wherein the fitting degree comprises the fitting degree of the power generation, the calculation speed and the generalization performance.
When the Cross validation is applied, a sample space is divided into a training set and a testing set, then a part of the training set is divided into the validation set, most samples in the training set are used for model training, and a small part of the samples are used for model validation.
There are three common methods for cross-validation, Holdout, K-fold, and Leave-one-out. The method adopts the Leave-one-out method, and has the advantages that the method still has good effect on the condition of small sample amount, most samples participate in training during model training, and the sample utilization rate is highest. When using Leave-one-out, one item is taken out of the training set for validation, and most of the remaining samples are used for model training, looping through the process and ensuring that eventually each sample is over-trained.
Another way of evaluating this is to evaluate its fitness and check if it meets economic indicators as well as operational indicators. Evaluating the fitting degree of the scheduling rule extracted based on the GRNN neural network and deterministic optimal scheduling, wherein the fitting degree is evaluated, and whether the fitting degree meets economic benefit indexes and operation indexes is checked, and the method mainly comprises the following aspects:
1) index of generated energy
The training set adopted by the model is a deterministic optimization process with the maximum generated energy as a target, so that the generated energy of the power station in a scheduling period measures the rule fitting degree of the model to the optimized scheduling.
2) Calculating a velocity index
One of the greatest advantages of the GRNN neural network model is its computational speed, which is therefore an indicator of comparison with other models.
3) Model fitting degree index
To quantitatively evaluate the generalization ability of the model, the present invention employs Mean Absolute Error (MAE), Mean Absolute Percent Error (MAPE), mean square error (RMSE), and coefficient of membership (R)2) Describing the generalization performance index of the regression model, which is shown as the following formula; wherein the MAE measures the difference between the predicted value and the measured value; MAPE considers the relative error between the predicted value and the actual value; RMSE describes the loss of regression; r2The goodness of fit of the model is measured.
Wherein the content of the first and second substances,
Figure BDA0002330907330000101
Figure BDA0002330907330000102
in the formula, yiThe power generation amount actual value is represented,
Figure BDA0002330907330000103
is used as a power generation amount predicted value,
Figure BDA0002330907330000104
represents the average value of the power generation amount, and i is the data sample number.
Preferably, the above-mentioned optimization of the smoothing factor of the generalized regression network by using the particle swarm optimization based on the center of gravity of the colony specifically includes:
step 131, randomly generating values of initial smoothing factors, and constructing a plurality of smoothing factor communities;
step 132, obtaining an optimal smoothing factor of each community cut to the current iteration based on the training error minimum objective function;
and step 133, determining a center of gravity of the community based on an optimal training error value corresponding to the current iteration of each community, and using the center of gravity of the community as a global optimal position of each community for the next iteration until a training error calculated based on the center of gravity of the community reaches a preset value, where the center of gravity of the community is a smoothing factor of the generalized regression network.
For example, discretizing the smoothing factor in the value range, and constructing a plurality of smoothing factor communities; optimizing each community by respectively adopting a particle swarm optimization method to obtain an optimal smooth factor of the current iteration of the community; based on the optimal smooth factor of each community and the known input flow and reservoir water level corresponding to each time interval of each reservoir, calculating to obtain the optimal annual energy production corresponding to the community by adopting a generalized regression function and an objective function in turn, and based on the optimal solution of the output flow, obtaining the optimal annual energy production to calculate the annual energy production error of each community after the current iteration; and determining the center of gravity of the community based on all annual power generation errors corresponding to the current iteration, and taking the center of gravity of the community as the global optimal position of each community for the next iteration until the annual power generation errors obtained by calculation based on the center of gravity of the community reach a preset value, wherein the center of gravity of the community is a smoothing factor of the generalized regression network.
Preferably, step 134 includes:
calculating the sum of all community training errors corresponding to the current iteration, and calculating the gravity coefficient of each community by adopting the sum based on the universal gravity distance and the gravity value negative correlation characteristic;
