Energy efficiency optimization method for renewable energy power generation side power system

Abstract

The invention relates to the technical field of power system energy efficiency, and discloses a renewable energy power generation side power system energy efficiency optimization method. The method comprises the following steps: acquiring renewable energy power generation output data, load demand data and system stability parameters in different operation scenes; for each operation scene, determining an associated reference operation scene group according to the load demand of the operation scene, and calculating the energy efficiency influence degree of each evaluation time point; integrating the energy efficiency and load distribution data of all the operation scenes, calculating a preliminary energy efficiency correlation index, and performing correction in combination with the energy efficiency influence degree and the stability parameter to obtain a final energy efficiency correlation index; non-energy efficiency factor weights of all time points are deduced according to load requirements and stability parameter distribution; identifying a core evaluation time point from all evaluation time points according to the final energy efficiency association index and the non-energy efficiency factor weight; and establishing an energy efficiency optimization model by using the energy efficiency data and the stability parameters of the core point, and adjusting the operation state of the power system.

Description

Technical Field

[0001] This invention relates to the field of power system energy efficiency technology, specifically to a method for optimizing the energy efficiency of a power system on the renewable energy generation side. Background Technology

[0002] Power system energy efficiency optimization is a key step in improving energy utilization efficiency and reducing operating costs. With the large-scale integration of renewable energy into the grid, the operating characteristics of the power system have changed significantly. Renewable energy generation output is intermittent, fluctuating, and uncertain; for example, wind power and photovoltaic power are significantly affected by weather conditions, leading to unstable power output on the generation side. Traditional power system energy efficiency optimization methods are mainly aimed at fossil fuel-dominated power structures, where power generation output is highly controllable, and energy efficiency optimization focuses on static or quasi-static issues such as unit combination and economic dispatch. These methods are ill-suited to the dynamic operating environment under a high proportion of renewable energy integration.

[0003] Current energy efficiency assessments mostly employ global or long-term average indicators, such as plant power consumption rate and line loss rate, lacking a detailed characterization of short-term and localized energy efficiency states. Power system operation is a continuous dynamic process, with energy efficiency levels fluctuating over time and depending on operating modes. Especially in scenarios with high renewable energy penetration, the system needs to frequently adjust the output of conventional units to balance renewable energy fluctuations, a process that may be accompanied by instantaneous reductions in energy efficiency. Traditional methods cannot effectively identify these critical transient processes and their impact on overall energy efficiency.

[0004] The spatiotemporal matching between load demand and renewable energy output has a significant impact on system energy efficiency. If renewable energy output is insufficient during peak load periods, less efficient peak-shaving units must be activated, leading to a decrease in overall system energy efficiency. Conversely, if renewable energy output is excessive and load demand is insufficient, wind and solar power curtailment may occur, resulting in energy waste. Existing optimization models often consider energy efficiency and stability objectives separately or perform simple weighting, failing to deeply reveal the inherent relationship and trade-off mechanisms between the two under complex operating scenarios.

[0005] Energy efficiency optimization measures may affect system stability. For example, excessive pursuit of energy efficiency may reduce the system's spinning reserve capacity and jeopardize frequency stability. Ensuring the safe and stable operation of the system while improving energy efficiency is the core challenge in achieving efficient and reliable operation of renewable energy power systems. Therefore, a new method is needed that can comprehensively consider the characteristics of renewable energy, load demand, and system stability, and perform fine-grained optimization across multiple time scales. Summary of the Invention

[0006] The purpose of this invention is to provide a method for optimizing the energy efficiency of a power system on the renewable energy generation side, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a method for optimizing the energy efficiency of a renewable energy generation-side power system, the method comprising: Data on renewable energy power generation output, load demand, and system stability parameters are collected under different operating scenarios of the power system. Each operating scenario corresponds to a specific time window, and multiple evaluation time points are evenly set within the time window. For each operating scenario, its associated reference operating scenario group is determined based on the load demand data of that operating scenario. For each assessment time point, the energy efficiency impact of that operating scenario at that assessment time point is calculated based on the energy efficiency index distribution and load demand distribution of the associated reference operating scenario group. Integrate the energy efficiency index distribution and load demand distribution of all operating scenarios, and calculate the preliminary energy efficiency correlation index for each assessment time point; By combining the energy efficiency impact and system stability parameters of each operating scenario, the preliminary energy efficiency correlation indicators are revised to obtain the final energy efficiency correlation indicators at each evaluation time point. Based on the load demand distribution and stability parameter distribution of all operating scenarios, the weights of non-energy efficiency factors at each evaluation time point are derived. Based on the final energy efficiency related indicators and the weights of non-energy efficiency factors, the core assessment time points are identified from all assessment time points; By utilizing energy efficiency index data and system stability parameters at key assessment time points, an energy efficiency optimization model is established, and the energy efficiency optimization model is applied to adjust the operating status of the power system.

[0008] Preferably, the calculation of the energy efficiency impact of the operating scenario at the evaluation time point includes: The variation of the energy efficiency index of the associated reference operating scenario group for this operating scenario at the evaluation time point is normalized to obtain the energy efficiency fluctuation coefficient. Extract the load demand variation of the associated reference operating scenario group for this operating scenario as the load fluctuation coefficient; The energy efficiency fluctuation coefficients of all operating scenarios are aggregated to form an energy efficiency fluctuation sequence, and the load fluctuation coefficients of all operating scenarios are aggregated to form a load fluctuation sequence. Elements with the same index in the two sequences correspond to the same operating scenario. Analyze the correlation strength between energy efficiency fluctuation series and load fluctuation series, and calculate the weight of energy efficiency impact; The energy efficiency impact weight is obtained by multiplying the energy efficiency fluctuation coefficient by the energy efficiency impact weight.

[0009] Preferably, the calculation of the preliminary energy efficiency correlation index at each evaluation time point includes: Obtain the energy efficiency index values ​​of all operating scenarios at the target evaluation time point, and construct a set of energy efficiency index values; The information entropy index of the set of energy efficiency index values ​​is used as the first uncertainty measure. Obtain the load demand values ​​for all operating scenarios and construct a load demand value set; The information entropy index of the set of load demand values ​​is used as a second uncertainty measure. Establish a joint probability distribution of the set of energy efficiency index values ​​and the set of load demand values; Extract marginal distribution features and covariance features from the joint probability distribution; Input the first uncertainty measure and the second uncertainty measure into the feature fusion network; The features of the joint probability distribution are input into the same feature fusion network, and the preliminary energy efficiency correlation index is output through the feature fusion network.

