Source network load storage cooperative scheduling method and system based on multi-time scale cost optimization

Abstract

The invention discloses a source-grid-load-storage cooperative scheduling method and system based on multi-time scale cost optimization, and the method comprises the steps: outputting a dynamic coupling map containing short-term and long-term cost evolution paths according to the source-side power generation cost, the grid-side transmission loss, the load-side load demand and the storage-side life attenuation data of an energy system; based on the dynamic coupling map, outputting an uncertainty quantization parameter containing probability distribution; according to the uncertainty quantization parameters, outputting a collaborative scheduling framework containing space-time correlation constraints; and based on the collaborative scheduling framework, performing multi-objective optimization solution by adopting a hybrid algorithm, and outputting a global optimal scheduling strategy. By utilizing the embodiment of the invention, the global optimal decision can be realized to balance the economy and safety of the system and the service life of the equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of source-grid-load-storage scheduling, and in particular to a source-grid-load-storage coordinated scheduling method and system based on multi-time-scale cost optimization. Background Art

[0002] Coordinated scheduling of sources, grids, loads, and storage faces challenges such as multi-time-scale coupling and increased uncertainty. Traditional scheduling methods typically use single-time-scale optimization, making it difficult to reconcile the contradiction between short-term operating costs and long-term equipment losses, and lack the ability to quantitatively analyze uncertainties such as meteorological changes and load fluctuations. In existing technologies, short-term scheduling focuses on day-ahead economic scheduling, while long-term planning focuses on annual investment decisions. The separation of the two leads to accelerated degradation of energy storage life or imbalance in source-load matching. In addition, existing uncertainty modeling methods are mostly based on static probability distributions and cannot dynamically respond to the spatiotemporal correlation characteristics of extreme weather events and market electricity price fluctuations. Summary of the Invention

[0003] The purpose of the present invention is to provide a source-grid-load-storage coordinated scheduling method and system based on multi-time-scale cost optimization to address the deficiencies in the existing technology and achieve global optimal decision-making to balance system economy, safety and equipment life.

[0004] An embodiment of the present application provides a source-grid-load-storage coordinated scheduling method based on multi-time-scale cost optimization, the method comprising: Based on the energy system's source-side power generation costs, grid-side transmission losses, load-side load demand, and storage-side life attenuation data, a time series prediction algorithm is used to conduct a spatiotemporal correlation analysis between electricity price fluctuations and energy storage cycle life, outputting a dynamic coupling map containing short-term and long-term cost evolution paths. Based on the dynamic coupling map, combined with meteorological forecast data and historical load fluctuation characteristics, the influencing factors of weather changes and load fluctuations are quantified through dynamic uncertainty map generation technology, and uncertainty quantification parameters containing probability distribution are output; Based on the uncertainty quantification parameters, a multi-time-scale coupled scheduling framework is designed. A multi-scale coupled factor decomposition model is used to decompose the scheduling task into hourly short-term optimization and monthly long-term optimization sub-problems, and a collaborative scheduling framework with spatiotemporal correlation constraints is output. Based on the collaborative scheduling framework, a hybrid algorithm is adopted to solve multi-objective optimization problems. The short-term optimization problem is handled by the co-evolutionary mechanism of genetic algorithm and reinforcement learning, and the fuzzy logic decision tree is introduced to optimize the long-term strategy and output the global optimal scheduling strategy.

[0005] Optionally, based on the source-side power generation cost, grid-side transmission loss, load-side load demand, and storage-side life attenuation data of the energy system, a time series prediction algorithm is used to perform spatiotemporal correlation analysis on electricity price fluctuations and energy storage cycle life, and output a dynamic coupling map containing short-term and long-term cost evolution paths, including: Based on the source-side power generation cost time series data and the storage-side life decay curve, multi-scale wavelet transform is used to extract the power generation cost fluctuation characteristics and generate spatiotemporal correlation feature vectors. The spatiotemporal correlation feature vector is integrated with the topological distribution data of the network-side transmission loss, and the source-grid-load-storage coupling relationship is modeled through the spatiotemporal attention mechanism to output a four-dimensional dynamic correlation tensor. Based on the four-dimensional dynamic correlation tensor, a tensor decomposition algorithm is used to separate short-term electricity price sensitivity factors and long-term energy storage life influencing factors, generating a dual-time scale prediction feature set consisting of a short-term prediction feature set and a long-term prediction feature set. The short-term prediction feature set is input into the long short-term memory network to predict the electricity price fluctuation curve in the next 72 hours. At the same time, the long-term prediction feature set is input into the Prophet model to predict the 30-day energy storage cycle life attenuation trajectory, and the two are fused to generate a dynamic coupling map.

[0006] Optionally, based on the dynamic coupling map, combined with meteorological forecast data and historical load fluctuation characteristics, the influencing factors of weather changes and load fluctuations are quantified through dynamic uncertainty map generation technology, and uncertainty quantification parameters containing probability distribution are output, including: According to the long-term energy storage life trajectory in the dynamic coupling map, meteorological sensitive features are extracted to generate meteorological impact coding vectors; The meteorological impact encoding vector is input into the dynamic Bayesian network, combined with the historical load fluctuation characteristics, to model the weather-load joint probability distribution and output the uncertainty propagation map; Based on the uncertainty propagation graph, Monte Carlo sampling of extreme weather scenarios is embedded in the dynamic coupling map to generate a risk scenario set; Based on the risk scenario set, the kernel density estimation algorithm is used to quantify the joint impact of weather and load fluctuations on system costs and output a multidimensional probability density function. The multidimensional probability density function is aligned with the spatiotemporal features of the dynamic coupling map to generate an uncertainty quantification parameter matrix with confidence intervals.

[0007] Optionally, based on the uncertainty quantification parameters, a multi-time-scale coupled scheduling framework is designed, and a multi-scale coupled factor decomposition model is used to decompose the scheduling task into hourly short-term optimization and monthly long-term optimization sub-problems, and output a collaborative scheduling framework containing spatiotemporal correlation constraints, including: Based on the uncertainty quantification parameter matrix, an hourly scheduling objective function is constructed to define dynamic constraints on power generation cost, transmission loss, and load deviation. Based on the long-term energy storage attenuation trajectory in the dynamic coupling map, a monthly optimization objective function is established to define the energy storage life balance and investment return constraints; A two-layer decomposition algorithm is used to decompose hourly and monthly targets into independent sub-problems, and cross-timescale parameter interaction is achieved through the coupling factor transfer matrix. According to the coupling factor transmission results, a spatiotemporal correlation constraint propagation mechanism is introduced to ensure the consistency of short-term scheduling and long-term strategies in key parameters such as energy storage charging and discharging depth, and output a coordinated scheduling framework.

[0008] Optionally, based on the collaborative scheduling framework, a hybrid algorithm is used to perform multi-objective optimization solutions, wherein short-term optimization problems are handled by a co-evolutionary mechanism of genetic algorithm and reinforcement learning, and a fuzzy logic decision tree is introduced to optimize long-term strategies, thereby outputting a global optimal scheduling strategy, including: According to the hourly objective function in the coordinated scheduling framework, a genetic algorithm is used to initialize the population, encode the real-time electricity price, load demand and energy storage SOC into chromosome genes, and generate the initial short-term solution set; Based on the initial short-term solution set, a deep reinforcement learning agent is embedded to dynamically adjust the crossover and mutation probability, and the Q-learning algorithm is used to optimize the solution set convergence speed and output the Pareto frontier short-term candidate solutions. The short-term candidate solutions of the Pareto front are input into the fuzzy logic decision tree and combined with the uncertainty parameters in the monthly optimization objective to generate a long-term strategy fuzzy rule base; Based on the long-term strategy fuzzy rule base, the long-term strategy is iteratively optimized through the dynamic programming algorithm to output a monthly charging and discharging plan that is decoupled from the short-term candidates of the Pareto front; The short-term candidate solutions of the Pareto front and the monthly charging and discharging plan are input into the non-dominated sorting genetic algorithm, and the comprehensive cost and risk score are calculated by combining the entropy weight method to screen out the global optimal scheduling strategy.