and calculating the gravity center of the community based on the gravity coefficient of each community, replacing the optimal position of the community of a speed updating formula in the particle swarm algorithm by the gravity center of the community, and respectively iterating for the next time until the training error obtained based on the gravity center calculation of the community reaches a preset value.
Preferably, the attraction coefficient of each community is expressed as:
Figure BDA0002330907330000111
wherein r isijGravitation coefficient for the jth community of the ith iteration, SiIs the sum, n is the number of colonies, CijThe annual energy production error of the jth community of the ith iteration is calculated;
the center of gravity of the community is expressed as:
Figure BDA0002330907330000112
wherein p is·Is center of gravity of the community, PijThe optimal position of the jth community for the ith iteration.
Utilizing a CPSO algorithm to optimize a smoothing coefficient vector of GRNN (generalized regression network, namely a scheduling function O ═ f (I, L; lambda, β, sigma), wherein I and L are respectively matrixes formed by each time interval of each reservoir, and output flow O is in a vector form),
Figure BDA0002330907330000121
and n is the characteristic number (namely the number of the hydropower stations) input by the model.
Due to the fact that
Figure BDA0002330907330000122
The solution space is a continuous real number domain, and a continuous value optimization algorithm is used, so the invention provides an improved PSO algorithm to optimize the coefficient vector. Different smoothing factors are tried through traversal of a CPSO algorithm in a value range (real number interval) of the smoothing factor, a detection value of output flow is calculated based on a target function, I and L, and optimization is performed by taking the minimum training error of the GRNN as a target (the optimal solution obtained by POA solution is taken as a reference, the smoothing factor corresponding to the minimum error is determined, and one smoothing factor corresponds to one error).
It should be noted that the CPSO algorithm is obtained by improving a Particle Swarm Optimization (PSO) by introducing the concepts of "community" and "community center of gravity". The original particle swarm algorithm utilizes the positions of 'individual optimal' and 'global optimal' to update the speed and the position of the original particle swarm algorithm by a certain mechanism, so that a better solution is obtained, and the strategy is as follows:
Figure BDA0002330907330000123
the defect of the mechanism is obvious, all particles in the population share one set of optimization parameters, and when the diversity of the population is reduced, the population is easy to fall into local optimum, wherein:
Figure BDA0002330907330000124
representing the velocity of the kth iteration of the ith particle;
Figure BDA0002330907330000125
representing the position of the kth iteration of the ith particle;
Figure BDA0002330907330000126
representing the optimal solution found by the previous k iterations of the ith particle;
Figure BDA0002330907330000127
representing the optimal position searched by the previous k times of optimization; w represents the particle inertial weight; c1Represents the individual learning factor of the particle, C2Representing a social learning factor; r is1、r2Represents [0,1 ]]The random number of (2). It should be noted that each particle represents a smoothing factor vector.
The CPSO algorithm divides a population into a plurality of communities, each community is an independent particle swarm, and the number of particles and optimization parameters among the communities are relatively independent. Firstly, calculating the sum of the optimal fitness values of all communities:
Figure BDA0002330907330000128
where i is the number of iterations and j is the population number.
Then theCalculating the gravity coefficient of each community:
Figure BDA0002330907330000131
the coefficient determines the influence of each community on the global gravity center when SiWhen the gravity coefficient is equal to 0, the gravity coefficients of the respective colonies are the same
Figure BDA0002330907330000132
Calculating the center of gravity of the community:
Figure BDA0002330907330000133
replacing the original global optimal position with the center of gravity of the community, and then transforming the speed updating strategy as follows, wherein the process is called social consciousness updating:
Figure BDA0002330907330000134
in the formula, n represents the number of colonies; cijRepresenting the global optimal annual energy production error of the jth community in the ith iteration, SiThe total error for all populations; r isijThe gravity coefficient of the jth community of the ith iteration is; pijRepresenting the optimal position of the jth community of the ith iteration; p*Indicating the center of gravity of the community.
As shown in fig. 3, when each iteration is finished, each community elects its own global optimal position and shares it with other communities, then calculates the barycentric position of the community as the global optimal position of the whole community, and each community uses the value to perform velocity update in the next iteration. Based on the method, the optimal position of each community contributes to the global optimal position of the community, and the phenomenon that all communities evolve towards the optimal value of a certain community to lose diversity is avoided.
