Multi-objective decision-making technology for equity investment of provincial management industry of power grid

Abstract

The invention relates to the technical field of multi-objective decision-making, and discloses a power grid provincial management industry stock right investment multi-objective decision-making technology, which constructs an incremental power distribution network investment risk assessment index system considering a planning risk, a management risk, a market risk and an economical efficiency risk, constructs an incremental power distribution network investment risk assessment model based on MGARA, and provides a power grid provincial management industry stock right investment multi-objective decision-making method. Constructing an investment capability prediction model based on deep learning and an optimization algorithm; according to the model, an appropriate constraint condition and an objective function are set, a benefit function of multiple subjects is combined, an equilibrium solution is deduced, multi-subject game optimization of a genetic algorithm nested particle swarm optimization algorithm is adopted, and model solving is carried out, so that an optimal investment strategy is obtained, and optimal collaboration and benefit balance among the subjects are achieved; the overall optimal investment decision of the system is achieved; according to the method, collaboration and benefit maximization among different subjects can be realized, and an innovative solution is provided for stock right investment, resource allocation and risk management of the power grid.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-objective decision-making for power grid equity investment, and more specifically, to a multi-objective decision-making technology for power grid provincial industry equity investment. Background Art

[0002] With the continuous growth of electricity demand and the transformation of energy structure, power grid investment and operation are facing more complex challenges. In particular, the construction and investment of incremental distribution networks involve the collaboration and decision-making of multiple parties, including traditional power grid operators, microgrid systems, energy storage service providers and other roles. These parties often have different goals and risk tolerance when facing market changes, technological advances and policy guidance. How to coordinate the interests of all parties and optimize investment decisions has become a key issue that needs to be solved in the current power grid investment field.

[0003] To this end, we propose a multi-objective decision-making technique for equity investment in provincially managed power grid industries. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-objective decision-making technology for equity investment in power grid provincial management industries to overcome the technical problems existing in the prior art.

[0005] In order to achieve the above technical objectives and the above technical effects, the present invention provides the following technical solutions:

[0006] A multi-objective decision-making technology for power grid equity investment based on multi-agent game includes the following steps:

[0007] Step 1: Use the load forecast results as input indicators to build an incremental distribution network investment risk assessment indicator system that considers planning risk, management risk, market risk, and economic risk;

[0008] Step 2: The subjective and objective weights are calculated using the order relationship method and the coefficient of variation method, and the combined weight optimization model is solved based on the QIPSO algorithm. Finally, an incremental distribution network investment risk assessment model based on MGARA is constructed.

[0009] Step 3: Systematically identify factors affecting investment capacity and use correlation analysis to determine the correlation between investment capacity and each key influencing factor. Combined with the selected main influencing factors, an investment capacity prediction model based on deep learning and optimization algorithms is constructed;

[0010] Step 4: Use game theory as a framework to consider the strategic choices and game behaviors of different entities when facing investment choices. The model sets appropriate constraints and objective functions, combines the benefit functions of multiple entities, and derives the equilibrium solution;

[0011] Step 5: Use the multi-agent game optimization of genetic algorithm nested particle swarm algorithm to solve the model, so as to obtain the optimal investment strategy, achieve the best coordination and interest balance among agents, and reach the optimal investment decision of the overall system.

[0012] Preferably, in a multi-objective decision-making technology for equity investment in power grid provincial management industries, in Step 1, the load forecasting result is used as an input index, and an investment risk assessment index system for incremental distribution networks considering planning risk, management risk, market risk, and economic risk is constructed.

[0013] In a market-oriented environment, power grid companies investing in incremental distribution network parks face risks in many aspects such as social economy, policies, management, and the market. These factors run through the entire process of incremental distribution network investment. Therefore, planning risk, management risk, market risk, and economic risk are selected as first-level indicators to construct an evaluation system for power grid companies' investment in incremental distribution networks considering various risk factors.

[0014] Preferably, in a multi-objective decision-making technology for equity investment in power grid provincial management industries, in Step 2, the order relation method and the coefficient of variation method are used to calculate the subjective and objective weights, the combined weight optimization model is solved based on the QIPSO algorithm, and finally an investment risk assessment model for incremental distribution networks based on MGARA is constructed.

[0015] The order relation method is a new method that does not require consistency testing. It is a subjective weighting method that first qualitatively ranks the evaluation indicators, then makes a rational judgment on adjacent indicators, and finally conducts quantitative calculations. The order relation method does not require the construction of a judgment matrix, nor does it require consistency testing. Compared with constructing an AHP judgment matrix, the calculation amount is reduced by several times, and there is no limit to the number of schemes. Since the order relation given fully expresses the will of experts, its results are completely trustworthy. This method is simple to calculate, has good order preservation, and strong applicability.

[0016] The coefficient of variation method is a statistical index commonly used in statistics to measure data differences. This method assigns weights to each indicator according to the degree of variation of the observed values of each indicator on all evaluated objects. The basic principle of the coefficient of variation method is that the greater the degree of variation of an indicator, the greater its impact on the comprehensive evaluation, and the size of the weight reflects the discrimination ability of the indicator. However, it cannot reflect the independence of the indicator and the understanding of the importance of the indicator by the evaluator, lacking comprehensiveness and scientificity.

[0017] The PSO algorithm is a stochastic search algorithm based on group cooperation. In the process of each iterative search of the random particles in the group, the optimal solution is found by tracking two "extreme values". Suppose the size of a certain population is N, and in the D-dimensional search space, the position of particle i in the t-th generation is The speed is The individual historical optimal position is The global optimal position of the t-th generation Then the optimal solution formula is as follows:

[0018]

[0019] In the formula: c 1 and c 2 are learning factors; r 1 and r 2 are random numbers uniformly distributed between [0, 1]; k is the current iteration number, I tera is the maximum iteration number, ω is the inertia weight, which linearly decreases from the maximum ω max to ω min .

