Offshore wind turbine selection method based on fuzzy bayesian network decision

Abstract

The present application relates to a kind of offshore wind turbine selection method based on fuzzy bayesian network decision-making, comprising the following steps: obtaining the influencing factor of offshore wind turbine selection and analysis, based on the analysis result, construct three-layer bayesian structure network;Each influencing factor variable is input fuzzy;Parameter setting and optimization of conditional probability table are carried out based on simulated annealing algorithm;Based on the optimized conditional probability table, inference is carried out on the influencing factor of offshore wind turbine selection using three-layer bayesian structure network;Based on expert knowledge and default utility value respectively, the inference result is defuzzified, and first evaluation score and second evaluation score are obtained;According to first evaluation score and second evaluation score, the best offshore wind turbine selection is obtained by comparison and selection.Compared with prior art, the present application can cope with the challenge faced by traditional wind turbine decision-making method in the face of complex and changeable influencing factors, and realize comprehensive, comprehensive and systematic wind turbine selection scheme decision-making.

Description

Technical Field

[0001] This invention relates to the field of offshore wind turbine selection decision-making, and in particular to an offshore wind turbine selection method based on fuzzy Bayesian network decision-making. Background Technology

[0002] With the widespread application of wind power, the selection of offshore wind turbines has become a key issue in the wind power industry. Factors influencing the selection of offshore wind turbines include turbine parameters, economic efficiency, reliability, and navigation safety.

[0003] The technical problems existing in the traditional offshore wind turbine selection decision-making method include: inaccuracy of probability allocation, the fixed nature of conditional probability tables limiting the flexibility of the model, and difficulty in handling complex and ever-changing influencing factors. Summary of the Invention

[0004] The purpose of this invention is to overcome the defects in the prior art and provide a method for selecting offshore wind turbines based on fuzzy Bayesian network decision-making, so as to achieve efficient and feasible wind turbine selection under complex and variable influencing factors.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A method for selecting offshore wind turbines based on fuzzy Bayesian network decision-making includes the following steps:

[0007] S1. Obtain and analyze the influencing factors of offshore wind turbine selection, and construct a three-layer Bayesian structure network based on the analysis results.

[0008] S2, fuzzify the input for each influencing factor variable;

[0009] S3, Parameter setting and optimization of conditional probability tables based on simulated annealing algorithm;

[0010] S4. Based on the optimized conditional probability table, a three-layer Bayesian structure network is used to infer the influencing factors of offshore wind turbine selection.

[0011] S5, based on expert knowledge and default utility value respectively, the reasoning results are defuzzified to obtain the first evaluation score and the second evaluation score;

[0012] S6. The best offshore wind turbine is selected by comparing the first evaluation score and the second evaluation score.

[0013] The influencing factors include offshore wind turbine performance parameters, offshore wind turbine economic indicators, offshore wind turbine reliability indicators, and related waterway safety indicators.

[0014] The specific process of obtaining and analyzing the influencing factors for the selection of offshore wind turbines is as follows: Obtain the influencing factors for the selection of offshore wind turbines, establish a correlation matrix between the variables of each influencing factor, and identify the principal components to determine the main and important influencing factors and classify them according to the degree of correlation.

[0015] The specific process of setting and optimizing the parameters of the conditional probability table based on the simulated annealing algorithm is as follows: Select the optimal type sample data of the wind turbine according to information retrieval and expert data, and update to obtain the optimized parameter settings of the conditional probability table based on the simulated annealing algorithm with the sample data as the annealing target.

[0016] Step S3 includes the following steps:

[0017] S31, Set the initial temperature and the initial value of the conditional probability table;

[0018] S32, Calculate the initial solution based on the initial temperature and the initial value of the conditional probability table;

[0019] S33, Lower the temperature. Generate a new parameter value of the conditional probability table based on the simulated annealing algorithm according to the current temperature and the current parameter solution, and give a new solution based on the new conditional probability table;

[0020] S34, Calculate the energy function according to the new solution and judge whether to accept the new solution according to the optimization objective. If the new solution is accepted, return to step S33 for iterative repetition. Otherwise, judge whether the search is sufficient. If the search is sufficient, execute step S35. Otherwise, return to step S33 for iterative repetition;

[0021] S35, Judge whether the current temperature has dropped to the target temperature. If the current temperature has dropped to the target temperature, take the current obtained conditional probability table as the optimized conditional probability table. Otherwise, return to step S33 for iterative repetition.