Preferably, the smoothing factor of the generalized regression network is optimized, specifically: and (4) computing the particle swarm algorithm based on the center of gravity of the community by adopting a parallel computing method to complete the optimization of the smoothing factor.
The method provides a distributed parallel hyper-parameter optimization framework to complete the optimization process, each community completes the training process on an independent server node, and the result after each iteration is shared, so that the parallelism is realized, and the hyper-parameter optimization process is accelerated.
The method adopts a master-slave node architecture mode, a master node adopts an intelligent algorithm to be responsible for scheduling of the whole process, and a slave node is responsible for calculation. Because each sub-node operates independently and does not interfere with each other, the main node only needs to collect the calculation results after the sub-nodes finish the calculation (namely, the P is collected)ijFor calculation of center of gravity of the community). The specific operation steps are as follows:
1) the main node loads a cluster computing node list and initializes the population of an optimization algorithm (CPSO);
2) and the main node broadcasts confirmation information to all the computing nodes according to the node list, and if the node does not respond, the node is removed from the node list. When the node list is empty, namely all the nodes cannot work, prompting a user to check node information and terminating the calculation;
3) if available nodes exist, the main node starts multithreading, balances the calculation tasks of the population to all the calculation nodes, then starts thread waiting, sets timeout time and waits for all the calculation nodes to return calculation results. When the node calculation is overtime, the node calculation result is abandoned;
4) when all the nodes are calculated, if the termination condition is not met, selecting an optimization result, and entering the step 3);
5) and when the termination condition is reached, performing model training by using the optimal parameters, and outputting the trained model.
For better explanation of the present invention, the above-mentioned qingjiang cascade water buffalo power station is taken as an example to explain: programming by adopting a python language to realize the hyper-parameter optimization of the CPSO algorithm to the GRNN model; the method specifically comprises the following substeps;
1) three typical years of peaceful wither in 1998, 1999 and 2005 are extracted as test sets, and the rest 54 years are used as training sets to train GRNN models;
2) the improved performance of the CPSO algorithm is verified by using several typical test functions of rastrigin, ackley, griewangk, sphere and the like, and as can be seen from the following figure 4, the CPSO algorithm has higher convergence speed, higher convergence precision and more excellent performance compared with the PSO algorithm, and can be used for super-parameter optimization;
3) and traversing in the value range of the smoothing factor through a CPSO algorithm, trying different smoothing factor combinations, and optimizing by taking the minimum training error of the GRNN as a target. In order to improve the generalization capability of the model, a Cross-validation (Cross-validation) mode is introduced to evaluate the performance of the model, so that the model is improved according to the Cross-validation mode, and the model has good adaptability to the input except the sample.
In order to further verify the index conditions of the models, the basic GRNN network, the BP neural network and the RBF neural network are adopted to extract the scheduling rules of the same data set, and the relevant calculation parameters of the models are shown in the following table 1.
TABLE 1 calculation parameters associated with each model
Figure BDA0002330907330000151
As can be seen from table 2, in terms of power generation, in different ages, the power generation amount generated by the model provided by the invention and the conventional models such as GRNN, BP and RBF models is less than the result of POA optimization scheduling, but compared with the conventional models, the power generation benefit of the model provided by the invention is closer to the optimal benefit; in terms of calculation time, GRNN only uses 5s because of the smoothness factor, and CPSO-GRNN uses 30s more because of one-step hyper-parameter optimization. However, compared with the BP and RBF neural networks, the time consumption is less, because the BP is a globally updated neural network, the weight of the whole network is updated by all the sample inputs, and the calculation time is greatly increased. Although the RBF neural network adopts a local updating mode, a lot of time is still spent on training the weight.
TABLE 2 comparison table of CPSO-GRNN and conventional model generated energy and calculation time index
Figure BDA0002330907330000152
