An improved reservoir optimal scheduling method based on adaptive genetic algorithm
By optimizing reservoir scheduling through an adaptive genetic algorithm, time intervals are divided according to the inflow status. Shortened chromosome encoding and adaptive genetic operations are used to solve the problems of high computational resource consumption and frequent gate changes in reservoir scheduling, thus achieving efficient reservoir optimization scheduling.
Patent Information
- Application Number
- CN202111655949.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-30
- Publication Date
- 2026-02-27
- Estimated Expiration
- 2041-12-30
AI Technical Summary
Existing reservoir optimization scheduling methods consume significant computational resources during data reading, require extensive calculations for regular inflows, and frequent changes in gate opening make it difficult to meet scheduling requirements.
An adaptive genetic algorithm is adopted to divide the time interval according to the rise, fall and steady state of the incoming water. The discharge scheme is represented by a shortened chromosome code. The fitness function is optimized by combining amplitude limiting filter and anti-shaking filter. The genetic operation is adaptively adjusted by controlling the parameters.
It reduces computational resource consumption, improves convergence speed and adaptability, reduces the frequency of gate opening changes, and meets the optimization requirements of reservoir scheduling.
Smart Images
Figure SMS_1 
Figure SMS_2 
Figure SMS_3
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of reservoir regulation, and particularly relates to a reservoir optimal regulation method based on an improved adaptive genetic algorithm. BACKGROUND
[0002] The optimization algorithm for reservoir regulation mainly includes conventional optimization algorithms and intelligent optimization algorithms.
[0003] The conventional optimization algorithm has limitations such as curse of dimensionality, poor convergence, and slow calculation speed; therefore, the conventional optimization algorithm is rarely used at home and abroad.
[0004] With the continuous rise of artificial intelligence, more and more researches on optimization regulation by using neural networks and genetic algorithms are carried out; the model of the reservoir optimization regulation algorithm by using neural networks is relatively complex, and therefore the genetic algorithm for reservoir optimization regulation has become a relatively mainstream algorithm.
[0005] However, the simple genetic algorithm mainly has two problems: one is slow convergence of data; and the other is easy to produce premature phenomena. The local search capability of the simple genetic algorithm is not strong, and the factors affecting the search performance of the algorithm mainly include initial population, coding mode, genetic operator, and control parameters. In view of the case that the control parameters (population size, crossover probability, and mutation probability) are unchanged in the evolution process, the existing scholars propose an adaptive genetic algorithm, that is, in the initial stage of the genetic algorithm, the individuals with relatively low fitness are concentrated, and if small crossover probability and mutation probability are used, the population is difficult to produce excellent new individuals, resulting in slow convergence. In the later stage of the genetic algorithm, the individuals with relatively high fitness are concentrated, and if large crossover probability and mutation probability are used, the excellent individuals are easily destroyed, and the algorithm is trapped in a local optimal solution. The adaptive genetic algorithm is used to automatically change the crossover probability and mutation probability with the increase of the genetic generation.
[0006] The mainstream idea for improving the coding mode is to change the binary coding into floating-point number coding, Gray code and the like. In view of the fact that the discharge scheme is closely related to the water level of the reservoir in the reservoir optimization regulation, some scholars propose that the N+1 discrete water level points (Z1, Z2, …, Z n , Z n+1 ) are obtained by equally dividing the reservoir between the highest water level and the maximum water level, each sub-interval represents a discharge, and 0 or 1 is used to form a binary string with a length of L, thereby forming a chromosome with a total length of N*L for each period, and the total length of the chromosome for M periods is M*N*L.
[0007] It is found through practice that if the number of periods to be regulated is large, such a coding mode will result in a long chromosome, and in the calculation process, the following four problems will be caused:
[0008] 1) Chromosome length, resource consumption is high: because each period corresponds to a segment of chromosome, resulting in the length of the chromosome is very long; in the process of data reading will consume large CPU, memory and other computing resources.
[0009] 2) Chromosome length, slow genetic convergence: due to the length of the chromosome is too long, in the process of genetic will lead to cross, variation of small value change, slow convergence data.