[0020] Currently, the particle swarm optimization algorithm has been applied in many fields. However, when dealing with complex models, the algorithm is prone to falling into the local optimal problem, that is, the "premature problem". Aiming at the obvious deficiencies of the particle swarm optimization algorithm in dealing with complex problems and avoiding the algorithm falling into the local optimum during the iteration process, a quadratic interpolation factor is introduced to improve it, and the QIPSO algorithm is proposed. This improvement improves the optimization speed of the algorithm, enhances the operation efficiency of the algorithm and the global optimization ability, thus improving the rationality of the algorithm in risk assessment. The essence of the QIPSO algorithm is to introduce a quadratic interpolation operator into the particle swarm optimization algorithm to improve the convergence speed and calculation efficiency of the algorithm, and then find the global optimal position of the model.

[0021] The risk assessment model based on MGARA is a comprehensive risk assessment method that combines multi-objective optimization and grey relational analysis, aiming to systematically evaluate and optimize decisions on risk factors in a complex investment environment. Through the multi-objective optimization framework, this method comprehensively considers the interrelationships of multiple risk factors, and uses grey relational analysis technology to conduct sensitivity analysis and weight assignment for each risk factor, so as to achieve accurate risk assessment.

[0022] Preferably, in a multi-objective decision-making technology for equity investment in power grid provincial management industries, in step three, the system identifies the influencing factors of investment ability, uses the correlation analysis method to judge the correlation degree between investment ability and each key influencing factor, and combines the selected main influencing factors to construct an investment ability prediction model based on deep learning and optimization algorithms.

[0023] Identification of influencing factors of investment ability. The quantifiable indicators of the power grid investment ability include: the accuracy of power grid load forecasting, measured by the deviation rate between the forecasted load and the actual load; the growth rate of market demand, expressed as the annual growth rate percentage of electricity demand; the impact of power market reform policies, evaluated by the amplitude of electricity price adjustment or the marketization policy impact index; the technical level of the power grid, measured by the smart grid coverage rate, the energy storage system capacity, and the proportion of new technology applications; the financial health status, including the capital adequacy ratio, the debt ratio, and the return on equity (ROE); the capital market financing environment, evaluated by the financing cost (such as interest rate) and the stock market volatility; the government subsidy and policy support intensity, quantified by the annual government subsidy amount or the policy support intensity index; the power grid operation efficiency, reflected by the power loss rate and the power supply reliability index; the social and economic development level, measured by the annual GDP growth rate or the per capita GDP; and the power grid asset structure and return on investment, evaluated by the debt-to-asset ratio and the return on investment. Through these quantifiable indicators, the investment ability and risks of the power grid can be comprehensively analyzed.

[0024] Screening of key factors of investment ability. In the process of evaluating the power grid investment ability, using the correlation analysis method to screen key factors is an effective technical means. Correlation analysis can help us identify the factors highly correlated with the power grid investment ability, so as to make accurate investment decisions.

[0025] The main purpose of correlation analysis (CorrelationAnalysis) is to study the degree of closeness of the relationship between variables, such as the relationship between height and weight, the amount of wire and the amount of tower materials, the capacity of main transformers and the distribution device, etc. In statistical analysis, correlation generally refers to "linear correlation", and its closeness is expressed by the correlation coefficient. The correlation coefficient is usually denoted as, and its value ranges from -1 to +1. The closer the absolute value is to 1, the closer the relationship between variables. When the absolute value is equal to 1, it means that the two variables are completely correlated, and the value of variable B can be obtained from the value of variable A. When the correlation coefficient is positive, it means that when variable A increases, variable B also increases, and the two are in a positive correlation relationship; on the contrary, it means that when variable A increases, variable B decreases, and the two are in a negative correlation relationship. This patent uses the Pearson correlation coefficient for correlation analysis.

[0026] Construction of investment ability prediction model. Build a deep belief network model (DBN), and apply the grey wolf algorithm (GWO) to optimize the magic core parameters; form a training set with several groups of preprocessed sample data to obtain the optimal carbon emissions. Use the improved grey wolf algorithm to optimize the DBN network connection weights, improve the classification performance of the network, and have a better fitting effect.

[0027] As a derivative model of a multi-layer neural training network, the deep learning model is different in that it abstracts the low-level features of model data, excavates the internal distribution features of the data, and uses fewer training data samples to obtain the essential features of the data. At the same time, the deep learning model well inherits the robust characteristics of the neural network training model and has the computing ability to process complex functions under the condition of considering fewer data samples.

[0028] The deep learning model mainly organically integrates RBM and adaptive intelligent algorithms. Its training idea: 1. Extract the underlying data feature quantities of the deep learning model for the input variables of the top-level learning of the model design. At the same time, adopt a mode of training layer by layer from the bottom layer to the top layer of the model; 2. After training to the top layer of the model, then use the adaptive particle swarm algorithm to adaptively optimize and adjust the entire training network to ensure that the model training result can jump out of the local solution.

[0029] Preferably, in a multi-objective decision-making technology for equity investment in the provincial-owned industries of the power grid, in step four, game theory is used as a framework to consider the strategic choices and game behaviors of different subjects when facing investment choices. The model derives the equilibrium solution by setting appropriate constraint conditions and objective functions and combining the benefit functions of multiple parties.

[0030] The game theory model derives the optimal investment strategy by analyzing the strategic choices and game behaviors of different subjects when facing investment decisions. In order to effectively implement the game theory model, game theory equilibrium analysis is required, including factors such as the type of game, the strategies of participants, the objective function, and the constraint conditions.

