CN104156779A - Cascade power generation flood control optimal scheduling method based on analog data field mechanism - Google Patents

Abstract

The invention discloses a cascade power generation flood control optimal scheduling method based on an analog data field mechanism. The method comprises the steps of calculating, based on a calculation method of the decision space range, a decision space range at a scheduling time d for each power station according to the scheduling water level, storage runoff, discharged flow constraint, output constraint and water level constraint of each power station at a given initial time of power generation scheduling of cascade power stations; randomly generating a decision variable initial value at the scheduling time d by the decision space range; calculating the corresponding mass value of each individual, the resultant force received by each individual from all the directions, the generated acceleration magnitude and the speed updating weight by a DFO algorithm; conducting local optimization for each individual by use of a local search algorithm; and obtaining the individual of the best quality in the group, that is the best running water level for each reservoir in this calculation of optimal scheduling. The beneficial effects of the invention are that the DFO algorithm is provided, and the cascade power generation flood control optimal scheduling method has the advantages of excellent optimization capability and high calculation speed.

Description

A kind of basin step generating Flood Optimal Scheduling method based on intending data fields mechanism

Technical field

The invention belongs to basin step generating Flood Optimal Scheduling technical field, relate to a kind of basin step generating Flood Optimal Scheduling method based on intending data fields mechanism.

Background technology

In recent years, a kind of new natural Heuristic Method: the global optimization method of analogies mechanism of science has caused scholars' extensive concern [1], at present, the physics mechanism of simulation mainly contains take the gravity field that the electromagnetic field that Coulomb law is physical basis and the law of universal gravitation of take be physical basis, the former mainly contains electromagnetism-like mechanism (EM) and improves algorithm [2-4], the latter mainly contains central force algorithm (CFO) [5,6], gravity field searching algorithm (GSA) and improvement algorithm [7,8] thereof.These methods are by the interaction of data object in problem analysis solution space, self-adaptation attraction/exclusion rule in simulation universal gravitation or electromagnetic force, from the thinking of different physics visual angle with explained emerging in large numbers of biocenose intelligence, the never mapping relations to optimization problem space with physical space have been set up, for the research of Heuristic Method provides new approaches.

Said as No Free Lunch theorem [9], do not have a kind of method can solve most effectively all optimization problem nature heuristic optimization research theoretical and method and remain a problem that has challenge, the solution of a large amount of hot issues still awaits further consistent efforts, for example: natural system is to the mapping of optimization problem; From physics law and biotic population self-organization, extract effective rule application in search strategy of optimization method etc.Therefore,, from the angle of development optimum theory and innovation optimization method, explore new optimization Self-configuring, study new high-performance optimization method and remain an open question.

The concept of the reference physics midfields such as Li Deyi has proposed data fields [10,11,12,13], for data analysis and process provides a kind of new plan physics method, has set up relatively complete theoretical system.Data fields is introduced abstract number field space by the interaction between material particle and a describing method thereof, each object is equivalent to have in space particle or the nucleon of certain mass, there is an applied field around in it, the any object that is positioned at field all will be subject to the synergy of other objects, has determined thus a data fields on whole space.Complete data fields theory provides theoretical foundation for the global optimization's method research based on data fields.

Summary of the invention

The object of the present invention is to provide a kind of basin step generating Flood Optimal Scheduling based on intending data fields mechanism, the basin step Optimized Operation optimization method that has solved existing complexity exists optimizing indifferent, the bad problem that is easily absorbed in local optimum of convergence.

Application data of the present invention field concept, each object is regarded as to particle or the nucleon in space with certain mass, there is an applied field around in it, any object that is positioned at field all will be subject to the synergy of other objects, has determined thus a data fields on whole space.The present invention proposes a kind of global optimization method of new plan physics mechanism: the global optimization method based on data fields (DFO).DFO is associated suitable applied field source strength with objective function to be optimized, problem of implementation solution space to the data field of force, the mapping of data potential field; Each feasible solution in decision space is considered as to a data particle, there is an applied field around in it, the any data particle that is positioned at field is all subject to the synergy of other data particle, in solution space, form a data field of force thus, self-organization by simulated data particle under data force field is moved, induction data particle moves towards the data particle in colony with high fitness, finally approaches globally optimal solution.First the present invention proposes the optimizing mechanism based on data fields global optimization (DFO) method, then provides the structure that realizes DFO method, finally the optimizing ability of DFO method is carried out to comprehensive evaluation analysis.

The technical solution adopted in the present invention is:

Step1: according to the scheduling water level in given each power station of initial time of step hydropower station power generation dispatching, warehouse-in runoff, letdown flow constraint, units limits, and restriction of water level, according to the computing method of decision space scope, calculate each power station in the decision space scope of scheduling moment d;

Step2: decision space scope generates the scheduling decision variable initial value of d constantly at random;

Step3: by step gross generation PE, as the fitness value f of DFO algorithm _k(t), calculate each individual corresponding mass value M _k(t), calculate the suffered accekeration of making a concerted effort and producing of individual all directions;

Step4: the weight that computing velocity is upgraded;

Step5: upgrade the decision variable of individual all directions, and do the processing of crossing the border;

Step6: call local search algorithm each individuality is carried out to local optimal searching;

Step7: return to Step3 and carry out iterative computation, until t > T;

Step8: draw the individuality of optimal quality in colony, be the operating water level of each reservoir optimum of this Optimized Operation calculating.

The DFO algorithm that the invention has the beneficial effects as follows proposition, has optimizing ability strong, the advantage that computing velocity is fast.

Accompanying drawing explanation

Fig. 1 is the comparison diagram of the method convergence processs such as DFO of the present invention;

Fig. 2 is DFO algorithm of the present invention is 100,500 and 1000 o'clock constringency performance comparison diagrams for unimodal function at decision variable dimension;

Fig. 3 be DFO algorithm of the present invention for unimodal function the constringency performance comparison diagram at different range decision space;

Fig. 4 is Flood Dispatching Optimization of the present invention and adjusts flood to calculate earial drainage process (0.5% flood in 1998) figure.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is described in detail.

1. this method solves following optimization problem:

minf(x)，s.t.x∈[x _min，x _max] (1)

Wherein, x is n dimension decision variable, x _minand x _maxbe the border of decision space, f (x) is the objective function that minimizes of problem to be optimized, and the value of f (x) is called as " fitness ".Formula 1 is to want a mathematical description of Solve problems below, and the X of follow-up formula is decision variable namely.

1.2 machine-processed with global optimizing

1.2.1 the concept of field

The concept of field is by English physicist faraday, to be proposed for 1837 the earliest, he thinks the interactional generation of the noncontact between object, as the electrostatic force between universal gravitation, electrified body and the magneticaction between magnet etc., all must could realize by the transmission of certain intermediate medium, and the interactional medium of this transmission is exactly field.Along with the development of field theory thought, people by its abstract be a mathematical concept.

Definition: if each in the Ω of space put a definite value of corresponding certain physical quantity or mathematical function, claimed to determine a field of this physical quantity or mathematical function on Ω.Consider that Gaussian function has good mathematical property and universality, we adopt nucleoid field of force potential function, and Gauss potential function is described the character of field.In hypothesis space, having N quality is M _iparticle X _i, i=1,2 ..., N, the potential function of the particle X ∈ Ω that in given space, any point quality is M is:

Wherein, n ∈ N is the number of particle in a space, and M >=0 represents the quality of particle, can be used as the fitness value of each feasible solution in decision space in optimization problem, and σ ∈ (0 ,+∞) controls the interaction distance between particle, is called factor of influence.E is nature Changshu.