[0022] The setting rule of the initial value of the conditional probability table is as follows: The conditional probability table represents the probability relationship between the sum ζ of the evaluation levels of the influencing factors and the evaluation levels. Among them, when the number of influencing factors is m and the number of evaluation levels is n (n>2), the sum ζ of the evaluation levels has a total of m(n - 1) + 1 values, which are m, m + 1, …, nm. When ζ = m, the conditional probability sequence corresponding to the evaluation level sequence (1, 2, 3, …, m) is (1, 0, 0, …, 0). When ζ = nm, the conditional probability sequence corresponding to the evaluation level sequence (1, 2, 3, …, m) is (0, 0, 0, …, 1). The conditional probability sequence when m < ζ < nm is determined according to the conditional probability transfer rule.

[0023] The conditional probability transfer rule is that each transfer simultaneously satisfies the following conditions:

[0024] a) The probability value of each transfer is 1 / m;

[0025] b) Only one probability corresponding to an evaluation level is transferred at a time, and it can only be transferred to an adjacent evaluation level;

[0026] c) Both the conditional probability sequences before and after the transition are unimodal sequences;

[0027] d) Except for the first and last evaluation levels, the probability difference between two adjacent evaluation levels in the middle is no greater than (n-2) / m.

[0028] The evaluation level is set to 4 levels.

[0029] In step S33, through an iterative process based on the simulated annealing algorithm, the parameters of the conditional probability table are adjusted according to the set state transition equation and the current temperature. Specifically, let the current temperature be τ, and according to the conditional probability table, the probabilities of the current inference target corresponding to the multi-level evaluation level are p1, p2, p3…p n Satisfying p1+p2+p3+…+p n If the value is 1, then the state transition method is as follows:

[0030] make

[0031]

[0032] q′ j =q j +r×τ

[0033] Where r is a random number whose value depends on the initial temperature, when q′ j <0 or q′ j When the value of r is greater than 1, r is randomly re-evaluated and q′ is recalculated. j until 0≤q′ j ≤1;

[0034] {q′ j Sort the values ​​in ascending order to get the ascending sorted array {q′1,q′2,q′3,…q′}. n}, then {p1,p2,p3,…,p n}Transfer to

[0035]

[0036] In step S34, the sum of the absolute values ​​of the differences between each inference result obtained from the current conditional probability table and the expected corresponding inference result in the sample data is used as the energy function. When the energy function value E′ of the new solution is less than the energy function value E of the old solution, the new solution is accepted and the old solution is updated to the new solution; otherwise, the optimized Metropolis criterion is used to accept the new solution, that is, the new solution is accepted when the following probability P(τ) is greater than a random number within the preset range of 0-1:

[0037]

[0038] Where τ is the current temperature and a is a random number.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. This invention utilizes the simulated annealing algorithm to globally optimize the conditional probability table, thereby improving the adaptability of the conditional probability table and making it more consistent with actual observation data.

[0041] 2. The invention has good flexibility and is suitable for comprehensive analysis of complex decision-making environments and diverse influencing factors. Its feasibility has been verified in actual cases of offshore wind turbine selection.

[0042] 3. Based on the fast convergence of the simulated annealing algorithm, this invention can obtain the required conditional probability table faster and better when the amount of sample data is relatively scarce. Attached Figure Description

[0043] Figure 1 The flowchart shows the offshore wind turbine selection method based on fuzzy Bayesian network decision-making according to the present invention.