[0031] The participants in the game include three subjects:

[0032] Leader (distribution network): The decision maker with the active investment decision-making power.

[0033] Follower 1 (microgrid system): A coalition composed of multiple microgrid systems, whose decisions depend on the decisions of the leader.

[0034] Follower 2 (shared energy storage operator): Conduct energy storage system investment and operation according to the decisions of the leader and follower 1.

[0035] Objective function. In the multi-agent game model, the objective function is usually the benefit function of each subject in the investment decision. For the leader and the follower, their respective objective functions can be expressed as:

[0036] Leader's objective function: Maximize the total return on investment of the distribution network, that is, optimize the investment structure, improve the power grid efficiency, and ensure power supply stability.

[0037] Follower 1 objective function: Maximize the return on investment of the microgrid system, taking into account the synergy between energy storage facilities and distributed energy sources to achieve optimal economic benefits.

[0038] Follower 2 objective function: Maximize the benefits of the shared energy storage operator, mainly focusing on energy storage scheduling optimization to reduce costs and improve profitability.

[0039] Constraints. The constraints in the model include resource constraints, technical constraints, financial constraints, etc. Among them, the investment capacity is an important constraint condition to ensure the feasibility and rationality of investment decisions. The specific constraints are as follows:

[0040] Investment capacity constraint: The investment decision of each entity must be within its investment capacity to ensure that the investment amount does not exceed its affordable financial burden and capital sources. For example, the investment of the power grid enterprise cannot exceed its financing capacity or capital limit, and the microgrid system and the shared energy storage operator must also follow similar financial constraints.

[0041] Technical constraints: The technical feasibility of the power grid and energy storage systems to ensure that the selected investment plan can be achieved at the technical level, such as the capacity of the smart grid and energy storage equipment.

[0042] Market constraints: Consider factors such as the supply-demand balance and market fluctuations in the power market to ensure that investment decisions meet market demand and the stability of the power system.

[0043] Policy constraints: The decisions of each entity need to comply with the government's policies and regulations, such as the price policy of the power market and the use of government subsidies.

[0044] Preferably, in a multi-objective decision-making technology for equity investment in the provincial-level management industry of the power grid, in step five, a multi-agent game optimization using a genetic algorithm nested particle swarm algorithm is adopted to solve the model to obtain the optimal investment strategy, achieve the best synergy and interest balance among entities, and reach the optimal investment decision for the overall system.

[0045] The solution algorithm first solves the internal multi-objective optimization model of Follower 1 through the genetic algorithm (NSGA-II). Since the objective function of Follower 1 mainly aims at minimizing values, the decision-making theory of the right-skewed fuzzy set is adopted to construct the fuzzy membership function. By calculating the membership values of each optimal solution on the Pareto front, the solution with the highest satisfaction degree is finally selected as the optimal compromise solution. Next, the optimization result of the genetic algorithm is substituted into a one-leader-multiple-followers Stackelberg game model. In this model, the decisions of the leader, Follower 1, and Follower 2 are adjusted based on game theory. The decision-making strategy of the leader is optimized through the particle swarm optimization (PSO) algorithm, and the CPLEX solver is used to solve the decisions of Follower 1 and Follower 2. Finally, the game equilibrium solution is obtained through the benefit functions of multiple parties. The whole process ensures that the decisions among all parties can achieve optimal coordination and interest balance within the framework of the game.

[0046] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0047] The present invention combines the multi-agent game theory with the multi-objective optimization algorithm. By comprehensively considering planning risks, management risks, market risks, and economic risks, an investment risk assessment model for the incremental distribution network is constructed. Innovatively, deep learning is combined with the optimization algorithm for investment capacity prediction, and the nested optimization of the genetic algorithm and the particle swarm optimization algorithm is used to improve the decision-making accuracy, ensuring the interest balance of multiple parties. Finally, the optimal coordination and systematic optimization of the power grid investment decision-making are realized;

[0048] The present invention constructs an investment risk assessment and investment capacity prediction model for the incremental distribution network based on multi-agent game optimization, which not only effectively improves the scientificity and accuracy of the decision-making for the incremental distribution network investment, but also through the multi-agent game optimization model, can achieve the coordination and interest maximization among different parties, providing an innovative solution for the equity investment, resource allocation, and risk management of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for use in the description of the specific embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0050] Figure 1 is the overall flowchart of the present invention;

[0051] Figure 2 is the index schematic diagram of the investment risk assessment index system for the incremental distribution network;

[0052] Figure 3It is a flow chart for solving the game optimization model. Specific implementation manners

[0053] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts belong to the scope of protection of the present invention.

[0054] This embodiment is a multi-objective decision-making technology for power grid equity investment based on multi-agent game, including the following steps:

[0055] Step S1: Use the load forecasting result as an input index to construct an incremental distribution network investment risk assessment index system considering planning risk, management risk, market risk, and economic risk.

[0056] As Figure 2 shown, in a market-oriented environment, the power grid company investing in an incremental distribution network park faces risks in multiple aspects such as social economy, policy, management, and market. These factors run through the entire process of incremental distribution network investment. Therefore, select planning risk, management risk, market risk, and economic risk as the first-level indicators to construct an evaluation system for the power grid company's investment in incremental distribution networks considering multiple risk factors.

[0057] Step two: Use the order relation method and the coefficient of variation method to calculate the subjective and objective weights, solve the combined weight optimization model based on the QIPSO algorithm, and finally construct an incremental distribution network investment risk assessment model based on MGARA.

[0058] 1. Order relation method

[0059] The order relation method is a new method that does not require consistency testing. It is a subjective weighting method that first qualitatively ranks the evaluation indicators, then makes a rational judgment on adjacent indicators, and finally conducts quantitative calculations. The order relation method does not require constructing a judgment matrix, nor does it require consistency testing. Compared with constructing an AHP judgment matrix, the calculation amount is reduced by several times, and there is no limit to the number of schemes. Since the order relation given fully expresses the will of the experts, its results are completely trustworthy. This method is simple to calculate, has good order preservation, and strong applicability.