According to the gradient of potential function, be the field intensity function in the corresponding field of force, in given space any point x ∈ Ω field intensity vector be:

Wherein, being one does not affect the constant that field intensity vector distributes, and therefore, F (x) can be abbreviated as

1.2.2 the global optimizing based on data fields is machine-processed:

(a) position of initialization colony and velocity;

The basic thought of the global optimization method based on data fields is that each feasible solution in decision space is considered as to a particle, by introducing virtual data field, interaction individual in decision space is described, by simulate the individual newtonian motion theorem of following under the effect of data fields, to the particle in colony with high fitness, move, finally converge to globally optimal solution.Specifically, under the effect that there is no external force, in order to accelerate individual movement velocity, individuality is only subject to certain given individual attraction collecting and move toward one another, and such as the individuality collection larger than its quality, this given individual collection is referred to as Attracting Set of individual k.Algorithm iteration operation is until meet the condition of convergence.

If individual k is designated as O at t Attracting Set constantly _k(t), position vector is x _k(t), quality is M _k(t), O _k(t) quality individual in is M _j(t), position vector is x _j(t), j=1,2 ..., | O _k(t) |, | O _k(t) | represent Attracting Set O _k(t) number of individuals in.According to formula (4), the t field force effect F that individual k is subject in field constantly _k(t) be:

$F_k\left(t\right)=M_k\left(t\right)\·\underset{j=1}{\overset{\left|O_k\left(t\right)\right|}{\mathrm{\Σ}}}((x_k\left(t\right)-x_j\left(t\right))\·M_j\left(t\right){\·e}^{-{\left(\frac{\left|\right|x_k\left(t\right)-x_j\left|\right|}{\mathrm{\σ}}\right)}^2})---\left(5\right)$

Wherein || x _k(t)-x _j(t) || represent vector x _kand x (t) _j(t) Euclidean distance.

According to newton's second law of motion, t is the transient acceleration vector a of individual k constantly _k(t) be:

$a_k\left(t\right)=\frac{F_k\left(t\right)}{M_k\left(t\right)}---\left(6\right)$

When enough hour of the time interval of calculating, we think individual and in the time period, do uniform variable motion at [t, t+1], and making individual k is v in t velocity constantly _k(t), the t+1 velocity v of individual k constantly _kand position vector x (t+1) _k(t+1) be:

v _k(t+1)＝v _k(t)+a _k(t) (7)

X _k(t+1)＝X _k(t)+v _k(t+1) (8)

1.3 intend the realization of the Global Optimization Algorithm For Analysis (DFO) of data fields:

(b) with formula (10), calculate the quality of colony;

In DFO algorithm, the position of body is corresponding to a solution of optimization problem one by one, and individual quality is by its fitness evaluation, and quality is larger, and fitness value is larger, and individual performance is just better.All individualities attract each other under field action, the individuality that quality is larger, and it attracts other individual field forces larger, and it is mobile that all individualities are made the overall situation to the larger individuality of quality under the effect of field force, searches for optimum solution in decision space.By the iteration of certain hour, the individuality in colony is attracted compared with large individuality by quality and restrains, and the individuality of quality maximum represents the optimum solution of decision space.The individuality that quality is larger, the individuality that its translational speed is little compared with quality is slow, to guarantee the global convergence performance of DFO algorithm.

The given colony that has N individuality, its decision space dimension is p, the position vector of individual k is x _k=(x _{k, 1}, x _{k, 2}..., x _{k, p}), k=1,2 ... N, its fitness value f _k(t) be that individual k is at the target function value of moment t, f _k(t) according to particular problem, by formula (1), calculated.Individual k at the Mass Calculation formula of moment t is:

$M_k\left(t\right)=\frac{1-f_k\left(t\right)}{\left|f_{\mathrm{best}}\left(t\right)\right|+\left|f_{\mathrm{worst}}\left(t\right)\right|}---\left(9\right)$

Relative mass M ｀ at the individual k of moment t in colony _k(t) by following formula, calculated:

$M_k^{\′}\left(t\right)=\frac{M_k\left(t\right)}{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{M}_i\left(t\right)}---\left(10\right)$

M wherein _i(t) be that individual i is in the quality of moment t.

For minimum optimization problem, f _bestand f (t) _worst(t) be defined as follows:

$f_{\mathrm{best}}\left(t\right)=\underset{j=\{1,...,N\}}{\min }{f}_j\left(t\right)---\left(11\right)$

$f_{\mathrm{worst}}\left(t\right)=\underset{j=\{1,...,N\}}{\max }{f}_j\left(t\right)---\left(12\right)$

For largest optimization problem, (11), (12) formula become:

$f_{\mathrm{best}}\left(t\right)=\underset{j=\{1,...,N\}}{\max }{f}_j\left(t\right)---\left(13\right)$

$f_{\mathrm{worst}}\left(t\right)=\underset{j=\{1,...,N\}}{\min }{f}_j\left(t\right)---\left(14\right)$

Wherein, f _j(t) for being the fitness value of individual j at moment t.

Likely there is negative value in the target function value of considering some optimization problem, denominator f in formula (9) _bestand f (t) _worst(t) take absolute value.

(c) with formula (15) and formula (6), calculate the suffered acceleration of making a concerted effort and producing of individual all directions; In order to allow DFO algorithm there is random character, in formula (5), calculate the make a concerted effort F of individual k in the field of force _k(t) time, increase random weight, F now makes a concerted effort _k(t) by following formula, calculated:

$F_k\left(t\right)=M_k\left(t\right)\·\underset{j=1}{\overset{\left|O_k\left(t\right)\right|}{\mathrm{\Σ}}}R(1,p)\·((x_k\left(t\right)-x_j\left(t\right))\·M_j\left(t\right){\·e}^{-{\left(\frac{\left|\right|x_k\left(t\right)-x_j\left|\right|}{\mathrm{\σ}}\right)}^2})---\left(15\right)$

Wherein, and R (1, p) be the random vector between p dimension [0,1].In order to improve convergence of algorithm speed, an effective method is O in choose reasonable (15) formula _k(t) number of individuals, therefore, DFO algorithm, when calculating individual suffered making a concerted effort, is only selected the individual collection of part that quality is large, rather than the whole individualities in colony.In order to reduce algorithm, avoid being absorbed in the possibility of local optimum, algorithm must have stronger space exploration ability at the iteration initial stage; In order to improve the search precision of algorithm, in the iteration later stage, must there is good local search ability.DFO algorithm is by controlling O _k(t) individuality carrys out the balance of the developing of the implementation algorithm overall situation and local search ability.First to the individuality in colony according to its relative mass M ' _k(t) descending sort, selects front K wherein _k(t) individuality is as O _k(t) individuality.K _k(t) computing formula is as follows:

$K_k\left(t\right)=[K_0+(1-\frac{t}{T})\·(100-K_0)]\·\frac{N_k}{100}---\left(16\right)$

Wherein, T represents greatest iteration time, N _krepresent that quality in colony is better than the number of individuals of individual k, K ₀represent the last O of iteration _k(t) individual amount in accounts for N _knumber percent.For individual k, K _k(t) from certain initial value, start to reduce with t is linear, be i.e. at the algorithm iteration initial stage, have more individuality to making a concerted effort, to have contribution, to add the overall situation developing ability of strong algorithms, along with the carrying out of iteration, only there is individual participation of minority that quality is large to make a concerted effort to calculate, to improve convergence of algorithm speed.