[0044] Figure 2 This is a flowchart for setting and optimizing the parameters of a conditional probability table. Detailed Implementation

[0045] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0046] This embodiment provides a method for selecting offshore wind turbines based on fuzzy Bayesian network decision-making. First, influencing factors are identified: by identifying influencing factors and decision rules, a three-layer Bayesian decision framework is obtained by combining expert knowledge and principal component analysis, thus fuzzifying the input and output variables of the influencing factors. Then, Bayesian network inference is performed: based on simulated annealing and data retrieval, the parameters of the conditional probability table are set and optimized, enabling the Bayesian network to infer the influencing factors for offshore wind turbine selection. Finally, optimal scheme comparison is conducted: based on expert knowledge and fuzzy logic, the inference results are defuzzified to obtain a first wind turbine evaluation score, and based on the default utility value, the inference results are defuzzified again to obtain a second wind turbine evaluation score. Finally, the optimal wind turbine selection is obtained through comparison.

[0047] Specifically, such as Figure 1 As shown, it includes the following steps:

[0048] S1. Obtain and analyze the influencing factors of offshore wind turbine selection, and construct a three-layer Bayesian structure network based on the analysis results.

[0049] Before embarking on the specific implementation process of offshore wind turbine selection, data preparation is essential. This includes collecting data on various parameters and influencing factors related to offshore wind turbine selection, such as offshore wind turbine performance parameters, offshore wind turbine economic indicators, offshore wind turbine reliability indicators, and relevant waterway safety indicators.

[0050] After collecting sufficient data, it is necessary to perform certain feature analysis on the data. In this embodiment, principal component analysis is used to establish a correlation matrix between various influencing factor variables and identify their principal components, so as to determine the main and important influencing factors and classify them according to their correlation. Then, a three-layer Bayesian structure network is established based on the analysis results.

[0051] S2, fuzzify the input for each influencing factor variable.

[0052] In this embodiment, the fuzzification method of triangular fuzzy numbers is used to transform specific variable values ​​into linguistic evaluations and their membership degrees, making the evaluation of variables more flexible and operable.

[0053] S3, parameter setting and optimization of the conditional probability table based on simulated annealing algorithm.

[0054] This embodiment uses an extended IF-THEN confidence rule to obtain the conditional probability table required for inference, and optimizes the parameters of the conditional probability table based on the simulated annealing algorithm. Specifically, it selects optimal wind turbine model sample data based on information retrieval and expert data, and updates the optimized conditional probability table parameters using the sample data as the annealing target based on the simulated annealing algorithm. For example... Figure 2 As shown, it includes the following steps:

[0055] S31, Set the initial temperature and initial values ​​for the conditional probability table;

[0056] S32, calculate the initial solution based on the initial temperature and the initial values ​​of the conditional probability table;

[0057] S33, Cooling down, based on the current temperature and current parameter solution, a new conditional probability table parameter value is generated based on the simulated annealing algorithm, and a new solution is given based on the new conditional probability table;

[0058] S34. Calculate the energy function based on the new solution and determine whether to accept the new solution based on the optimization objective. If the new solution is accepted, return to step S33 for repeated iteration. Otherwise, determine whether the search is sufficient. If the search is sufficient, proceed to step S35. Otherwise, return to step S33 for repeated iteration.

[0059] In S35, it is determined whether the current temperature has dropped to the target temperature. If the current temperature has dropped to the target temperature, the currently obtained conditional probability table is used as the optimized conditional probability table; otherwise, return to step S33 for repeated iteration.

[0060] In step S31, the setting rule for the initial value of the conditional probability table is as follows: The conditional probability table represents the probability relationship between the sum ζ of the evaluation levels of influencing factors and the evaluation levels. Among them, when the number of influencing factors is m and the number of evaluation levels is n (n > 2), the sum ζ of the evaluation levels has a total of m(n - 1) + 1 values, which are m, m + 1,..., nm. When ζ = m, the conditional probability sequence corresponding to the evaluation level sequence (1, 2, 3,..., m) is (1, 0, 0,..., 0). When ζ = nm, the conditional probability sequence corresponding to the evaluation level sequence (1, 2, 3,..., m) is (0, 0, 0,..., 1). When m < ζ < nm, the conditional probability sequence is determined according to the conditional probability transition rule.