[0010] 3) fixed encoding, poor flexibility: due to the fixed encoding method, cannot be flexible according to the water characteristics of coding; resulting in a large number of calculations to solve the conventional water.
[0011] 4) every period randomly, sawtooth number: because each period has a division of water level interval, decoding is equivalent to each period is a random value; such a solution to the solution of sawtooth serious, no better to sawtooth algorithm, resulting in can not meet the principle of scheduling gate opening degree can't change too frequently. SUMMARY
[0012] The technical problems to be solved by the present application are to provide a reservoir optimization scheduling method based on an improved adaptive genetic algorithm, to solve the technical problems of the prior art reservoir optimization scheduling method, such as the reading of data in the calculation process consuming large CPU, memory and other computing resources; a large number of calculations are required to solve the conventional water; and the principle of gate opening degree cannot be changed too frequently in scheduling.
[0013] The technical solution of the present application is:
[0014] A reservoir optimization scheduling method based on an improved adaptive genetic algorithm, comprising:
[0015] Step 1, calculating the cross section water without storage and scheduling;
[0016] Step 2, dividing the discharge flow;
[0017] Step 3, identifying the rising, falling and flat period interval of the water;
[0018] Step 4, encoding each interval in order;
[0019] Step 5, determining the time interval of each period, and decoding the corresponding chromosome segment, combining the predicted water process and the starting optimization of the starting water level, to obtain the discharge process and the water level change process;
[0020] Step 6, fitness function calculation; according to the discharge process, the fitness of each individual is calculated;
[0021] Step 7, convergence judgment;
[0022] Step 8, genetic manipulation;
[0023] Step 9, control parameters;
[0024] Step 10, repeat steps 5 to 9 until the conditions are met.
[0025] The method for calculating the cross-section inflow without storage and scheduling is: according to the previous cross-section inflow, using Muskingum to calculate the inflow of the control section.
[0026] The method for dividing the discharge is: set a minimum discharge lower limit, when the inflow is less than the lower limit value, it is all divided as a steady state; the method of setting the lower limit is: if there is a cross-section average flow value for more than one year, use the average flow, otherwise use the average value of the inflow forecast.
[0027] The method for identifying the rising, falling and flat period interval of the inflow is: the rising water period is the period from the set discharge lower limit value or valley value to the peak value; the falling water period is the period from the peak value to the set discharge lower limit value or valley value; the flat period is the period less than or equal to the set discharge lower limit value.
[0028] The method for encoding each interval in sequence is: according to the period interval divided in step 3, sequentially encode each interval, divide the reservoir allowed minimum water level and allowed maximum water level into N parts, get N+1 discrete water level points, each interval represents a discharge, use 0 or 1 to form a string with length L, N subintervals form a chromosome with total length N*L.
[0029] The method for judging the time interval of each period, obtaining the corresponding chromosome segment for decoding, combining the predicted inflow process and the starting regulation water level to start optimization, and obtaining the discharge process and water level change process includes:
[0030] Judging the time interval of each period and obtaining the corresponding chromosome segment; then converting the chromosome binary to decimal, then through the reservoir water level and discharge relationship curve, finding the maximum discharge capacity (q1max, q2max, q3max…qnmax) of each water level corresponding to the water level interval, and then converting the decimal number to variable (q1, q2, q3,…qn) according to the decoding formula; the decoding formula is as follows:
[0031]
[0032] The optimization is started in combination with the prediction of the incoming water process and the starting regulation water level; in the first period, it is assumed that the reservoir does not discharge flood water in this period, the reservoir storage at the end of this period is calculated according to the incoming water, and the corresponding water level is found according to the water level-storage capacity relation curve; through the decoding process above, the water level interval corresponding to the water level is found; the discharge flow q(t) is calculated, which is assumed to be the average flow, the reservoir storage V(t) is recalculated according to the water balance formula, the corresponding water level Z(t) is found, the corresponding q(t) is found by searching the q-V curve, if q(t) and q(t) are equal, the assumption is correct, and q(t) and V(t) are taken as the state of the next period; if not, the average value of q(t) and q(t) is taken as the new q(t) to recalculate; the same operation process is performed on the next period to obtain the discharge process and the water level change process.