Except utilizing O _k(t) the individual external of calculating of making a concerted effort participated in control, uses for reference adaptive weighting PSO algorithm [10], and DFO algorithm is introduced weight in speed renewal is calculated, with the overall situation and the local search ability of balanced algorithm.The computing formula that speed in formula (7) more newly increases after weight is as follows:

$v_k(t+1)={\mathrm{\ω}}_1^k\left(t\right)\·v_k\left(t\right)+{\mathrm{\ω}}_2\left(t\right)\·a_k\left(t\right)---\left(17\right)$

Weight be used for controlling the influence degree of the historical speed of individual k to present speed, one larger value can be accelerated the new region of individual search, and therefore, it is suitable to choose the overall situation and the local search ability of value energy balance DFO algorithm, thus better separated.In DFO algorithm, for quality, be inferior to the individuality of average colony quality, a larger weight is set and induces this individuality to draw close to the good field of search; The individuality that is better than colony's average quality for quality; less weight is set; thereby protect this individuality; meanwhile, when individual quality is tending towards global optimum, its weight is less; to strengthen its local optimal searching precision; otherwise when individual quality is overstepping the bounds of propriety loose, weight is larger, strengthens this individual space exploration ability. definition suc as formula shown in (18).

${\mathrm{\&omega;}}_1^k\left(t\right)=\left\{\begin{array}{cc}{\mathrm{\&omega;}}_{\max }-({\mathrm{\&omega;}}_{\max }-{\mathrm{\&omega;}}_{\min })\&CenterDot;\frac{M_k^{\&prime;}\left(t\right)-M_{\min }\left(t\right)}{M_{\max }\left(t\right)-M_{\min }\left(t\right)}& M_k^{\&prime;}\left(t\right)\&GreaterEqual;M_{\mathrm{avg}}\left(t\right)\\ {\mathrm{\&omega;}}_{\max }& M_k^{\&prime;}\left(t\right)<M_{\mathrm{avg}}\left(t\right)\end{array}\right.---\left(18\right)$

Wherein, M _max(t) be the maximal value of relative mass in colony, m _min(t) be the minimum value of relative mass in colony, m _avg(t) be the mean value of colony's relative mass, ω _maxand ω _minfor given maximal value and minimum value.

ω ₂(t) be used for controlling acceleration proportion in speed is upgraded, ω ₂(t) value is larger, and the ability in algorithm developing new search space is stronger, otherwise the local convergence precision of algorithm is just higher.At the DFO algorithm initial stage, larger ω is set ₂(t) value, ω ₂(t) along with iterative steps reduces with exponential form.ω ₂(t) be calculated as follows shown in formula.

${\mathrm{\&omega;}}_2\left(t\right)={\mathrm{\&omega;}}_0\&CenterDot;e^{-\frac{\mathrm{\&alpha;}}{T}}---\left(19\right)$

Wherein, ω ₀with α be the given constant that is greater than zero, T is the algorithm greatest iteration time.

In order to improve the search performance of DFO algorithm, we identify by the self-adaptation to decision space, and guiding individuality is drawn close to optimal spatial fast; By the local search algorithm of variable boundary, improve the convergence precision in DFO algorithm iteration later stage.

The self-adaptation identification (DSR) of decision space:

Under the effect on the scene of the colony of DFO algorithm at whole decision space [x _min, x _max] middle search optimum solution, finally to optimal region, draw close, therefore, in algorithm iteration process, effectively dwindle region of search and identify optimal region and can obviously accelerate convergence of algorithm speed and precision.If the t constantly border of decision space is [x _min(t), x _max(t)], in colony, the position vector of optimum individual is x _opt(t), the t+1 border [x of decision space constantly _min(t+1), x _max(t+1) be]:

$x_{\max }(t+1)=x_{\max }\left(t\right)-\frac{x_{\max }\left(t\right)-x_{\mathrm{opt}}\left(t\right)}{\mathrm{\&eta;}\left(t\right)}---\left(20\right)$

$x_{\min }(t+1)=x_{\min }\left(t\right)-\frac{x_{\mathrm{opt}}\left(t\right)-x_{\min }\left(t\right)}{\mathrm{\&eta;}\left(t\right)}---\left(21\right)$

Wherein, η (t) is the space contraction factor, n _s(t) be the iterations that in algorithm computation process, in colony, optimum individual objective function does not improve continuously.η (t) is interior with n in [1,2] scope _s(t) index reduces, as convergence of algorithm speed, i.e. n _s(t) hour, η (t) gets larger value with convergence speedup precision, otherwise η (t) gets larger value so that algorithm, in larger space search optimum solution, improves the ability that it opens up new space.

DSR algorithm in use must be careful, otherwise algorithm is easily absorbed in local optimum.The method that we adopt is the time (such as 50 steps) of every computational rules to call DSR algorithm once, and if n _s(t) be greater than certain setting (such as 20), illustrate that algorithm is in certain local optimum region in computation process, now never call DSR algorithm, make algorithm in larger space search optimum solution, to increase the possibility of jumping out local optimum.

Local search algorithm (LS):

Heuristic Stochastic Optimization Algorithms has stronger global optimizing ability, but in the convergence precision in iteration later stage often not high [10], in conjunction with the feature of heuristic Stochastic Optimization Algorithms, designs the optimizing precision that effective local search algorithm can effectively improve algorithm.

First LS algorithm shrinks the border of local optimal searching, then carries out the Local Search of stipulated number.Suppose that individual k is x at t position vector constantly _k(t), each search comprises 2 steps: the first step is from x _k(t) with upwards search of a fixed step size, if objective function is improved, stop this search, otherwise from x _k(t) with a fixed step size, search for downwards.

If individual k at t Local Search outgoing position vector is constantly its initial value equals x _k(t).LS is implemented as follows:

1) calculate the t border [x of LS algorithm constantly _min(t), x _max(t)].Wherein, x _max(t)=x _max(t)/1.2 ', x _min(t)=x _min(t)/1.2 '.

2) carry out the Local Search of predetermined number of times, establishing search step number is l=1,2 ..., N _l, N _lfor given total searching times.Upwards search for according to the following formula l step:

$x_i\left(t\right)=x_k^{\mathrm{out}}\left(t\right)+R(1,p)\&CenterDot;\frac{x_{\max }\left(t\right)-x_k^{\mathrm{out}}\left(t\right)}{2^{(l-1)}}---\left(22\right)$

If objective function is improved, use x _k(t) upgrade and exit this Local Search, otherwise press following formula search l step downwards:

$x_i\left(t\right)=x_k^{\mathrm{out}}\left(t\right)-R(1,p)\&CenterDot;\frac{x_k^{\mathrm{out}}\left(t\right)-x_{\min }\left(t\right)}{2^{(l-1)}}---\left(23\right)$

If objective function is improved, use x _k(t) upgrade and exit this Local Search, otherwise return to (2), carry out l+1 step Local Search, until l=N _l.

The specific implementation step of DFO algorithm is as follows.

(a) position of initialization colony and velocity;

(b) by (10) formula, calculate the quality of colony;

(c) by (15) formula and formula (6) formula, calculate the suffered acceleration of making a concerted effort and producing of individual all directions;

(d) by (18) formula and formula (19) formula computing velocity, upgrade weight;

(e) by (17) formula, upgrade the speed of individual all directions, and the processing of crossing the border;

(f) by (8) formula, upgrade the position of individual all directions, and the processing of crossing the border;

(g) call local search algorithm;

(h) call decision space adaptive algorithm;

(i) return to (b) and carry out t+1 step iterative computation, until meet algorithm end condition;

(j) individuality of optimal quality in output colony.

