High-voltage special transformer remote cost control parameter adjusting method based on particle swarm optimization

Abstract

The invention discloses a high-voltage special transformer remote cost control parameter adjustment method based on particle swarm optimization, and the method comprises the following steps: S1, collecting the multi-source operation data of a high-voltage transformer, and forming a data set; s2, performing multi-level layered preprocessing on the data set; s3, constructing a self-adaptive multi-target cost control parameter optimization model, and generating an optimal solution set by using a multi-level self-adaptive collaborative particle swarm optimization algorithm; s4, generating an optimized cost control parameter combination in combination with a multi-dimensional quantum entropy coding optimization algorithm; s5, fine optimization and cross validation are carried out through a deep adaptive enhanced particle swarm optimization algorithm; s6, before application, verifying a parameter combination by a dynamic robustness evaluation method based on quantum particle filtering; s7, the parameter combination is applied to actual operation, and real-time monitoring and self-adaptive optimization are carried out; and S8, combining real-time and historical data, and dynamically adjusting an optimization strategy by adopting a hybrid multi-modal particle swarm optimization method. By optimizing the cost control parameters, the operation efficiency and stability of the transformer are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system optimization, and in particular to a method for adjusting parameters of high-voltage special transformer remote cost control based on particle swarm optimization. Background Art

[0002] With the continuous development and complexity of power systems, high-voltage special transformers play a key role in power transmission and distribution. However, with the diversification of transformer operating environments and the continuous changes in load conditions, how to efficiently and accurately adjust remote cost control parameters has become an important technical problem in power system optimization.

[0003] In the existing technology, traditional cost control parameter adjustment methods mainly rely on manual experience and fixed rule settings. These methods have obvious limitations and defects when facing increasingly complex operating conditions:

[0004] 1. Lack of adaptive capabilities: Traditional methods are usually based on fixed parameter settings and rule bases, and cannot adapt to changes in transformer operating conditions in real time. When the environment or load conditions change, the existing parameter settings may not meet the new requirements, resulting in reduced system efficiency or unnecessary energy consumption.

[0005] 2. Insufficient optimization capability: Most existing cost control parameter optimization methods adopt simple single-objective optimization strategies, ignoring the need for multi-objective optimization. These methods cannot effectively balance multiple conflicting objectives, resulting in unsatisfactory optimization results. For example, it is difficult to achieve complex multi-objective optimization requirements such as ensuring the safe operation of transformers and minimizing energy costs through existing methods.

[0006] 3. Limited global search capabilities: Traditional optimization methods such as gradient descent and simulated annealing can perform local optimization to a certain extent, but when faced with complex multi-dimensional search spaces, it is often difficult to find the global optimal solution. This optimization method that lacks a global perspective is prone to falling into local optimality and cannot fully tap the optimization potential of the system.

[0007] 4. Slow dynamic response: Traditional fee control systems usually rely on periodic data collection and batch processing, which makes it difficult to achieve real-time monitoring and adjustment of transformer operating status. When the system status changes rapidly, the response speed of existing methods often lags behind, and fee control parameters cannot be adjusted in time, resulting in reduced system operating efficiency.

[0008] 5. Lack of robustness evaluation mechanism: In existing technologies, the verification process after parameter adjustment is often simple and lacks depth, and fails to fully consider the robustness under different operating conditions. This means that when the system faces abnormal conditions or emergencies, the existing parameter adjustment may not guarantee the safe and stable operation of the system.

[0009] 6. Ignoring the use of historical data: Most of the existing cost control parameter adjustment methods rely only on current operating data, and fail to make full use of historical operating data for comprehensive analysis and optimization. This method of ignoring historical data often leads to the inability to make effective decisions quickly when dealing with similar operating scenarios, thereby reducing the intelligence level of the system.

[0010] Therefore, how to provide a high-voltage special transformer remote cost control parameter adjustment method based on particle swarm optimization is a problem that technical personnel in this field urgently need to solve. Summary of the invention

[0011] One purpose of the present invention is to propose a method for adjusting remote cost control parameters of a high-voltage special transformer based on particle swarm optimization. The present invention makes full use of advanced algorithms such as particle swarm optimization, quantum entropy coding, and deep adaptive reinforcement learning, and describes in detail the process of realizing dynamic cost control parameter optimization in a complex power system. It has the advantages of high optimization efficiency, strong real-time adaptability and good system stability.

[0012] A method for adjusting parameters of a high-voltage dedicated transformer remote cost control based on particle swarm optimization according to an embodiment of the present invention comprises the following steps:

[0013] S1. Collect multi-source operation data of high-voltage dedicated transformers to form a multi-source data set;

[0014] S2, perform multi-level hierarchical preprocessing on multi-source data sets, and generate preprocessed data sets using a burst data fluctuation response method based on multidimensional spatiotemporal correlation analysis;

[0015] S3. Based on the preprocessed data set, an adaptive multi-objective cost control parameter optimization model is constructed, and multiple objective functions are optimized through a multi-level adaptive collaborative particle swarm optimization algorithm, and the weights and constraints of the cost control parameters are dynamically adjusted to generate a Pareto optimal solution set;

[0016] S4. Obtain the generated Pareto optimal solution set, combine it with the multi-dimensional quantum entropy coding optimization algorithm to conduct further global search, divide and encode the search space in multiple dimensions through quantum entropy coding technology, and generate a further optimized fee control parameter combination;

[0017] S5. The generated cost control parameter combination is refined optimized using the deep adaptive enhanced particle swarm optimization algorithm. The strategy combination is dynamically evaluated and selected through a deep neural network to generate the final optimization parameters for practical application. The optimization parameter combination is cross-validated with the preprocessed data set.