[0033] The fitness function calculation in step 6; the method for calculating the fitness of each individual according to the discharge process is to remove the sawtooth in the discharge process by using "amplitude limiting filtering method" and "debouncing filtering method"; the water balance formula is used to recalculate and solve; the solution corresponding to the chromosome is taken as the input for fitness calculation; the fitness of each individual is calculated according to the set objective function, including maximum peak clipping, shortest disaster duration or minimum over-standard water volume.
[0034] The method for judging convergence is:
[0035] Step 7.1, the precision requirement is met, i.e. the solution of an individual in the population satisfies the objective function
[0036] Step 7.2, the difference between the fitness values of the optimal individuals of several consecutive generations is small and the difference between the fitness values of all individuals in the population is small;
[0037] Step 7.3, the set iteration number is exceeded;
[0038] Step 7.5, the set calculation time is exceeded.
[0039] The genetic operation in step 8 includes:
[0040] Step 8.1, selection: combination of optimal saving strategy and proportional selection;
[0041] Step 8.2, single-point crossover: a crossover point is randomly set in the individual coding string, and then the partial chromosomes of the two paired individuals are exchanged at the point;
[0042] Step 8.3, uniform mutation: the original gene values at each locus in the individual coding string are replaced by a random function uniformly distributed within a certain range with a certain small probability.
[0043] The method for controlling parameters is to set the probabilities of selection, crossover and mutation and the fitness value to be scaled with the change in the genetic algorithm, so as to realize the self-adaptation of the control parameters.
[0044] The present application has the following advantages:
[0045] In the process of reservoir optimal scheduling, the same water level interval is different when it is in a stable, rising, falling state and the desired discharge scheme, so the present application proposes to divide the time interval according to the rising, falling and stable state of the incoming water, and to use a chromosome to encode the discharge scheme of a time interval to achieve the goal of shortening the code and reducing the sawtooth of the discharge process.
[0046] Assuming that the incoming water time interval is M, and according to the existing general coding method, the water level is set to N intervals, and the length of each interval is L, the total length of the chromosome is: M*N*L; this paper proposes that only the incoming water needs to be divided into a stable period before rising, a rising period, a falling period and a stable period after falling m intervals, and the coding length is: m*N*L; the multi-peak incoming water process is also divided into time intervals according to the rising, falling and stable characteristics of the incoming water, and then coded. Due to the hydrological characteristics of the incoming water process, the value of m will be less than the value of M, and such a coding method will greatly shorten the length of the chromosome. Therefore, the consumption of computing resources is reduced, the convergence speed is increased, and the adaptability to different incoming water is improved; and since the water level is in the same interval within a certain period of time, the number of sawtooths is reduced and it is easy to eliminate the sawtooths through the outflow process of water balance trial calculation.
[0047] The present application solves the technical problems of the prior art, such as the consumption of large CPU, memory and other computing resources in the data reading process of the reservoir optimal scheduling method, the need for a large amount of calculation to solve the conventional incoming water, and the inability to meet the principle that the gate opening degree cannot be changed too frequently in the scheduling. DETAILED DESCRIPTION
[0048] The present application improves the coding method. In the scheduling process, the same water level interval is different when it is in a stable, rising, falling state and the desired discharge scheme, so the present application proposes to divide the time interval according to the rising, falling and stable state of the incoming water, and to use a chromosome to encode the discharge scheme of a time interval to achieve the goal of shortening the code and reducing the sawtooth of the discharge process.
[0049] The implementation steps of the present application are consistent with the application of the self-adaptive genetic algorithm in reservoir optimal scheduling, and the specific steps are as follows:
[0050] Step 1: Calculate the incoming water of the section under the condition of no storage and scheduling
[0051] According to the incoming water of the section, the incoming water of the section to be controlled is calculated by using the Muskingum method.