[0009] Another embodiment of the present application provides a source-grid-load-storage coordinated scheduling system based on multi-time-scale cost optimization, the system comprising: The analysis module is used to conduct spatiotemporal correlation analysis between electricity price fluctuations and energy storage cycle life based on the energy system's source-side power generation costs, grid-side transmission losses, load-side load demand, and storage-side life decay data using a time series prediction algorithm. The module then outputs a dynamic coupling map containing short-term and long-term cost evolution paths. A generation module is used to quantify the influencing factors of weather changes and load fluctuations based on the dynamic coupling map, combined with meteorological forecast data and historical load fluctuation characteristics, through dynamic uncertainty map generation technology, and output uncertainty quantification parameters including probability distribution; A decomposition module is used to design a multi-time-scale coupled scheduling framework based on the uncertainty quantification parameters, decompose the scheduling task into hourly short-term optimization and monthly long-term optimization sub-problems using a multi-scale coupled factor decomposition model, and output a collaborative scheduling framework with spatiotemporal correlation constraints; The output module is used to perform multi-objective optimization solutions based on the collaborative scheduling framework using a hybrid algorithm, wherein short-term optimization problems are handled through the collaborative evolution mechanism of genetic algorithm and reinforcement learning, and fuzzy logic decision trees are introduced to optimize long-term strategies to output the global optimal scheduling strategy.

[0010] Yet another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, wherein the computer program is configured to execute any of the above methods when run.

[0011] Yet another embodiment of the present application provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the above methods.

[0012] Compared with the existing technology, the present invention provides a source-grid-load-storage collaborative scheduling method based on multi-time-scale cost optimization. According to the source-side power generation cost, grid-side transmission loss, load-side load demand and storage-side life attenuation data of the energy system, the method outputs a dynamic coupling map containing short-term and long-term cost evolution paths; based on the dynamic coupling map, the method outputs uncertainty quantification parameters containing probability distribution; based on the uncertainty quantification parameters, the method outputs a collaborative scheduling framework containing time-space correlation constraints; based on the collaborative scheduling framework, a hybrid algorithm is used to perform multi-objective optimization solution, and the global optimal scheduling strategy is output, so that the global optimal decision can be achieved to balance the system economy, safety and equipment life. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A hardware structure block diagram of a computer terminal for a source-grid-load-storage collaborative scheduling method based on multi-time-scale cost optimization provided by an embodiment of the present invention; Figure 2 A flow chart of a source-grid-load-storage collaborative scheduling method based on multi-time-scale cost optimization provided by an embodiment of the present invention; Figure 3 A schematic structural diagram of a source-grid-load-storage collaborative scheduling system based on multi-time-scale cost optimization provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limiting the present invention.

[0015] The embodiment of the present invention first provides a source-grid-load-storage coordinated scheduling method based on multi-time-scale cost optimization. The method can be applied to electronic devices, such as computer terminals, specifically ordinary computers.

[0016] The following describes it in detail by taking running on a computer terminal as an example. Figure 1 The hardware structure block diagram of a computer terminal for a source-grid-load-storage coordinated scheduling method based on multi-time-scale cost optimization provided by an embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0017] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, which, when executed, can cause a processor to execute any source-grid-load-storage coordinated scheduling method based on multi-time-scale cost optimization.

[0018] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0019] The internal memory provides an environment for the operation of computer programs in non-volatile storage media. When the computer program is executed by the processor, the processor can execute any source-grid-load-storage coordinated scheduling method based on multi-time-scale cost optimization.

[0020] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0021] It should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0022] See also Figure 2 The embodiment of the present invention provides a source-grid-load-storage coordinated scheduling method based on multi-time-scale cost optimization, which may include the following steps: S201: Based on the energy system's source-side power generation costs, grid-side transmission losses, load-side load demand, and storage-side lifespan decay data, a time series prediction algorithm is used to perform spatiotemporal correlation analysis between electricity price fluctuations and energy storage cycle lifespan, outputting a dynamic coupling map containing short-term and long-term cost evolution paths. Specifically, we can use multi-scale wavelet transform to extract the fluctuation characteristics of power generation cost based on the source-side power generation cost time series data and the storage-side life decay curve to generate a spatiotemporal correlation feature vector. The multi-scale wavelet transform uses the Daubechies 4 wavelet basis (DB4) to perform a five-layer decomposition, decomposing the source-side power generation cost time series data (such as the 15-minute sampling data of the wind farm) into different frequency bands: High-frequency detail layer (D1-D3): captures minute-level fluctuations (such as power fluctuations caused by sudden changes in wind speed) and eliminates noise through threshold filtering (hard threshold λ = 0.1*max|D|); Low-frequency approximation layer (A5): extracts daily cycle trends (such as load peak and valley patterns) and uses moving average smoothing (window width = 96 sampling points).

[0023] The storage side life decay curve (such as the capacity decay data of lithium-ion batteries) is processed by Morlet wavelet transform: Scale parameter: Set 10 scales (s=1~10), corresponding to different charge and discharge cycle periods (for example, s=5 maps 200 cycles); Wavelet coefficients: Calculate the energy density at each scale and quantify the decay rate (e.g., a coefficient > 0.8 indicates accelerated decay).

[0024] Feature fusion: The high-frequency detail energy of power generation cost (unit: yuan2 / MWh) and the energy storage attenuation coefficient (unit: % / cycle) are spliced ​​into a mixed vector; Add spatial location coding (such as the GeoHash value of the longitude and latitude coordinates of the wind farm) to form a 128-dimensional spatiotemporal correlation feature vector.

[0025] For example, the single-day feature vector of a wind farm is [0.32, 0.15, ..., 0.07 | 0.83, 0.21, ... |w21jr], where the first 64 dimensions are cost fluctuation energy, the middle 50 dimensions are attenuation coefficients, and the last 14 dimensions are position codes.

[0026] The spatiotemporal correlation feature vector is integrated with the topological distribution data of the network-side transmission loss, and the source-grid-load-storage coupling relationship is modeled through the spatiotemporal attention mechanism to output a four-dimensional dynamic correlation tensor. The network-side transmission loss topology data includes: Electrical distance matrix: per-unit impedance between nodes (e.g., 0.05 to 0.2 pu); Power flow distribution heat map: generated based on historical SCADA data (resolution 1km×1km).

[0027] Fusion and attention mechanism: Topological embedding: The electrical distance matrix is ​​input into a graph convolutional network (GCN, number of layers = 2), which outputs a node feature vector (dimension 32); The power flow heat map is input into CNN (convolution kernel 3×3) and outputs a spatial feature map (16×16×64).

[0028] Spatiotemporal Attention: Temporal attention head: Calculates the correlation of feature vectors along the time axis (e.g., the correlation weight between the load peak at 7:00 and the peak at 19:00 is 0.9). Spatial attention head: calculates the energy flow association between grid nodes (e.g., the weight from wind farm A to substation B is 0.75); Cross-modal attention head: Modeling source-storage dynamic interaction (e.g., the weight of wind power output and battery SOC is 0.6).

[0029] Tensor construction: Dimension 1 (time): 96 time slices (24 hours × 4 sampling points); Dimension 2 (space): 50 power grid nodes; Dimension 3 (feature): 128-dimensional mixed features; Dimension 4 (coupling relationship): attention weight matrix (50 × 50).

[0030] Example: At a certain moment, the tensor slice is [node 1 features: (0.4, 0.2,...), node 2 to node 1 coupling weight: 0.83].

[0031] Based on the four-dimensional dynamic correlation tensor, a tensor decomposition algorithm is used to separate short-term electricity price sensitivity factors and long-term energy storage life influencing factors, generating a dual-time scale prediction feature set consisting of a short-term prediction feature set and a long-term prediction feature set. Tensor decomposition uses the Tucker decomposition model: Core Tensor Extraction: The original tensor X∈R^{96×50×128×50} is decomposed into: Core tensor G∈R^{24×10×32×10} (compression rate 85%); Factor matrix: time matrix A∈R^{96×24}, space matrix B∈R^{50×10}, feature matrix C∈R^{128×32}, coupling matrix D∈R^{50×10}.

[0032] Factor separation: Short-term electricity price sensitivity factor: Take the first 12 time patterns of the core tensor G (corresponding to 0-48 hours) and the first 16 dimensions of the feature matrix C (such as wind power fluctuation rate and load change rate); Long-term energy storage life factor: take the last 12 time modes of G (corresponding to 7 to 30 days) and the last 16 dimensions of C (such as cycle depth DoD and average charge and discharge rate).

[0033] Feature set construction: Short-term feature set: 24-dimensional vector (12 time patterns × 2 key features). For example, [0.35, -0.12, ...] represents the sensitivity to electricity prices in the next 6 hours. Long-term feature set: 24-dimensional vector (12 time patterns × 2 key features). For example, [0.08, 0.21, ...] represents the lifespan decay rate in the next 20 days.

[0034] The short-term prediction feature set is input into the long short-term memory network to predict the electricity price fluctuation curve in the next 72 hours. At the same time, the long-term prediction feature set is input into the Prophet model to predict the 30-day energy storage cycle life attenuation trajectory, and the two are fused to generate a dynamic coupling map.