[0010] Preferably, the correction of the preliminary energy efficiency correlation indicators includes: All operating scenarios are subjected to K-means clustering analysis based on load demand values ​​to generate load cluster groups; The operation scenario numbers within each load cluster group are sorted to form a group number sequence; All operating scenarios are clustered according to their energy efficiency index values ​​to generate energy efficiency cluster groups; The operating scenario numbers within each energy efficiency cluster are sorted to form an energy efficiency number sequence; Calculate the overlap index between the group number sequence and the energy efficiency number sequence, and extract the energy efficiency impact data of the operation scenario corresponding to the non-overlapping numbers; Construct an energy efficiency impact matrix and perform singular value decomposition. Take the largest singular value as the energy efficiency impact weight coefficient. Multiply the overlap index with the energy efficiency impact weight coefficient to obtain the correction factor. Use the correction factor to adjust the weight of the preliminary energy efficiency correlation index to obtain the weight of non-energy efficiency factors.

[0011] Preferably, the derivation of the non-energy efficiency factor weights at each evaluation time point includes: Kernel density estimation is performed on the load demand values ​​of the associated reference operating scenario group for the target operating scenario; Obtain the probability density function curve of load demand; The same kernel density estimation is performed on the load demand values ​​for all operating scenarios; Obtain the global load demand probability density function curve; Calculate the Fréchet distance between two probability density function curves; The Frescher distance is input into the sigmoid activation function for transformation; Calculate the arithmetic mean of the transformation results for all running scenarios, and then perform linear normalization on the mean. The complement of the normalized result is used as the weight of non-energy efficiency factors.

[0012] Preferably, the identification of core evaluation time points includes: Calculate the ratio of the final energy efficiency-related indicators to the weights of non-energy efficiency factors at each assessment time point, and standardize the ratio to obtain the time point importance score. Evaluation time points whose importance scores exceed a set threshold are identified as core evaluation time points.

[0013] Preferably, the establishment of the energy efficiency optimization model includes: The energy efficiency index data, load demand data, and system stability parameters at the core assessment time points are input into the multivariate optimization algorithm to train the energy efficiency optimization model, which includes the minimization of system operating costs and stability constraints.

[0014] Preferably, the application of the energy efficiency optimization model to adjust the operating state of the power system includes: Real-time acquisition of renewable energy output, load demand, and stability parameters of the power system; Extract the feature values ​​of real-time data at the core assessment time points and input them into the energy efficiency optimization model; Based on the optimization strategy output by the energy efficiency optimization model, adjust the resource allocation on the power generation side and the system operating parameters.

[0015] Preferably, determining the associated reference operating scenario group for each operating scenario based on the load demand data of that operating scenario includes: A load similarity interval is defined centered on the load demand value of this operating scenario; Other operating scenarios whose load demand values ​​fall within this range, excluding this operating scenario, are selected to form a group of associated reference operating scenarios.

[0016] Preferably, the method further includes: Generate a set of verification scenarios based on historical operational data; The system processes the verification scenario set through an energy efficiency optimization model, outputs the optimized energy efficiency index, compares the changes in energy efficiency index before and after optimization, and records the system operation status data.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention achieves dynamic and precise evaluation of power system energy efficiency by setting multiple time windows and evaluation time points. Traditional methods often focus on long-term average energy efficiency, while this method evenly distributes evaluation points within a specific operating scenario's time window, capturing instantaneous fluctuations and key changes in energy efficiency, providing detailed time-section data for accurate optimization.

[0018] By introducing associated reference operating scenario groups and energy efficiency impact calculation, the contextual relevance of energy efficiency assessment is enhanced. Energy efficiency performance under different operating scenarios is comparable. This method, by associating scenario groups with similar load demands and analyzing energy efficiency distribution within each group, makes the energy efficiency assessment results of the current scenario more consistent with its operating environment, thus improving the rationality and accuracy of the assessment.

[0019] The preliminary energy efficiency indicators are revised by comprehensively considering the impact of energy efficiency and system stability parameters, making the energy efficiency assessment results closer to actual operational needs. High energy efficiency alone does not necessarily represent the optimal operating state; system stability constraints must also be considered. This method uses stability parameters as correction factors to ensure that the final energy efficiency-related indicators reflect both the energy efficiency level and the requirements for safe system operation.

[0020] By deriving the weights of non-energy-efficiency factors and identifying key evaluation time points, this method highlights crucial aspects of optimization decision-making. Power system operation is influenced by various factors; by quantifying the weights of non-energy-efficiency factors, this method helps identify the time points that have the greatest impact on overall system performance, enabling optimization resources to be concentrated on the most critical aspects.

[0021] The established energy efficiency optimization model, based on detailed data from key assessment time points, enables precise adjustments to the operational status. The model fully utilizes energy efficiency and stability information from these key time points, generating more specific and targeted optimization strategies. This effectively guides adjustments to power generation output and the configuration of reserve capacity, thereby significantly improving the overall energy efficiency of renewable energy generation while ensuring system stability. This method provides a scientific and practical technical means for energy efficiency management of high-proportion renewable energy power systems. Attached Figure Description

[0022] Figure 1 This is a graph showing the changes in energy efficiency-related indicators; Figure 2 A flowchart for calculating the energy efficiency impact of the operating scenario at the evaluation time point; Figure 3 A flowchart for the initial revision of energy efficiency-related indicators; Figure 4 This is a comparison chart of load demand kernel density estimation and distribution. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figure 1 This invention provides a method for optimizing the energy efficiency of a power system on the renewable energy generation side. The method includes: collecting renewable energy generation output data, load demand data, and system stability parameters of the power system under different operating scenarios. Each operating scenario corresponds to a specific time window, and multiple evaluation time points are evenly set within the time window. For each operating scenario, a group of associated reference operating scenarios is determined based on the load demand data of that operating scenario. For each evaluation time point, the energy efficiency impact of that operating scenario at that evaluation time point is calculated based on the energy efficiency index distribution and load demand distribution of the associated reference operating scenario group. The energy efficiency index distribution and load demand distribution of all operating scenarios are integrated to calculate the preliminary energy efficiency correlation index for each evaluation time point. The preliminary energy efficiency correlation index is corrected by combining the energy efficiency impact and system stability parameters of each operating scenario to obtain the final energy efficiency correlation index for each evaluation time point. The weights of non-energy efficiency factors at each evaluation time point are derived based on the load demand distribution and stability parameter distribution of all operating scenarios. The core evaluation time point is identified from all evaluation time points based on the final energy efficiency correlation index and the weights of non-energy efficiency factors. By utilizing energy efficiency index data and system stability parameters at key assessment time points, an energy efficiency optimization model is established, and the energy efficiency optimization model is applied to adjust the operating status of the power system.