[0052] Step 2: Set the division of the discharge
[0053] In order to avoid the rise and fall state division of some conventional water, the present application will set a minimum discharge lower limit, when the water is less than the value, all as a stable state division. The setting value algorithm principle is: the multi-year average of the section is used, and the average value of the water forecast is used.
[0054] Step 3: Identify the water rise, fall, and flat period interval
[0055] Rising water section: set the discharge lower limit value or the time period from the valley to the peak;
[0056] Falling water section: the time period from the peak to the set discharge lower limit value or valley value;
[0057] Flat section: the period less than or equal to the set discharge lower limit value;
[0058] Divide the subscript at the end of each period interval, such as dividing a period of 72 into 4 period intervals, and the corresponding period subscript is [10, 16, 34, 72].
[0059] Step 4: Encoding
[0060] According to the 4 period intervals divided in step 3, each interval is sequentially divided into N parts between the lowest water level allowed by the reservoir (flood control water level) and the highest water level allowed (flood control high water level), obtaining N+1 discrete water level points, each interval representing a discharge, and using 0 or 1 to form a string of length L, N subintervals form a total length of N*L chromosome. For example, the flood control water level of a reservoir is 106m, the flood control high water level is 110.5m, and it is divided into 5 parts, Z1=106, Z2=106.9, Z3=107.8, Z4=108.7, Z5=109.6, Z6=110.5; length L=10, randomly generate chromosome code: 4*N*L, indicating the discharge (q1, q2, q3, q4, q5) corresponding to the water level in the above 5 water level intervals. The coding accuracy is:
[0061] (Umax-Umin) / (pow(2,L)-1). The coding length is changed from the existing coding form of 72*M*L to 4*M*L, which is shortened by 18 times.
[0062] Step 5: Decoding
[0063] Determine the time interval of each period, and get the corresponding chromosome segment; then convert the chromosome binary to decimal, and then find the maximum discharge capacity (q1max, q2max, q3max…qnmax) of each water level in (Z1, Z2, Z3,…Zn) corresponding to the water level interval through the reservoir water level and discharge relationship curve, and then convert the decimal number to variable (q1, q2, q3,…qn) according to the decoding formula. The decoding formula is as follows:
[0064]
[0065] Combine the forecast inflow process and the starting regulation water level (not provided, then use the flood control limit water level of the reservoir as the starting regulation) to start optimization. In the first period, assuming that the reservoir does not discharge flood water in this period, calculate the reservoir storage at the end of this period according to the inflow, and find the corresponding water level according to the water level-storage relationship curve. Through the above decoding process, the water level interval corresponding to the water level is calculated; the discharge q(t) is calculated as an average flow, and the reservoir storage V(t) is recalculated according to the water balance equation, and the corresponding water level Z(t) is found, and the corresponding q(t) is found by searching the q-V curve. If q(t) and q(t) are equal, it is assumed to be correct, and q(t) and V(t) are used as the state of the next period. If it is not correct, the average value of q(t) and q(t) is used as the new q(t) to recalculate. The same operation process is performed on the next period to obtain the discharge process and the water level change process.
[0066] Use "amplitude limiting filter method" and "anti-jitter filter method" to remove the sawtooth in the discharge process; recalculate and solve using the water balance equation. As the corresponding solution of the chromosome (discharge process), further as the input of the fitness calculation.
[0067] Step 6: Fitness function calculation
[0068] According to the set target function (maximum peak clipping, shortest disaster duration, minimum excess water, etc.), calculate the fitness of each individual according to the discharge process.