[0035] Short-term electricity price prediction (LSTM): Network structure: three-layer LSTM (128 hidden units) + fully connected layer (output dimension 24); Input data: short-term feature set (24 dimensions) + historical electricity prices (lagged 24 time points); Training mechanism: Using Seq2Seq architecture, the encoder inputs 72 hours of data and the decoder outputs predictions for the next 72 hours; Output: Electricity price curve at 24 time points (e.g., [0.45, 0.52, ..., 0.38] yuan / kWh).

[0036] Long-term lifespan prediction (Prophet): Model configuration: Growth trend: piecewise linear (automatic detection of turning points); Seasonal term: Fourier series (order = 5, period = 30 days); Input data: long-term feature set (24 dimensions) + historical capacity decay rate (e.g., 0.5% per month); Output: 30-day decay trajectory (e.g. [100%, 99.7%, ..., 98.2%]).

[0037] Dynamic coupling map generation: Time axis alignment: Map the 72-hour electricity price curve and the 30-day life trajectory to a unified time axis (1-hour resolution); Coupling relationship visualization: X-axis: time (0-720 hours); left Y-axis: electricity price (yuan / kWh); right Y-axis: energy storage capacity (%); correlation line: marks the causal relationship between electricity price peak and capacity drop (for example, when the electricity price is greater than 0.5 yuan, the battery DoD deepens by 5%).

[0038] Example graph: shows that the peak electricity price in the 120th hour (5th day) is 0.58 yuan, corresponding to the battery capacity dropping from 99.1% to 98.9%.

[0039] By integrating key parameters of various links in the energy system, including the operating costs of the power generation side, the transmission efficiency of the power grid, the user load demand, and the life decay characteristics of the energy storage equipment, a spatiotemporal correlation model between electricity price fluctuations and energy storage life is established through time series analysis technology. This can simultaneously capture the dynamic laws of short-term electricity price response and long-term energy storage performance degradation, forming a coupling map that reflects the cost changes of the system throughout its life cycle. It constructs a spatiotemporal framework for system-level cost analysis and organically combines traditional isolated short-term scheduling with long-term planning. The generation of dynamic coupling maps not only reveals the interactive influence mechanism between electricity prices and energy storage life, but also provides a quantitative basis for subsequent multi-time scale optimization, avoiding decision-making bias caused by the fragmentation of the time dimension.

[0040] S202, based on the dynamic coupling map, combined with meteorological forecast data and historical load fluctuation characteristics, quantify the influencing factors of weather changes and load fluctuations through dynamic uncertainty map generation technology, and output uncertainty quantification parameters including probability distribution; Specifically, the meteorological sensitive features can be extracted based on the long-term energy storage life trajectory in the dynamic coupling map to generate the meteorological impact coding vector; Meteorological sensitive feature extraction focuses on the correlation analysis between energy storage life and meteorological parameters. First, a multi-factor response model for energy storage battery life degradation is established, and three core meteorological indicators, temperature, humidity, and irradiance, are selected as input dimensions. For example: Temperature sensitivity: Using a piecewise linearization model, when the ambient temperature exceeds 25°C, the capacity decay rate of lithium batteries increases exponentially with increasing temperature. A temperature-attenuation coefficient mapping table is determined based on historical data statistics (e.g., the attenuation coefficient is 1.15 at 30°C and 1.32 at 35°C).

[0041] Humidity sensitivity: Based on the electrochemical corrosion model, when the humidity is >70%, the battery internal resistance growth rate increases to 1.8 times that in a dry environment.

[0042] Irradiance sensitive features: for photovoltaic supporting energy storage, high irradiance (>800W / m 2 ) leads to an increase in the frequency of charge and discharge, and defines the charge and discharge cycle acceleration factor (such as the irradiance increases by 100W / m 2 , daily cycle number + 0.2).

[0043] The feature extraction process is as follows: Data alignment: The weather station data (sampling frequency 5 minutes) and the energy storage life monitoring data (daily capacity detection) are aligned on the time axis through cubic spline interpolation.

[0044] Feature calculation: Calculate the daily temperature fluctuation range (DTTR), which is formulated as the difference between the highest and lowest temperatures of the day; The high humidity duration (HHD) was calculated, defined as the sum of consecutive periods with humidity >70%; The equivalent irradiation dose (EID) was calculated by integration, and the irradiance curve was numerically integrated using a trapezoidal method.

[0045] Feature encoding: The temperature feature is encoded as a 3D vector: [daily average temperature, DTTR, proportion of time > 30°C]; Humidity features are encoded as a 2-dimensional vector: [average daily humidity, HHD]; The irradiance feature is encoded as a 2D vector: [peak irradiance, EID].

[0046] Finally, a 7-dimensional meteorological impact encoding vector is generated, for example: [28.5℃, 12.3℃, 15%, 65%, 4.2h, 850W / m 2 , 5.6kWh / m 2 ].

[0047] The meteorological impact encoding vector is input into the dynamic Bayesian network, combined with the historical load fluctuation characteristics, to model the weather-load joint probability distribution and output the uncertainty propagation map; Dynamic Bayesian Network (DBN) adopts a three-layer spatiotemporal topology: Input layer: weather impact encoding vector (7 nodes); Hidden layer: load fluctuation characteristics (5 nodes), including: industrial load fluctuation rate (standard deviation σ_industry); commercial load peak-to-valley ratio (P / C_ratio); residential load seasonal sensitivity (S_season); base load stability index (BSI); load surge probability (P_surge); Output layer: system cost fluctuation (3 nodes), including: power generation cost variation coefficient; network loss fluctuation amplitude; energy storage life attenuation acceleration factor.

[0048] Network construction process: Conditional Probability Table (CPT) training: Parameter learning is performed using five years of historical data (a total of 43,800 hours of recording). The expectation-maximization (EM) algorithm is used to iteratively optimize CPT parameters, with a convergence threshold Δ<0.001. For example, when the temperature is greater than 30°C and the humidity is greater than 75%, the probability of a load mutation increases to 1.5 times the baseline value.

[0049] Spatiotemporal correlation modeling: Industrial load is negatively correlated with temperature (the correlation coefficient is -0.7 when the proportion of air-conditioning load increases); commercial load is positively correlated with irradiance (shopping mall traffic increases on sunny days, with a correlation coefficient of +0.6).

[0050] Joint probability distribution generation: Gibbs sampling is used to simulate 100,000 state transitions. The output is a three-dimensional joint probability distribution: [P(cost fluctuation|weather, load), P(grid loss|load characteristics), P(life decay|weather)].

[0051] The final generated uncertainty propagation graph contains: node topology structure (15-node directed graph); conditional probability matrix (15×15 sparse matrix); joint distribution heat map (resolution 0.1 probability unit).

[0052] Based on the uncertainty propagation graph, Monte Carlo sampling of extreme weather scenarios is embedded in the dynamic coupling map to generate a risk scenario set; Monte Carlo Sampling (MCS) is used to generate low-probability, high-risk extreme weather scenarios to assess their impact on the power system. The specific process is as follows: ‌Defining extreme weather event criteria‌: High temperature event: The maximum daily temperature exceeds the historical 95th percentile (e.g. 40°C) for three consecutive days.

[0053] Low temperature event: The minimum daily temperature is below the historical 5% quantile (e.g. -10°C) for two consecutive days.

[0054] High wind speed event: The instantaneous wind speed exceeds 20m / s and lasts for more than 6 hours.

[0055] Latin Hypercube Sampling (LHS) Generate 1000 sets of meteorological parameter samples (SAMPLES=1000) within the extreme event window (e.g., days 15-17). Each set of samples includes the temperature perturbation (TEMPERATURE_SHIFT), humidity perturbation (HUMIDITY_SHIFT), and wind speed perturbation (WIND_SHIFT). For example, the high temperature perturbation for sample 50 is +2°C (TEMPERATURE_SAMPLE_50 = TEMPERATURE_PREDICTED + 2°C).

[0056] The disturbance range is set based on historical extreme values. For example, the temperature disturbance range is [-3°C, +5°C], and the wind speed disturbance range is [-5m / s, +10m / s].

[0057] ‌Scene embedding and correction‌: Based on the perturbed meteorological parameters, the energy storage life trajectory and load forecast in the dynamic coupling map are revised: in high-temperature scenarios, the energy storage capacity decay rate is increased to 1.5 times the baseline value (for example, from 0.8% per month to 1.2%). Load demand is adjusted based on the DBN inference results. For example, for every 1°C increase in temperature, the air conditioning load increases by 5%.