[0018] S6. Before applying the final optimization parameters, based on the generated final optimization parameters and the preprocessed data set, a dynamic robustness evaluation method based on quantum particle filtering is used to perform a multi-dimensional dynamic robustness evaluation on the high-dimensional parameter space to verify the optimal parameter combination;

[0019] S7. Apply the verified cost control parameter combination to the actual operation of the high-voltage dedicated transformer, and conduct real-time monitoring and adaptive optimization of the transformer's operating conditions through a multi-modal dynamic adaptive particle swarm optimization algorithm;

[0020] S8. Combine the data generated by real-time monitoring with historical data sets, adopt a multimodal deep learning intelligent optimization method based on hybrid multimodal particle swarm optimization, learn and analyze historical data sets through particle swarm optimization and deep reinforcement learning models, and dynamically adjust future optimization strategies to adapt to the ever-changing operating conditions and needs of high-voltage dedicated transformers.

[0021] Optionally, S3 includes the following steps:

[0022] S31, extracting features from the preprocessed multi-source data set, using a holographic evolutionary adaptive optimization algorithm, using holographic storage technology to record and retrieve historical optimization paths, and generating feature vector sets required for each objective function;

[0023] S32. Based on the feature vector set, an adaptive multi-objective cost control parameter optimization model is constructed, in which the objective function set includes multiple objective functions, each objective function is parameterized by a specific weight coefficient, and the weight coefficient is adjusted in real time by a holographic evolutionary adaptive mechanism;

[0024] S33, inputting the optimization model into a multi-level adaptive collaborative particle swarm optimization algorithm, setting an initial population, the initial population consisting of a plurality of particles, and the position of each particle is randomly initialized and generated by a eigenvalue in a eigenvector set;

[0025] S34. During the optimization process, the initial population is divided into multiple sub-populations by introducing a multi-level collaborative mechanism. Each sub-population is independently optimized and evolved in its corresponding objective function subspace, and collaborative optimization among the sub-populations is achieved through an information sharing mechanism.

[0026] S35. In each iteration, the weight coefficients of each subpopulation are dynamically adjusted according to the optimization progress of each objective function.

[0027]

[0028] in, is the weight coefficient of the objective function in the next generation t+1, is the current fitness value of the objective function in the tth generation, is the objective function value with the highest fitness value among all objective functions of the current generation, is the average fitness value of all objective functions in the current generation, α and β are learning rates, γ is the exponential adjustment factor, δ is the historical weight adjustment factor, exp is the exponential function, m and n are the number of objective functions, represents the historical optimization path weight of the objective function, It is the sum of the weights of the historical optimization paths of all objective functions in the tth generation;

[0029] S36, evaluate the fitness of each generation of population, select the particle with the highest fitness as the current optimal solution, and update it to the global optimal solution set P * Among them represents the Pareto optimal solution set;

[0030] S37, after reaching the preset termination condition, the final Pareto optimal solution set P * Output, which serves as the basis for optimization in subsequent steps.

[0031] Optionally, S4 includes the following steps:

[0032] S41. Obtain the final Pareto optimal solution set P * , and further global search for the optimal solution set is performed in combination with the multi-dimensional quantum entropy coding optimization algorithm;

[0033] S42, through the quantum entropy coding technology, the search space of each particle position in the initial population is divided and encoded in multiple dimensions to generate a particle position vector with quantum state characteristics

[0034] S43, in each iteration of the multidimensional quantum entropy coding optimization algorithm, based on the dynamic weight adjustment formula Particle position vector Update the weights and generate new optimization parameter combinations:

[0035]

[0036] Among them, p j Represents particles The probability distribution in the jth dimension, θ j is the quantum state angle, is the phase difference between the particle in dimension j and dimension k, α and γ are adjustment factors, and are the weight coefficients in the current and previous generations, respectively;

[0037] S44. In each iteration, a multi-level collaborative mechanism of adaptive quantum states is introduced, so that the quantum states of different particles can influence and co-evolve each other, and dynamically adjust the optimization direction and strategy of particles through the interaction of quantum entangled states;

[0038] S45. According to quantum information entropy The particles are screened and optimized based on the calculation results, the particle position vectors with higher information entropy values ​​are selected, and the selected particle position vectors are input into the adaptive deep learning decoding network;

[0039] S46. Through a complex nonlinear decoding process, the particle position vector is converted into a high-dimensional feature representation:

[0040]

[0041] Among them, A decode is the final output weight matrix, W decode is the weight matrix of the decoding network, b decode is the bias vector, d final is the bias vector of the output layer, tanh is the nonlinear activation function, To optimize the parameter combination;

[0042] S47. Perform cluster analysis on the decoded particle positions, calculate the similarity between the particle positions and the current cluster center, merge particles with higher similarity according to the similarity of quantum states, adjust the structure and distribution of the population, and output the optimized cost control parameter combination.