[0069] Step 7: Convergence judgment
[0070] 1) Reach the accuracy requirement, i.e. the solution of the individual in the population satisfies the target function
[0071] 2) The difference between the optimal individual fitness values of several consecutive generations is small, and the difference between the fitness values of all individuals in the population is small
[0072] 3) Exceed the set iteration number
[0073] 4) Exceed the set calculation time
[0074] Step 8: Genetic operation
[0075] 1) Selection: The best individual is selected and saved to the next generation without crossover and mutation. The fitness of each individual is normalized, and the relative fitness value pi is calculated. The population size is N, and the fitness value of each individual is fi. The relative fitness value is pi = fi / sum(fi). The sum of the relative fitness values is 1. The relative fitness values of each chromosome are proportionally distributed in a segment of varying length. A random number A between 0.0 and 1.0 is generated. If A falls within a certain interval, the individual in that interval is selected to be passed to the next generation. This process is repeated (N-1) times to select N individuals to be passed to the next generation.
[0076] 2) Single-point crossover: A crossover point is randomly set in the individual's code string, and then the parts of the two paired individuals are exchanged at this point. The process is to designate each locus in the individual's code string as a mutation point. For each mutation point, a random number between 0.0 and 1.0 is generated and compared with the mutation probability pc. If it is less than pc, the individual is crossed. The adaptive genetic algorithm adjusts the crossover probability pc according to the fitness value. The adjustment is as follows:
[0077]
[0078] fmax: the maximum fitness value in the population;
[0079] favg: the average fitness value of the population in each generation;
[0080] f`: the larger fitness value of the two individuals to be crossed;
[0081] Pc1=0.9 Pc2=0.6
[0082] 3) Uniform mutation: Each locus in the individual's code string is designated as a mutation point. For each mutation point, a random number between 0.0 and 1.0 is generated and compared with the mutation probability pm. If it is less than pm, the original gene value at this locus is replaced with a new value generated by a uniform distribution function. The adaptive genetic algorithm adjusts the mutation probability pm according to the fitness value. The adjustment is as follows:
[0083]
[0084] fmax: the maximum fitness value in the population;
[0085] favg: the average fitness value of the population in each generation;
[0086] f: the fitness value of the individual to be mutated;
[0087] Pm1 = 0.1 Pm2 = 0.001
[0088] Step 9: Control parameter
[0089] In order to avoid premature phenomenon and local optimal solution and balance between convergence speed as far as possible. The fitness should be "scaling method". The probability of selection, crossover, mutation and the scaling of fitness value change with the change of the genetic, so as to realize the self-adaptation of control parameter. That is: when the fitness difference between individuals is large (initial), the scaling value is set smaller, so the fitness of the individual close to the average fitness value is "expansion" trend, and the fitness far from the average fitness value is "shrink" trend, which maintains the diversity, so as to avoid premature phenomenon; when the fitness difference between individuals is small (later), the scaling value is set larger, so the fitness of the individual close to the average fitness value is "expansion" to "shrink" trend, and the fitness far from the average fitness value is "shrink" to "expansion" trend. The fitness difference between individuals is increased, and the convergence speed is accelerated.
[0090] Step 10: Cycle step 5 to step 9 until the condition is met.