[0058] Generate a risk scenario set (RISK_SCENARIOS), each scenario contains: Scenario ID (such as SCENARIO_100), meteorological parameters, energy storage attenuation correction factor, and load correction factor.

[0059] Based on the risk scenario set, the kernel density estimation algorithm is used to quantify the joint impact of weather and load fluctuations on system costs and output a multidimensional probability density function. Kernel Density Estimation (KDE) is used to fit the joint probability distribution of weather, load, and cost. The specific steps are as follows: Cost simulation and sample collection: Cost calculation is performed on 1000 samples in the risk scenario set. The total cost (TOTAL_COST) includes: Source-side power generation cost (GEN_COST): fuel cost of coal-fired units + start-up and shutdown costs of gas-fired units.

[0060] Grid-side transmission loss (LOSS_COST): The economic cost corresponding to line resistance loss.

[0061] Storage side life depreciation (STORAGE_COST): Equipment replacement cost caused by capacity decay.

[0062] Load side deviation penalty (LOAD_PENALTY): Penalty for the deviation between the actual load and the predicted value.

[0063] For example, in a certain scenario, GEN_COST = 500,000 yuan, LOSS_COST = 50,000 yuan, STORAGE_COST = 100,000 yuan, LOAD_PENALTY = 20,000 yuan, and TOTAL_COST = 670,000 yuan.

[0064] Multidimensional kernel density estimation: Use a Gaussian kernel function to calculate the probability density of the four-dimensional cost space (GEN_COST, LOSS_COST, STORAGE_COST, LOAD_PENALTY). The bandwidth (BANDWIDTH) is selected using the Silverman criterion. For example, the bandwidth of the generation cost dimension is h = 1.06 × σ × n^(-1 / 5), where σ is the sample standard deviation and n = 1000.

[0065] To reduce the computational complexity, the Hoeffding Projection Pursuit is used to decompose the four-dimensional space into two sets of two-dimensional subspaces (such as GEN_COST-LOSS_COST and STORAGE_COST-LOAD_PENALTY), and the joint distribution is estimated separately and then synthesized.

[0066] Generate a multidimensional probability density function: Outputs a four-dimensional joint probability density function (JPDF) (gen_cost, loss_cost, storage_cost, load_penalty), allowing you to query the probability of any cost combination. For example, the probability of total costs exceeding 800,000 yuan is 5%, primarily due to high power generation costs and high energy storage depreciation in high-temperature scenarios.

[0067] The multidimensional probability density function is aligned with the spatiotemporal features of the dynamic coupling map to generate an uncertainty quantification parameter matrix with confidence intervals.

[0068] Spatiotemporal alignment requires ensuring that the probability density function and the dynamic coupling map have consistent temporal resolution (hourly level) and spatial partitioning (grid topology). This is achieved as follows: ‌Time Dimension Alignment‌: The Dynamic Time Warping (DTW) algorithm is used to decompose the monthly JPDF (30 days) into hourly slices. For example, the 30 days are divided into six periods (5 days each), and the JPDF parameters (mean μ, standard deviation σ) of each period are mapped to the corresponding weekly intervals in the dynamic coupling map.

[0069] The confidence interval for each hour's cost is calculated using bootstrap resampling. For example, the 95% confidence interval for the total cost at the 500th hour (5:00 PM on a certain day) is [550,000 RMB, 650,000 RMB].

[0070] ‌Spatial Dimension Alignment‌: The power grid is divided into multiple zones (e.g., ZONE_01 to ZONE_10), each associated with data from neighboring weather stations. Regional weather influence weights are calculated using Inverse Distance Weighting (IDW). For example, the weight of zone ZONE_03 is determined by the three nearest weather stations (distances d1 = 5 km, d2 = 8 km, and d3 = 10 km), using the formula w_i = 1 / (d_i^2).

[0071] Combining regionalized meteorological impact weights with the JPDF generates a regional cost probability distribution. For example, ZONE_05, due to its proximity to an industrial area, has a wider confidence interval for its load cost ([500,000 yuan, 700,000 yuan]).

[0072] Generate uncertainty quantization parameter matrix: The matrix fields include timestamp (TIMESTAMP), zone ID (ZONE_ID), cost mean (COST_MEAN), standard deviation (COST_STD), and confidence interval (COST_CI). For example: TIMESTAMP: 2024-07-01T14:00; ZONE_ID: ZONE_07; COST_MEAN: 580,000 yuan; COST_STD: 60,000 yuan; COST_CI: [520,000 yuan, 640,000 yuan].

[0073] This matrix supports the dynamic adjustment of the dispatch system strategy. For example, when the confidence interval width in a certain area exceeds a threshold (such as 200,000 yuan), the backup power capacity needs to be increased to hedge the risk.

[0074] Based on the dynamic coupling map, weather forecasts and historical load data are introduced as external disturbance variables. Probabilistic statistical methods are used to analyze the combined impact of extreme weather events and random load fluctuations on system costs. Uncertainties are converted into quantifiable probability distribution parameters, and a mathematical representation of risk transmission is established. This enables explicit modeling of uncertainties and extends traditional deterministic optimization to probabilistic optimization. By quantifying the risk contributions of weather and load fluctuations, robustness is ensured for scheduling strategies, enhancing the system's ability to cope with extreme events.

[0075] S203: Design a multi-time-scale coupled scheduling framework based on the uncertainty quantification parameters, decompose the scheduling task into hourly short-term optimization and monthly long-term optimization sub-problems using a multi-scale coupled factor decomposition model, and output a collaborative scheduling framework that includes spatiotemporal correlation constraints. Specifically, we can construct an hourly scheduling objective function based on the uncertainty quantification parameter matrix and define the dynamic constraints of power generation cost, transmission loss and load deviation; The hourly scheduling objective function aims to minimize the total operating cost from the current moment to the next 72 hours while satisfying the grid's real-time operating constraints. The objective function consists of three components: source-side generation cost (GEN_COST), grid-side transmission loss (LOSS_COST), and load-side load deviation penalty (LOAD_PENALTY).

[0076] Objective function construction: Power Generation Cost (GEN_COST): The power generation cost of source-side equipment such as coal-fired units, gas-fired units, and photovoltaic power plants is calculated based on output (POWER_GEN). For example, the cost of a coal-fired unit is a linear function: GEN_COST = a * POWER_COAL + b, where a = 200 yuan / MWh (unit power generation cost) and b = 5,000 yuan (start-up and shutdown fixed cost).

[0077] ‌Transmission loss (LOSS_COST)‌: According to the line resistance (R_LINE) and the square of the current (I 2 ) Calculate the power loss (LOSS=I 2 R_LINE), converted to economic cost. For example, if a line R_LINE = 0.1Ω and current I = 100A, the loss cost per hour is LOSS_COST = (100 2 0.1) / 1000 * electricity price (assuming the electricity price is 0.5 yuan / kWh) = 5 yuan.

[0078] Load Deviation Penalty (LOAD_PENALTY): When the actual load (LOAD_ACTUAL) deviates from the predicted value (LOAD_PREDICTED) by more than a threshold (e.g., ±5%), a penalty is imposed proportional to the load deviation. For example, a 10% deviation results in a penalty of 50 RMB / MWh.

[0079] Dynamic constraint definition: Power Constraint: Each unit's output must be within its technical output range. For example, the output of a coal-fired unit is limited to 50MW at the lower limit and 300MW at the upper limit, i.e., 50MW ≤ POWER_COAL ≤ 300MW.

[0080] Energy storage charge and discharge constraints: The state of charge (SOC) of energy storage systems (such as lithium batteries) must be maintained within a safe range (e.g., 20% ≤ SOC ≤ 90%), and the charge and discharge power is limited by the device rating (e.g., the maximum charge and discharge power is 50MW).

[0081] Load balancing constraint: The total power generation plus the energy storage discharge must equal the load demand plus transmission losses, i.e. SUM(POWER_GEN) + POWER_STORAGE_DISCHARGE = LOAD_DEMAND + LOSS.

[0082] Parameter association and optimization weights: The confidence intervals in the uncertainty quantification parameter matrix (e.g., the cost fluctuation range [550,000 yuan, 650,000 yuan]) are converted into robustness weights for the objective function. For example, the power generation cost weight during periods of high uncertainty is increased by 20% to prioritize stable output solutions.

[0083] Based on the long-term energy storage attenuation trajectory in the dynamic coupling map, a monthly optimization objective function is established to define the energy storage life balance and investment return constraints; The monthly optimization objective function focuses on the life attenuation balance and return on investment (ROI) of the energy storage system within 30 days, and requires a coordinated charging and discharging strategy to avoid local excessive losses.