[0060] The specific steps of the order relation method are as follows:

[0061] (1) Determine the order relation

[0062] If the importance degree of evaluation index x i relative to a certain evaluation criterion is greater than (or not less than) x j , then it is denoted as

[0063] For the evaluation index set {x 1 , x 2 , …, x m}, the order relation can be established according to the following steps:

[0064] 1) Among the index set {x 1 , x 2 , …, x m}, the expert selects the most important index and marks it as

[0065] 2) Among the remaining m - 1 indexes, the expert selects the most important index and marks it as

[0066] 3) Repeat the above steps;

[0067] 4) Among the remaining m - (k - 1) indexes, the expert selects the most important index and marks it as

[0068] In this way, an order relation is uniquely determined.

[0069]

[0070] (2) Give the comparison judgment of the relative importance between x k-1 and x k

[0071] Suppose the rational judgments of the ratio w k-1 / w k of the importance of the evaluation indexes x k-1 / w k by the expert are respectively

[0072] w k-1 / w k = r k (k = m, m - 1, …, 2)

[0073] The assignment of r k can refer to the following table

[0074] rk Description 1.0 The index xk-1 is as important as the index xk 1.2 The index xk-1 is slightly more important than the index xk 1.4 The index xk-1 is significantly more important than the index xk 1.6 The index xk-1 is strongly more important than the index xk 1.8 The index xk-1 is extremely more important than the index xk

[0075] (3) Calculation of the weight coefficient w k

[0076] If the rational assignment of r k given by the expert satisfies the relation r k-1 > 1 / r k (k = m, m - 1, …, 2), then

[0077] ​​

[0078] w k-1 = r k w k (k = m - 1, …, 2)

[0079] (4) The subjective weight obtained by the order relation method is

[0080] w' = (w' 1 , w' 2 , …, w' m )

[0081] 2. Coefficient of variation method

[0082] The coefficient of variation method is a statistical index commonly used in statistics to measure data differences. This method assigns weights to each index according to the degree of variation in the observed values of all evaluated objects for each index. The basic principle of the coefficient of variation method is that the greater the degree of variation of an index, the greater its impact on the comprehensive evaluation, and the weight size reflects the discrimination ability of the index. However, it cannot reflect the degree of independence of the index and the evaluator's understanding of the importance of the index, lacking comprehensiveness and scientificity.

[0083] The specific method is as follows:

[0084] 1) Calculate the standard deviation σ of each index k , representing the absolute degree of variation of each index, and its calculation formula is

[0085]

[0086] 2) Calculate the coefficient of variation c of each index k , which reflects the relative degree of variation of each index, and its calculation formula is

[0087]

[0088] 3) Normalize the coefficient of variation of each index to obtain the weight of each index as

[0089]

[0090] Then the weight vector of each index is obtained as

[0091] w” = (w' 1 ', w' 2 ', …, w” m )

[0092] 3. Combined weight calculation

[0093] The PSO algorithm is a stochastic search algorithm based on swarm cooperation. During each iterative search of the random particles in the swarm, the optimal solution is found by tracking two "extreme values". Suppose the size of a certain swarm is N, and in the D-dimensional search space, the position of particle i in the t-th generation is The velocity is The individual historical optimal position is The global optimal position in the t-th generation Then the optimal solution formula is:

[0094]

[0095] In the formula: c 1 , c 2 are learning factors; r 1 , r 2 are random numbers uniformly distributed between [0, 1]; k is the current iteration number, I tera is the maximum iteration number, ω is the inertia weight, and as the iteration calculation progresses, it linearly decreases from the maximum ω max to ω min .

[0096] Currently, the particle swarm algorithm has been applied in many fields. However, when dealing with complex models, this algorithm is prone to falling into the local optimal problem, that is, the "premature problem". Aiming at the obvious deficiencies of the particle swarm algorithm in dealing with complex problems and avoiding the algorithm from falling into the local optimum during the iterative process, a quadratic interpolation factor is introduced to improve it, and the QIPSO algorithm is proposed. This improvement has increased the optimization speed of the algorithm, enhanced the operation efficiency of the algorithm, and improved the global optimization ability, thereby improving the practicality of this algorithm in the optimal selection application of the transmission network reserve project library. The essence of the QIPSO algorithm is to introduce a quadratic interpolation operator into the particle swarm algorithm to increase the convergence speed and calculation efficiency of the algorithm, and then find the global optimal position of the model, that is:

[0097]

[0098] The point generated by the above formula is the minimum value point of the quadratic surface passing through the three points p l , p j and p g in the D-dimensional space. Therefore, the globally optimal position generated by iteration is always selected in the above formula. The globally optimal position in the t-th generation is p g =(p g1 , p g2 ,..., p gD ). Randomly select two positions p l , p j from the individual historical optimal positions, and i, j≠g. Among them, e is a very small positive number to make d non-zero.

[0099] Let Q i =(q i1, q i2, ..., q iD ), i = 1, 2,..., N, then we have:

[0100]

[0101] The QIPSO algorithm process is as follows.

[0102] Step 1. Set parameters such as population size n, learning factors c 1 , c 2 , maximum iteration number I tera , initial velocity v, number of indicators D, inertia weight, etc.;

[0103] Step 2. Generate an initial population, randomly generate the initial position and the initial velocity

[0104] Step 3. Calculate the fitness of the initial particles, and find the global optimal particle g best and the individual optimal particle P best ;

[0105] Step 4. Update the historical optimal position P i of the particles and the global optimal position P g ;

[0106] Step 5. Eliminate the particle with the lowest fitness, and update the velocity v and the position x of the particles;

[0107] Step 6. Update the global optimal particle and the individual optimal particle;

[0108] Step 7. Repeat Step 3 to Step 6. If the maximum iteration number I tera is reached and the convergence condition is satisfied, the algorithm optimization ends, and the combined weighting weights are output.