[0025] Example 1: See Figure 2For each operating scenario, the energy efficiency impact at the evaluation time point is calculated. Specifically, this involves normalizing the variability of energy efficiency indicators of the associated reference operating scenario group at the evaluation time point to obtain the energy efficiency fluctuation coefficient. The normalization process uses a minimum-maximum scaling method to map the variability to a value range of zero to one. The variability is obtained by calculating the standard deviation of the energy efficiency indicator values ​​of each operating scenario within the associated reference operating scenario group at the target evaluation time point. The minimum-maximum scaling formula uses the global minimum and maximum values ​​of the variability of the energy efficiency indicators in this group for linear transformation, making the energy efficiency fluctuation coefficients of different operating scenarios comparable and eliminating analytical biases caused by differences in dimensions. The load demand variability of the associated reference operating scenario group is extracted as the load fluctuation coefficient. The load demand variability is also obtained by calculating the standard deviation of the load demand data of the operating scenarios within the group. The load demand data comes from the average load power measurement value of each operating scenario within the corresponding time window. The load fluctuation coefficient characterizes the load fluctuation intensity of the operating scenario group, reflecting the degree of impact of external demand uncertainty on system operation. Energy efficiency fluctuation coefficients from all operating scenarios are aggregated to form an energy efficiency fluctuation sequence, and load fluctuation coefficients from all operating scenarios are aggregated to form a load fluctuation sequence. Strict index consistency is maintained during sequence construction; that is, the i-th element in the energy efficiency fluctuation sequence corresponds to the i-th element in the load fluctuation sequence, both within the same operating scenario. The sequence data structure uses a one-dimensional array, with elements arranged in ascending order of operating scenario numbers to facilitate subsequent correlation analysis between sequences. The correlation strength between the energy efficiency fluctuation sequence and the load fluctuation sequence is analyzed, and the energy efficiency impact weight is calculated. The correlation strength analysis uses the Pearson correlation coefficient algorithm, calculating the ratio of the product of the covariance and standard deviation of the two sequences. The Pearson correlation coefficient ranges from -1 to +1; a larger absolute value indicates a stronger linear correlation, and the sign indicates the direction of the correlation. The energy efficiency impact weight is multiplied by the energy efficiency fluctuation coefficient to obtain the energy efficiency impact degree. The multiplication operation is performed at the element level. The energy efficiency impact weight is used as a scaling factor to adjust the contribution of the energy efficiency fluctuation coefficient. A larger energy efficiency impact degree indicates a more significant impact of the energy efficiency fluctuation of that operating scenario on the overall energy efficiency status of the system at the evaluation time point.

[0026] The preliminary energy efficiency correlation indicators for each assessment time point are calculated. Specifically, this includes obtaining the energy efficiency indicator values ​​for all operating scenarios at the target assessment time point, constructing an energy efficiency indicator value set, which contains the real-time energy efficiency measurement results for each operating scenario at the target assessment time point. The energy efficiency indicator is defined as the ratio of renewable energy power generation output to the total system input energy. The set data is organized in a list structure, with each element index corresponding one-to-one with the operating scenario number. The information entropy index of the energy efficiency indicator value set is calculated as the first uncertainty measure. The information entropy calculation is based on the probability distribution estimation of the energy efficiency indicator values. The probability distribution is obtained by statistically analyzing the frequency of occurrence of energy efficiency indicator values ​​in each numerical interval. A higher information entropy value indicates higher randomness of the energy efficiency indicator values ​​and a more unstable system energy efficiency state. The load demand values ​​for all operating scenarios are obtained, and a load demand value set is constructed. The load demand value set originates from the load power sampling values ​​defined in the operating scenario. The set construction method is similar to that of the energy efficiency indicator value set, maintaining the same data structure and indexing rules. The information entropy index of the load demand value set is calculated as the second uncertainty measure. The load demand information entropy uses the same calculation method as the energy efficiency indicator information entropy to measure the dispersion and unpredictability of the load demand values.

[0027] A joint probability distribution of the energy efficiency index value set and the load demand value set is established. This joint probability distribution is constructed using a two-dimensional histogram statistical method, dividing the ranges of both energy efficiency index values ​​and load demand values ​​into several discrete intervals. The frequency of data points occurring within each two-dimensional interval is statistically analyzed as a joint probability estimate. Marginal distribution features and covariance features are extracted from the joint probability distribution. Marginal distribution features are obtained by summing the rows and columns of the joint probability distribution matrix, describing the probability distribution characteristics of each variable (energy efficiency index and load demand). Covariance features are calculated by averaging the products of the deviations of the two variable values ​​from their respective means, characterizing the degree of linear correlation between energy efficiency index and load demand. The first and second uncertainty measures are input into a feature fusion network. This feature fusion network employs a three-layer fully connected neural network structure. The input layer has two neurons that receive the first and second uncertainty measures respectively, the hidden layer contains ten neurons using the ReLU activation function, and the output layer has one neuron that generates an intermediate feature vector. The features of the joint probability distribution are input into the same feature fusion network. The marginal distribution features and covariance features are flattened and then concatenated into a one-dimensional vector as the network input. The network output layer weights and fuses the intermediate feature vector with the joint probability distribution feature vector. A preliminary energy efficiency correlation index between zero and one is generated by the sigmoid activation function. This index comprehensively reflects the correlation strength between energy efficiency and load at the evaluation time point.

[0028] Example 2: See Figure 3The preliminary energy efficiency correlation indicators are revised, specifically by performing K-means clustering analysis on all operating scenarios according to their load demand values ​​to generate load cluster groups. The K-means clustering algorithm calculates the Euclidean distance based on the load demand vector of the operating scenarios. The initial cluster centers are randomly selected, and the cluster center positions are continuously updated during the iteration process until the center point moves less than a set threshold. The load clustering grouping results group operating scenarios with similar load characteristics into the same cluster. The number of groups is determined by the silhouette coefficient method, and the number of groups corresponding to the maximum silhouette coefficient is selected as the final number of clusters. The operating scenario numbers within each load cluster group are sorted to form a group number sequence. The sorting is based on the magnitude of the load demand value of the operating scenarios in ascending order. Operating scenarios with the same load demand value are arranged according to the order of their data collection timestamps. The group number sequence records the numbering order of the operating scenarios within the same load cluster group, and the sequence structure is stored in a linked list for easy dynamic adjustment. All operating scenarios are clustered by density peak values ​​according to their energy efficiency index values, generating energy efficiency cluster groups. The density peak clustering algorithm calculates the local density of each operating scenario and the minimum distance to higher density points. The local density is defined by the number of other operating scenarios contained in the neighborhood of the operating scenario in the energy efficiency index value space with a given cutoff distance as the radius. The minimum distance is the minimum distance from the operating scenario to all operating scenarios with higher local densities. The operating scenario with the larger product of local density and minimum distance is selected as the cluster center, and the remaining operating scenarios are assigned to the group to the nearest cluster center. The operating scenario numbers within each energy efficiency cluster group are sorted to form an energy efficiency number sequence. The sorting is based on the numerical value of the operating scenarios' energy efficiency index values ​​in descending order. Operating scenarios with the same energy efficiency index value are sorted according to their system stability parameter values ​​in descending order. The energy efficiency number sequence reflects the energy efficiency level distribution of the operating scenarios within the energy efficiency cluster group.