[0043] Optionally, S5 includes the following steps:

[0044] S51, obtain the optimized cost control parameter combination, and use it as input, apply the deep adaptive enhanced particle swarm optimization algorithm for fine optimization, and generate the optimized parameter vector

[0045] S52, updating and further optimizing the optimized parameter vector in multiple dimensions:

[0046]

[0047] Among them, η is the dynamic learning rate, is the weight updated by the dynamic adjustment formula, ⊙ represents element-by-element multiplication, λ is the adjustment factor, is the calculated value of quantum information entropy, Represents the parameter vector Gradient operation, V k is the adaptive perturbation vector, is the phase difference between particles in different dimensions;

[0048] S53, combined with the quantum perturbation mechanism in particle swarm optimization, the optimized parameter vector Used to adjust weight matrices and bias vectors in deep neural network models;

[0049] S54. Based on the adjusted weight matrix and bias vector, the deep reinforcement learning algorithm is used to dynamically evaluate and optimize the strategy combination, so that the model can be adaptively adjusted to cope with changes in the cost control strategy, and the final optimized cost control parameter combination R is generated through the continuously updated strategy combination. * ;

[0050] S55, the generated final optimized cost control parameter combination R * Combined with the preprocessed dataset, cross validation was performed.

[0051] Optionally, the S6 comprises the following steps:

[0052] S61. Obtain the final optimized cost control parameter combination R * , and combine it with the preprocessed data set to construct a high-dimensional parameter space P, where each dimension corresponds to an optimized cost control parameter, and initialize the quantum particle swarm;

[0053] S62. Introduce an adaptive quantum particle coordination mechanism to coordinately optimize the behavior of quantum particle groups and adaptively adjust the interaction between particles based on local and global information in the particle group:

[0054]

[0055] Where ρ is the synergy factor, represents the current position of the neighboring particles, represents the gradient calculation of the particle position vector, is the position of the ith particle in the kth dimension, and L represents the number of neighboring particles considered;

[0056] S63. After collaborative optimization, a dynamic robustness evaluation method based on quantum particle filtering is used to evaluate the performance of the optimization parameters. Get the final stability index S of the particle swarm i :

[0057]

[0058] Where m is the total number of particles, represents the distance between the particle and the optimized parameters, Indicates the distance between the optimized parameters and the final result;

[0059] S64. According to the multi-dimensional dynamic robustness evaluation results, select the one with the highest stability index S i The cost control parameter combination is determined and verified to obtain the optimal parameter combination.

[0060] Optionally, the S7 includes the following steps:

[0061] S71, the verified optimal cost control parameter combination R * Applied to the actual operation of high-voltage special transformers, a multi-modal dynamic adaptive particle swarm is initialized and a comprehensive data set is constructed, which includes real-time monitoring data and historical operation data;

[0062] S72, pre-process the real-time monitoring data, use a multi-modal feature extraction method to extract the operating characteristics of the high-voltage dedicated transformer, and form a multi-dimensional feature vector

[0063] S73. Introduce a dynamic optimization feedback mechanism based on adaptive prediction to predict future transformer operating conditions by establishing a prediction model based on deep learning. The predicted feature vector is used as input and compared with the real-time feature vector Together, input the multimodal dynamic adaptive particle swarm optimization algorithm to update the particle position:

[0064]

[0066] Among them, α is the global adaptation factor, which controls the particle to move toward the global optimal solution G (t) The movement in the direction, γ is the real-time adaptation factor, which adjusts the position of the particle based on the difference between the current and historical features, F historical represents the historical eigenvector, δ is the prediction feedback factor, which is used to adjust the position of the particle according to the future prediction eigenvector, ∈ is the random perturbation factor, is a random perturbation vector used for particle exploration;

[0067] S74. Real-time monitoring of fee control parameters after each iteration update And combine it with the historical optimization parameters R * For comparison, the fitness function is calculated using a back-analysis method based on historical data.

[0068]

[0069] Where N is the number of samples in the historical data set, is the distance between the current feature vector and the historical feature vector, is the distance between the current cost control parameters and the historical optimal parameter combination;

[0070] S75, according to the evaluation result of the fitness function, dynamically adjust the cost control parameter combination in the particle swarm and update the global optimal solution G of the particle swarm (t+1) , and continue to iterate until the preset convergence conditions are met, and finally output the real-time adaptive optimization cost control parameter combination R final , used in long-term operation monitoring and fee control adjustment of high-voltage special transformers.

[0071] The beneficial effects of the present invention are:

[0072] (1) The present invention significantly improves the optimization efficiency and accuracy of remote cost control parameter adjustment of high-voltage transformers by combining technologies such as particle swarm optimization, multi-dimensional quantum entropy coding, deep adaptive reinforcement learning, and quantum particle filtering. In particular, in terms of multi-objective optimization and global search, the present invention overcomes the defects of insufficient adaptive ability and local optimization in traditional methods, and can dynamically adjust cost control parameters to generate an optimal solution set that adapts to complex operating conditions.