[0061] In this embodiment, the conditional probability transition rule satisfies the following conditions simultaneously for each transfer:

[0062] a) The probability value for each transfer is 1 / m;

[0063] b) Only the probability corresponding to one evaluation level is transferred each time and can only be transferred to an adjacent evaluation level;

[0064] c) The conditional probability sequences before and after the transfer are both unimodal sequences;

[0065] d) Except for the first and last evaluation levels, the probability difference between two adjacent middle evaluation levels is not greater than (n - 2) / m.

[0066] Preferably, the evaluation level is set to 4 levels in this embodiment.

[0067] Suppose there are 2 influencing factors, and each influencing factor has 4 levels of evaluation from poor to excellent. The evaluation of this unit obtained by reasoning is also 4 levels of evaluation from 1 to 4. Then, according to the sum ζ of the evaluation levels of the two given influencing factors, ζ has a total of 7 values, and the initial conditional probability table is shown in Table 1 below.

[0068] Table 1 Initial conditional probability table when m = 2 and n = 4

[0069]

[0070]

[0071] Suppose there are 3 influencing factors, and there are also 4 levels of evaluation from 1 to 4. Then, according to the sum ζ of the evaluation levels of the three influencing factors, ζ has a total of 10 values, and the initial conditional probability table is shown in Table 2 below.

[0072] Table 2 Initial conditional probabilities when m=3, n=4

[0073]

[0074] When the number of influencing factors exceeds 3, the same principle applies, and this embodiment will not be elaborated further here.

[0075] In step S33, through an iterative process based on the simulated annealing algorithm, the parameters of the conditional probability table are adjusted according to the set state transition equation and the current temperature. Specifically, let the current temperature be τ, and according to the conditional probability table, the probabilities of the current inference target corresponding to the multi-level evaluation level are p1, p2, p3…p n Satisfying p1+p2+p3+…+p n If the value is 1, then the state transition method is as follows:

[0076] make

[0077]

[0078] q′ j =q j +r×τ

[0079] Where r is a random number whose value depends on the initial temperature, when q′ j <0 or q′ j When the value of r is greater than 1, r is randomly re-evaluated and q′ is recalculated. k until 0≤q′ j ≤1.

[0080] {q′ j Sort the values ​​in ascending order to get the ascending sorted array {q′, q′2, q′3, ..., q′}. n}, then {p1,p2,p3,…,p n}Transfer to

[0081]

[0082] Taking an evaluation level of 4 as an example, if the current temperature is τ, and the probabilities of the current reasoning target corresponding to the fourth evaluation level are p1, p2, p3, and p4 (satisfying p1+p2+p3+p4=1), then the state transition method is as follows:

[0083] make

[0084]

[0085] q′ j =q j +r×τ

[0086] {q′ jSort the values ​​in ascending order to get the ascending sorted {q′1,q′2,q′3,q′4}. Then {p1,p2,p3,p4} is transformed into...

[0087]

[0088] In this embodiment, if the initial value of τ is set to τ0 at the beginning of the algorithm, the random number r will be set as follows:

[0089]

[0090] Where r0 is a random real number between -1 and 1, excluding 0.

[0091] In step S34, the energy function is used to evaluate the fit between the conditional probability table and the actual observed data. Simultaneously, the algorithm allows for the acceptance of degraded solutions with a certain probability to avoid getting trapped in local optima. Specifically, the energy function is the sum of the absolute values ​​of the differences between each inference result obtained from the current conditional probability table and the expected corresponding inference result in the sample data. When the energy function value E′ of the new solution is less than the energy function value E of the old solution, i.e., ΔE=E′=E<0, the new solution is accepted, and the old solution is updated to the new solution. Otherwise, the optimized Metropolis criterion is used to accept the new solution, i.e., the new solution is accepted when the following probability P(τ) is greater than the preset random number rand:

[0092]

[0093] Where τ is the current temperature, a is a random number (its range is related to the specific data of the instance), and rand∈(0,1).

[0094] S4. Based on the optimized conditional probability table, a three-layer Bayesian structure network is used to infer the influencing factors of offshore wind turbine selection.