[0084] ‌Objective function construction‌: Energy Storage Life Balance (LIFE_BALANCE): This balance is achieved by minimizing the capacity decay differences among storage units. For example, the difference indicator is defined as: LIFE_BALANCE = MAX(CAPACITY_DECAY_RATE_i) - MIN(CAPACITY_DECAY_RATE_i), where i is the storage unit number.

[0085] Return on Investment (ROI): Calculates the ratio of the energy storage system's net revenue to its initial investment. For example, if the initial investment is 10 million yuan (COST_INVEST), and monthly revenue includes peak-valley arbitrage (PROFIT_ARBITRAGE = 2 million yuan) and ancillary service revenue (PROFIT_ANC = 500,000 yuan), then the ROI is (200 + 50) / 1000 = 25%.

[0086] ‌Constraint Definition‌: Cycle Constraint: The monthly charge and discharge cycle count of an energy storage unit must not exceed the rated value. For example, a lithium battery is rated for 5,000 cycles, with a monthly limit of 200 cycles.

[0087] Depth of charge and discharge (DOD) limit: The single depth of charge and discharge (e.g. DOD = 80%) must meet the manufacturer's requirements, and the monthly average DOD must not exceed 60%.

[0088] ‌Return on Investment Minimum‌: ROI ≥ 15%, otherwise it is considered an economically unfeasible solution.

[0089] Fusion of long-term energy storage decay trajectories: The energy storage capacity attenuation prediction curve (e.g., the capacity attenuates by 0.8% per month) is extracted from the dynamic coupling map and discretized into a 30-day daily attenuation (e.g., DAY_DECAY=0.0267%) as a reference benchmark for life balance optimization.

[0090] The decay rate is associated with the charge and discharge strategy. For example, if the number of charge and discharge times on a certain day increases by 10%, the decay rate for that day is corrected to 0.0293%.

[0091] A two-layer decomposition algorithm is used to decompose hourly and monthly targets into independent sub-problems, and cross-timescale parameter interaction is achieved through the coupling factor transfer matrix. The two-level decomposition algorithm splits the complex optimization problem into hourly (lower level) and monthly (upper level) sub-problems, and realizes two-way interaction through the coupling factor transfer matrix (COUPLING_MATRIX).

[0092] ‌Sub-problem breakdown‌: Hourly subproblem: Using the next 72 hours as the optimization window, solve the hourly power generation plan, energy storage charging and discharging power, and load distribution, with the goal of minimizing total operating costs.

[0093] Monthly sub-problem: Optimize the distribution of energy storage charge and discharge times and return on investment over a 30-day period, with the goal of minimizing lifetime imbalance and maximizing ROI.

[0094] ‌Coupling factor transfer matrix design‌: The rows of the matrix represent the time scale (1 to 72 hours at the hour level and 1 to 30 days at the month level), and the columns represent the coupling parameter type (such as energy storage SOC and number of charge and discharge times).

[0095] ‌Example matrix structure‌: COUPLING_MATRIX = [ [DAY1_HOUR1_SOC, DAY1_HOUR1_DISCHARGE_COUNT, ...], [DAY1_HOUR2_SOC, DAY1_HOUR2_DISCHARGE_COUNT, ...], ... [DAY30_HOUR24_SOC, DAY30_HOUR24_DISCHARGE_COUNT, ...] ]. Parameter passing rules: The monthly level transmits the monthly balance target of the energy storage SOC to the hourly level (e.g., daily SOC fluctuation does not exceed ±10%).

[0096] The hourly layer feeds back the actual number of charges and discharges to the monthly layer to update the life attenuation model.

[0097] ‌Iterative Optimization Process‌: Initial Iteration: The monthly layer sets the initial charge and discharge plan, and the hourly layer executes the plan and returns the actual SOC and cost.

[0098] Parameter Correction: If the actual SOC at the hourly level deviates from the monthly target (e.g., deviation > 5%), the monthly level will adjust the charge and discharge depth limits for subsequent days.

[0099] Convergence condition: When the rate of change of the objective function is less than 1% for three consecutive iterations, it is considered converged.

[0100] According to the coupling factor transmission results, a spatiotemporal correlation constraint propagation mechanism is introduced to ensure the consistency of short-term scheduling and long-term strategies in key parameters such as energy storage charging and discharging depth, and output a coordinated scheduling framework.

[0101] The spatiotemporal correlation constraint propagation mechanism resolves policy conflicts across time scales through rule reasoning and dynamic priority adjustment.

[0102] Constraint conflict detection: Charge-discharge depth conflict: For example, hourly optimization requires a single discharge depth of 90% for a storage unit to meet peak load requirements, but the monthly strategy requires an average DOD ≤ 60%.

[0103] Lifespan balance conflict: The monthly attenuation rate of a certain energy storage unit exceeds the average by 20% due to frequent use, violating the lifespan balance goal.

[0104] Constraint Propagation and Precedence Rules: ‌Hard constraints take precedence‌: Safety-related constraints (such as SOC ≥ 20%) unconditionally take precedence over economic constraints.

[0105] Dynamic Weight Adjustment: Penalties are imposed on conflicting parameters (e.g., DOD). For example, if the DOD of the hourly plan exceeds the limit, a penalty term PENALTY = 1000 * (DOD - 60%) is added to the objective function, forcing the algorithm to re-search for a feasible solution.

[0106] ‌Rule inference example‌: If the predicted load peak on a particular day exceeds the historical extreme by 10%, the DOD is temporarily allowed to increase to 70%, but must be compensated and reduced to 50% on a subsequent date.

[0107] ‌Collaborative framework output‌: The final scheduling framework includes hourly and monthly optimization models, coupling factor transfer interface and constraint rule library.

[0108] ‌Framework operation process‌: The monthly layer generates a draft charge and discharge plan and passes it to the hourly layer; The hourly layer performs real-time optimization according to the draft and returns the actual parameters; The constraint propagation mechanism detects and corrects conflicts; Repeat the iteration until all constraints are satisfied.

[0109] Key technology examples and parameter descriptions: ‌GEN_COST‌: Source-side power generation cost, including fuel costs, operation and maintenance costs, and start-up and shutdown losses.

[0110] ‌LOSS_COST‌: The economic cost of grid-side transmission loss, calculated by multiplying the line loss power by the real-time electricity price.

[0111] ‌SOC‌ (State of Charge): The state of charge of the energy storage system, indicating the percentage of the current remaining capacity to the total capacity.

[0112] ‌DOD‌ (Depth of Discharge): Depth of charge and discharge refers to the ratio of single discharge capacity to total capacity.

[0113] ‌COUPLING_MATRIX‌: Coupling factor transfer matrix, used for parameter interaction across time scales. The matrix elements include key parameters such as SOC and charge and discharge times.

[0114] Based on the results of uncertainty analysis, a hierarchical optimization architecture was constructed. Coupled factor decomposition techniques were used to identify the interaction parameters between hourly scheduling and monthly planning. A cross-timescale constraint propagation mechanism was established to ensure that short-term operations and long-term strategies are coordinated across key variables such as energy storage charge and discharge depth. This framework addresses the curse of dimensionality in multi-timescale optimization and reduces computational complexity through intelligent decomposition. The introduction of spatiotemporal constraints ensures the temporal consistency of scheduling solutions, preventing short-term optimization from overdrawing long-term resources.

[0115] S204, based on the collaborative scheduling framework, a hybrid algorithm is used to perform multi-objective optimization solutions, wherein the short-term optimization problem is handled by the collaborative evolution mechanism of genetic algorithm and reinforcement learning, and a fuzzy logic decision tree is introduced to optimize the long-term strategy, and the global optimal scheduling strategy is output.

[0116] Specifically, based on the hourly objective function in the collaborative scheduling framework, a genetic algorithm can be used to initialize the population, encode the real-time electricity price, load demand, and energy storage SOC into chromosome genes, and generate an initial short-term solution set; The initial population of a genetic algorithm (GA) is the basis for searching for feasible solutions. Hourly scheduling parameters must be encoded as a chromosome structure. The chromosome genes correspond to scheduling decision variables, including real-time price (RTP), load demand (LD), and energy storage state of charge (SOC).

[0117] ‌Chromosome coding design‌: Gene Structure: Each chromosome represents the schedule for the next 72 hours and contains 216 genes (3 parameters x 72 hours). For example, the genome for hour 1 is [RTP_1, LD_1, SOC_1], and for hour 72 it is [RTP_72, LD_72, SOC_72].

[0118] Coding format: Real-coding is used. For example, the RTP value range is [0.1 yuan / kWh, 1.0 yuan / kWh], the LD range is [100MW, 300MW], and the SOC range is [20%, 90%].