[0073] (2) The present invention optimizes the real-time monitoring and adaptive adjustment process of the high-voltage transformer by introducing a multi-modal dynamic adaptive particle swarm optimization algorithm and a hybrid multi-modal deep learning method, and improves the system's real-time response capability to operating conditions. Combined with the dynamic robustness evaluation of the quantum particle filter, the present invention ensures the stability and safety of the final parameter combination, and can maintain efficient operation even under abnormal conditions.

[0074] (3) The present invention provides an intelligent, efficient and reliable method for adjusting the remote cost control parameters of a high-voltage special transformer by comprehensively applying a variety of advanced optimization and evaluation technologies, so that the system can adapt to the changing operating environment and automatically handle complex cost control adjustment tasks, thereby significantly improving the operating efficiency and stability of the transformer. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0076] Figure 1 This is a flow chart of a method for adjusting parameters of high-voltage dedicated transformer remote cost control based on particle swarm optimization proposed by the present invention;

[0077] Figure 2 A schematic diagram of generating an optimal solution set by the multi-level adaptive collaborative particle swarm optimization algorithm proposed in the present invention;

[0078] Figure 3 This is a global search flow chart based on the multidimensional quantum entropy coding optimization algorithm proposed by the present invention. DETAILED DESCRIPTION

[0079] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0080] refer to Figure 1-3 A method for adjusting parameters of high-voltage transformer remote cost control based on particle swarm optimization includes the following steps:

[0081] S1. Collect multi-source operation data of high-voltage dedicated transformers to form a multi-source data set;

[0082] S2, perform multi-level hierarchical preprocessing on multi-source data sets, and generate preprocessed data sets using a burst data fluctuation response method based on multidimensional spatiotemporal correlation analysis;

[0083] S3. Based on the preprocessed data set, an adaptive multi-objective cost control parameter optimization model is constructed, and multiple objective functions are optimized through a multi-level adaptive collaborative particle swarm optimization algorithm, and the weights and constraints of the cost control parameters are dynamically adjusted to generate a Pareto optimal solution set;

[0084] S4. Obtain the generated Pareto optimal solution set, combine it with the multi-dimensional quantum entropy coding optimization algorithm to conduct further global search, divide and encode the search space in multiple dimensions through quantum entropy coding technology, and generate a further optimized fee control parameter combination;

[0085] S5. The generated cost control parameter combination is refined optimized using the deep adaptive enhanced particle swarm optimization algorithm. The strategy combination is dynamically evaluated and selected through a deep neural network to generate the final optimization parameters for practical application. The optimization parameter combination is cross-validated with the preprocessed data set.

[0086] S6. Before applying the final optimization parameters, based on the generated final optimization parameters and the preprocessed data set, a dynamic robustness evaluation method based on quantum particle filtering is used to perform a multi-dimensional dynamic robustness evaluation on the high-dimensional parameter space to verify the optimal parameter combination;

[0087] S7. Apply the verified cost control parameter combination to the actual operation of the high-voltage dedicated transformer, and conduct real-time monitoring and adaptive optimization of the transformer's operating conditions through a multi-modal dynamic adaptive particle swarm optimization algorithm;

[0088] S8. Combine the data generated by real-time monitoring with historical data sets, adopt a multimodal deep learning intelligent optimization method based on hybrid multimodal particle swarm optimization, learn and analyze historical data sets through particle swarm optimization and deep reinforcement learning models, and dynamically adjust future optimization strategies to adapt to the ever-changing operating conditions and needs of high-voltage dedicated transformers.

[0089] In this implementation, S3 includes the following steps:

[0090] S31, extracting features from the preprocessed multi-source data set, using a holographic evolutionary adaptive optimization algorithm, using holographic storage technology to record and retrieve historical optimization paths, and generating feature vector sets required for each objective function;

[0091] S32. Based on the feature vector set, an adaptive multi-objective cost control parameter optimization model is constructed, in which the objective function set includes multiple objective functions, each objective function is parameterized by a specific weight coefficient, and the weight coefficient is adjusted in real time by a holographic evolutionary adaptive mechanism;

[0092] S33, inputting the optimization model into a multi-level adaptive collaborative particle swarm optimization algorithm, setting an initial population, the initial population consisting of a plurality of particles, and the position of each particle is randomly initialized and generated by a eigenvalue in a eigenvector set;

[0093] S34. During the optimization process, the initial population is divided into multiple sub-populations by introducing a multi-level collaborative mechanism. Each sub-population is independently optimized and evolved in its corresponding objective function subspace, and collaborative optimization among the sub-populations is achieved through an information sharing mechanism.

[0094] S35. In each iteration, the weight coefficients of each subpopulation are dynamically adjusted according to the optimization progress of each objective function.

[0095]

[0096] in, is the weight coefficient of the objective function in the next generation t+1, is the current fitness value of the objective function in the tth generation, is the objective function value with the highest fitness value among all objective functions of the current generation, is the average fitness value of all objective functions in the current generation, α and β are learning rates, γ is the exponential adjustment factor, δ is the historical weight adjustment factor, exp is the exponential function, m and n are the number of objective functions, represents the historical optimization path weight of the objective function, It is the sum of the weights of the historical optimization paths of all objective functions in the tth generation;

[0097] S36, evaluate the fitness of each generation of population, select the particle with the highest fitness as the current optimal solution, and update it to the global optimal solution set P * Among them represents the Pareto optimal solution set;

[0098] S37, after reaching the preset termination condition, the final Pareto optimal solution set P * Output, which serves as the basis for optimization in subsequent steps.

