Source-grid-load-storage collaborative scheduling method and system based on multi-time scale cost optimization

Abstract

The application discloses a source-network-load-storage collaborative scheduling method and system based on multi-time scale cost optimization, and the method comprises the following steps: according to the source-side power generation cost, the network-side transmission loss, the load-side load demand and the storage-side life attenuation data of an energy system, a dynamic coupling graph containing short-term and long-term cost evolution paths is output; based on the dynamic coupling graph, uncertainty quantization parameters containing probability distribution are output; according to the uncertainty quantization parameters, a collaborative scheduling framework containing space-time correlation constraints is output; based on the collaborative scheduling framework, a mixed algorithm is used for multi-objective optimization solution, and a global optimal scheduling strategy is output. By using the embodiment of the application, global optimal decision can be realized to balance the system economy, safety and equipment life.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of source network load storage scheduling, and particularly relates to a source network load storage collaborative scheduling method and system based on multi-time scale cost optimization. BACKGROUND

[0002] The source network load storage collaborative scheduling faces challenges such as multi-time scale coupling and enhanced uncertainty. Traditional scheduling methods usually adopt single-time scale optimization, which is difficult to coordinate the contradiction between short-term operation cost and long-term equipment wear and tear, and lacks quantitative analysis capability for uncertain factors such as weather changes and load fluctuations. In the prior art, short-term scheduling is mostly focused on day-ahead economic scheduling, while long-term planning focuses on annual investment decision-making, and the separation of the two leads to accelerated wear and tear of energy storage life or imbalance of source-load matching. In addition, existing uncertainty modeling methods are mostly based on static probability distribution, and cannot dynamically respond to the spatiotemporal correlation characteristics of extreme weather events and market electricity price fluctuations. SUMMARY

[0003] The purpose of the present application is to provide a source network load storage collaborative scheduling method and system based on multi-time scale cost optimization to solve the problems in the prior art and achieve global optimal decision-making to balance system economy, safety and equipment life.

[0004] One embodiment of the present application provides a source network load storage collaborative scheduling method based on multi-time scale cost optimization, which comprises the following steps:

[0005] According to the source-side power generation cost, network-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 a dynamic coupling graph containing short-term and long-term cost evolution paths is outputted;

[0006] Based on the dynamic coupling graph, combining meteorological prediction data and historical load fluctuation characteristics, a dynamic uncertainty graph generation technology is used to quantify the influence factors of weather changes and load fluctuations, and an uncertainty quantization parameter containing a probability distribution is outputted;

[0007] According to the uncertainty quantization parameter, a multi-time scale coupling scheduling framework is designed, a multi-scale coupling factor decomposition model is used to decompose the scheduling task into hour-level short-term optimization and month-level long-term optimization sub-problems, and a collaborative scheduling framework containing spatiotemporal correlation constraints is outputted;

[0008] Based on the collaborative scheduling framework, a hybrid algorithm is used for multi-objective optimization solution, wherein a genetic algorithm and a reinforcement learning co-evolution mechanism are used to process the short-term optimization problem, and a fuzzy logic decision tree is introduced to optimize the long-term strategy, and a global optimal scheduling strategy is outputted.

[0009] Optionally, according to the source-side power generation cost, the network-side transmission loss, the load-side load demand and the storage-side life attenuation data of the energy system, time series prediction algorithm is adopted to perform space-time correlation analysis on electricity price fluctuation and storage cycle life, and a dynamic coupling graph containing short-term and long-term cost evolution paths is output, including:

[0010] According to the time sequence data of the source-side power generation cost and the storage-side life attenuation curve, multi-scale wavelet transform is adopted to extract the power generation cost fluctuation characteristics, and a space-time correlation feature vector is generated;

[0011] The space-time correlation feature vector is fused with the topological distribution data of the network-side transmission loss, and the source-network-load-storage coupling relationship is modeled through the space-time attention mechanism, and a four-dimensional dynamic correlation tensor is output;

[0012] Based on the four-dimensional dynamic correlation tensor, a tensor decomposition algorithm is adopted to separate the short-term electricity price sensitive factors and the long-term storage life influence factors, and a double-time-scale prediction feature set containing a short-term prediction feature set and a long-term prediction feature set is generated;

[0013] The short-term prediction feature set is input into the long short-term memory network to predict the future 72-hour electricity price fluctuation curve, and the long-term prediction feature set is input into the Prophet model to predict the 30-day storage cycle life attenuation trajectory, and a dynamic coupling graph is fused and generated.