1.4 comparative studies:

In order to evaluate the performance of nucleoid field of force mechanism optimization method, this section, by the calculating to 13 typical trial functions, is carried out analysis and comparison to the optimization method of nucleoid field of force mechanism.Trial function provides at 5.1 joints, and NSA and GSA, the comparative analysis of CFO aspect computational accuracy and speed of convergence are at 5.2 and 5.3 joints.

1.4.1 trial function:

The trial function of algorithm is from document [7], as shown in table 1, table 2 and table 3.F in table 1 ₁～f ₇for unimodal function, table 2, the f in table 3 ₈～f ₁₃, f ₁₄～f ₂₃for Solving Multimodal Function, f wherein ₁₄～f ₂₃for fixing dimension function.[x in table _min, x _max] be the border of decision variable x, f _minfor minimum of a function value.In table 1 and table 2, n is function dimension.Function f ₅, f ₁₂and f ₁₃optimal location be [1] ⁿ, f ₈be 420.9687, all the other are [0] ⁿ.The detailed description of trial function is shown in shown in appendix A.

The unimodal trial function of table 1

Table 2 multimodal trial function

Table 3 has the fixedly multimodal trial function of dimension

1.4.2DFO the relatively calculating of optimizing precision:

The relative parameters setting of DFO algorithm is: population size N is 50, and maximum iteration time T is 5000, δ=(x in formula (5) _max-x _min)/2, K in formula (16) ₀=2, ω in formula (18) _max=0.9, ω _min=0.01, ω in formula (19) ₀=100, α=20, in LS algorithm, maximum search number of times is 20.The end condition of DFO algorithm is that the iterations that optimum individual objective function does not improve continuously surpasses 200.SGA population size is also 50, and maximum iteration time is also 5000.Trial function f ₁～f ₁₃dimension p be 30.DFO and CFO, SGA independent operating 30 times, the best values (best) of optimum individual objective function in 30 independent calculating, mean value (mean), standard deviation (std.) is in Table 5, shown in table 6 and table 7.Wherein the operation result of CFO algorithm is from document [5], and CFO determines type algorithm, therefore only has an optimum results.

From the result of calculation of table 4, can find out, for trial function f ₁～f ₄and f ₇the best result of calculation of NSA algorithm in 30 times are calculated can both obtain very high computational accuracy, meanwhile, the best values that DFO algorithm calculates, mean value and standard deviation are all much better than other two kinds of algorithms, illustrate that DFO can obtain better optimizing precision and robustness for above-mentioned trial function; For f ₅, the result of DFO algorithm is slightly better than CFO and GSA, but convergence precision is not high, and the robustness of result of calculation is not high; For f ₆, 3 kinds of algorithms can obtain optimum solution.

Table 4 function f ₁-f ₇optimum results

Therefore Solving Multimodal Function has a lot of local minimum points, is used for evaluation algorithms to jump out the ability of local optimum and search global optimum, and final computational accuracy is an important evaluation index.We use function f ₈-f ₁₃as multimodal higher-dimension trial function, its local minimum point is with dimension exponent increase.From the result of calculation of table 5, can find out, for f ₉, f ₁₀and f ₁₁, NSA has obtained good convergence precision, obviously be better than other two kinds of algorithms; For f ₈, CFO is optimum; f ₁₂and f ₁₃, GSA is slightly better than other 2 two kinds of algorithms.On the whole, DFO processing Solving Multimodal Function ability needs further to be improved.

Table 5 function f ₈-f ₁₃optimum results

Multimodal low-dimensional function f ₁₄-f ₂₃comparison result of calculation be shown in Table 6, table 6 shows, DFO, SGA, CFO have obtained the result almost identical with theoretical optimum solution to all functions, therefore to multimodal low-dimensional function, above heuristic has good optimizing ability.

Table 6 function f ₁₄-f ₂₃optimum results

1.4.3 the relatively calculating of algorithm the convergence speed:

We carry out the speed of convergence of testing algorithm with unimodal function.DFO, EM, PSO and GSA are for trial function f ₁～f ₆in 30 independent calculating the convergence process of a best result of calculation as shown in Figure 1, (a) in Fig. 1 (b) (c) (d) (e) represent respectively trial function f ₁～f ₆convergence process, in figure, horizontal ordinate represents the iterative computation number of times of algorithm, ordinate represents the optimum solution that algorithm converges to when the corresponding number of times of technology repeatly.Result of calculation shows, for all trial functions, DFO convergence of algorithm speed is optimum.For f ₅, there is the problem of Premature Convergence in each algorithm.

1.4.4 the relatively calculating of algorithm Large-scale Optimization Problems space optimizing ability:

In order to evaluate DFO algorithm for the optimizing ability of extensive decision variable optimization problem, we use DFO and GSA algorithm to unimodal function f ₁-f ₇with Solving Multimodal Function f ₈-f ₁₃at decision variable dimension p, be respectively 100,500 and 1,000 three kinds of situations and compare calculating.In 30 independent calculating preferably individual target function value in Table 7 and table 8 shown in.The result of calculation of the unimodal function of table 7 shows, for function f ₅, under 2 kinds of algorithms, all do not search for optimum solution, function f ₂at decision variable, be within 1000 o'clock, not search near optimum solution, DFO has showed the optimization ability of good extensive function to other functions.For Solving Multimodal Function, f ₉, f ₁₁and f ₁₂function, DFO and GSA algorithm have all been obtained good result, but other functions all do not search near optimum solution.

DFO algorithm in p is respectively 100,500 and 1000: 30 independent calculating preferably individual convergence situation see Fig. 2 and shown in, in figure, horizontal ordinate represents the iterative computation number of times of algorithm, ordinate represents the optimum solution that algorithm converges to when the corresponding number of times of technology repeatly.。Fig. 2 is the convergence situation of unimodal function, (a) in Fig. 2 (b) (c) (d) (e) represent respectively trial function f ₁, f ₂, f ₃, f ₄, f ₆and f ₇convergence process.Result shows for function f ₁, f ₂, f ₃, f ₄, f ₆and f ₇, DFO algorithm is along with the increase of decision variable dimension, and constringency performance does not significantly reduce, and shows that DFO algorithm has the optimizing ability of good Large-scale Optimization Problems for this class function.

Table 7 function f ₁-f ₇optimum results at different dimension decision spaces

Table 8 function f ₈-f ₁₃optimum results at different dimension decision spaces

1.4.5 the algorithm relatively calculating of optimization problem space optimizing ability on a large scale:

In order to evaluate DFO algorithm for the optimizing ability of decision space optimization problem on a large scale, we are unimodal function f ₁-f ₇with Solving Multimodal Function f ₈-f ₁₃decision space scope expand respectively 10 times, 100 times and 500 times, i.e. the x of table 1 and table 2 _minand x _maxdistribution is multiplied by 10,100 and 500, then with DFO and GSA algorithm, calculates respectively, and all function decision variable dimension p=30 of table 1 and table 2, other parameters of DFO algorithm arrange with 5.2.In 30 independent calculating preferably individual target function value in Table 9 and table 10 shown in.The result of table 9 shows, for unimodal function, for the decision-making of different sizes, except f ₅outward, DFO has obtained the solution of degree of precision, and meanwhile, the optimizing ability of DFO decision space is on a large scale better than SGA.The Multiple hump function optimization result of table 10 shows, for function f ₉, f ₁₁, f ₁₂and f ₁₃, the optimizing ability of DFO decision space is on a large scale better than SGA; For function f ₁₀, the result obtaining of two kinds of algorithms is close; For function f ₈, two kinds of algorithms all fail to search in decision space on a large scale optimum solution.