[0119] Population size: The initial population consists of 100 individuals (chromosomes), which are randomly generated. For example, individual 1 has RTP_1 = 0.5 yuan / kWh, LD_1 = 150MW, and SOC_1 = 60%.

[0120] Fitness function definition: The fitness value reflects the quality of the solution and is calculated as the inverse of the total cost (Fitness = 1 / TOTAL_COST), with the goal of minimizing the cost.

[0121] The total cost (TOTAL_COST) includes the generation cost (GEN_COST), transmission loss (LOSS_COST), and load deviation penalty (LOAD_PENALTY). For example, if the total cost of an individual is 1.2 million yuan, the fitness value is 1 / 120 = 0.0083.

[0122] ‌ Initial solution set generation and screening‌ : After randomly generating 100 individuals, the fitness value of each individual is calculated, and individuals with fitness lower than the average value (such as the average value 0.005) are eliminated, and the top 50 high-quality individuals are retained.

[0123] The remaining individuals are subjected to an "elite retention strategy" (Elitism), which is directly copied to the next generation to ensure that excellent genes are not lost.

[0124] Based on the initial short-term solution set, a deep reinforcement learning agent is embedded to dynamically adjust the crossover and mutation probability, and the Q-learning algorithm is used to optimize the solution set convergence speed and output the Pareto frontier short-term candidate solutions. The Deep Reinforcement Learning (DRL) agent dynamically adjusts the crossover rate (CR) and mutation rate (MR) of the genetic algorithm through the Q-learning algorithm to accelerate convergence to the Pareto Front.

[0125] ‌State-Action Space Definition‌: State: Current population diversity indicators, including fitness variance (FV) and gene similarity (GS). For example, an FV of 0.002 indicates that the population fitness variance is small and the mutation probability needs to be increased.

[0126] Action: Adjust the crossover probability CR and mutation probability MR. For example, the action space is [CR+0.1, CR-0.1, MR+0.05, MR-0.05].

[0127] Reward: Based on the population's evolutionary speed, if the optimal fitness of the new generation improves by 10%, the reward is +10; if it degrades, the penalty is -5.

[0128] ‌Q-learning algorithm process‌: ‌Q table initialization‌: Build a state-action value table with initial Q values ​​of 0.

[0129] ‌Exploration and Exploitation‌: Adopting the ε-greedy strategy, the first 50 generations are random exploration actions with a probability of ε=0.3, and then gradually reducing it to ε=0.1.

[0130] ‌Q-value update‌: Update the Q-table based on the reward. For example, if the fitness improves after executing the action "CR+0.1" in a certain state, the Q-value is updated to Q(s,a) = Q(s,a) + α*(reward + γ*max(Q(s',a')) - Q(s,a)), where the learning rate α = 0.1 and the discount factor γ = 0.9.

[0131] Pareto front candidate solution generation: After 100 generations of iteration, non-dominated solutions are screened from the final population, that is, no other solution is better than this solution in all objectives.

[0132] For example, solution A has a total cost of 1 million yuan and a lifespan balance index of 5%, while solution B has a cost of 1.05 million yuan and a balance index of 3%. If solution A is better than solution B in terms of cost, and solution B is better than solution A in terms of balance, then both are candidate solutions on the Pareto front.

[0133] The short-term candidate solutions of the Pareto front are input into the fuzzy logic decision tree and combined with the uncertainty parameters in the monthly optimization objective to generate a long-term strategy fuzzy rule base; Fuzzy Logic Decision Tree (FLDT) maps short-term candidate solutions to long-term strategy rules, which requires the definition of fuzzy sets and inference rules.

[0134] ‌Fuzzy set definition‌: Input variables: Load demand (LD): Fuzzified into “Low”, “Medium”, and “High”, and divided into intervals of [100MW, 150MW], [150MW, 250MW], and [250MW, 300MW].

[0135] Real-time electricity price (RTP): Fuzzy into "Low", "Mid", and "High", with intervals of [0.1 yuan, 0.4 yuan], [0.4 yuan, 0.7 yuan], and [0.7 yuan, 1.0 yuan].

[0136] Energy storage SOC: Fuzzy into "Low", "Mid", and "High", corresponding to [20%, 50%], [50%, 80%], and [80%, 90%].

[0137] Output variables: Long-term charge and discharge strategies, including "Aggressive Charge", "Conservative Charge and Discharge", and "Aggressive Discharge".

[0138] Fuzzy rule base construction: Generate rules based on expert experience and historical data. For example: Rule 1: If LD = High and RTP = Low, then strategy = Aggressive Charge (charge during low-price periods to meet high load demand).

[0139] Rule 2: If SOC = High and RTP = High, then strategy = Aggressive Discharge (discharge to profit during high price periods).

[0140] The rule base contains a total of 27 rules (3 input variables × 3 fuzzy levels^3).

[0141] Uncertainty parameter fusion: Combined with uncertainty quantification parameters (such as the probability of weather impact) in monthly optimization, the confidence level of the rules is modified. For example, if the probability of high temperature in a certain month is 70%, the weight of Rule 1 is increased to 1.2 times.

[0142] Based on the long-term strategy fuzzy rule base, the long-term strategy is iteratively optimized through the dynamic programming algorithm to output a monthly charging and discharging plan that is decoupled from the short-term candidates of the Pareto front; The dynamic programming (DP) algorithm decomposes the 30-day scheduling problem into daily decision-making stages to minimize the long-term cost and satisfy the lifetime equilibrium constraint.

[0143] State variables and decision variables: ‌State variables‌: Energy storage SOC (range 20%~90%), discretized into 10% intervals (such as 20%, 30%, ..., 90%).

[0144] The energy storage decay rate (DR) is discretized into three levels: 0.5%, 0.8%, and 1.0% per month.

[0145] Decision variables (Action): Daily charge and discharge strategies, including charge power (CP) and discharge power (DP).

[0146] State transition equation: SOC update: SOC_{t+1} = SOC_t + (CP - DP) / CAPACITY, where CAPACITY is the total energy storage capacity (e.g. 100MWh).

[0147] Attenuation rate update: If the number of charge and discharge cycles in a single day exceeds 2 times, the DR will be increased by 0.1%.

[0148] ‌Value Function and Iterative Optimization‌: Value Function: V(s) = Minimum Total Cost (Monthly GEN_COST + STORAGE_COST) + Life Balance Penalty.

[0149] Bellman equation: V(s_t) = min_{a_t} [ Cost(s_t, a_t) + γ * V(s_{t+1}) ], where the discount factor γ=0.95.

[0150] ‌Reverse Iteration‌: Work backward from day 30 to day 1, calculating the optimal strategy day by day. For example, SOC needs to reach 50% on day 30 to meet the month-end equilibrium target.

[0151] The short-term candidate solutions of the Pareto front and the monthly charging and discharging plan are input into the non-dominated sorting genetic algorithm, and the comprehensive cost and risk score are calculated by combining the entropy weight method to screen out the global optimal scheduling strategy.

[0152] The Non-Dominated Sorting Genetic Algorithm (NSGA-II) is used to integrate the short-term and long-term solution sets, and the Entropy Weight Method (ENM) quantifies the weights of multiple objectives to screen the optimal solution.

[0153] ‌Non-dominated sorting and congestion calculation‌: Non-dominated sorting: The solution set is divided into layers according to dominance. The first layer contains the Pareto front solutions, the second layer contains the solutions dominated by the first layer, and so on.

[0154] Crowding Distance: Calculates the density of solutions in the target space, prioritizing solutions in sparse areas to maintain diversity. For example, if a solution has a distance of 100,000 yuan from its neighbor on the cost axis and a distance of 2% on the balance axis, then the crowding distance is 10 + 2 = 12.

[0155] ‌Entropy weight method weight distribution‌: Data normalization: Normalize the cost (unit: 10,000 yuan) and risk score (unit: %) to [0, 1].

[0156] ‌Information entropy calculation‌: Cost entropy E_COST = -Σ(p_i * ln p_i), where p_i is the cost proportion of the i-th solution.

[0157] Similarly, calculate the risk entropy E_RISK.

[0158] Weight Allocation: W_COST = (1 - E_COST) / [(1 - E_COST) + (1 - E_RISK)]. For example, if E_COST = 0.2 and E_RISK = 0.5, then W_COST = 0.8 / 1.3 ≈ 61.5%.

[0159] ‌Global optimal solution screening‌: Calculate the composite score for each solution: SCORE = W_COST * Normalized Cost + W_RISK * Normalized Risk.

[0160] Select the solution with the highest overall score. For example, if solution X has a standardized cost of 0.3 and a risk of 0.2, then SCORE = 0.6150.3 + 0.3850.2 = 0.254. Select the solution with the highest score.