[0014] Optionally, based on the dynamic coupling graph, meteorological prediction data and historical load fluctuation characteristics are combined, and the influence factors of weather changes and load fluctuations are quantified through dynamic uncertainty graph generation technology, and uncertainty quantization parameters containing probability distribution are output, including:

[0015] According to the long-term storage life trajectory in the dynamic coupling graph, meteorological sensitive features are extracted, and a weather influence coding vector is generated;

[0016] The weather influence coding vector is input into the dynamic Bayesian network, and the historical load fluctuation characteristics are combined to model the weather-load joint probability distribution, and an uncertainty propagation graph is output;

[0017] Based on the uncertainty propagation graph, an extreme weather scenario set is generated by embedding a Monte Carlo sampling in the dynamic coupling graph;

[0018] According to the risk scenario set, a kernel density estimation algorithm is adopted to quantify the joint influence of weather and load fluctuation on system cost, and a multi-dimensional probability density function is output;

[0019] The multi-dimensional probability density function is aligned with the space-time features of the dynamic coupling graph, and an uncertainty quantization parameter matrix with a confidence interval is generated.

[0020] Optionally, the scheduling framework is designed according to the uncertainty quantification parameter, a multi-scale coupling factor decomposition model is used to decompose the scheduling task into a short-term optimization at the hour level and a long-term optimization at the month level, and a collaborative scheduling framework containing spatiotemporal correlation constraints is output, including:

[0021] According to the uncertainty quantification parameter matrix, a short-term scheduling objective function is constructed, and dynamic constraint conditions of generation cost, transmission loss and load deviation are defined;

[0022] Based on the long-term energy storage attenuation trajectory in the dynamic coupling graph, a month-level optimization objective function is established, and life balance and return on investment constraints are defined;

[0023] A double-layer decomposition algorithm is used to decompose the short-term objective and the month-level objective into independent sub-problems, and cross-time scale parameter interaction is achieved through a coupling factor transfer matrix;

[0024] According to the coupling factor transfer result, a spatiotemporal correlation constraint propagation mechanism is introduced to ensure the consistency of short-term scheduling and long-term strategy in key parameters such as energy storage charging and discharging depth, and a collaborative scheduling framework is output.

[0025] Optionally, based on the collaborative scheduling framework, a hybrid algorithm is used for multi-objective optimization, wherein a co-evolution mechanism of genetic algorithm and reinforcement learning is used to process short-term optimization problems, and a fuzzy logic decision tree is used to optimize long-term strategy, and a global optimal scheduling strategy is output, including:

[0026] According to the short-term objective function in the collaborative scheduling framework, a genetic algorithm is used to initialize a population, real-time electricity price, load demand and energy storage SOC are coded as chromosome genes, and an initial short-term solution set is generated;

[0027] Based on the initial short-term solution set, a deep reinforcement learning agent is embedded to dynamically adjust the crossover and mutation probabilities, and a Q-learning algorithm is used to optimize the convergence speed of the solution set, and a Pareto frontier short-term candidate solution is output;

[0028] The Pareto frontier short-term candidate solution is input into a fuzzy logic decision tree, combined with the uncertainty parameters in the month-level optimization objective, and a long-term strategy fuzzy rule base is generated;

[0029] Based on the long-term strategy fuzzy rule base, a dynamic programming algorithm is used to iteratively optimize the long-term strategy, and a month-level charging and discharging plan decoupled from the Pareto frontier short-term candidate solution is output;

[0030] The Pareto frontier short-term candidate solution and the month-level charging and discharging plan are input into a non-dominated sorting genetic algorithm, and the comprehensive cost and risk score are calculated by entropy weight method, and the global optimal scheduling strategy is selected.

[0031] A further embodiment of the present application provides a source-grid-load-storage collaborative scheduling system based on multi-time scale cost optimization, comprising:

[0032] An analysis module is configured to perform spatio-temporal correlation analysis on electricity price fluctuation and energy storage cycle life by using a time series prediction algorithm according to source-side power generation cost, grid-side transmission loss, load-side load demand and storage-side life attenuation data of an energy system, and output a dynamic coupling graph containing short-term and long-term cost evolution paths.

[0033] A generation module is configured to quantify influence factors of weather changes and load fluctuations by using a dynamic uncertainty graph generation technology based on the dynamic coupling graph, in combination with meteorological prediction data and historical load fluctuation characteristics, and output uncertainty quantization parameters containing probability distribution.