Tables 3-5 show various fitting indexes of the model provided by the invention and the conventional model under different water years, and it is easy to see that the indexes of the model provided by the invention are similar to those of the GRNN model, but are partially improved, which shows the beneficial effect of the super-parameter optimization on parameter optimization; the BP and RBF models have smaller data set quantity, so that the fitting precision is influenced. Comparing different year shares, wherein the three fitting errors of the dry year are all minimum values, and the error of the rich year is maximum; secondly, the goodness of fit of the full water year is higher and is 0.90.
TABLE 3 comparison table of the indexes of the CPSO-GRNN and the conventional model for the year of the
Figure BDA0002330907330000161
TABLE 4 comparison table of CPSO-GRNN and conventional model horizontal year fitting indexes
Figure BDA0002330907330000162
TABLE 5 comparison table of dry year and water year fitting indexes of CPSO-GRNN and conventional model
Figure BDA0002330907330000163
In order to further verify the practicability of the model, the extracted scheduling rule is used for simulation scheduling. And similarly, incoming water of the representative year of rich, moderate and dry is used as input, 376m is used as starting water level, the initial water level of the faced time interval and the runoff in a warehouse are input, then the flow out of the warehouse is forecasted by a dispatching rule, the end water level of the time interval is calculated according to the flow out of the warehouse and is used as the starting water level of the next time interval, and the process is repeatedly operated until the whole dispatching period is completely calculated.
The scheduling rules simulate scheduling, and because the GRNN neural network is essentially a nonlinear regression process, constraints cannot be added inside the model. The output value may differ significantly from the actual value, subject to the training set length and fluctuations. Therefore, the relative error of each ten days is counted, and when the number proportion of the ten days with the relative error smaller than 25% is not less than 80%, the model precision is considered to meet the requirement. And simultaneously, carrying out constraint correction on the ten days with the relative error not meeting the requirement. The result of the scheduling simulation is shown in FIG. 5, the upper graph in FIG. 5 is a comparison graph of the simulated scheduling processes of POA and CPSO-GRNN in Rich water years (1998); the middle graph is a comparison (1999) graph of the POA and CPSO-GRNN simulation scheduling process in the horizontal year; the lower graph is a comparison (2005) graph of the POA and CPSO-GRNN simulation scheduling processes in the dry water, and the simulation yield in different years is shown in Table 6. According to the simulation scheduling result, in the rich water year, the simulation ex-warehouse yield of the provided model is 75%, which is slightly lower than the precision standard, but the ex-warehouse flow and water level process are basically consistent with the trend of the POA scheduling result, so that the constraint correction is needed. The ex-warehouse pass rate of open water years and dry water years is high, but the simulation precision of ex-warehouse in flood season is poor, so that the accumulative effect of the water level is large, and therefore the ex-warehouse simulation result in the flood season needs to be corrected to reduce the accumulative error, so that the simulation result is more in line with the engineering practice. In conclusion, when the scheduling rule model is used in open water years and dry water years, the flood season needs to be properly corrected according to the actual scheduling condition, and other seasons are basically consistent with the optimal scheduling, so that the reference is high.
TABLE 6 comparison table of dry year and water year fitting indexes of CPSO-GRNN and conventional model
Figure BDA0002330907330000171
Example two
A hydropower station group power generation dispatching rule is extracted by any one of the hydropower station group power generation dispatching rule extraction methods in the embodiment I.
The hydropower station group power generation scheduling rule extraction method is obtained by combining deep learning and cascade hydropower station power generation scheduling based on the current artificial intelligence hotspot technology, a generalized regression network corresponding to a scheduling function is established, the optimal solution of output flow determined by combining a target function and constraint conditions based on a community gravity-center particle swarm is used for fast iterative optimization of generalized regression network parameters, a hydropower station scheduling function is determined, and cascade hydropower station power generation scheduling rules based on the community gravity-center particle swarm and the generalized regression network (CPSO-GRNN) are extracted. Therefore, the scheduling rule has high reliability and strong robustness.
The related technical solution is the same as the first embodiment, and is not described herein again.
EXAMPLE III
A storage medium having instructions stored therein, which when read by a computer, cause the computer to execute any one of the hydropower station group power generation scheduling rule extraction methods according to the above embodiments.
The related technical solution is the same as the first embodiment, and is not described herein again.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (10)