[0099] In this implementation manner, S4 includes the following steps:

[0100] S41. Obtain the final Pareto optimal solution set P * , and further global search for the optimal solution set is performed in combination with the multi-dimensional quantum entropy coding optimization algorithm;

[0101] S42, through the quantum entropy coding technology, the search space of each particle position in the initial population is divided and encoded in multiple dimensions to generate a particle position vector with quantum state characteristics

[0102] S43, in each iteration of the multidimensional quantum entropy coding optimization algorithm, based on the dynamic weight adjustment formula Particle position vector Update the weights and generate new optimization parameter combinations:

[0103]

[0104] Among them, p j Represents particles The probability distribution in the jth dimension, θ j is the quantum state angle, is the phase difference between the particle in dimension j and dimension k, α and γ are adjustment factors, and are the weight coefficients in the current and previous generations, respectively;

[0105] S44. In each iteration, a multi-level collaborative mechanism of adaptive quantum states is introduced, so that the quantum states of different particles can influence and co-evolve each other, and dynamically adjust the optimization direction and strategy of particles through the interaction of quantum entangled states;

[0106] S45. According to quantum information entropy The particles are screened and optimized based on the calculation results, the particle position vectors with higher information entropy values ​​are selected, and the selected particle position vectors are input into the adaptive deep learning decoding network;

[0107] S46. Through a complex nonlinear decoding process, the particle position vector is converted into a high-dimensional feature representation:

[0108]

[0109] Among them, A decode is the final output weight matrix, W decode is the weight matrix of the decoding network, b decode is the bias vector, d final is the bias vector of the output layer, tanh is the nonlinear activation function, To optimize the parameter combination;

[0110] S47. Perform cluster analysis on the decoded particle positions, calculate the similarity between the particle positions and the current cluster center, merge particles with higher similarity according to the similarity of quantum states, adjust the structure and distribution of the population, and output the optimized cost control parameter combination.

[0111] In this implementation manner, S5 includes the following steps:

[0112] S51, obtain the optimized cost control parameter combination, and use it as input, apply the deep adaptive enhanced particle swarm optimization algorithm for fine optimization, and generate the optimized parameter vector

[0113] S52, updating and further optimizing the optimized parameter vector in multiple dimensions:

[0114]

[0115] Among them, η is the dynamic learning rate, is the weight updated by the dynamic adjustment formula, ⊙ represents element-by-element multiplication, λ is the adjustment factor, is the calculated value of quantum information entropy, Represents the parameter vector Gradient operation, V k is the adaptive perturbation vector, is the phase difference between particles in different dimensions;

[0116] S53, combined with the quantum perturbation mechanism in particle swarm optimization, the optimized parameter vector Used to adjust weight matrices and bias vectors in deep neural network models;

[0117] S54. Based on the adjusted weight matrix and bias vector, the deep reinforcement learning algorithm is used to dynamically evaluate and optimize the strategy combination, so that the model can be adaptively adjusted to cope with changes in the cost control strategy, and the final optimized cost control parameter combination R is generated through the continuously updated strategy combination. * ;

[0118] S55, the generated final optimized cost control parameter combination R * Combined with the preprocessed dataset, cross validation was performed.

[0119] In this implementation manner, S6 includes the following steps:

[0120] S61. Obtain the final optimized cost control parameter combination R * , and combine it with the preprocessed data set to construct a high-dimensional parameter space P, where each dimension corresponds to an optimized cost control parameter, and initialize the quantum particle swarm;

[0121] S62. Introduce an adaptive quantum particle coordination mechanism to coordinately optimize the behavior of quantum particle groups and adaptively adjust the interaction between particles based on local and global information in the particle group:

[0122]

[0123] Where ρ is the synergy factor, represents the current position of the neighboring particles, represents the gradient calculation of the particle position vector, is the position of the ith particle in the kth dimension, and L represents the number of neighboring particles considered;

[0124] S63. After collaborative optimization, a dynamic robustness evaluation method based on quantum particle filtering is used to evaluate the performance of the optimization parameters. Get the final stability index S of the particle swarm i :

[0125]

[0126] Where m is the total number of particles, represents the distance between the particle and the optimized parameters, Indicates the distance between the optimized parameters and the final result;

[0127] S64. According to the multi-dimensional dynamic robustness evaluation results, select the one with the highest stability index S i The cost control parameter combination is determined and verified to obtain the optimal parameter combination.

[0128] In this implementation manner, the S7 includes the following steps:

[0129] S71, the verified optimal cost control parameter combination R * Applied to the actual operation of high-voltage special transformers, a multi-modal dynamic adaptive particle swarm is initialized and a comprehensive data set is constructed, which includes real-time monitoring data and historical operation data;

[0130] S72, pre-process the real-time monitoring data, use a multi-modal feature extraction method to extract the operating characteristics of the high-voltage dedicated transformer, and form a multi-dimensional feature vector

[0131] S73. Introduce a dynamic optimization feedback mechanism based on adaptive prediction to predict future transformer operating conditions by establishing a prediction model based on deep learning. The predicted feature vector is used as input and compared with the real-time feature vector Together, input the multimodal dynamic adaptive particle swarm optimization algorithm to update the particle position:

[0132]

[0133] Among them, α is the global adaptation factor, which controls the movement of particles towards the global optimal solution G(t), γ is the real-time adaptation factor, which adjusts the position of particles based on the difference between current and historical features, and F historical represents the historical eigenvector, δ is the prediction feedback factor, which is used to adjust the position of the particle according to the future prediction eigenvector, ∈ is the random perturbation factor, is a random perturbation vector used for particle exploration;

[0134] S74. Real-time monitoring of fee control parameters after each iteration update And combine it with the historical optimization parameters R * For comparison, the fitness function is calculated using a back-analysis method based on historical data.