[0161] Key technology examples and parameter descriptions: ‌SOC (State of Charge)‌: The percentage of the remaining capacity of the energy storage system to the total capacity, which is the core state variable of the charging and discharging strategy.

[0162] NSGA-II: Non-dominated sorting genetic algorithm, used to solve multi-objective optimization problems, ensuring the diversity and convergence of the solution set through hierarchical sorting and crowding calculation.

[0163] Entropy Weight Method: An objective weight allocation method based on information entropy to avoid the influence of subjective preferences on decision-making.

[0164] To address the unique characteristics of optimization problems at different timescales, the advantages of intelligent algorithms are combined: genetic algorithms handle discrete decision variables, while reinforcement learning dynamically adjusts search directions to accelerate short-term optimization convergence. Fuzzy logic handles the linguistic rules of long-term strategies, transforming expert experience into computable decision trees. Hybrid algorithms leverage the complementary nature of each technology, improving computational efficiency while ensuring solution accuracy. This global strategy achieves a multi-objective balance between economy, reliability, and equipment lifespan, providing optimal decision support for the coordinated operation of power generation, grid, load, and storage.

[0165] It can be seen that according to the source-side power generation cost, grid-side transmission loss, load-side load demand and storage-side life attenuation data of the energy system, a dynamic coupling map containing short-term and long-term cost evolution paths is output; based on the dynamic coupling map, uncertainty quantification parameters containing probability distribution are output; according to the uncertainty quantification parameters, a collaborative scheduling framework containing time-space correlation constraints is output; based on the collaborative scheduling framework, a hybrid algorithm is used to perform multi-objective optimization solution, and the global optimal scheduling strategy is output, so that the global optimal decision can be achieved to balance the system economy, safety and equipment life.

[0166] Another embodiment of the present invention provides a source-grid-load-storage coordinated scheduling system based on multi-time-scale cost optimization, see Figure 3 , the system may include: Analysis module 301 is used to perform spatiotemporal correlation analysis between electricity price fluctuations and energy storage cycle life using a time series prediction algorithm based on the energy system's source-side power generation costs, grid-side transmission losses, load-side load demand, and storage-side life decay data, and output a dynamic coupling map containing short-term and long-term cost evolution paths; A generation module 302 is configured to quantify the influencing factors of weather changes and load fluctuations based on the dynamic coupling map, combined with meteorological forecast data and historical load fluctuation characteristics, using a dynamic uncertainty map generation technology, and output uncertainty quantification parameters including probability distribution; Decomposition module 303 is used to design a multi-time-scale coupled scheduling framework based on the uncertainty quantification parameters, decompose the scheduling task into hourly short-term optimization and monthly long-term optimization sub-problems using a multi-scale coupled factor decomposition model, and output a collaborative scheduling framework containing spatiotemporal correlation constraints; The output module 304 is used to perform multi-objective optimization solutions based on the collaborative scheduling framework using a hybrid algorithm, wherein short-term optimization problems are handled through the collaborative evolution mechanism of genetic algorithm and reinforcement learning, and fuzzy logic decision trees are introduced to optimize long-term strategies to output the global optimal scheduling strategy.

[0167] It can be seen that according to the source-side power generation cost, grid-side transmission loss, load-side load demand and storage-side life attenuation data of the energy system, a dynamic coupling map containing short-term and long-term cost evolution paths is output; based on the dynamic coupling map, uncertainty quantification parameters containing probability distribution are output; according to the uncertainty quantification parameters, a collaborative scheduling framework containing time-space correlation constraints is output; based on the collaborative scheduling framework, a hybrid algorithm is used to perform multi-objective optimization solution, and the global optimal scheduling strategy is output, so that the global optimal decision can be achieved to balance the system economy, safety and equipment life.

[0168] An embodiment of the present invention further provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps of any one of the above method embodiments when running.

[0169] Specifically, in this embodiment, the above-mentioned storage medium may be configured to store a computer program for performing the following steps: S201: Based on the energy system's source-side power generation costs, grid-side transmission losses, load-side load demand, and storage-side lifespan decay data, a time series prediction algorithm is used to perform spatiotemporal correlation analysis between electricity price fluctuations and energy storage cycle lifespan, outputting a dynamic coupling map containing short-term and long-term cost evolution paths. S202, based on the dynamic coupling map, combined with meteorological forecast data and historical load fluctuation characteristics, quantify the influencing factors of weather changes and load fluctuations through dynamic uncertainty map generation technology, and output uncertainty quantification parameters including probability distribution; S203: Design a multi-time-scale coupled scheduling framework based on the uncertainty quantification parameters, decompose the scheduling task into hourly short-term optimization and monthly long-term optimization sub-problems using a multi-scale coupled factor decomposition model, and output a collaborative scheduling framework that includes spatiotemporal correlation constraints. S204, based on the collaborative scheduling framework, a hybrid algorithm is used to perform multi-objective optimization solutions, wherein the short-term optimization problem is handled by the collaborative evolution mechanism of genetic algorithm and reinforcement learning, and a fuzzy logic decision tree is introduced to optimize the long-term strategy, and the global optimal scheduling strategy is output.

[0170] It can be seen that according to the source-side power generation cost, grid-side transmission loss, load-side load demand and storage-side life attenuation data of the energy system, a dynamic coupling map containing short-term and long-term cost evolution paths is output; based on the dynamic coupling map, uncertainty quantification parameters containing probability distribution are output; according to the uncertainty quantification parameters, a collaborative scheduling framework containing time-space correlation constraints is output; based on the collaborative scheduling framework, a hybrid algorithm is used to perform multi-objective optimization solution, and the global optimal scheduling strategy is output, so that the global optimal decision can be achieved to balance the system economy, safety and equipment life.

[0171] An embodiment of the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any one of the above method embodiments.

[0172] Specifically, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0173] Specifically, in this embodiment, the processor may be configured to execute the following steps through a computer program: S201: Based on the energy system's source-side power generation costs, grid-side transmission losses, load-side load demand, and storage-side lifespan decay data, a time series prediction algorithm is used to perform spatiotemporal correlation analysis between electricity price fluctuations and energy storage cycle lifespan, outputting a dynamic coupling map containing short-term and long-term cost evolution paths. S202, based on the dynamic coupling map, combined with meteorological forecast data and historical load fluctuation characteristics, quantify the influencing factors of weather changes and load fluctuations through dynamic uncertainty map generation technology, and output uncertainty quantification parameters including probability distribution; S203: Design a multi-time-scale coupled scheduling framework based on the uncertainty quantification parameters, decompose the scheduling task into hourly short-term optimization and monthly long-term optimization sub-problems using a multi-scale coupled factor decomposition model, and output a collaborative scheduling framework that includes spatiotemporal correlation constraints. S204, based on the collaborative scheduling framework, a hybrid algorithm is used to perform multi-objective optimization solutions, wherein the short-term optimization problem is handled by the collaborative evolution mechanism of genetic algorithm and reinforcement learning, and a fuzzy logic decision tree is introduced to optimize the long-term strategy, and the global optimal scheduling strategy is output.

[0174] It can be seen that according to the source-side power generation cost, grid-side transmission loss, load-side load demand and storage-side life attenuation data of the energy system, a dynamic coupling map containing short-term and long-term cost evolution paths is output; based on the dynamic coupling map, uncertainty quantification parameters containing probability distribution are output; according to the uncertainty quantification parameters, a collaborative scheduling framework containing time-space correlation constraints is output; based on the collaborative scheduling framework, a hybrid algorithm is used to perform multi-objective optimization solution, and the global optimal scheduling strategy is output, so that the global optimal decision can be achieved to balance the system economy, safety and equipment life.

[0175] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the scope of implementation of the present invention is not limited to what is shown in the drawings. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which do not exceed the spirit covered by the description and drawings, should be within the scope of protection of the present invention.

Claims

1. A source-grid-load-storage coordinated scheduling method based on multi-time-scale cost optimization, characterized in that: The method comprises: Based on the energy system's source-side power generation costs, grid-side transmission losses, load-side load demand, and storage-side life attenuation data, a time series prediction algorithm is used to conduct a spatiotemporal correlation analysis between electricity price fluctuations and energy storage cycle life, outputting a dynamic coupling map containing short-term and long-term cost evolution paths. Based on the dynamic coupling map, combined with meteorological forecast data and historical load fluctuation characteristics, the influencing factors of weather changes and load fluctuations are quantified through dynamic uncertainty map generation technology, and uncertainty quantification parameters containing probability distribution are output; Based on the uncertainty quantification parameters, a multi-time-scale coupled scheduling framework is designed. A multi-scale coupled factor decomposition model is used to decompose the scheduling task into hourly short-term optimization and monthly long-term optimization sub-problems, and a collaborative scheduling framework with spatiotemporal correlation constraints is output. Based on the collaborative scheduling framework, a hybrid algorithm is adopted to solve multi-objective optimization problems. The short-term optimization problem is handled by the co-evolutionary mechanism of genetic algorithm and reinforcement learning, and the fuzzy logic decision tree is introduced to optimize the long-term strategy and output the global optimal scheduling strategy.