1. A hydropower station group power generation scheduling rule extraction method is characterized by comprising the following steps:
establishing an objective function with the maximum annual energy production of the hydropower station group as a target and a constraint condition thereof, and solving the objective function to obtain an optimal solution of the output flow of each reservoir in each period;
establishing a generalized regression network as a dispatching function model by taking the input flow and the reservoir water level corresponding to each period of each reservoir as input and output flows as output;
and optimizing a smoothing factor of the generalized regression network by adopting a swarm optimization method based on the known input flow and reservoir water level corresponding to each time interval of each reservoir and the output flow optimal solution and taking the minimum training error as a target to obtain a scheduling function and finish the extraction of the scheduling rule.
2. The method according to claim 1, wherein the solving of the objective function is specifically: and solving the objective function by adopting a POA algorithm.
3. The method of claim 1, wherein the method further comprises:
and evaluating the scheduling function by adopting a cross verification method so as to adjust the scheduling function.
4. The method of claim 1, wherein the method further comprises:
and evaluating the fitting degree of the deterministic optimal scheduling corresponding to the optimal solution of the output flow by the scheduling function so as to adjust the scheduling rule, wherein the fitting degree comprises the fitting degree of the output flow, the calculation speed and the generalization performance.
5. The method for extracting the hydropower station group power generation regulation rule according to any one of claims 1 to 4, wherein the smoothing factor of the generalized regression network is optimized by adopting a particle swarm optimization method based on the center of gravity of the community, and specifically comprises the following steps:
s1, randomly generating values of initial smoothing factors, and constructing a plurality of smoothing factor communities;
s2, obtaining an optimal smoothing factor of each community cut to the current iteration based on the training error minimum objective function;
and S3, determining the center of gravity of the community based on the optimal training error value corresponding to the current iteration of each community, and taking the center of gravity of the community as the global optimal position of each community for the next iteration until the training error calculated based on the center of gravity of the community reaches a preset value, wherein the center of gravity of the community is the smoothing factor of the generalized regression network.
6. The method for advancing regulation rules of hydropower station group power generation according to claim 5, wherein the S3 comprises:
calculating the sum of all community training errors corresponding to the current iteration, and calculating the gravity coefficient of each community by adopting the sum based on the universal gravity distance and the gravity value negative correlation characteristic;
and calculating the gravity center of the community based on the gravity coefficient of each community, replacing the optimal position of the community of a speed updating formula in the particle swarm algorithm by the gravity center of the community, and respectively iterating for the next time until the training error obtained by calculation based on the gravity center of the community reaches a preset value.
7. The method according to claim 6, wherein the gravitational coefficient of each community is expressed as:
Figure FDA0002330907320000021
wherein r isijGravitation coefficient for the jth community of the ith iteration, SiIs the sum, n is the number of colonies, CijThe annual energy production error of the jth community of the ith iteration is calculated;
the center of gravity of the community is expressed as:
Figure FDA0002330907320000022
wherein p is*Is center of gravity of the community, PijAnd (4) the optimal smoothing factor of the jth community for the ith iteration.
8. The method for extracting the hydropower station group power generation scheduling rule according to any one of claims 1 to 4, wherein the optimizing the smoothing factor of the generalized regression network specifically comprises: and (4) computing the particle swarm algorithm based on the center of gravity of the community by adopting a parallel computing method to complete the optimization of the smoothing factor.
9. A hydropower station group power generation scheduling rule, which is extracted by the method for extracting the hydropower station group power generation scheduling rule according to any one of claims 1 to 8.
10. A storage medium having stored therein instructions which, when read by a computer, cause the computer to carry out a method of extracting a hydropower station group power generation schedule according to any one of claims 1 to 8.
CN201911335948.1A 2019-12-23 2019-12-23 Hydropower station group power generation electric scheduling rule extraction method Pending CN111008790A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911335948.1A CN111008790A (en) 2019-12-23 2019-12-23 Hydropower station group power generation electric scheduling rule extraction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911335948.1A CN111008790A (en) 2019-12-23 2019-12-23 Hydropower station group power generation electric scheduling rule extraction method