[0135]

[0136] Where N is the number of samples in the historical data set, is the distance between the current feature vector and the historical feature vector, is the distance between the current cost control parameters and the historical optimal parameter combination;

[0137] S75, according to the evaluation result of the fitness function, dynamically adjust the cost control parameter combination in the particle swarm and update the global optimal solution G of the particle swarm (t+1) , and continue to iterate until the preset convergence conditions are met, and finally output the real-time adaptive optimization cost control parameter combination R final , used in long-term operation monitoring and fee control adjustment of high-voltage special transformers.

[0138] Embodiment 1:

[0139] In order to verify the feasibility of the present invention in implementation, the present invention is applied to the main substation of a city in East China. The substation is responsible for providing power guarantee for a number of important industrial areas. The operating stability of the transformer and the precise adjustment of the cost control parameters are crucial to the reliability of the power supply. In the traditional cost control parameter adjustment method, the setting of the cost control parameters is usually based on historical data and experience, which makes it difficult to cope with the drastic fluctuations in power demand and changes in operating status caused by equipment aging. By introducing a high-voltage dedicated transformer remote cost control parameter adjustment method based on particle swarm optimization of the present invention, it is hoped that real-time optimization and adjustment of the transformer cost control parameters can be achieved under a complex power load environment, thereby improving the operating efficiency and stability of the transformer.

[0140] First, the method of the present invention installs multiple sensors and remote terminals in the main substation to collect multi-source operation data of high-voltage dedicated transformers in real time. These data include key parameters such as voltage, current, load rate, power factor, temperature, etc., and the collection frequency is once every 10 seconds. Through multi-level hierarchical preprocessing technology, the data is cleaned and denoised, and after multi-dimensional spatiotemporal correlation analysis, a preprocessing data set is formed after processing. These data sets provide reliable input for the subsequent adaptive multi-objective cost control parameter optimization model.

[0141] Next, the system uses a multi-level adaptive collaborative particle swarm optimization algorithm to build an adaptive multi-objective cost control parameter optimization model based on the preprocessed data set. Through this model, the system dynamically adjusts the weights and constraints of the cost control parameters under different power load conditions to generate the optimal solution set. In order to further improve the global search capability, the system combines the multi-dimensional quantum entropy coding optimization algorithm to conduct further global search on the generated optimal solution set and generate an optimized cost control parameter combination.

[0142] In practical applications, these optimized parameter combinations are refined and optimized through the deep adaptive enhanced particle swarm optimization algorithm. Using the deep neural network model, the system dynamically evaluates and selects the strategy combination, and finally generates the cost control parameter combination for practical application. This parameter combination is verified by the dynamic robustness evaluation method of quantum particle filtering to ensure its stability and adaptability under different operating conditions.

[0143] Table 1: Comparison of the implementation effects of the remote cost control parameter adjustment method for high-voltage transformers based on particle swarm optimization

[0144] Time period Summer peak electricity season 3-month trial period Load rate fluctuation range under traditional methods 40％-90％ 40％-90％ Load rate fluctuation range of the method of the present invention 50％-80％ 50％-80％ The frequency of load factor deviation from target value by traditional method is reduced none none The method of the present invention reduces the frequency of load rate deviation from the target value More than 60% More than 60% Transformer operation efficiency improvement 15％ 15％ Transformer average operating temperature 85℃ 85℃ Average operating temperature of transformer under the method of the present invention 82.5℃ 82.5℃ Response time 15 minutes 15 minutes Response time under the method of the present invention Within 5 minutes Within 5 minutes Improved fault response rate none 40％ Fewer power outages none 0.6 times / month

[0145] During the peak period of electricity demand in summer, the system effectively responded to sudden load fluctuations. Data show that after implementing the method of the present invention, the load rate fluctuation range of the transformer has been reduced from the original 40%-90% to 50%-80%, and the frequency of the load rate deviating from the target value has been reduced by more than 60%. During a sudden peak in electricity demand, the system improved the operating efficiency of the transformer by about 15% through real-time optimization and adjustment, avoiding equipment damage caused by overload. In addition, during the three-month trial operation period, the system effectively reduced the average operating temperature of the transformer from the original average of 85°C to 82.5°C by dynamically adjusting the cost control parameters, thereby extending the service life of the equipment.

[0146] Through application verification in actual power systems, the effectiveness of the method of the present invention in improving the operating efficiency and stability of high-voltage dedicated transformers has been proven. The system can not only adjust the cost control parameters in real time to cope with the changing power load, but also maintain the optimal working state of the transformer during long-term operation, extend the life of the equipment, and reduce maintenance and operating costs.