[0034] A decomposition module is configured to design a multi-time scale coupled scheduling framework according to the uncertainty quantization parameters, decompose the scheduling task into hour-level short-term optimization and month-level long-term optimization sub-problems by using a multi-scale coupling factor decomposition model, and output a collaborative scheduling framework containing spatio-temporal correlation constraints.

[0035] An output module is configured to perform multi-objective optimization solving by using a hybrid algorithm based on the collaborative scheduling framework, wherein a co-evolution mechanism of a genetic algorithm and reinforcement learning is used to process short-term optimization problems, and a fuzzy logic decision tree is introduced to optimize long-term strategies, and a globally optimal scheduling strategy is output.

[0036] A further embodiment of the present application provides a storage medium having a computer program stored therein, wherein the computer program is configured to execute the method described in any of the above embodiments when running.

[0037] A further embodiment of the present application provides an electronic device comprising a memory and a processor, wherein the memory has a computer program stored therein, and the processor is configured to execute the computer program to perform the method described in any of the above embodiments.

[0038] Compared with the prior art, the source-grid-load-storage collaborative scheduling method based on multi-time scale cost optimization provided by the present application can output a dynamic coupling graph containing short-term and long-term cost evolution paths according to source-side power generation cost, grid-side transmission loss, load-side load demand and storage-side life attenuation data of an energy system, output uncertainty quantization parameters containing probability distribution based on the dynamic coupling graph, output a collaborative scheduling framework containing spatio-temporal correlation constraints according to the uncertainty quantization parameters, and output a globally optimal scheduling strategy by performing multi-objective optimization solving based on the collaborative scheduling framework by using a hybrid algorithm, so as to balance system economy, safety and equipment life. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 A hardware structure block diagram of a computer terminal of a source-grid-load-storage collaborative scheduling method based on multi-time scale cost optimization provided by the embodiment of the present application is provided.

[0040] Figure 2 A flowchart of a source-grid-load-storage collaborative scheduling method based on multi-time scale cost optimization provided by the embodiment of the present application is provided.

[0041] Figure 3 A structure diagram of a source-grid-load-storage collaborative scheduling system based on multi-time scale cost optimization provided by the embodiment of the present application is provided. DETAILED DESCRIPTION

[0042] The embodiments described below with reference to the drawings are exemplary and are only used to explain the present application, and cannot be explained as a limitation of the present application.

[0043] The embodiment of the present application first provides a source-grid-load-storage collaborative scheduling method based on multi-time scale cost optimization, which can be applied to electronic equipment, such as a computer terminal, specifically, a general computer, etc.

[0044] The following will be described in detail by taking a computer terminal as an example. Figure 1 A hardware structure block diagram of a computer terminal of a source-grid-load-storage collaborative scheduling method based on multi-time scale cost optimization provided by the embodiment of the present application is provided. As shown in the figure, Figure 1 The computer equipment includes a processor, a memory and a network interface connected through a system bus, wherein the memory can include a non-volatile storage medium and an internal memory.

[0045] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, which, when executed, can make the processor execute any kind of source-grid-load-storage collaborative scheduling method based on multi-time scale cost optimization.

[0046] The processor is used to provide computing and control capabilities to support the operation of the entire computer equipment.

[0047] The internal memory provides an environment for the running of the computer program in the non-volatile storage medium, which, when executed by the processor, can make the processor execute any kind of source-grid-load-storage collaborative scheduling method based on multi-time scale cost optimization.

[0048] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that, Figure 1The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0049] It should be understood that the processor can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0050] Referring to Figure 2 The embodiment of the present application provides a source-network-load-storage collaborative scheduling method based on multi-time scale cost optimization, which can include the following steps:

[0051] S201, according to the source-side power generation cost, the network-side transmission loss, the load-side load demand and the storage-side life attenuation data of the energy system, a time series prediction algorithm is used to perform space-time correlation analysis on the electricity price fluctuation and the storage cycle life, and a dynamic coupling graph containing short-term and long-term cost evolution paths is output.

[0052] Specifically, according to the source-side power generation cost time series data and the storage-side life attenuation curve, a multi-scale wavelet transform is used to extract the power generation cost fluctuation characteristics, and a space-time correlation feature vector is generated.