DFO algorithm expands 10 times at decision space, when 100 times and 500 times, in 30 independent calculating, preferably individual convergence situation is as shown in Figure 3, in figure, horizontal ordinate represents the iterative computation number of times of algorithm, and ordinate represents the optimum solution that algorithm converges to when the corresponding number of times of technology repeatly.Fig. 3 is the convergence situation of unimodal function, (a) in Fig. 3 (b) (c) (d) (e) represent respectively trial function f ₁, f ₂, f ₃, f ₄, f ₆and f ₇convergence process.DFO algorithm is for function f ₁-f ₄3 kinds of situations, almost identical convergence process, function f have been obtained ₆in less than 250 iterative computation, all searched optimum solution, this shows that DFO algorithm all has good optimizing ability for this class function at different range decision space.

Table 9 function f ₁-f ₇optimum results under different range decision space

Table 10 function f ₈-f ₁₃optimum results under different range decision space

1.5 brief summaries:

In recent years, the global optimization method of analogies mechanism of science more and more receives people's concern, and seeking analogies more pervasive, science Self-configuring of science is the key of research high-performance optimization method.We adopt virtual data field as the simulation Self-configuring of optimization method, have set up the mapping relations of data fields space to optimization problem space, have designed a kind of global optimization approach of intending data fields mechanism.Be similar to the optimization method of the plan physics mechanism such as GSA, the method is considered as the feasible solution of optimization problem to have in nuclear field space the fictitious point mass of certain mass, particle is followed newtonian motion theorem and is moved to the particle in colony with high fitness under the effect of nuclear field, finally converges to globally optimal solution.In addition, the optimization spatially adaptive recognition technology of the method can effectively be dwindled region of search and identify optimal region, can obviously accelerate convergence of algorithm speed and precision.

In order to evaluate the superiority of intending data fields Self-configuring, by one group of trial function, compared the performance of DFO algorithm and other several typical heuristic intelligent optimization methods, result shows, to most of trial functions, DFO method can search the solution that precision is higher at faster speed.

2 Three Gorges cascade based on DFO generating Flood Optimal Schedulings.

2.1 foreword

HYDROELECTRIC ENERGY is not only cleaning, cheapness, the reproducible environmental protection energy, is also the desirable peak regulation of electric system, frequency modulation, emergency auxiliary power simultaneously, and the safe and stable operation of electrical network is had to vital role.Optimized Scheduling of Hydroelectric Power is an important process in the productive technology management of hydropower enterprise and power marketing, is performance power station potentiality, makes full use of the multiple clean electric energy of water power, reduces the effective measures of other energy resource consumption.Practice at home and abroad shows, Optimized Operation generally can increase by 1%～7% power benefit.Step power station upstream and downstream waterpower, electric power contact complexity, have compensation and the coordination ability between power station, make its method of operation flexible and changeable, bringing into play very important effect in the economical operation of electric system.The Optimized Operation of step hydropower station not only can bring huge economic benefit for Hydropower Enterprise ', but also to alleviating rich withered, the peak valley contradiction of electrical network, the safe and stable operation abilities such as peak regulation, frequency modulation and emergency duty of raising electrical network and flood control, ecologic environment have great impact.

Step hydropower station Optimal Scheduling is large-scale, dynamic, the non-protruding non-linear scheduling decision problem under the constraint set conditions such as marketing rule and hydrologic cycle, Generation Control, the method for operation, scheduling method, power grid security, power requirement and electricity consumption behavior, has proposed new challenge to scheduling decision theory with method.Lot of domestic and foreign scholar is devoted to the whole bag of tricks that research can effectively address the above problem always.Yet, in the Hydro Power Systems with Cascaded Reservoirs of basin, the mutual conflict of complex target and the coupling of constraint condition make the very difficulty that solves of the description of problem and model, so far almost there is no gratifying solution, urgently further develop new theory and explore its Implementation Technology.Therefore, the theory of the complicated Hydropower energy system scheduling decision of basin step and the research of method are the hot issue of new academic frontier all the time.

By the comparison to one group of trial function, DFO optimization method has been shown superior constringency performance.Generating Optimal Scheduling and the Flood Dispatching Optimization problem of Three-Gorge Cascade Hydropower Station of take is below object, the engineering application of research DFO algorithm, and complicated basin step Optimized Operation is asked a new solution route is provided.

2.2 generating Optimized Operations:

This section is target to the maximum with step generated energy, the step hydropower station generating Optimal Operation Model of foundation based on DFO algorithm, take Three-Gorge Cascade Hydropower Station Optimal Scheduling as application background, obtain the generation schedule of pipe of reinforced concrete at Three Gorges Power Station and Gezhouba Hydropower Station runoff process in given schedule periods.

2.2.1 mathematical model

In given schedule periods, the objective function of the total power benefit maximum of step of runoff process is as follows:

$\mathrm{PE}=\max \underset{d=1}{\overset{D}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{S}{\mathrm{\&Sigma;}}}{P}_{i,d}\&CenterDot;N_{i,d}\&CenterDot;\mathrm{\&Delta;t}---\left(24\right)$

In formula, PE is the total power benefit in the schedule periods of step hydropower station, and D is the total moment number in schedule periods, and S is step hydropower station number, P _{i, d}and N _{i, d}be respectively the electricity price of period d power station i and exert oneself, segment length when Δ t is.The objective function of formula (24) is that power benefit is maximum, works as P _{i, d}equal at 1 o'clock, this objective function is the gross generation maximum of step hydropower station.

The constraint condition of step power generation dispatching is as follows:

(1) power station letdown flow constraint

${\underset{\&OverBar;}{Q}}_{j,d}\&le;Q_{j,d}\&le;{\stackrel{\&OverBar;}{Q}}_{j,d}---\left(25\right)$

In formula, with be respectively minimum, the maximal value of letdown flow in the scheduling slot d of j power station.This constraint comprises: the restriction to letdown flow such as flood control in the discharge capacity constraint of reservoir and schedule periods, shipping, get constrained common factor part.

(2) output of power station constraint

${\underset{\&OverBar;}{N}}_{j,d}\&le;N_{j,d}\&le;{\stackrel{\&OverBar;}{N}}_{j,d}---\left(26\right)$

In formula, with be respectively minimum, the maximum constrained of in the scheduling slot d of j power station, exerting oneself.This constraint comprises: in the maximum installed capacity constraint in power station, schedule periods, to the requirement of output of power station etc., get constrained common factor part.

(3) storage capacity (water level) bound constraint

${\underset{\&OverBar;}{Z}}_{j,d}\&le;Z_{j,d}\&le;{\stackrel{\&OverBar;}{Z}}_{j,d}---\left(27\right)$

In formula, with be respectively minimum, the maximum constrained of water level in the scheduling slot d of j power station.This constraint comprises that the restriction of adjusting storage capacity of setting in the maximum of reservoir, the restriction of minimum storage capacity and schedule periods is, constrained common factor part is got in the restriction of the range of stage of shipping request etc.

(4) water balance equation

$\begin{array}{c}{V}_{j,d}=V_{j,d-1}+I_{j,d}+Q_{j-1,d-{\mathrm{\&tau;}}_j}+S_{j-1,d-{\mathrm{\&tau;}}_i}-Q_{j,d}-S_{j,d}\\ j=\mathrm{1,2}..N,d=\mathrm{1,2}...,D\end{array}----\left(28\right)$

In formula: τ _jrepresent that current stream reaches the time; V _{j, d}, I _{j, d}and S _{j, d}represent respectively power station j at storage capacity constantly of d, put runoff in storage and abandon discharge.