[0147] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for adjusting parameters of high-voltage special transformer remote cost control based on particle swarm optimization, characterized in that: The steps include: S1. Collect multi-source operation data of high-voltage dedicated transformers to form a multi-source data set; S2, perform multi-level hierarchical preprocessing on multi-source data sets, and generate preprocessed data sets using a burst data fluctuation response method based on multidimensional spatiotemporal correlation analysis; S3. Based on the preprocessed data set, an adaptive multi-objective cost control parameter optimization model is constructed, and multiple objective functions are optimized through a multi-level adaptive collaborative particle swarm optimization algorithm, and the weights and constraints of the cost control parameters are dynamically adjusted to generate a Pareto optimal solution set; S4. Obtain the generated Pareto optimal solution set, combine it with the multi-dimensional quantum entropy coding optimization algorithm to conduct further global search, divide and encode the search space in multiple dimensions through quantum entropy coding technology, and generate a further optimized fee control parameter combination; S5. The generated cost control parameter combination is refined optimized using the deep adaptive enhanced particle swarm optimization algorithm. The strategy combination is dynamically evaluated and selected through a deep neural network to generate the final optimization parameters for practical application. The optimization parameter combination is cross-validated with the preprocessed data set. S6. Before applying the final optimization parameters, based on the generated final optimization parameters and the preprocessed data set, a dynamic robustness evaluation method based on quantum particle filtering is used to perform a multi-dimensional dynamic robustness evaluation on the high-dimensional parameter space to verify the optimal parameter combination; S7. Apply the verified cost control parameter combination to the actual operation of the high-voltage dedicated transformer, and conduct real-time monitoring and adaptive optimization of the transformer's operating conditions through a multi-modal dynamic adaptive particle swarm optimization algorithm; S8. Combine the data generated by real-time monitoring with historical data sets, adopt a multimodal deep learning intelligent optimization method based on hybrid multimodal particle swarm optimization, learn and analyze historical data sets through particle swarm optimization and deep reinforcement learning models, and dynamically adjust future optimization strategies to adapt to the ever-changing operating conditions and needs of high-voltage dedicated transformers.

2. According to the particle swarm optimization-based high-voltage transformer remote cost control parameter adjustment method of claim 1, it is characterized in that: The S3 specifically includes: S31, extracting features from the preprocessed multi-source data set, using a holographic evolutionary adaptive optimization algorithm, using holographic storage technology to record and retrieve historical optimization paths, and generating feature vector sets required for each objective function; S32. Based on the feature vector set, an adaptive multi-objective cost control parameter optimization model is constructed, in which the objective function set includes multiple objective functions, each objective function is parameterized by a specific weight coefficient, and the weight coefficient is adjusted in real time by a holographic evolutionary adaptive mechanism; S33, inputting the optimization model into a multi-level adaptive collaborative particle swarm optimization algorithm, setting an initial population, the initial population consisting of a plurality of particles, and the position of each particle is randomly initialized and generated by a eigenvalue in a eigenvector set; S34. During the optimization process, the initial population is divided into multiple sub-populations by introducing a multi-level collaborative mechanism. Each sub-population is independently optimized and evolved in its corresponding objective function subspace, and collaborative optimization among the sub-populations is achieved through an information sharing mechanism. S35. In each iteration, the weight coefficients of each subpopulation are dynamically adjusted according to the optimization progress of each objective function. in, is the weight coefficient of the objective function in the next generation t+1, is the current fitness value of the objective function in the tth generation, is the objective function value with the highest fitness value among all objective functions of the current generation, is the average fitness value of all objective functions in the current generation, α and β are learning rates, γ is the exponential adjustment factor, δ is the historical weight adjustment factor, exp is the exponential function, m and n are the number of objective functions, represents the historical optimization path weight of the objective function, It is the sum of the weights of the historical optimization paths of all objective functions in the tth generation; S36, evaluate the fitness of each generation of population, select the particle with the highest fitness as the current optimal solution, and update it to the global optimal solution set P * Among them represents the Pareto optimal solution set; S37, after reaching the preset termination condition, the final Pareto optimal solution set P * Output, which serves as the basis for optimization in subsequent steps.

3. According to the particle swarm optimization-based high-voltage transformer remote cost control parameter adjustment method of claim 1, it is characterized in that: The S4 specifically includes: S41. Obtain the final Pareto optimal solution set P * , and further global search for the optimal solution set is performed in combination with the multi-dimensional quantum entropy coding optimization algorithm; S42, through the quantum entropy coding technology, the search space of each particle position in the initial population is divided and encoded in multiple dimensions to generate a particle position vector with quantum state characteristics S43, in each iteration of the multidimensional quantum entropy coding optimization algorithm, based on the dynamic weight adjustment formula Particle position vector Update the weights and generate new optimization parameter combinations: Among them, p j Represents particles The probability distribution in the jth dimension, θ j is the quantum state angle, is the phase difference between the particle in dimension j and dimension k, α and γ are adjustment factors, and are the weight coefficients in the current and previous generations, respectively; S44. In each iteration, a multi-level collaborative mechanism of adaptive quantum states is introduced, so that the quantum states of different particles can influence and co-evolve each other, and dynamically adjust the optimization direction and strategy of particles through the interaction of quantum entangled states; S45. According to quantum information entropy The particles are screened and optimized based on the calculation results, the particle position vectors with higher information entropy values ​​are selected, and the selected particle position vectors are input into the adaptive deep learning decoding network; S46. Through a complex nonlinear decoding process, the particle position vector is converted into a high-dimensional feature representation: Among them, A decode is the final output weight matrix, W decode is the weight matrix of the decoding network, b decode is the bias vector, d final is the bias vector of the output layer, tanh is the nonlinear activation function, To optimize the parameter combination; S47. Perform cluster analysis on the decoded particle positions, calculate the similarity between the particle positions and the current cluster center, merge particles with higher similarity according to the similarity of quantum states, adjust the structure and distribution of the population, and output the optimized cost control parameter combination.