[0029] The overlap index between the group number sequence and the energy efficiency number sequence is calculated using the Jaccard similarity coefficient, which is defined as the ratio of the number of elements in the intersection to the number of elements in the union of two sequences. A coefficient closer to one indicates a higher similarity between the two sequences. The overlap index reflects the consistency between the load clustering and energy efficiency clustering results. Energy efficiency impact data for non-overlapping numbers is extracted. Non-overlapping numbers refer to operating scenario identifiers located at different positions in the group number sequence and the energy efficiency number sequence. These operating scenarios exhibit atypical correlations between load characteristics and energy efficiency characteristics. The energy efficiency impact data is derived from the energy efficiency impact value matrix of each operating scenario at each evaluation time point, calculated previously. An energy efficiency impact matrix is ​​constructed and singular value decomposition (SVD) is performed. The row dimension of the energy efficiency impact matrix corresponds to the total number of operating scenarios, and the column dimension corresponds to the total number of evaluation time points. The matrix elements are the energy efficiency impact values ​​of a specific operating scenario at a specific evaluation time point. The matrix is ​​stored using a two-dimensional array data structure. SVD decomposes the energy efficiency impact matrix into a product of three matrices: a left singular vector matrix, a singular value diagonal matrix, and the transpose of the right singular vector matrix. The largest singular value is taken as the energy efficiency impact weight coefficient. The largest singular value is located at the top left corner of the main diagonal of the singular value diagonal matrix. The magnitude of the value reflects the concentration of the main energy contained in the energy efficiency impact matrix. The energy efficiency impact weight coefficient characterizes the dominant influence of energy efficiency fluctuations on the overall energy efficiency status of the system.

[0030] The correction factor is obtained by multiplying the overlap index by the energy efficiency impact weighting coefficient. This multiplication is performed at the scalar level. The correction factor comprehensively considers the consistency of the load-energy efficiency correlation characteristics and the dominant influence of energy efficiency fluctuations. The correction factor value fluctuates between zero and one; a larger value indicates a greater adjustment required for the initial energy efficiency correlation index. The correction factor is then used to weight the initial energy efficiency correlation index to obtain the weights of non-energy efficiency factors. This weighting adjustment uses a multiplicative correction mode, multiplying the initial energy efficiency correlation index by the correction factor to obtain the adjusted value. The weights of non-energy efficiency factors ultimately reflect the comprehensive influence of system parameters other than energy efficiency factors on the energy efficiency assessment results. A larger weight value indicates a higher proportion of non-energy efficiency factors need to be considered in energy efficiency optimization decisions.

[0031] The generation process of load clustering requires setting an appropriate number of clusters. The silhouette coefficient method determines the optimal number of clusters by calculating the average silhouette width of all operating scenarios under different numbers of clusters. An average silhouette width close to 1 indicates that the clustering results have high cohesion and high separability. The iterative convergence condition of the K-means clustering algorithm is set to the cluster center movement distance being less than one part per million between two consecutive iterations, ensuring the stability of the clustering results. The cutoff distance parameter of density peak clustering is determined by trying different values ​​of clustering quality, and the cutoff distance value that produces the clearest decision map is selected. The decision map uses the local density of each operating scenario as the x-axis and the minimum distance as the y-axis. The cluster center points should have both high local density and high minimum distance characteristics. The singular value decomposition of the energy efficiency influence matrix is ​​implemented using the Jacobi iterative algorithm. The algorithm transforms the original matrix into a bidiagonal matrix through a series of orthogonal transformations before decomposition. The iterative process continues until the sum of the absolute values ​​of the off-diagonal elements is less than a set tolerance value, ensuring the decomposition accuracy. The calculation of the correction factor introduces the product relationship between the overlap index and the energy efficiency impact weight coefficient. This calculation method can capture the need to correct the preliminary energy efficiency correlation index when the load clustering and energy efficiency clustering results are inconsistent. When the two clustering results are highly consistent, the correction factor is close to one, and the preliminary energy efficiency correlation index remains basically unchanged. When the two clustering results are significantly different, the correction factor deviates from one, and the preliminary energy efficiency correlation index needs to be adjusted significantly.

[0032] The acquisition of non-energy efficiency factor weights marks the completion of the preliminary energy efficiency correlation index correction process. These weights will be used in the subsequent core assessment time point identification stage. The accuracy of the weight values ​​directly affects the effectiveness of the final energy efficiency optimization model. The entire correction process systematically integrates multi-dimensional information such as load characteristics, energy efficiency characteristics, and system stability through multi-step cluster analysis and matrix operations, providing a more comprehensive data foundation for energy efficiency optimization of the renewable energy generation side of the power system. Comparative analysis of load clustering and energy efficiency clustering reveals the complex correlation between load and energy efficiency during system operation. Singular value decomposition technology extracts the main influence patterns from the energy efficiency influence matrix. The introduction of correction factors enables the preliminary energy efficiency correlation index to more accurately reflect the energy efficiency status of the actual system. The calculation of non-energy efficiency factor weights provides a quantitative basis for distinguishing energy efficiency dominant factors from other influencing factors.

[0033] Example 3: Deriving the weights of non-energy efficiency factors at each evaluation time point, specifically including kernel density estimation of the load demand values ​​of the associated reference operating scenario group for the target operating scenario. The kernel density estimation uses a Gaussian kernel function to smooth the load demand data to construct a probability density distribution. The mathematical expression of the Gaussian kernel function is as follows: , Where: symbol This represents the deviation of the standardized load demand value, calculated as the difference between the original load demand value and the sample mean, divided by the sample standard deviation; (symbol) Pi represents a mathematical constant; symbol The base of the natural logarithm is an irrational number. The kernel density estimation process treats each data point as a distribution center, forming a smooth and continuous probability density curve by superimposing the Gaussian distributions of all points. The bandwidth parameter is chosen using an empirical rule based on the data standard deviation; the bandwidth value directly affects the smoothness of the curve. Too small a bandwidth will result in an overly rugged curve, while too large a bandwidth will obscure distribution details. The probability density function curve of load demand is obtained. This curve, with load demand value on the x-axis and probability density on the y-axis, visually displays the distribution of load demand within the associated reference operating scenario group. The curve peak corresponds to the range of values ​​where load demand is most concentrated, and the width of the curve waveform reflects the dispersion of the data.