According to above DFO algorithm, a kind of basin step generating Flood Optimal Scheduling method based on intending data fields mechanism of the present invention is as follows:

2.2.2 algorithm is realized:

2.2.2.1 the calculating of decision space scope:

Cascade Hydropower Stations on River Basin Optimal Scheduling has complicated waterpower and electric power contact, and the mutual conflict coupling of all multi-constraint conditions, makes to become very difficult based on solving of heuristic random optimization method.The present invention is by method below, numerous constraint conditions such as the waterpower of the step Optimal Scheduling based on DFO, electric power are transformed to the restriction to decision variable feasible zone, finally Complex Constraints optimization problem is changed into unconstrained optimization problem, greatly simplify the difficulty solving, improved the efficiency solving.Constraint condition transform method is as follows:

Based on DFO step Optimization scheduling algorithm, decision variable X is the upper water place value of reservoir day part.First calculate the feasible zone of each decision variable, then individual optimizing within the scope of this.The computing method of decision variable feasible zone are as follows.

Step1: the minimum letdown flow in power station, upstream of considering lower station letdown flow restriction

${\underset{\&OverBar;}{Q}}_{j,d}=(V_{j+1,d+{\mathrm{\&tau;}}_j}\left(Z_{j+1,d+{\mathrm{\&tau;}}_j}\right)-V_{j+1,d}\left({\underset{\&OverBar;}{Z}}_{j+1,d}\right))/\mathrm{\&Delta;T}+(\underset{\&OverBar;}{Q_{j+1,d}}-Q_{q,d})$

Wherein, V _{j+1, d+ τ}, and V _{j+1, d}be respectively j+1 power station (being lower station) at d+ τ _jthe moment and d storage capacity constantly, and Q _{q, d}be respectively j+1 power station (being lower station) in d lowest water level, minimum letdown flow and local inflow constantly, for lower station is at d+ τ _jwater level constantly, τ _jfor stream reaches the time, the length that Δ T is scheduling slot.

Step2: calculate the upper scheduling water level Z of (d-1 constantly) constantly _{j, d-1}water level limitation scope to scheduling moment d with

Known Z _{j, d-1}, the letdown flow scope of establishing d moment reservoir j is with according to water balance equation:

${\underset{\&OverBar;}{V}}_{j,d}=V_{j,d-1}\left(Z_{j,d-1}\right)+(I_{j,d}-{\stackrel{\&OverBar;}{Q}}_{j,d})\mathrm{\&Delta;T},$

${\stackrel{\&OverBar;}{V}}_{j,d}=V_{j,d-1}\left(Z_{j,d-1}\right)+(I_{j,d}-{\underset{\&OverBar;}{Q}}_{j,d})\mathrm{\&Delta;T},$

Calculate the maximal value of the storage capacity of the constantly corresponding power station j of d and minimum value the storage-capacity curve of searching given power station j obtains corresponding water level range with

Wherein, V _{j, d-1}, Z _{j, d-1}for power station j is at last scheduling storage capacity and water level constantly, with for minimum value and the maximal value of power station j at the storage capacity of scheduling moment d, with be respectively power station j at maximal value and the minimum value of scheduling moment d letdown flow, I _{j, d}for power station j is at d warehouse-in runoff constantly.

Step3: calculate next scheduling water level Z of (d+1 constantly) constantly _{j, d+1}limited field to the water level of scheduling moment d with

Known Z _{j, d+1}, the letdown flow scope of establishing d+1 moment reservoir is with according to water balance equation

${\stackrel{\&OverBar;}{V}}_{j,d}=V_{j,d+1}\left(Z_{j,d+1}\right)-(I_{j,d+1}-{\stackrel{\&OverBar;}{Q}}_{j,d+1})\mathrm{\&Delta;T}$

${\underset{\&OverBar;}{V}}_{j,d}=V_{j,d+1}\left(Z_{j,d+1}\right)+(I_{j,d+1}-{\underset{\&OverBar;}{Q}}_{j,d+1})\mathrm{\&Delta;T}$

Calculate the d maximal value of the corresponding storage capacity of power station j constantly and minimum value the storage-capacity curve of searching given power station j obtains the maximal value of corresponding water level and minimum value

Wherein, V _{j, d+1}, Z _{j, d+1}, I _{j, d+1}be respectively power station j at storage capacity, water level and the warehouse-in runoff of next scheduling moment d+1, I _{j, d}for power station j is at d warehouse-in runoff constantly.

Step4: the least commitment of exerting oneself by power station j at scheduling moment d and maximum constrained by calculating corresponding water level range with

Step5: the scope of establishing t given water level of period is

Step6: by result of calculation above, calculate power station j at the water level range of scheduling moment d.Maximal value wherein ${\stackrel{\&OverBar;}{Z}}_{j,d}=\min ({\stackrel{\&OverBar;}{Z}}_{j,0},{\stackrel{\&OverBar;}{Z}}_{j,1},{\stackrel{\&OverBar;}{Z}}_{j,2},{\stackrel{\&OverBar;}{Z}}_{j,3}),$ Minimum value ${\underset{\&OverBar;}{Z}}_{j,d}=\max ({\underset{\&OverBar;}{Z}}_{j,0},{\underset{\&OverBar;}{Z}}_{j,1},{\underset{\&OverBar;}{Z}}_{j,2},{\underset{\&OverBar;}{Z}}_{j,3}).$

The span x of decision variable X in formula (1) _minand x _maxbe the border of decision space, in the scheduling of the step hydropower station based on DFO, different scheduling decision space scope is constantly different, and the decision space scope of establishing scheduling moment d is [x _{d, min}, x _{d, max}], decision space is in the maximal value of scheduling moment d minimum value $x_{d,\min }={\underset{\&OverBar;}{Z}}_{j,d}.$

Correlated variables based on DFO step power generation dispatching algorithm is described as follows:

2.2.2.2 the realization based on DFO step power generation dispatching algorithm (invention claim):

Based on being implemented as follows of DFO step power generation dispatching algorithm:

Step1: according to the scheduling water level in given each power station of initial time of step hydropower station power generation dispatching, warehouse-in runoff, letdown flow constraint (being provided by formula (25)), units limits (being provided by formula (26)) and restriction of water level (being provided by formula (27)) value, according to the computing method of the decision space scope of 2.2.2.1, calculate each power station at the decision space scope [x of scheduling moment d _{d, min}, x _{d, max}], d=1,2 ..., p.

Wherein, p is decision space dimension, and its value equals the product of power station number S and scheduling hop count D when total.

Step2: at [x _{d, min}, x _{d, max}] scope generates the constantly decision variable initial value x of d of scheduling at random _k(t)={ x ₁(t), x ₂(t) ..., x _d(t) ..., x _p(t) }, t=1, k=1,2 ..., N.

Wherein, x _d(t)=rand (x _{d, min}, x _{d, max}), d=1,2 ..., p, rand () is random number generation function, N is the particle sum of DFO algorithm.

Step3: calculate x by formula (24) _k(t) corresponding step gross generation PE, as the fitness value f of DFO algorithm _k(t), and by formula (9) to (14) calculate each individual corresponding mass value M _k(t), by formula (15) and (6) formula, calculate the suffered accekeration of making a concerted effort and producing of individual all directions.k＝1，2，…，N。

Step4: the weight of upgrading by formula (18) and formula (19) computing velocity;

Step5: upgrade the decision variable of individual all directions with formula (8), and do the processing of crossing the border.The disposal route of crossing the border is as follows:

Step6: call local search algorithm each individuality is carried out to local optimal searching;

Step7: return to Step3 and carry out t+1 step iterative computation, until t > T.T is the greatest iteration time of DFO algorithm.

Step8: draw the individuality of optimal quality in colony, be the operating water level of each reservoir optimum of this Optimized Operation calculating.