4. According to the particle swarm optimization-based high-voltage transformer remote cost control parameter adjustment method of claim 1, it is characterized in that: The S5 specifically includes: S51, obtain the optimized cost control parameter combination, and use it as input, apply the deep adaptive enhanced particle swarm optimization algorithm for fine optimization, and generate the optimized parameter vector S52, updating and further optimizing the optimized parameter vector in multiple dimensions: Among them, η is the dynamic learning rate, is the weight updated by the dynamic adjustment formula, ⊙ represents element-by-element multiplication, λ is the adjustment factor, is the calculated value of quantum information entropy, Represents the parameter vector Gradient operation, V k is the adaptive perturbation vector, is the phase difference between particles in different dimensions; S53, combined with the quantum perturbation mechanism in particle swarm optimization, the optimized parameter vector Used to adjust weight matrices and bias vectors in deep neural network models; S54. Based on the adjusted weight matrix and bias vector, the deep reinforcement learning algorithm is used to dynamically evaluate and optimize the strategy combination, so that the model can be adaptively adjusted to cope with changes in the cost control strategy, and the final optimized cost control parameter combination R is generated through the continuously updated strategy combination. * ; S55, the generated final optimized cost control parameter combination R * Combined with the preprocessed dataset, cross validation was performed.

5. According to the particle swarm optimization-based high-voltage transformer remote cost control parameter adjustment method of claim 1, it is characterized in that: The S6 specifically includes: S61. Obtain the final optimized cost control parameter combination R * , and combine it with the preprocessed data set to construct a high-dimensional parameter space P, where each dimension corresponds to an optimized cost control parameter, and initialize the quantum particle swarm; S62. Introduce an adaptive quantum particle coordination mechanism to coordinately optimize the behavior of quantum particle groups and adaptively adjust the interaction between particles based on local and global information in the particle group: Where ρ is the synergy factor, represents the current position of the neighboring particles, represents the gradient calculation of the particle position vector, is the position of the ith particle in the kth dimension, and L represents the number of neighboring particles considered; S63. After collaborative optimization, a dynamic robustness evaluation method based on quantum particle filtering is used to evaluate the performance of the optimization parameters. Get the final stability index S of the particle swarm i : Where m is the total number of particles, represents the distance between the particle and the optimized parameters, Indicates the distance between the optimized parameters and the final result; S64. According to the multi-dimensional dynamic robustness evaluation results, select the one with the highest stability index S i The cost control parameter combination is determined and verified to obtain the optimal parameter combination.

6. According to the particle swarm optimization-based high-voltage transformer remote cost control parameter adjustment method of claim 1, it is characterized in that: The S7 specifically includes: S71, the verified optimal cost control parameter combination R * Applied to the actual operation of high-voltage special transformers, a multi-modal dynamic adaptive particle swarm is initialized and a comprehensive data set is constructed, which includes real-time monitoring data and historical operation data; S72. Preprocess the real-time monitoring data, use a multi-modal feature extraction method to extract the operating characteristics of the high-voltage dedicated transformer, and form a multi-dimensional feature vector S73. Introduce a dynamic optimization feedback mechanism based on adaptive prediction to predict future transformer operating conditions by establishing a prediction model based on deep learning. The predicted feature vector is used as input and compared with the real-time feature vector Together, input the multimodal dynamic adaptive particle swarm optimization algorithm to update the particle position: Among them, α is the global adaptation factor, which controls the particle to move toward the global optimal solution G t The movement in the direction, γ is the real-time adaptation factor, which adjusts the position of the particle based on the difference between the current and historical features, F historical represents the historical eigenvector, δ is the prediction feedback factor, which is used to adjust the position of the particle according to the future prediction eigenvector, ∈ is the random perturbation factor, is a random perturbation vector used for particle exploration; S74. Real-time monitoring of fee control parameters after each iteration update And combine it with the historical optimization parameters R * For comparison, the fitness function is calculated using a back-analysis method based on historical data. Where N is the number of samples in the historical data set, is the distance between the current feature vector and the historical feature vector, is the distance between the current cost control parameters and the historical optimal parameter combination; S75, according to the evaluation result of the fitness function, dynamically adjust the cost control parameter combination in the particle swarm and update the global optimal solution G of the particle swarm t+1 , and continue to iterate until the preset convergence conditions are met, and finally output the real-time adaptive optimization cost control parameter combination R final , used in long-term operation monitoring and fee control adjustment of high-voltage special transformers.

Images