[0109] 4. Evaluation model construction

[0110] (1) Single-layer MGARA comprehensive evaluation

[0111]

[0112] In the formula: η ij is the correlation coefficient of the first-level indicator i of the evaluation object j; x ij is the comparison sequence value, that is, the specific value of the indicator of the first-level indicator i of the evaluation object j, is the maximum absolute difference, is the minimum absolute difference, ρ is the discrimination coefficient, S vTo compare the mean of the area related to the reference sequence.

[0113] Let the first-level index be \(x\) i The correlation coefficient matrix is \(R\) i , and the weight vector is Then the single-layer comprehensive evaluation result of the index \(x\) i | is shown in the following formula.

[0114]

[0115] (2) Multi-layer MGARA comprehensive evaluation

[0116] Taking the single-layer MGARA comprehensive evaluation result as the grey correlation matrix \(R\) of the multi-layer, and denoting the weight vector of the first-level index as The formula for the multi-layer is:

[0117]

[0118] (3) Investment strategy based on the comprehensive evaluation result

[0119] Through the comprehensive evaluation result, the risk assessment value of each incremental distribution network can be obtained. According to the investment threshold determined by expert scoring, if the risk assessment value of an incremental distribution network is greater than or equal to the threshold, it means that the risk of the project is within the controllable range and is suitable for investment; if the risk assessment value is lower than the threshold, it means that the risk of the project is relatively high and not suitable for investment. This judgment method provides a clear basis for decision-makers to make reasonable investment decisions on the premise of ensuring controllable risks.

[0120] Step 3: The system identifies the influencing factors of investment ability and uses correlation analysis to judge the correlation degree between investment ability and each key influencing factor. Combining the selected main influencing factors, an investment ability prediction model based on deep learning and optimization algorithms is constructed.

[0121] 1. Identification of influencing factors of investment ability

[0122] Quantifiable indicators of the power grid investment capacity include: the accuracy of power grid load forecasting, measured by the deviation rate between the forecasted load and the actual load; the growth rate of market demand, expressed as the annual growth rate percentage of electricity demand; the impact of power market reform policies, evaluated by the extent of electricity price adjustment or the marketization policy impact index; the technical level of the power grid, measured by the smart grid coverage rate, the energy storage system capacity, and the proportion of new technology applications; the financial health status, including the capital adequacy ratio, the debt ratio, and the return on equity (ROE); the capital market financing environment, evaluated by the financing cost (such as interest rate) and the stock market volatility; the government subsidy and policy support strength, quantified by the annual government subsidy amount or the policy support strength index; the power grid operation efficiency, reflected by the power loss rate and the power supply reliability index; the social and economic development level, measured by the annual GDP growth rate or the per capita GDP; and the power grid asset structure and return on investment, evaluated by the debt ratio and the return on investment. Through these quantifiable indicators, the investment capacity and risks of the power grid can be comprehensively analyzed.

[0123] 2. Screening of Key Factors for Investment Capacity

[0124] In the process of evaluating the power grid investment capacity, using the correlation analysis method to screen key factors is an effective technical means. Correlation analysis can help us identify factors highly correlated with the power grid investment capacity, thus making accurate investment decisions.

[0125] The main purpose of correlation analysis (CorrelationAnalysis) is to study the degree of closeness of the relationship between variables, such as the relationship between height and weight, the amount of wire and the amount of tower materials, the capacity of main transformers and the distribution device, etc. In statistical analysis, correlation generally refers to "linear correlation", and its closeness is represented by the correlation coefficient. The correlation coefficient is usually denoted as \(r\), and its value ranges from -1 to +1. The closer the absolute value is to 1, the closer the relationship between the variables. When the absolute value is equal to 1, it means that the two variables are completely correlated, and the value of variable B can be obtained given the value of variable A. When the correlation coefficient is positive, it means that when variable A increases, variable B also increases, and the two have a positive correlation; conversely, it means that when variable A increases, variable B decreases, and the two have a negative correlation. This patent uses the Pearson correlation coefficient for correlation analysis.

[0126] The Pearson simple correlation coefficient is used to measure the linear correlation relationship of interval variables and is the most widely used in calculating the correlation coefficient. Its calculation formula is:

[0127]

[0128] where is the number of samples, and are the values of two variables in different samples. Since the calculation formula of the Pearson simple correlation coefficient is exactly in the form of matrix product, it is also called the product-moment correlation coefficient. After transforming the formula, it is found that the correlation coefficient can be expressed as the product of the standardized and respectively, and then the average of the products is calculated.

[0129] The four most relevant quantifiable indicators of the power grid investment ability include the power grid load forecasting accuracy, market demand growth rate, power grid technology level, and financial health status, which comprehensively reflect the market potential, technological innovation, and financial robustness of the power grid.

[0130] 3. Construction of the investment ability prediction model

[0131] The four most relevant quantifiable indicators of the power grid investment ability (including the power grid load forecasting accuracy, market demand growth rate, power grid technology level, and financial health status) will be used as the training data of the deep belief network (DBN) model. These indicators will extract their underlying features and be used as the input variables for the top-level learning of the model design. Then, deeper intrinsic features will be extracted through layer-by-layer training, and the connection weights of the DBN network will be optimized using the improved grey wolf optimization (GWO) algorithm to improve the classification performance and fitting effect of the model. In addition, the training network will be optimized by combining the adaptive particle swarm optimization algorithm to ensure that the prediction results can avoid falling into local optimal solutions, thereby obtaining more accurate carbon emissions predictions.