The present invention will be described to enumerate specific embodiment below:

Embodiment 1:

Three Gorges cascade power generation dispatching:

The Three Gorges Reservoir modular design runoff of take is reservoir inflow, with step annual electricity generating capacity, is target to the maximum, and Three Gorges cascade power generation dispatching problem is studied.

(1) parameter setting

The parameter of DFO algorithm is shown in " 1.4.2 ", and the correlation parameter of Three-Gorge Cascade Hydropower Station is shown in Table 11.

Table 11 Three Gorges cascade correlation parameter

(2) scheduling achievement

It is warehouse-in runoff that 1947～1976 years actual monthly average waters of Three Gorges Reservoir are take in the present invention, has carried out calculating with the Three Gorges power generation dispatching of generated energy maximum target.

A) reservoir is not less than 155m in the overboard position of disappearing of low water season, and the operating water level of other period reservoir is 145m～175m.Scheduling result is shown in Table 12, Optimized Operation result of calculation shows, average year generated energy 951.09 hundred million KWH of nineteen forty-seven～1976 year Three Gorges Hydropower Plant, scheduling scheme calculating achievement average year generated energy 849.31 hundred million KWH in the Three Gorges Key Water Project cascade operation rules that provide with Yangtze River Water Conservancy Commission compare, on average issue additional for many years electric weight 101.78 hundred million KWH, increasing degree reaches 10.7%.In table 12, the unit of generated energy is hundred million KWH.

B) flood season (6～August), Three Gorges Reservoir Area water level limitation was at 145m, be to start retaining September, reservoir is not less than 155m in the overboard position of disappearing of low water season, in Optimized Operation result, the many average annual energy outputs in year Three Gorges, nineteen forty-seven～1976 are 912.17 hundred million KWH, on average issue additional for many years electric weight 62.86 hundred million KWH, increasing degree is 7.4%.

Table 121947 year～Three Gorges Hydropower Plant Optimized Operation comparison of results table in 1976

(3) interpretation of result:

In the Three Gorges cascade single goal power generation dispatching optimization of power benefit maximum is calculated, warehouse-in runoff adopts the actual water of nineteen forty-six～1976 year, result of calculation shows, between nineteen forty-six～1976 year, many average annual energy outputs in Three Gorges are issued additional respectively electric energy 10.7% and 7.4% for the scheduling scheme providing than design schedule regulation under 2 kinds of power generation dispatching rules.

Embodiment 2:

Flood Optimal Scheduling:

Three Gorges cascade key water control project integrates the comprehensive utilization functions such as flood control, retaining, generating, shipping, and the main task in flood season is flood control, and the Dispatching Flood of science, when obtaining huge Benefit of Flood Preventation, also can obtain considerable power benefit.

We are object with DFO algorithm to Three Gorges cascade Flood Control Dispatch, under modular design flood, be target to the maximum carry out the research of Three Gorges cascade Dispatching Flood with peak clipping rate.

The model description of Flood Control Dispatch:

(1) objective function:

The target of Three Gorges cascade reservoir regulation for flood control substantially can be divided into and guarantee dam safety, downstream submerge loss and dyke safety, upstream submerge loss.Safety of dam body is main relevant with reservoir storage capacity with upstream submerge loss, and downstream submerge loss is mainly relevant with minute magnanimity, and dyke safety is relevant with river course flood flow (or water level).The present invention take by flood control reference mark flood peak discharge minimum be criterion, the peak rate that disappears of a flood events is to the maximum to regulation goal.Objective function is:

minL＝Q _max (29)

In formula: Q _maxthe crest discharge maximal value of passing through for flood control reference mark, downstream.

Constraint condition:

The constraint condition of Three Gorges cascade Flood Control Dispatch is divided into 2 classes, and the one, the physical constraint of step reservoir, is shown in " the constraint bar of step power generation dispatching ".Another kind of is Three Gorges cascade characteristic water level of reservoir and traffic constraints, shown in 13.In table, flux unit is m ³/ s, water level unit is m.

Table 13 Three Gorges cascade characteristic water level of reservoir and flow

Realization based on DFO Flood Control Dispatch algorithm:

Based on DFO Flood Control Dispatch algorithm, take power station upper pond level as decision variable, the formula (29) of take is regulation goal, take DFO algorithm as framework calculates, the flood control program of the peak rate maximum that obtains disappearing.Being implemented as follows of algorithm:

Step1: according to Three Gorges Projects plan for flood control and given Three Gorges cascade characteristic water level of reservoir and the flow of table 15, according to the method for 2.2.2.1, calculate each scheduling decision space scope [x of d constantly _{d, min}, x _{d, max}], d=1,2 ..., D.

Step2: at [x _{d, min}, x _{d, max}] scope is the random scheduling decision variable initial value x of d constantly that generates _k(t)={ x ₁(t), x ₂(t). ..., x _p(t) }, t=1, k=1,2 ..., N.

Step3: calculate decision variable x by formula (29) _k(t) the crest discharge maximal value that corresponding flood control reference mark, downstream is passed through (is fitness value f _k(t)), and by formula (9) to (14) calculate each individual corresponding mass value M _k(t), by formula (15) and (6) formula, calculate the suffered accekeration of making a concerted effort and producing of individual all directions.k＝1，2，…，N。

Step4: the weight of upgrading by formula (18) and formula (19) computing velocity;

Step5: upgrade the decision variable of individual all directions with formula (8), and do the processing of crossing the border.The disposal route of crossing the border is as follows:

Step6: call local search algorithm each individuality is carried out to local optimal searching;

Step7: return to Step3 and carry out t+1 step iterative computation, until meet algorithm end condition;

Step8: the individuality of optimal quality in output colony, by the scheduling scheme of flood control reference mark, downstream crest discharge minimum.

Scheduling result:

Using 0.5% design flood in 1998 as flood into reservoir, and analysis and comparison flood is based on DFO Optimized Operation and adjust the big vast Flood Control Dispatch scheme calculating.Optimized Operation and adjust that flood calculates play water transfer position and scheduling end of term water level is flood control 145m, it is 40000m that rising of adjusting that flood calculates adjusted a flow ³/ s, the water level range of Optimized Operation is 135m～180m, Three Gorges letdown flow scope is 15800m ³/ s～98800m ³/ s, Three Gorges gate opening/closing is sequentially followed successively by flood discharge deep hole (maximum hole count is 23 holes), bottom hole for sand flushing (maximum hole count is 7 holes), floating row hole (maximum hole count is 2 holes), flood discharge table hole (maximum hole count is 22 holes), and Ge Zhou Ba gate opening/closing is sequentially followed successively by two river sluice gates, great river sluice gate and three river sluice gates.According to adjusting big vast scheduling rule open-close sluice gate.

Optimized Operation and adjust earial drainage graph that flood calculates as shown in Figure 4, hop count when horizontal ordinate is in figure, ordinate is flow, the risk indicator statistics of scheduling result is shown in Table 14.Can find out, in two kinds of water situations, Optimized Operation based on DFO algorithm and the last water level of adjusting flood to calculate can be got back to flood control 145m, the highest reservoir level of Three Gorges Reservoir is all in rational scope, the peak rate that disappears of Optimized Operation scheme will be apparently higher than the scheme of adjusting flood scheduling, and its earial drainage graph is also smooth than the scheme of adjusting flood to calculate.Optimized Operation scheme is significantly better than adjusting big vast numerical procedure to the Benefit of Flood Preventation in downstream.