[0034] The same kernel density estimation is performed on the load demand values ​​of all operating scenarios. The estimation process uses the exact same kernel function type and bandwidth parameters as the associated reference operating scenario group of the target operating scenario, ensuring the consistency of the estimation method and making the probability density functions of different operating scenario groups comparable. A global load demand probability density function curve is obtained. This curve is constructed based on the load demand data of all operating scenarios, reflecting the overall distribution characteristics of load demand during the entire system operation. The difference between the global curve and the curve of a specific operating scenario group reflects the degree of deviation between local load characteristics and global load characteristics. The Friesian distance between the two probability density function curves is calculated. The Friesian distance is a mathematical tool for measuring the similarity between two curves. It can be figuratively understood as the shortest leash length required for a person to walk a dog along a curve. The calculation process uses a dynamic programming algorithm on a discretized sequence of curve points. The algorithm constructs a distance matrix to store the Euclidean distance between all pairs of points and finds a path from the starting point to the ending point that minimizes the maximum distance between any adjacent pairs of points along the path. This minimized maximum distance is the Friesian distance. A smaller Fraser distance value indicates that the load demand distribution of the target operating scenario group is more similar to the global distribution, while a larger distance value indicates that the load characteristics of the group are more unique.

[0035] The Friesian distance is input into the sigmoid activation function for transformation. The sigmoid function has an S-shaped curve characteristic, capable of mapping any real number to the interval between zero and one. The function's output value gradually approaches one as the input value increases. This transformation converts the absolute value of the Friesian distance into a relative proportional value, eliminating dimensional differences and facilitating subsequent calculations. The mathematical expression of the sigmoid function is: ,in This represents the Fraser distance value.

[0036] The arithmetic mean of the transformation results for all running scenarios is calculated. This is achieved by summing the sigmoid output values ​​for all running scenarios and dividing by the total number of running scenarios. The average value represents the central tendency of the load distribution of all running scenarios after the sigmoid transformation, reflecting the average deviation from the global distribution. The average value is then linearly normalized, scaling the arithmetic mean to a range of zero to one. The formula is as follows: , in: Represents the original arithmetic mean. This represents the minimum value among the average values ​​of all running scenarios. Represents the maximum value. This represents the normalization result.

[0037] The complement of the normalized result is used as the weight of non-energy efficiency factors. The complement is calculated by subtracting the normalized average from one. The final weight of non-energy efficiency factors ranges from zero to one. The closer the weight value is to one, the lower the influence of non-energy efficiency factors such as load demand distribution on the system energy efficiency assessment; the closer the weight value is to zero, the higher the influence of non-energy efficiency factors. The derivation process of the non-energy efficiency factor weights establishes a complete mapping relationship from the statistical characteristics of load data to the weight values. The weight values ​​will serve as key parameters in the identification and decision-making of subsequent core assessment time points.

[0038] The Gaussian kernel function for kernel density estimation is chosen based on its favorable mathematical properties. The kernel function exhibits symmetry and smoothness, effectively reflecting the probability distribution characteristics of the load data. The bandwidth parameter directly affects the shape of the probability density function curve. Optimal bandwidth selection requires a balance between bias and variance; excessive bandwidth increases estimation bias, while insufficient bandwidth increases estimation variance. The accuracy of the Fréchet distance calculation depends on the granularity of the curve discretization. Smaller intervals between discrete points result in higher accuracy but also greater computational cost; a suitable discretization level must be selected based on actual needs. The transformation characteristics of the sigmoid function give moderately sized Fréchet distances high sensitivity, while extremely large or small distance values ​​are compressed into the saturation region. This nonlinear transformation enhances the discriminative power of distance values ​​within the normal range. The arithmetic mean calculation assumes equal contributions from each operating scenario. Normalization eliminates the influence of data range, giving the weight values ​​a standardized scale. The complement operation achieves an inverse representation of the degree of influence, making the weight values ​​negatively correlated with the influence intensity, consistent with the physical meaning of non-energy efficiency factor weights.

[0039] See Figure 4A key step in deriving the weights of non-energy efficiency factors is analyzing the distribution characteristics of load demand. The top figure focuses on the probability density distribution of load demand within a related reference operating scenario group, using load demand as the horizontal axis and probability density as the vertical axis to visually present the numerical clustering of load demand within the group, reflecting the consistency of load characteristics. The bottom figure compares the probability density distribution of global load and related reference group load using overlaid histograms. Light gray represents global load, and dark gray represents related reference group load, clearly showing the differences in distribution between the two. The related reference group load has a higher density in the 250-300MW range, while the global load has a more significant proportion in the 200-250MW range. This distribution comparison is the core basis for calculating the "Freche distance": by quantifying the similarity of the two load distribution curves, combined with sigmoid transformation, linear normalization, and complement operations, the weights of non-energy efficiency factors are ultimately derived. This provides intuitive distribution characteristic support for subsequent steps, ensuring that the derivation of non-energy efficiency factor weights not only conforms to the statistical regularity of load data but also accurately reflects the degree of deviation between local and global load characteristics, serving as a key visual basis for comprehensive consideration.

[0040] Example 4: Identifying core assessment time points specifically involves calculating the ratio of the final energy efficiency-related index to the weight of non-energy efficiency factors at each assessment time point. The ratio calculation uses division, dividing the final energy efficiency-related index value at the same assessment time point by the non-energy efficiency factor weight value to obtain the raw ratio value. The ratio is then standardized to obtain a time point importance score. Standardization uses the Z-score method, calculating the difference between the ratio value at each assessment time point and the mean ratio of all assessment time points, then dividing by the standard deviation of the ratio values ​​at all assessment time points. The Z-score standardization formula is: , in: Represents the original ratio value. The representative ratio is the mean. The standard deviation of the representative ratio is used. Evaluation time points whose importance scores exceed a set threshold are identified as core evaluation time points. The set threshold value is determined by analyzing the percentile distribution of the importance scores of time points in historical operating data. The 75th percentile is selected as the discrimination threshold. Evaluation time points whose importance scores are higher than this threshold are considered core evaluation time points that have a significant impact on the optimization of system energy efficiency.

[0041] An energy efficiency optimization model is established, specifically by inputting energy efficiency index data, load demand data, and system stability parameters at core assessment time points into a multivariate optimization algorithm to train the model. The multivariate optimization algorithm employs linear programming, with the objective function set as minimizing system operating costs. Constraints include system stability constraints. System operating cost calculation considers power generation fuel costs, equipment maintenance costs, environmental governance costs, and renewable energy curtailment penalty costs. Power generation fuel costs are directly proportional to the generator output; equipment maintenance costs are calculated based on accumulated equipment operating time; environmental governance costs are related to carbon emissions from fossil fuel power generation; and renewable energy curtailment penalty costs are set for the curtailment of wind and solar power generation. Stability constraints include voltage stability constraints, frequency stability constraints, and power angle stability constraints. Voltage stability constraints require that the voltage deviation at each node does not exceed ±5% of the rated value; frequency stability constraints stipulate that the system frequency fluctuation range is between 49.8 Hz and 50.2 Hz; and power angle stability constraints set that the relative power angle difference of the generators does not exceed 120 degrees.