[0132] The deep learning model mainly organically integrates the RBM and the adaptive intelligent algorithm. Its training idea: 1. Extract the underlying data feature quantities of the deep learning model for the input variables of the top-level learning of the model design. At the same time, adopt the mode of layer-by-layer training from the bottom layer to the top layer of the model; 2. After training to the top layer of the model, use the adaptive particle swarm optimization algorithm to perform adaptive optimization and adjustment on the entire training network to ensure that the training results of the model can jump out of the local solution.

[0133] The RBM consists of the visible layer v i and the hidden layer h i The last layer is the BP network, which is mainly used for fine-tuning the weights of the network from top to bottom.

[0134] In the RBM, according to the given (v, h), the energy function is:

[0135]

[0136] In the above formula, θ = {w, a, b} are the network parameters, w is the weight between the visible layer and the hidden layer, a and b are the biases of the visible layer and the hidden layer, and m and n are the numbers of neurons in the visible layer and the hidden layer.

[0137] According to the energy function, the following joint probability distribution function can be obtained:

[0138]

[0139] In the above formula, is expressed as a normalization factor, representing the algebraic sum of all variable energy functions.

[0140] When the v state of the visible layer is determined, the activation probability of the hidden layer unit is:

[0141]

[0142] When the h state of the hidden layer is determined, the activation probability of the visible layer unit is:

[0143]

[0144] When the number of training samples is K, the parameter θ can be determined by solving the problem of maximizing the log-likelihood function. The objective function of the problem of maximizing the log-likelihood function is given by the formula:

[0145]

[0146] In the above formula, maxL(θ) is obtained by the stochastic gradient method.

[0147] By repeating Gibbs sampling, the update rule of the RBM parameters can be obtained as the formula:

[0148] Δw ij = ε(<v i h j > data - <v i h j > recon )

[0149] Δa i = ε(<v i > data - <v i > recon )

[0150] Δb i = ε(<h j > data - <h j > recon )

[0151] In the formula, ε is the RBM learning rate, <·> data and <·> recon are the mathematical expectations of the input data and the reconstructed data respectively.

[0152] The training steps of the DBN model are as follows:

[0153] 1) Input the original data as the input layer vector into the first-layer RBM to complete unsupervised training;

[0154] 2) When the first-layer RBM has completed the feature learning of the original data, feature data will be obtained, and these feature data will be regarded as the input vector of the new layer and input into the next-layer RBM for continuous unsupervised training;

[0155] 3) Continuously repeat steps (1) and (2) until each layer of RBM has been trained and learned. The features obtained from the last-layer RBM can be used as output features, and the local optimal parameters of each layer of RBM can be obtained;

[0156] 4) Use the error backpropagation algorithm to perform top-down supervised fine-tuning on the RBM, adjust the parameters of each layer of RBM, and finally the global optimal parameters of the entire DBN network model can be obtained. The DBN network uses several RBM units to form the basic network architecture, enabling the DBN network to have both unsupervised pre-training and supervised fine-tuning. This combination of unsupervised and supervised not only solves the gradient dispersion problem in traditional methods but also solves the problem that the network is prone to falling into local optima.

[0157] Apply the Grey Wolf Optimization (GWO) algorithm to optimize the DBN model. The specific principle of the GWO algorithm is as follows:

[0158] In the standard GWO algorithm, grey wolf individuals are represented by α, β, δ, and ω, where α represents the individual that makes decisions and manages the wolf pack, β and δ have lower fitness than α, and ω is an ordinary individual. The specific behaviors of the GWO algorithm include encircling, hunting, and attacking.

[0159] 1) Encircling behavior

[0160] The data model of grey wolves encircling prey can be expressed as an equation.

[0161] D = |C · X p (t) - X(t)|

[0162] X(t + 1) = X p (t) - A · D

[0163] In the equation, D represents the distance between the wolf pack and the prey, A = 2α · r 1 - α, C = 2 · r 2 , t represents the number of iterations, X p and X respectively represent the positions of the prey and the wolf pack, r 1 、r 2 are random quantities, and their value ranges are [0, 1], and the value range of α is [0, 2].

[0164] 2) Hunting behavior

[0165] Suppose α, β, and δ represent the global optimal solution, the second solution, and the third solution of the gray wolf individuals, and their optimization positions are located. The distances are expressed as equations respectively.

[0166] D α = |C 1 ·X α -X

[0167] D β = |C 2 ·X β -X

[0168] D δ = |C 2 ·X δ -X

[0169] In the equations, D α , D β , D δ represent the approximate distances between individuals α, β, δ and the current position X. X α , X β , X δ represent the positions of the global optimal solution, the second solution, and the third solution in sequence; C 1 , C 2 , C 3 represent random vectors, and their value ranges are [0, 1]. X and X(t + 1) are expressed as equations respectively.

[0170] X 1 = X α - A 1 ·(D α )

[0171] X 2 = X β - A 2 ·(D β )

[0172] X 3 = X δ - A 3 ·(D δ )

[0173]

[0174] In the equations, X(t + 1) represents the updated solution, and A 1 , A 2 , A 3 represent random quantities.

[0175] 3) Attack behavior

[0176] The attack is the final stage of the wolf pack's predation behavior, and the attack can be achieved by adjusting the parameter α.

[0177] If A ≤ 1, the wolf pack approaches the prey and concentrates on attacking the prey (X * , Y * ); otherwise, the wolf pack gradually moves away from the prey.