[0095] The process of reasoning based on Bayesian network structures is a conventional setup in this field, and will not be elaborated here to avoid obscuring the purpose of this application.

[0096] S5, based on expert knowledge and default utility value respectively, defuzzify the reasoning results to obtain the first evaluation score Q1 and the second evaluation score Q2.

[0097] Specifically, for the evaluation score Q1 obtained based on expert knowledge, this embodiment comprehensively evaluates the scores given by experts at each level of evaluation to ultimately obtain the corresponding utility value of each level of evaluation. Specifically, first, the given experts at each level are assigned weights, and then the overall judgment is obtained by the experts' judgments on a specific evaluation level. For example, if there are n experts, then the i-th expert is assigned a relative weight β. i ,satisfy and β i>0. And according to experts, the linguistic description of a specific variable should be x. i =(a i ,b i ,c i Therefore, the linguistic description of the triangular fuzzy number A = (a, b, c) in expert evaluation can be defined as the equation: After obtaining (a, b, c), the center-of-gravity method is used to obtain the utility value corresponding to this evaluation level. The expert knowledge assessment score Q1 is obtained by multiplying the utility values ​​at each level by the corresponding probabilities of the final evaluation at each level and summing the results. For the score Q2 obtained based on the default utility value, when taking the four-level evaluation standard, if the probabilities of the four levels of the final result are x1, x2, x3, and x4, then the default utility values ​​are generally taken as (0.35, 0.55, 0.85, 1.00) based on existing experience. Thus, Q2 = 0.35*x1 + 0.55*x2 + 0.85*x3 + 1.00*x4.

[0098] S6. The best offshore wind turbine is selected by comparing the first evaluation score Q1 and the second evaluation score Q2.

[0099] After obtaining Q1 and Q2, rank them according to Q1 as the primary keyword and Q2 as the secondary keyword, and select the one with the highest score as the optimal offshore wind turbine.

[0100] For example, in the following case of selecting a wind turbine WTG1-6, the input variables are as follows:

[0101]

[0102]

[0103] After being upgraded to Level 4 evaluation, the indicators for each wind turbine are as follows:

[0104]

[0105] The final evaluation and its score were obtained by taking the average:

[0106]

[0107] Therefore, the WTG-6 fan was ultimately selected.

[0108] The final evaluation and its score obtained by analyzing the method of this invention are as follows:

[0109]

[0110] Therefore, the WTG-5 fan was ultimately selected.

[0111] Based on the actual situation analysis, in the actual project of this case, after expert analysis and discussion, it was finally determined that the optimal model for the project was WTG-5 instead of WTG-6, which is consistent with the result obtained by the method of the present invention, thus demonstrating the effectiveness of the method of the present invention.