[0042] An energy efficiency optimization model is applied to adjust the operating status of the power system. Specifically, this involves real-time acquisition of renewable energy output, load demand, and stability parameters. Data acquisition is achieved through monitoring equipment deployed on the generation, transmission, and distribution sides, with the sampling frequency consistent with the temporal resolution of the assessment time point. Feature values ​​of real-time data at core assessment time points are extracted. For each core assessment time point, a data segment within the corresponding time window is extracted from the real-time data stream, and energy efficiency indicators, average load demand, and stability parameter statistics are calculated. These are input into the energy efficiency optimization model, which performs optimization calculations based on the input feature values ​​to find the generation-side resource allocation scheme that minimizes system operating costs while satisfying stability constraints. Based on the optimization strategy output by the energy efficiency optimization model, generation-side resource allocation and system operating parameters are adjusted. Generation-side resource allocation adjustments include hydropower unit output settings, thermal power unit start-up and shutdown plans, wind farm power control, and photovoltaic power plant generation plans. System operating parameter adjustments involve transformer tap positions, reactive power compensation device switching, and protection relay setting adjustments. The identification of core assessment time points directly impacts the quality of input data for the energy efficiency optimization model. Z-score standardization of the time point importance scores eliminates dimensional differences in the ratio values ​​of different assessment time points, ensuring comparability of the scores. The selection of thresholds is based on historical data analysis; the 75th percentile threshold ensures sufficient representativeness of the core assessment time points while avoiding excessive model complexity due to too many time points. The linear programming solution for the energy efficiency optimization model employs the simplex algorithm, which iteratively searches for the optimal solution at the vertices of the feasible region. The convergence condition is set as the rate of change of the objective function value being less than 0.02%.

[0043] The coefficient matrix of the objective function for minimizing system operating costs needs to be dynamically updated based on real-time energy prices. Fuel cost coefficients are linked to the market prices of primary energy sources such as coal and natural gas; environmental governance cost coefficients are determined based on carbon prices in the carbon emission trading market; and renewable energy curtailment penalty cost coefficients are set by government-mandated wind and solar curtailment assessment standards. Boundary values ​​for stability constraints consider system safety operation procedures and equipment technical specifications. The upper and lower limits of voltage stability constraints are determined based on the transformer turns ratio range and user tolerance. The fluctuation range of frequency stability constraints conforms to national standards for power grid frequency quality, and the angle limit of power angle stability constraints is set based on transient stability calculation and analysis results. The sensor accuracy of the real-time data acquisition system affects the accuracy of feature value extraction. The measurement error of voltage transformers should not exceed 0.2%, the measurement error of current transformers should not exceed 0.5%, and the accuracy of frequency measurement devices should reach ±0.001 Hz. Statistical calculations during feature value extraction include the arithmetic mean of energy efficiency indicators, the maximum and minimum values ​​of load demand, and the variance of stability parameters. These statistics characterize the system operating characteristics within the core assessment time period. The output optimization strategy of the energy efficiency optimization model is issued to the power generation side control system in the form of control commands. Thermal power generating units receive power setpoint commands, hydropower generating units receive opening degree adjustment commands, wind farms and photovoltaic power stations receive power control commands, transformer taps receive voltage adjustment commands, and reactive power compensation devices receive switching control commands.

[0044] Adjustments to power generation resource allocation must consider equipment operational limitations. The output adjustment rate of thermal power units is limited by boiler thermal inertia; the output adjustment of hydropower units is affected by turbine regulation characteristics; wind farm power control is constrained by the accuracy of wind energy forecasts; and photovoltaic power plant power generation plan adjustments are based on solar irradiance forecasts. The priority of system operating parameter adjustments is ranked according to their impact on system stability. Transformer tap changer adjustments take precedence over reactive power compensation device switching, with protection relay setting adjustments serving as a last resort. The application effect of the energy efficiency optimization model is evaluated using two indicators: system operating cost saving rate and energy efficiency improvement rate. The system operating cost saving rate is calculated as the ratio of the cost difference before and after optimization to the cost before optimization; the energy efficiency improvement rate is calculated as the ratio of the difference in energy efficiency indicators before and after optimization to the energy efficiency indicator value before optimization. See Table 1, which shows the key parameters in the identification process of the core evaluation time points.

[0045] Table 1: Parameter Table for Identifying Core Assessment Time Points Assessment time point number Final energy efficiency related indicators Non-energy efficiency factor weights Original ratio value Importance score of time points Is it a core assessment time point? T001 0.85 0.72 1.18 1.25 yes T002 0.63 0.81 0.78 -0.36 no T003 0.92 0.68 1.35 1.89 yes T004 0.71 0.75 0.95 0.42 no T005 0.88 0.65 1.35 1.89 yes The evaluation time points in the table correspond to specific moments on the system's operating timeline. A higher final energy efficiency correlation index value indicates a stronger correlation between energy efficiency and load. A lower weight value for non-energy efficiency factors indicates a greater impact from these factors. The original ratio values ​​reflect the dominance of energy efficiency factors relative to non-energy efficiency factors. The positive or negative sign of the time point importance score indicates the direction of deviation of that time point's importance from the average level. The determination of whether a time point is a core evaluation time point is based on a comparison of the time point importance score with a set threshold. Identifying core evaluation time points provides a precise time focus for the energy efficiency optimization model. The application of the optimization model enables the coordinated optimization of generation-side resource allocation and system operating parameters, improving the overall energy efficiency level of the renewable energy generation-side power system.

[0046] Example 5: For each operating scenario, a group of associated reference operating scenarios is determined based on the load demand data of the operating scenario. Specifically, this involves setting a load similarity interval centered on the load demand value of the operating scenario. The upper and lower boundaries of the load similarity interval are calculated as a percentage offset of the load demand value. For example, for an operating scenario with a load demand value of 150 MW, a 10% offset is set, resulting in a lower boundary of 135 MW and an upper boundary of 165 MW for the load similarity interval. Other operating scenarios whose load demand values ​​fall within the load similarity interval, excluding the operating scenario itself, are then selected to form the associated reference operating scenario group. The selection process is implemented based on database queries. The query conditions are set as a load demand value greater than or equal to 135 MW and less than or equal to 165 MW, while excluding the operating scenario itself with a load demand value of 150 MW. The query results return the record number, load demand value, and timestamp information of all operating scenarios that meet the conditions. The constituent elements of the associated reference operating scenario group include the operating scenario number, load demand value, renewable energy generation output data, and system stability parameters. The number of operating scenarios within the group depends on the width of the load similarity interval and the distribution density of the load demand in the original dataset.