Table 14 Flood Dispatching Optimization and the big vast result of calculation contrast table of tune

In sum, adopt DFO algorithm, take respectively modular design runoff and 1882～1998 years actual waters is warehouse-in runoff, has calculated the moon, the ten days generation schedule of Three Gorges and Gezhouba Hydropower Station, scheduling result is reasonable, and scheduling result is better than the scheduling scheme that design schedule regulation provides.

Take modular design flood as flood into reservoir, with peak clipping rate, be target to the maximum, carried out Flood Optimal Scheduling calculating, optimize Flood Control Dispatch scheme reasonable, the peak rate that disappears of Optimized Operation scheme will be apparently higher than the scheme of adjusting flood scheduling, and its earial drainage graph is also smooth than the scheme of adjusting flood to calculate.Optimized Operation scheme is significantly better than adjusting big vast numerical procedure to the Benefit of Flood Preventation in downstream.

By the calculating to Three Gorges cascade generating and Flood Control Dispatch, verified the engineering application of DFO algorithm, the while is also for the solution of complicated basin cascade operation provides a new effective way.

The present invention adopts DFO global optimization approach, solves basin step generating Flood Optimal Scheduling problem.For basin step generating Flood Optimal Scheduling problem provides an effective solution route, concrete manifestation is as follows:

(1) step hydropower station Optimal Scheduling is large-scale, dynamic, the non-protruding non-linear scheduling decision problem under the constraint set conditions such as marketing rule and hydrologic cycle, Generation Control, the method for operation, scheduling method, power grid security, power requirement and electricity consumption behavior, has proposed new challenge to existing scheduling theory and method.Lot of domestic and foreign scholar is devoted to the whole bag of tricks that research can effectively address the above problem always.Yet, in the Hydro Power Systems with Cascaded Reservoirs of basin, the mutual conflict of complex target and the coupling of constraint condition make the very difficulty that solves of the description of problem and model, so far almost there is no gratifying solution, urgently further develop new theory and explore its Implementation Technology.Therefore, the theory of the complicated Hydropower energy system scheduling decision of basin step and the research of method are the hot issue of new academic frontier all the time.

(2) modern heuristic intelligent optimization method provides a new approach for solving of complicated step Optimal Scheduling, at present main heuristic optimization algorithm is all successfully applied to the Optimal Scheduling of complicated step hydropower station as particle cluster algorithm (PSO), genetic algorithm (GA), differential evolution algorithm (DE), ant group algorithm (ACO), chaotic optimization algorithm etc., and has obtained good result.Along with the sharply expansion of power station scale, the Optimized model of structure is more and more meticulousr.Simultaneously, the raising that becomes more meticulous and require along with dispatching of power netwoks, especially when Short-term Optimal Operation, existing optimization method exists computing velocity slowly, easily to sink into the problems such as local optimum solution to some extent, also be difficult to meet the ageing requirement of model solution, therefore study new intelligent optimization method and solve day by day complicated basin step Optimal Scheduling, have most important theories and practical significance.

(3) the DFO algorithm that the present invention proposes, there is optimizing ability strong, the features such as computing velocity is fast, with Three Gorges cascade combined optimization, be scheduling to engineering background, basin step generating, the Flood Dispatching Optimization problem of research based on DFO algorithm, can solve the ageing problems such as local optimum that are not absorbed in by force and easily that existing optimization method exists to a certain extent.

(4) scheduling of basin step combined optimization generally can increase by 1%～7% power benefit, can bring the economic benefit of syntax to electricity power enterprise;

(5) Optimized Scheduling of Hydroelectric Power is an important process in the productive technology management of hydropower enterprise and power marketing, is performance power station potentiality, makes full use of the multiple clean electric energy of water power, reduces the effective measures of other energy resource consumption.China's energy-saving and emission-reduction are had to great meaning;

(6) step power station upstream and downstream waterpower, electric power contact complexity, have compensation and the coordination ability between power station, makes its method of operation flexible and changeable, bringing into play very important effect in the economical operation of electric system.The combined optimization scheduling of step hydropower station is to alleviating rich withered, the peak valley contradiction of electrical network, and the safe and stable operation abilities such as peak regulation, frequency modulation and emergency duty of raising electrical network and flood control, ecologic environment have great impact.

List of references (References)

[1]Liping Xie，Jianchao Zeng.The performance analysis of artificial physics optimization algorithm driven by different virtual forces.ICIC Express Letters，2010，4(1)：239-244

[2] Jiang Jianguo, Long Xiuping, field Min. a kind of electromagnetism-like mechanism algorithm based on good point set. Xian Electronics Science and Technology University's journal (natural science edition), 2011,38 (6): 167-172

Jiang Jianguo；Long Xiuping；Tian Min.Electromagnetism-like mechanism algorithm based on the good point set.Journal of Xidian University，2011，38(6)：167-172

[3]Yun Shang，Jian Chen，Qun Wang.Improved electromagnetism-like mechanism algorithm for constrained optimization problem.2010International Conference on Computational Intelligence and Security(CIS)，2010，165-169

[4]S.Ilker Birbil，Shu-cherng Fang.An Electromagnetism-like Mechanism fbr Global Optimization，Journal of Global Optimization25∶263-282，2003.

[5]R.A.Formato.Improved CFO algorithm fbr antenna optimization，Progress In Electromagnetics Research B.2010，19∶405-425

[6]R.A.Formato.On the utility of directional infbrmation for repositioning errant probes in central fbrce optimization.arXiv：1005.5490，http：//arXivorg(2010)

[7]Esmat Rashedi，Hossein Nezamabadi-pour，Saeid Saryazdi.GSA：A Gravitational Search Algorithm，Information Sciences，179∶2232-2248，2009.

[8]Chaoshun Li，Jianzhong Zhou.Parameters identification of hydraulic turbine governing system using improved gravitational search algorithm，Energy Conversion and Management.2011，52(1)：374-381

[9]David H.Wolpert，William G.Macready.No free lunch theorems fbr optimization.IEEE Transactions on Evolutionary Computation，1997，1∶67-82

[10] Li Deyi, Du Yi. uncertain artificial intelligence. Beijing: National Defense Industry Press, 2005.

[11] Gan literary composition swallow, He Nan, Li Deyi, etc. a kind of Web Community's discover method based on Topology Potential, Journal of Software, 20 (8): 2241-2254,2009.

[12]Tao Wu，Yan Gao，Image data field fbr edge detection，ICIC Express Letters，5(3)：733-740，2011.

[13] Gan literary composition swallow, He Nan, the firm a kind of Web Community's discover method based on Topology Potential of Li De, Journal of Software, 20 (8): 2241-2254,2009.

Claims (1)

1. the basin step generating Flood Optimal Scheduling method based on intending data fields mechanism, is characterized in that carrying out according to following steps:

Step1: according to the scheduling water level in given each power station of initial time of step hydropower station power generation dispatching, warehouse-in runoff, letdown flow constraint, units limits, and restriction of water level, according to the computing method of decision space scope, calculate each power station in the decision space scope of scheduling moment d;

Step2: decision space scope generates the scheduling decision variable initial value of d constantly at random;

Step3: by step gross generation PE, as the fitness value f of DFO algorithm _k(t), calculate each individual corresponding mass value M _k(t), calculate the suffered accekeration of making a concerted effort and producing of individual all directions;

Step4: the weight that computing velocity is upgraded;

Step5: upgrade the decision variable of individual all directions, and do the processing of crossing the border;

Step6: call local search algorithm each individuality is carried out to local optimal searching;

Step7: return to Step3 and carry out iterative computation, until t > T;

Step8: draw the individuality of optimal quality in colony, be the operating water level of each reservoir optimum of this Optimized Operation calculating.