Claims
1. A reservoir optimal scheduling method based on an improved adaptive genetic algorithm, comprising: Step 1, calculating the section inflow without storage and scheduling; Step 2, dividing the discharge flow; Step 3, identifying the rising, falling and flat period intervals of the inflow; The method for identifying the rising, falling and flat period intervals of the inflow is that the rising water section is the period from setting the discharge lower limit value or the valley value to the peak value, the falling water section is the period from the peak value to setting the discharge lower limit value or the valley value, and the flat section is the period less than or equal to the set discharge lower limit value; Step 4, sequentially coding each interval section; The method for sequentially coding each interval section is that according to the period intervals divided in Step 3, each interval section is sequentially coded, the reservoir allowed minimum water level and the allowed maximum water level are equally divided into N parts, N+1 discrete water level points are obtained, each interval represents a discharge flow, and 0 or 1 is used to form a string with a length of L, N sub-intervals form a total length of N*L, and a chromosome; Step 5, judging the period interval in which each period is located, obtaining the corresponding chromosome section for decoding, combining the predicted inflow process and the starting regulation water level to start optimization, and obtaining the discharge process and the water level change process; the implementation method comprises: judging the period interval in which each period is located and obtaining the corresponding chromosome section; then converting the chromosome binary into decimal, then searching for the maximum discharge capacity (q1max, q2max, q3max…qnmax) of each water level corresponding to the water level interval through the reservoir water level and discharge flow relationship curve, and then converting the decimal number into a variable (q1, q2, q3,…qn) according to the decoding formula; the decoding formula is as follows: ； wherein: is a decimal number; is a decimal number corresponding to a binary number; is is a lower limit value of is is an upper limit value of is the number of bits of a binary number; combining the predicted inflow process and the starting regulation water level to start optimization; in the first period, assuming that the reservoir does not discharge flood water in this period, calculating the reservoir storage at the end of this period according to the inflow, searching for the corresponding water level according to the water level-storage relationship curve; calculating the discharge flow q(t)false through the above decoding process from the water level corresponding to the water level interval; recalculating the reservoir storage V(t)false according to the water balance formula, searching for the corresponding water level Z(t)false, searching for q(t)check corresponding to the q-V curve, assuming that q(t)false is equal to q(t)check, the assumption is correct, and q(t)false and V(t)false are taken as the state of the next period; if it is not correct, the average value of q(t)false and q(t)check is taken as a new q(t)false for recalculation; the same operation process is performed on the next period to obtain the discharge process and the water level change process; Step 6, calculating the fitness function; calculating the fitness of each individual according to the discharge process; Step 7, judging the convergence; Step 8, genetic operation; Step 9, control parameters; Step 10, repeating Steps 5 to 9 until the conditions are met.
2. The improved reservoir optimal scheduling method based on adaptive genetic algorithm according to claim 1, characterized in that: The method for calculating the section inflow without storage and scheduling is that the section inflow is calculated according to the inflow of the previous section by using the Muskingum method.
3. The improved reservoir optimal scheduling method based on adaptive genetic algorithm according to claim 1, characterized in that: The method for dividing the discharge is as follows: setting a minimum discharge lower limit, when the incoming water is less than the lower limit value, all is divided as steady state; the method for setting the lower limit is as follows: using the average flow value of the section for more than one year, if there is no average flow value, using the average value of the incoming water forecast.
4. The improved reservoir optimal scheduling method based on adaptive genetic algorithm according to claim 1, characterized in that: The method for calculating the fitness function of each individual according to the discharge process is as follows: using the "limit filter method" and "anti-jitter filter method" to remove the sawtooth in the discharge process. The method for calculating the fitness function of each individual according to the discharge process is as follows: using the "limit filter method" and "anti-jitter filter method" to remove the sawtooth in the discharge process.
5. The improved reservoir optimal scheduling method based on adaptive genetic algorithm according to claim 1, characterized in that: The method for judging the convergence is as follows: Step 7.1, reaching the precision requirement, i.e. the solution of the individual in the population meets the target function Step 7.2, the difference between the optimal individual fitness values of several generations is small and the difference between the fitness values of all individuals in the population is small; Step 7.3, exceeding the set iteration number; Step 7.4, exceeding the set calculation time.
6. The improved reservoir optimal scheduling method based on adaptive genetic algorithm according to claim 1, characterized in that: The genetic operation in step 8 includes: Step 8.1, selection: optimal saving strategy combined with proportional selection; Step 8.2, single-point crossover: setting a crossover point in the individual code string at random, then exchanging the partial chromosomes of the two paired individuals at the point; Step 8.3, uniform mutation: respectively using a random function conforming to uniform distribution in a certain range to replace the original gene value at each locus in the individual code string with a certain small probability.
7. The improved reservoir optimal scheduling method based on adaptive genetic algorithm according to claim 1, characterized in that: The method for controlling the parameters is as follows: setting the probabilities of selection, crossover and mutation and the fitness value to change and scale in the genetic process, so as to realize the self-adaptation of the parameters. The method for controlling the parameters is as follows: setting the probabilities of selection, crossover and mutation and the fitness value to change and scale in the genetic process, so as to realize the self-adaptation of the parameters.