[0178] The population position initialization of the standard GWO algorithm usually adopts the random initialization method. This method may lead to too large a search range of the wolf pack, resulting in a longer search time. In this patent, while randomly generating the initial population, its opposite individual is generated, and the fitness of the opposite individual is compared with that of the original individual. If the fitness of the opposite individual is better than that of the original individual, the opposite individual is adopted; otherwise, the original individual is adopted. The expression of the opposite individual position vector is as follows.

[0179] X' = L b + U b - X

[0180] In the formula: X′ represents the opposite individual position vector; X represents the original individual position vector; Lb and Ub are the upper and lower bounds of X.

[0181] It can be seen from the company before improvement that the parameter A is determined by a, and a linearly decreases from 2 to 0. In the actual algorithm iteration process, the search space is large in the initial stage of iteration and global search is required, so the decreasing speed of a should be slowed down. In the later stage of iteration, the decreasing speed of a should be accelerated to improve local optimization and speed up the convergence speed.

[0182] The update formula of the convergence factor using the cosine variation law:

[0183]

[0184] In the formula: a max is the maximum value of the convergence factor, usually taken as 2; t max is the maximum iteration

[0185] times; n is the decreasing exponent, 0 < n < 1.

[0186] Step 4: Use game theory as the framework to consider the strategic choices and game behaviors of different entities when facing investment choices. The model derives the equilibrium solution by setting appropriate constraint conditions and objective functions and combining the benefit functions of multiple entities.

[0187] 1. Objective function

[0188] The objective function of the power grid company can be expressed as maximizing the net income, that is, the difference between the return obtained by the power grid company through equity investment and the investment cost, minus the loss caused by risks. Assume:

[0189] R ec (Iec ) represents the revenue obtained by the power grid company through investing in I ec .

[0190] C ec (I ec ) represents the cost invested by the power grid company.

[0191] γ ec is the risk aversion coefficient, measuring the sensitivity of the power grid company to risks.

[0192] σ ec is the risk measure of the power grid company's investment, usually represented by the variance or standard deviation of the investment project.

[0193] The objective function of the power grid company is:[[]]

[0194]

[0195] Follower 1 (multi-microgrid system alliance) has M objective functions. The NSGA-II algorithm can be used to solve the multi-objective optimization problem and obtain the Pareto front solution. Its objective function is:[[]]

[0196] U f1 = min(C f1 , σ f1 )

[0197] C f1 is the cost function of Follower 1, including operating costs, equipment investment, etc.

[0198] σ f1 is the risk function of Follower 1, representing the risk of the investment strategy.

[0199] The goal of Follower 2 is to maximize the economic benefits and dispatching flexibility of energy storage, considering the charging and discharging efficiency, costs and revenues of energy storage. Its objective function can be written as:[[]]

[0200]

[0201] where R f2 (I f2 ) represents the revenue obtained by the energy storage operator through charging and discharging.

[0202] C f2 (I f2 ) represents the cost of the energy storage operator (such as energy storage equipment, maintenance costs, etc.).

[0203] γ f2 is the risk aversion coefficient.

[0204] σ f2 is the risk measure of energy storage, reflecting the uncertainty of energy storage dispatching.

[0205] 2. Constraints

[0206] The power grid company must make investments within the scope of its available funds. Assume the maximum available investment funds are Imax.

[0207] L ec ≤I max

[0208] The power grid company must ensure that the investment can meet future electricity demands. Assume the future electricity demand forecast is D forecast , then the power grid company needs to invest sufficient funds to meet the demand.

[0209] I ec ≥f(D forecast ) The capacity of

[0210]

[0211] Power supply constraint: The investment and operation of the microgrid need to ensure that it can meet electricity demands and satisfy the grid load requirements.

[0212] I f1 ≥f(D forecast )

[0213] Energy storage capacity constraint: The charge and discharge amount of the energy storage device must be within the available capacity. Assume the maximum charge and discharge capacity is

[0214] P max .

[0215] I f2 ≤P max

[0216] Grid demand constraint: The charge and discharge of the energy storage system need to match the grid load demand and cannot exceed the grid load demand.

[0217] I f2 ≥f(D forecast )

[0218] Charge and discharge efficiency constraint: The charge efficiency η charge and discharge efficiency η discharge of the energy storage device need to meet certain constraint conditions:

[0219] η charge ≤η charge，max η discharge ≤η discharge，max

[0220] 3. Game equilibrium conditions

[0221] In this multi-agent game model, three parties (the leader, follower 1, and follower 2) need to achieve a game equilibrium through continuous decision-making interactions. This equilibrium can be described by the Nash equilibrium or the Stackelberg game, which requires that at the equilibrium point, changing the strategy of any party will not increase its benefits.

[0222]

[0223] Among them represents the optimal decision-making strategies of the three parties.

[0224] Step S5: Use the multi-agent game optimization of the genetic algorithm nested with the particle swarm algorithm to solve the model, so as to obtain the optimal investment strategy, achieve the best coordination and interest balance among the agents, and reach the optimal investment decision of the overall system.

[0225] The first step: Solve the multi-objective optimization model within follower 1 - genetic algorithm

[0226] Use the fast non-dominated multi-objective optimization algorithm with elitist retention strategy (non dominated sorting genetic algorithm, NSGA-II) to solve the multi-objective optimization model. Given that the objectives of follower 1 are all optimized with the minimum value as the goal, the partial-minimum fuzzy set decision theory is used to construct the fuzzy membership function, calculate the membership function values of each optimal solution on the Pareto front, and the greater the function value, the higher the satisfaction degree. Finally, the solution with the highest satisfaction degree is selected as the optimal compromise solution of the multi-objective optimization problem. The fuzzy membership function is as shown in the formula.

[0227]

[0228] In the formula: J m is the value of the m-th objective function; and are the minimum and maximum values of the m-th objective function respectively. Using the satisfaction degree as the evaluation index, the comprehensive satisfaction degree calculation formula is as shown in (2).