Publications (1)

Publication Number Publication Date
CN111008790A true CN111008790A (en) 2020-04-14

Family

ID=70117205

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911335948.1A Pending CN111008790A (en) 2019-12-23 2019-12-23 Hydropower station group power generation electric scheduling rule extraction method

Country Status (1)

Country Link
CN (1) CN111008790A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113326632A (en) * 2021-06-19 2021-08-31 南昌工程学院 Optimization correction method for backward-thrust reservoir warehousing flow process
CN113393119A (en) * 2021-06-11 2021-09-14 河海大学 Stepped hydropower short-term scheduling decision method based on scene reduction-deep learning
CN113419425A (en) * 2021-07-07 2021-09-21 哈尔滨理工大学 Intelligent regulation and control method for high-speed electric spindle water cooling system
CN113887154A (en) * 2021-10-21 2022-01-04 中国电建集团成都勘测设计研究院有限公司 Method for evaluating medium-and-long-term power generation capacity of cascade hydropower station group

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103942434A (en) * 2014-04-23 2014-07-23 天津大学 Hydropower station plant dam structure vibration response predication method based on SSPSO-GRNN
CN105138729A (en) * 2015-07-24 2015-12-09 南京信息工程大学 Filling method based on PSO-GRNN (Particle Swarm Optimization-Generalized Regression Neural Network) for defect wind speed values of wind turbines in wind power plant
CN106295801A (en) * 2016-08-01 2017-01-04 安徽农业大学 A kind of Folium Camelliae sinensis based on fruit bat algorithm optimization generalized regression nerve networks algorithm stores method chronological classification
CN108647829A (en) * 2018-05-16 2018-10-12 河海大学 A kind of Hydropower Stations combined dispatching Rules extraction method based on random forest
US20190251458A1 (en) * 2018-02-15 2019-08-15 Institute For Development And Research In Banking Technology System and method for particle swarm optimization and quantile regression based rule mining for regression techniques

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103942434A (en) * 2014-04-23 2014-07-23 天津大学 Hydropower station plant dam structure vibration response predication method based on SSPSO-GRNN
CN105138729A (en) * 2015-07-24 2015-12-09 南京信息工程大学 Filling method based on PSO-GRNN (Particle Swarm Optimization-Generalized Regression Neural Network) for defect wind speed values of wind turbines in wind power plant
CN106295801A (en) * 2016-08-01 2017-01-04 安徽农业大学 A kind of Folium Camelliae sinensis based on fruit bat algorithm optimization generalized regression nerve networks algorithm stores method chronological classification
US20190251458A1 (en) * 2018-02-15 2019-08-15 Institute For Development And Research In Banking Technology System and method for particle swarm optimization and quantile regression based rule mining for regression techniques
CN108647829A (en) * 2018-05-16 2018-10-12 河海大学 A kind of Hydropower Stations combined dispatching Rules extraction method based on random forest