[0112] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for selecting offshore wind turbines based on fuzzy Bayesian network decision-making, characterized in that, It includes the following steps: S1. Obtain the influencing factors for the selection of offshore wind turbines and analyze them, and construct a three-layer Bayesian structure network based on the analysis results; S2. Perform input fuzzification on each influencing factor variable; S3. Based on the simulated annealing algorithm, perform parameter setting and optimization of the conditional probability table, including the following steps: S31. Set the initial temperature and the initial value of the conditional probability table; S32. Calculate the initial solution based on the initial temperature and the initial value of the conditional probability table; S33. Decrease the temperature. Based on the current temperature and the current parameter solution, generate a new parameter value of the conditional probability table according to the simulated annealing algorithm, and give a new solution based on the new conditional probability table; S34. Calculate the energy function according to the new solution and judge whether to accept the new solution according to the optimization goal. If the new solution is accepted, return to step S33 for iterative repetition. Otherwise, judge whether the search is sufficient. If the search is sufficient, execute step S35. Otherwise, return to step S33 for iterative repetition; S35. Judge whether the current temperature has decreased to the target temperature. If the current temperature has decreased to the target temperature, use the current obtained conditional probability table as the optimized conditional probability table. Otherwise, return to step S33 for iterative repetition; The setting rule of the initial value of the conditional probability table is as follows: The conditional probability table represents the probability relationship between the sum ζ of the evaluation levels of the influencing factors and the evaluation levels. Among them, when the number of influencing factors is m and the number of evaluation levels is n (n>2), the sum ζ of the evaluation levels has a total of m(n - 1)+1 values, which are m, m + 1, …, nm respectively. When ζ = m, the conditional probability sequence corresponding to the evaluation level sequence (1, 2, 3, …, m) is (1, 0, 0, …, 0). When ζ = nm, the conditional probability sequence corresponding to the evaluation level sequence (1, 2, 3, …, m) is (0, 0, 0, …, 1). The conditional probability sequence when m < ζ < nm is determined according to the conditional probability transfer rule; The conditional probability transfer rule satisfies the following conditions simultaneously for each transfer: a) The probability value of each transfer is 1 / m; b) Only transfer the probability corresponding to one evaluation level each time and can only transfer to adjacent evaluation levels; c) The conditional probability sequences before and after the transfer are both unimodal sequences; d) Except for the head and tail evaluation levels, the probability difference between two adjacent evaluation levels in the middle is not greater than (n - 2) / m; S4. Based on the optimized conditional probability table, use the three-layer Bayesian structure network to perform inference on the influencing factors for the selection of offshore wind turbines; S5. Perform defuzzification on the inference results based on expert knowledge and default utility values respectively to obtain the first evaluation score and the second evaluation score; S6. Compare and select based on the first evaluation score and the second evaluation score to obtain the best selection of offshore wind turbines.

2. The method for selecting offshore wind turbines based on fuzzy Bayesian network decision-making as described in claim 1, characterized in that, The influencing factors include the performance parameters of offshore wind turbines, the economic indexes of offshore wind turbines, the reliability indexes of offshore wind turbines, and the relevant channel safety indexes.

3. The method for selecting offshore wind turbines based on fuzzy Bayesian network decision-making as described in claim 1, characterized in that, The specific operation of obtaining the influencing factors for the selection of offshore wind turbines and analyzing them is as follows: Obtain the influencing factors for the selection of offshore wind turbines, establish a correlation matrix between the influencing factor variables and identify the main components to determine the main and important influencing factors and classify them according to the correlation.

4. The method for selecting offshore wind turbines based on fuzzy Bayesian network decision-making as described in claim 1, characterized in that, The parameter setting and optimization of the conditional probability table based on the simulated annealing algorithm is as follows: Select the optimal sample data of the wind turbine based on information retrieval and expert data, and update the optimized conditional probability table parameter settings based on the sample data with the annealing target using the simulated annealing algorithm.

5. The method for selecting offshore wind turbines based on fuzzy Bayesian network decision-making as described in claim 1, characterized in that, The evaluation level is set to 4 levels.

6. The method for selecting offshore wind turbines based on fuzzy Bayesian network decision-making as described in claim 1, characterized in that, In step S33, through an iterative process based on the simulated annealing algorithm, the parameters of the conditional probability table are adjusted according to the set state transition equation and the current temperature. Specifically, let the current temperature be τ, and according to the conditional probability table, the probabilities of the current inference target corresponding to the multi-level evaluation level are as follows: , , … ,satisfy The state transition method is as follows: make in It is a random number whose value depends on the initial temperature. or Time Randomly select new values ​​and recalculate. until ; Will Sort by value in ascending order to obtain ascending sort. ,but Transfer to 。 7. The method for selecting offshore wind turbines based on fuzzy Bayesian network decision-making as described in claim 1, characterized in that, In step S34, the sum of the absolute values ​​of the differences between each inference result obtained from the current conditional probability table and the expected corresponding inference result in the sample data is used as the energy function. When the energy function value of the new solution... Less than the energy function value of the old solution If a new solution is accepted, the old solution is updated to the new solution; otherwise, the optimized Metropolis criterion is used to accept the new solution, i.e., if the following probability... A new solution will be accepted if the result is a random number greater than the preset range of 0-1. Where τ is the current temperature. It is a random number.

Images