2. The method according to claim 1, characterized in that Based on the energy system's source-side power generation costs, grid-side transmission losses, load-side load demand, and storage-side life attenuation data, a time series prediction algorithm is used to perform spatiotemporal correlation analysis on electricity price fluctuations and energy storage cycle life, outputting a dynamic coupling map containing short-term and long-term cost evolution paths, including: Based on the source-side power generation cost time series data and the storage-side life decay curve, multi-scale wavelet transform is used to extract the power generation cost fluctuation characteristics and generate spatiotemporal correlation feature vectors. The spatiotemporal correlation feature vector is integrated with the topological distribution data of the network-side transmission loss, and the source-grid-load-storage coupling relationship is modeled through the spatiotemporal attention mechanism to output a four-dimensional dynamic correlation tensor. Based on the four-dimensional dynamic correlation tensor, a tensor decomposition algorithm is used to separate short-term electricity price sensitivity factors and long-term energy storage life influencing factors, generating a dual-time scale prediction feature set consisting of a short-term prediction feature set and a long-term prediction feature set. The short-term prediction feature set is input into the long short-term memory network to predict the electricity price fluctuation curve in the next 72 hours. At the same time, the long-term prediction feature set is input into the Prophet model to predict the 30-day energy storage cycle life attenuation trajectory, and the two are fused to generate a dynamic coupling map.

3. The method according to claim 2, characterized in that Based on the dynamic coupling map, combined with meteorological forecast data and historical load fluctuation characteristics, the dynamic uncertainty map generation technology is used to quantify the influencing factors of weather changes and load fluctuations, and output uncertainty quantification parameters containing probability distribution, including: According to the long-term energy storage life trajectory in the dynamic coupling map, meteorological sensitive features are extracted to generate meteorological impact coding vectors; The meteorological impact encoding vector is input into the dynamic Bayesian network, combined with the historical load fluctuation characteristics, to model the weather-load joint probability distribution and output the uncertainty propagation map; Based on the uncertainty propagation graph, Monte Carlo sampling of extreme weather scenarios is embedded in the dynamic coupling map to generate a risk scenario set; Based on the risk scenario set, the kernel density estimation algorithm is used to quantify the joint impact of weather and load fluctuations on system costs and output a multidimensional probability density function. The multidimensional probability density function is aligned with the spatiotemporal features of the dynamic coupling map to generate an uncertainty quantification parameter matrix with confidence intervals.

4. The method according to claim 3, characterized in that Based on the uncertainty quantification parameters, a multi-time-scale coupled scheduling framework is designed. A multi-scale coupled factor decomposition model is used to decompose the scheduling task into hourly short-term optimization and monthly long-term optimization sub-problems, and a collaborative scheduling framework containing spatiotemporal correlation constraints is output, including: Based on the uncertainty quantification parameter matrix, an hourly scheduling objective function is constructed to define dynamic constraints on power generation cost, transmission loss, and load deviation. Based on the long-term energy storage attenuation trajectory in the dynamic coupling map, a monthly optimization objective function is established to define the energy storage life balance and investment return constraints; A two-layer decomposition algorithm is used to decompose hourly and monthly targets into independent sub-problems, and cross-timescale parameter interaction is achieved through the coupling factor transfer matrix. According to the coupling factor transmission results, a spatiotemporal correlation constraint propagation mechanism is introduced to ensure the consistency of short-term scheduling and long-term strategies in key parameters such as energy storage charging and discharging depth, and output a coordinated scheduling framework.

5. The method according to claim 4, characterized in that Based on the collaborative scheduling framework, a hybrid algorithm is used to solve multi-objective optimization problems. In this method, the short-term optimization problem is handled by the co-evolutionary mechanism of genetic algorithm and reinforcement learning, and the long-term strategy is optimized by introducing fuzzy logic decision tree to output the global optimal scheduling strategy, including: According to the hourly objective function in the coordinated scheduling framework, a genetic algorithm is used to initialize the population, encode the real-time electricity price, load demand and energy storage SOC into chromosome genes, and generate the initial short-term solution set; Based on the initial short-term solution set, a deep reinforcement learning agent is embedded to dynamically adjust the crossover and mutation probability, and the Q-learning algorithm is used to optimize the solution set convergence speed and output the Pareto frontier short-term candidate solutions. The short-term candidate solutions of the Pareto front are input into the fuzzy logic decision tree and combined with the uncertainty parameters in the monthly optimization objective to generate a long-term strategy fuzzy rule base; Based on the long-term strategy fuzzy rule base, the long-term strategy is iteratively optimized through the dynamic programming algorithm to output a monthly charging and discharging plan that is decoupled from the short-term candidates of the Pareto front; The short-term candidate solutions of the Pareto front and the monthly charging and discharging plan are input into the non-dominated sorting genetic algorithm, and the comprehensive cost and risk score are calculated by combining the entropy weight method to screen out the global optimal scheduling strategy.

6. A source-grid-load-storage coordinated dispatching system based on multi-time-scale cost optimization, characterized by: The system comprises: The analysis module is used to conduct spatiotemporal correlation analysis between electricity price fluctuations and energy storage cycle life based on the energy system's source-side power generation costs, grid-side transmission losses, load-side load demand, and storage-side life decay data using a time series prediction algorithm. The module then outputs a dynamic coupling map containing short-term and long-term cost evolution paths. A generation module is used to quantify the influencing factors of weather changes and load fluctuations based on the dynamic coupling map, combined with meteorological forecast data and historical load fluctuation characteristics, through dynamic uncertainty map generation technology, and output uncertainty quantification parameters including probability distribution; A decomposition module is used to design a multi-time-scale coupled scheduling framework based on the uncertainty quantification parameters, decompose the scheduling task into hourly short-term optimization and monthly long-term optimization sub-problems using a multi-scale coupled factor decomposition model, and output a collaborative scheduling framework with spatiotemporal correlation constraints; The output module is used to perform multi-objective optimization solutions based on the collaborative scheduling framework using a hybrid algorithm, wherein short-term optimization problems are handled through the collaborative evolution mechanism of genetic algorithm and reinforcement learning, and fuzzy logic decision trees are introduced to optimize long-term strategies to output the global optimal scheduling strategy.

7. The system according to claim 6, characterized in that The analysis module is specifically used to: Based on the source-side power generation cost time series data and the storage-side life decay curve, multi-scale wavelet transform is used to extract the power generation cost fluctuation characteristics and generate spatiotemporal correlation feature vectors. The spatiotemporal correlation feature vector is integrated with the topological distribution data of the network-side transmission loss, and the source-grid-load-storage coupling relationship is modeled through the spatiotemporal attention mechanism to output a four-dimensional dynamic correlation tensor. Based on the four-dimensional dynamic correlation tensor, a tensor decomposition algorithm is used to separate short-term electricity price sensitivity factors and long-term energy storage life influencing factors, generating a dual-time scale prediction feature set consisting of a short-term prediction feature set and a long-term prediction feature set. The short-term prediction feature set is input into the long short-term memory network to predict the electricity price fluctuation curve in the next 72 hours. At the same time, the long-term prediction feature set is input into the Prophet model to predict the 30-day energy storage cycle life attenuation trajectory, and the two are fused to generate a dynamic coupling map.

8. The system according to claim 7, characterized in that The generation module is specifically used to: According to the long-term energy storage life trajectory in the dynamic coupling map, meteorological sensitive features are extracted to generate meteorological impact coding vectors; The meteorological impact encoding vector is input into the dynamic Bayesian network, combined with the historical load fluctuation characteristics, to model the weather-load joint probability distribution and output the uncertainty propagation map; Based on the uncertainty propagation graph, Monte Carlo sampling of extreme weather scenarios is embedded in the dynamic coupling map to generate a risk scenario set; Based on the risk scenario set, the kernel density estimation algorithm is used to quantify the joint impact of weather and load fluctuations on system costs and output a multidimensional probability density function. The multidimensional probability density function is aligned with the spatiotemporal features of the dynamic coupling map to generate an uncertainty quantification parameter matrix with confidence intervals.

9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 5 when executed.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 5.

Images