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
JIEJIN CAI: "Predicting the critical heat flux in concentric-tube open", 《HEAT AND MASS TRANSFER》 *
宁必锋: "基于重心法的粒子群优化算法", 《电子世界》 *
舒生茂: "基于深度学习的梯级水电站发电调度及系统设计研究", 《万方数据》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113393119A (en) * 2021-06-11 2021-09-14 河海大学 Stepped hydropower short-term scheduling decision method based on scene reduction-deep learning
CN113393119B (en) * 2021-06-11 2022-08-30 河海大学 Stepped hydropower short-term scheduling decision method based on scene reduction-deep learning
CN113326632A (en) * 2021-06-19 2021-08-31 南昌工程学院 Optimization correction method for backward-thrust reservoir warehousing flow process
CN113326632B (en) * 2021-06-19 2022-09-23 南昌工程学院 Optimization correction method for backward-thrust reservoir warehousing flow process
CN113419425A (en) * 2021-07-07 2021-09-21 哈尔滨理工大学 Intelligent regulation and control method for high-speed electric spindle water cooling system
CN113887154A (en) * 2021-10-21 2022-01-04 中国电建集团成都勘测设计研究院有限公司 Method for evaluating medium-and-long-term power generation capacity of cascade hydropower station group
CN113887154B (en) * 2021-10-21 2023-05-26 中国电建集团成都勘测设计研究院有限公司 Assessment method for medium-long term power generation capacity of cascade hydropower station group

Similar Documents

Publication Publication Date Title
CN111783953B (en) 24-point power load value 7-day prediction method based on optimized LSTM network
Assareh et al. Application of PSO (particle swarm optimization) and GA (genetic algorithm) techniques on demand estimation of oil in Iran
CN111008790A (en) Hydropower station group power generation electric scheduling rule extraction method
CN108734355A (en) A kind of short-term electric load method of parallel prediction and system applied to power quality harnessed synthetically scene
CN108876001B (en) Short-term power load prediction method based on twin support vector machine
CN104636801A (en) Transmission line audible noise prediction method based on BP neural network optimization
CN109214449A (en) A kind of electric grid investment needing forecasting method
CN107133695A (en) A kind of wind power forecasting method and system
CN110837915B (en) Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning
CN111861013A (en) Power load prediction method and device
CN108171379A (en) A kind of electro-load forecast method
CN114362175B (en) Wind power prediction method and system based on depth certainty strategy gradient algorithm
CN109858665A (en) Photovoltaic short term power prediction technique based on Feature Selection and ANFIS-PSO
CN109934422A (en) Neural network wind speed prediction method based on time series data analysis
Akpinar et al. Forecasting natural gas consumption with hybrid neural networks—Artificial bee colony
CN107657349B (en) Method for extracting scheduling rules of staged power generation of reservoir
CN114119273A (en) Park comprehensive energy system non-invasive load decomposition method and system
CN112686693A (en) Method, system, equipment and storage medium for predicting marginal electricity price of electric power spot market
Xu et al. Short-term electricity consumption forecasting method for residential users based on cluster classification and backpropagation neural network
CN116415732A (en) User side power load data processing method based on improved ARNN
CN115345297A (en) Platform area sample generation method and system based on generation countermeasure network
CN113240181B (en) Rolling simulation method and device for reservoir dispatching operation
Liu et al. Short-Term Power Load Forecasting via Recurrent Neural Network with Similar Day Selection
CN114357865A (en) Hydropower station runoff and associated source load power year scene simulation and prediction method thereof
CN107330302A (en) The biological die body reconstructing method of joint Memetic algorithms and S 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
RJ01 Rejection of invention patent application after publication

Application publication date: 20200414

RJ01 Rejection of invention patent application after publication