[0047] A validation scenario set is generated based on historical operational data. This set selects representative operational scenarios from the historical operational database, with selection criteria covering various operating conditions such as typical days in different seasons, different weather types, and different load levels. Examples include peak load scenarios on summer weekdays, low load scenarios on winter weekends, photovoltaic power output limitations during cloudy / rainy weather, and wind power output fluctuations during windy weather. The data structure of the validation scenario set contains complete parameter records for each operational scenario, including time window identifiers, evaluation time point sequences, renewable energy generation output time series, load demand time series, and system stability parameter time series. The time series data is stored in a standard timestamp format with a time resolution of 15 minutes. The validation scenario set is processed by an energy efficiency optimization model. The model loads the input data from the validation scenario set and executes its internal optimization calculation process. The optimization calculation uses a linear programming algorithm to solve the system operating cost minimization problem, satisfying all stability constraints. The optimized energy efficiency index is output. The energy efficiency index is calculated using optimized generation-side configuration parameters and system operating parameters. The index value reflects the system energy efficiency level of the validation scenario after processing by the energy efficiency optimization model. The energy efficiency index is defined as the ratio of effective renewable energy generation output to total system energy consumption. The changes in energy efficiency indicators before and after optimization are compared. Change analysis calculates the optimization difference in energy efficiency indicators for each verification scenario. The difference is the energy efficiency indicator value after optimization minus the energy efficiency indicator value before optimization; a positive difference indicates improved energy efficiency, while a negative difference indicates decreased energy efficiency. System operating status data is recorded, including generator output curves, node voltage distribution, system frequency trajectory, and tie-line power fluctuations before and after optimization. Data records are in the form of structured database tables, with each record corresponding to an evaluation time point for a verification scenario.

[0048] The selection of the percentage offset for load similarity intervals affects the scale and quality of the associated reference operating scenario group. Too small an offset leads to insufficient number of operating scenarios within the group, lacking statistical representativeness; too large an offset introduces operating scenarios with significantly different load characteristics, reducing reference value. The representativeness of the verification scenario set is verified through cluster analysis. Historical operating data is subjected to K-means clustering, and operating scenarios close to the cluster center are selected from each cluster as verification scenarios, ensuring that the verification scenarios cover different types of operating states. The computational resource requirements for the energy efficiency optimization model to process the verification scenario set are proportional to the number of verification scenarios. The model processes the optimization calculations of multiple verification scenarios in parallel, utilizing high-performance computing clusters to shorten computation time. The analysis of optimized energy efficiency indicators not only focuses on numerical changes but also examines the spatial distribution characteristics of the indicators, comparing the energy efficiency improvement of different nodes and lines, and identifying weak links and advantageous areas in energy efficiency optimization. The system operating status data recording format includes four fields: timestamp, device identifier, parameter type, and parameter value. The timestamp is accurate to milliseconds, the device identifier adopts the power grid standard naming convention, the parameter type is classified into three categories: electrical quantity, status quantity, and control quantity, and the parameter value is stored with units. The boundary adjustment of the load similarity interval is adaptive. When the number of running scenarios in the associated reference running scenario group is less than the set threshold, the percentage offset is automatically increased until the number of running scenarios in the group reaches the minimum sample size requirement.

[0049] The generation frequency of the verification scenario set is synchronized with the cycle of system operation changes. The verification scenario set is updated at key time points such as seasonal transitions, equipment commissioning / retirement, and network structure changes to maintain consistency between the verification scenarios and actual operating conditions. The executability verification of the optimization strategy output by the energy efficiency optimization model is achieved through power flow calculation. Power flow simulation is performed on the optimized system state to verify whether node voltage and line power exceed limits and whether stability constraints are met. Long-term storage of system operating status data adopts a time-series database architecture. Data compression algorithms reduce storage space usage, and fast retrieval indexes support multi-dimensional query analysis. Example of constructing a related reference operating scenario group: Assume the current operating scenario number is S078, the load demand is 285 MW, the load similarity interval offset is set at 8%, the lower boundary of the load similarity interval is 262.2 MW, and the upper boundary is 307.8 MW. By querying the historical database, operating scenarios with load demands ranging from 262.2 MW to 307.8 MW were selected. Excluding operating scenario S078 itself, five operating scenarios were obtained: S045, S112, S189, S203, and S267. These five scenarios constitute the associated reference operating scenario group for operating scenario S078. The load demands of the operating scenarios within the associated reference operating scenario group are 265 MW, 290 MW, 295 MW, 305 MW, and 300 MW, respectively, with timestamps distributed across the same time period on different dates.

[0050] Example of processing the verification scenario set: Twelve verification scenarios, numbered V001 to V012, were selected from the historical operation database, covering typical daily data for spring, summer, autumn, and winter. The energy efficiency optimization model processed the twelve verification scenarios sequentially, outputting the optimized energy efficiency index for each scenario and comparing the changes in energy efficiency index before and after optimization. For example, the energy efficiency index of verification scenario V003 before optimization was 0.72, and after optimization it was 0.79, a difference of 0.07. For verification scenario V007, the energy efficiency index before optimization was 0.68, and after optimization it was 0.71, a difference of 0.03. System operating status data was recorded, including detailed values ​​of generator output, node voltage, and system frequency before and after optimization, forming a complete verification process record. The construction of the associated reference operating scenario group provided reference samples with similar load characteristics for each operating scenario. The setting of load similarity intervals ensured the comparability of the reference operating scenarios. The generation of the verification scenario set established the data foundation for the performance evaluation of the energy efficiency optimization model, and the representative selection ensured the reliability of the verification results. The process of processing the validation scenario set for the energy efficiency optimization model verifies the model's adaptability under different operating conditions, and the changes in optimized energy efficiency indicators reflect the model's optimization effect. Recording system operating status data provides data support for model parameter adjustment and algorithm improvement, and long-term accumulation forms a valuable operational experience library. The adaptive adjustment mechanism for load-similar intervals enhances the robustness of the associated reference operating scenario group construction; regular updates to the validation scenario set maintain the timeliness of model validation; the verification process of the energy efficiency optimization model ensures the safety and effectiveness of the optimization strategy; and the standardized management of system operating status data promotes the accumulation and sharing of operational experience.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing the energy efficiency of a renewable energy generation-side power system, characterized in that, The method includes the following steps: Data on renewable energy power generation output, load demand, and system stability parameters are collected under different operating scenarios of the power system. Each operating scenario corresponds to a specific time window, and multiple evaluation time points are evenly set within the time window. For each operating scenario, its associated reference operating scenario group is determined based on the load demand data of that operating scenario. For each assessment time point, the energy efficiency impact of that operating scenario at that assessment time point is calculated based on the energy efficiency index distribution and load demand distribution of the associated reference operating scenario group. Integrate the energy efficiency index distribution and load demand distribution of all operating scenarios, and calculate the preliminary energy efficiency correlation index for each assessment time point; By combining the energy efficiency impact and system stability parameters of each operating scenario, the preliminary energy efficiency correlation indicators are revised to obtain the final energy efficiency correlation indicators at each evaluation time point. Based on the load demand distribution and stability parameter distribution of all operating scenarios, the weights of non-energy efficiency factors at each evaluation time point are derived. Based on the final energy efficiency related indicators and the weights of non-energy efficiency factors, the core assessment time points are identified from all assessment time points; By utilizing energy efficiency index data and system stability parameters at key assessment time points, an energy efficiency optimization model is established, and the energy efficiency optimization model is applied to adjust the operating status of the power system.