[0229]

[0230] In the formula: M is the number of objective functions; ω m is the satisfaction weight of the m-th objective function.

[0231] The second step: Substitute the solution result of the first step into the optimization model of the one-leader multi-follower Stackelberg game - particle swarm algorithm. It can be expressed as:

[0232] G = {N; Y; {y 1 , y 2}; U; {μ 1 , μ 2}}

[0233] 1) Set of participants: The leader, Follower 1, and Follower 2 form the set of participants N in the game.

[0234] 2) Decision variables: The decision variable of the leader is Y; the decision variable of Follower 1 is the strategy y made based on the leader's decision variable 1 ; the decision variable of Follower 2 is the strategy y made based on the leader's decision variable 2 .

[0235] 3) Benefits: The benefit functions of the leader, Follower 1, and Follower 2 are U, μ 1 , μ 2 .

[0236] For a one - master - multi - slave Stackelberg game to reach equilibrium, the following conditions need to be met:

[0237]

[0238] In the above formula, represents the optimal strategies of the leader, Follower 1, and Follower 2 at the game equilibrium. This formula indicates that when the three take the optimal strategies, that is , if any party changes its strategy, its own benefit will not improve. Then is considered the equilibrium solution. The game model is solved by using the particle swarm optimization algorithm nested with the CPLEX solver. The upper - layer leader model is solved by the particle swarm optimization algorithm, and the decisions of the lower - layer Follower 1 and Follower 2 are solved using CPIEX.

[0239] The flow chart of solving the game optimization model is as Figure 3 shown.

[0240] Those skilled in the art can understand that all or part of the process of implementing the above - mentioned embodiment method can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer - readable storage medium. Among them, the computer - readable storage medium is a disk, an optical disc, a read - only memory, or a random access memory, etc.

[0241] It should be noted that relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, such that an article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising said element.

[0242] The above are only specific embodiments of the present application, enabling those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

[0243] It should be understood that the present application is not limited to what has been described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present application is only limited by the appended claims.

Claims

1. A multi-objective decision-making technology for equity investment in provincial power grid industries, characterized in that: The method comprises the following steps: Step 1: Use the load forecast results as input indicators to build an incremental distribution network investment risk assessment indicator system that considers planning risk, management risk, market risk, and economic risk; Step 2: Use the order relationship method and the coefficient of variation method to calculate the subjective and objective weights, solve the combined weight optimization model based on the QIPSO algorithm, and finally build an incremental distribution network investment risk assessment model based on MGARA; Step 3: Systematically identify factors affecting investment capacity and use correlation analysis to determine the correlation between investment capacity and each key influencing factor. Combined with the selected main influencing factors, build an investment capacity prediction model based on deep learning and optimization algorithms; Step 4: Use game theory as a framework to consider the strategic choices and game behaviors of different entities when facing investment choices. The model sets appropriate constraints and objective functions, combines the benefit functions of multiple entities, and derives the equilibrium solution; Step five, use the genetic algorithm nested particle swarm algorithm for multi-agent game optimization to solve the model to obtain the optimal investment strategy, achieve the best coordination and interest balance among the subjects, and reach the optimal investment decision for the system as a whole.

2. According to claim 1, a multi-objective decision-making technology for equity investment in power grid provincially managed industries is characterized by: In step 1, the incremental distribution network investment risk assessment index system selects planning risk, management risk, market risk, and economic risk as primary indicators.

3. According to claim 1, a multi-objective decision-making technology for equity investment in power grid provincially managed industries is characterized by: The construction of the incremental distribution network investment risk assessment model in step 2 includes the following contents: a. The subjective weighting method uses the order relationship method to qualitatively sort the evaluation indicators, then makes rational judgments on adjacent indicators, and finally performs quantitative calculations; b. Use the ordinal coefficient of variation method to assign weights to each indicator based on the degree of variation of its observed values ​​on all evaluated objects; c. Combine the weight optimization model and introduce the quadratic interpolation operator into the particle swarm algorithm to improve the convergence speed and computational efficiency of the algorithm, and then find the global optimal position of the model; d. The risk assessment model based on MGARA uses a multi-objective optimization framework to comprehensively consider the relationship between multiple risk factors, and uses grey correlation analysis technology to perform sensitivity analysis and weight assignment on each risk factor, thereby achieving accurate risk assessment.

4. According to claim 1, a multi-objective decision-making technology for equity investment in power grid provincially managed industries is characterized by: The factors affecting investment capacity in step three include grid load forecast accuracy, market demand growth rate, the impact of electricity market reform policies, grid technology level, financial health, capital market financing environment, government subsidies and policy support, grid operation efficiency, social and economic development level, and grid asset structure and investment return rate.

5. According to claim 1, a multi-objective decision-making technology for equity investment in power grid provincially managed industries is characterized by: The participants in the game in step 4 include the distribution network, microgrid system and shared energy storage operators. The objective function is the benefit function of each entity in the investment decision-making, and the constraints include investment capacity constraints, technical constraints, market constraints and policy constraints.

6. The multi-objective decision-making technology for equity investment in power grid provincially managed industries according to claim 1 is characterized by: The model solution in step 5 specifically includes the following contents: a. First, solve the internal multi-objective optimization model of follower 1 through genetic algorithm. b. Then, the optimization results of the genetic algorithm are substituted into the Stackelberg game model with one master and multiple followers. In this model, the decisions of the leader, follower 1, and follower 2 are adjusted based on game theory. c. Optimize the leader's decision strategy through the particle swarm algorithm (PSO), and use the CPLEX solver to solve the decisions of follower 1 and follower 2. d. Finally, the game equilibrium solution is obtained through the benefit functions of multiple parties.

Images