2. The method for optimizing the energy efficiency of a renewable energy generation-side power system according to claim 1, characterized in that, The calculation of the energy efficiency impact of the operating scenario at the evaluation time point includes: The variation of the energy efficiency index of the associated reference operating scenario group for this operating scenario at the evaluation time point is normalized to obtain the energy efficiency fluctuation coefficient. Extract the load demand variation of the associated reference operating scenario group for this operating scenario as the load fluctuation coefficient; The energy efficiency fluctuation coefficients of all operating scenarios are aggregated to form an energy efficiency fluctuation sequence, and the load fluctuation coefficients of all operating scenarios are aggregated to form a load fluctuation sequence. Elements with the same index in the two sequences correspond to the same operating scenario. Analyze the correlation strength between energy efficiency fluctuation series and load fluctuation series, and calculate the weight of energy efficiency impact; The energy efficiency impact weight is obtained by multiplying the energy efficiency fluctuation coefficient by the energy efficiency impact weight.

3. The method for optimizing the energy efficiency of a renewable energy generation-side power system according to claim 1, characterized in that, The calculation of the preliminary energy efficiency correlation indicators at each assessment time point includes: Obtain the energy efficiency index values ​​of all operating scenarios at the target evaluation time point, and construct a set of energy efficiency index values; The information entropy index of the set of energy efficiency index values ​​is used as the first uncertainty measure. Obtain the load demand values ​​for all operating scenarios and construct a load demand value set; The information entropy index of the set of load demand values ​​is used as a second uncertainty measure. Establish a joint probability distribution of the set of energy efficiency index values ​​and the set of load demand values; Extract marginal distribution features and covariance features from the joint probability distribution; Input the first uncertainty measure and the second uncertainty measure into the feature fusion network; The features of the joint probability distribution are input into the same feature fusion network, and the preliminary energy efficiency correlation index is output through the feature fusion network.

4. The method for optimizing the energy efficiency of a renewable energy generation-side power system according to claim 1, characterized in that, The revision of the preliminary energy efficiency correlation indicators includes: All operating scenarios are subjected to K-means clustering analysis based on load demand values ​​to generate load cluster groups; The operation scenario numbers within each load cluster group are sorted to form a group number sequence; All operating scenarios are clustered according to their energy efficiency index values ​​to generate energy efficiency cluster groups; The operating scenario numbers within each energy efficiency cluster are sorted to form an energy efficiency number sequence; Calculate the overlap index between the group number sequence and the energy efficiency number sequence, and extract the energy efficiency impact data of the operation scenario corresponding to the non-overlapping numbers; Construct an energy efficiency impact matrix and perform singular value decomposition. Take the largest singular value as the energy efficiency impact weight coefficient. Multiply the overlap index with the energy efficiency impact weight coefficient to obtain the correction factor. Use the correction factor to adjust the weight of the preliminary energy efficiency correlation index to obtain the weight of non-energy efficiency factors.

5. The energy efficiency optimization method for a renewable energy generation-side power system according to claim 3, characterized in that, The derived weights of non-energy efficiency factors at each evaluation time point include: Kernel density estimation is performed on the load demand values ​​of the associated reference operating scenario group for the target operating scenario; Obtain the probability density function curve of load demand; The same kernel density estimation is performed on the load demand values ​​for all operating scenarios; Obtain the global load demand probability density function curve; Calculate the Fréchet distance between two probability density function curves; The Frescher distance is input into the sigmoid activation function for transformation; Calculate the arithmetic mean of the transformation results for all running scenarios, and then perform linear normalization on the mean. The complement of the normalized result is used as the weight of non-energy efficiency factors.

6. The method for optimizing the energy efficiency of a renewable energy generation-side power system according to claim 1, characterized in that, The identified core evaluation time points include: Calculate the ratio of the final energy efficiency-related indicators to the weights of non-energy efficiency factors at each assessment time point, and standardize the ratio to obtain the time point importance score. Evaluation time points whose importance scores exceed a set threshold are identified as core evaluation time points.

7. The method for optimizing the energy efficiency of a renewable energy generation-side power system according to claim 1, characterized in that, The establishment of the energy efficiency optimization model includes: The energy efficiency index data, load demand data, and system stability parameters at the core assessment time points are input into the multivariate optimization algorithm to train the energy efficiency optimization model, which includes the minimization of system operating costs and stability constraints.

8. The method for optimizing the energy efficiency of a renewable energy generation-side power system according to claim 1, characterized in that, The application of the energy efficiency optimization model to adjust the operating state of the power system includes: Real-time acquisition of renewable energy output, load demand, and stability parameters of the power system; Extract the feature values ​​of real-time data at the core assessment time points and input them into the energy efficiency optimization model; Based on the optimization strategy output by the energy efficiency optimization model, adjust the resource allocation on the power generation side and the system operating parameters.

9. The method for optimizing the energy efficiency of a renewable energy generation-side power system according to claim 1, characterized in that, The process of determining the associated reference operating scenario group for each operating scenario based on its load demand data includes: A load similarity interval is defined centered on the load demand value of this operating scenario; Other operating scenarios whose load demand values ​​fall within this range, excluding this operating scenario, are selected to form a group of associated reference operating scenarios.

10. The method for optimizing the energy efficiency of a renewable energy generation-side power system according to claim 1, characterized in that, The method further includes: Generate a set of verification scenarios based on historical operational data; The system processes the verification scenario set through an energy efficiency optimization model, outputs the optimized energy efficiency index, compares the changes in energy efficiency index before and after optimization, and records the system operation status data.

Images