[0053] The multi-scale wavelet transform adopts Daubechies 4 wavelet basis (DB4) for five-layer decomposition, and decomposes the source-side power generation cost time series data (such as 15-minute sampling data of a wind farm) into different frequency bands:

[0054] High-frequency detail layer (D1-D3): capture minute-level fluctuations (such as power fluctuations caused by sudden changes in wind speed), and eliminate noise through threshold filtering (hard threshold λ=0.1*max|D|).

[0055] Low-frequency approximation layer (A5): extract the daily cycle trend (such as the load peak-valley rule), and use moving average smoothing (window width=96 sampling points).

[0056] The storage-side lifetime degradation curve (such as capacity degradation data of lithium-ion batteries) is processed using Morlet wavelet transform:

[0057] Scale parameters: Set 10 scales (s=1~10), corresponding to different charge and discharge cycle periods (e.g., s=5 maps to 200 cycles).

[0058] Wavelet coefficients: Calculate the energy density at each scale and quantize the decay rate (e.g., coefficient > 0.8 indicates accelerated decay).

[0059] Feature fusion:

[0060] High-frequency details of power generation costs (unit: yuan) 2 The energy storage attenuation coefficient (in % / cycle) is concatenated with the energy storage attenuation coefficient (in % / cycle) to form a hybrid vector;

[0061] Spatial location encoding (such as the GeoHash value of the latitude and longitude coordinates of the wind farm) is added to form a 128-dimensional spatiotemporal correlation feature vector.

[0062] Example: The daily 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 location codes.

[0063] By fusing spatiotemporal correlation feature vectors with topological distribution data of network-side transmission loss, and modeling the source-network-load-storage coupling relationship through a spatiotemporal attention mechanism, a four-dimensional dynamic correlation tensor is output.

[0064] Network-side transmission loss topology data includes:

[0065] Electrical distance matrix: per-unit impedance values ​​between nodes (e.g., 0.05~0.2 pu);

[0066] Heat map of power flow distribution: generated based on historical SCADA data (resolution 1km×1km).

[0067] Fusion and attention mechanisms:

[0068] Topological embedding:

[0069] The electrical distance matrix is ​​input to a graph convolutional network (GCN, number of layers=2), and the output is a node feature vector (dimension 32).

[0070] The heat map is input into a CNN (3×3 convolutional kernels) and outputs a spatial feature map (16×16×64).

[0071] Spatiotemporal attention:

[0072] Temporal attention head: compute the correlation of feature vectors along the time axis (e.g., the correlation weight between 7:00 and 19:00 is 0.9);

[0073] Spatial attention head: compute the energy flow correlation between grid nodes (e.g., the weight from wind farm A to substation B is 0.75);

[0074] Cross-modal attention head: model the source-storage dynamic interaction (e.g., the weight between wind power output and battery SOC is 0.6).

[0075] Tensor construction:

[0076] Dimension 1 (time): 96 time slices (24 hours x 4 sampling points); Dimension 2 (space): 50 grid nodes; Dimension 3 (feature): 128-dimensional mixed features; Dimension 4 (coupling relationship): attention weight matrix (50 x 50).

[0077] Example: At a certain time, the tensor slice is [Node 1 feature: (0.4, 0.2,...), coupling weight from Node 2 to Node 1: 0.83].

[0078] Based on the four-dimensional dynamic correlation tensor, a tensor decomposition algorithm is used to separate the short-term electricity price sensitive factor and the long-term storage life influence factor, generating a double-time scale prediction feature set containing a short-term prediction feature set and a long-term prediction feature set;

[0079] Tensor decomposition uses the Tucker decomposition model:

[0080] Core tensor extraction:

[0081] The original tensor X ∈ R^{96×50×128×50} is decomposed into:

[0082] Core tensor G ∈ R^{24×10×32×10} (compression rate 85%);

[0083] Factor matrices: time matrix A ∈ R^{96×24}, spatial matrix B ∈ R^{50×10}, feature matrix C ∈ R^{128×32}, and coupling matrix D ∈ R^{50×10}.

[0084] Factor separation:

[0085] Short-term electricity price sensitive factor: take the first 12 time modes 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, load change rate);

[0086] Long-term storage life factor: take the last 12 time modes of G (corresponding to 7~30 days) and the last 16 dimensions of C (such as cycle depth DoD, average charge and discharge rate).

[0087] Feature Set Construction:

[0088] Short-Term Feature Set: 24-dimensional vector (12 time patterns x 2 key features), example: [0.35, -0.12,...] represents future 6-hour price sensitivity;

[0089] Long-Term Feature Set: 24-dimensional vector (12 time patterns x 2 key features), example: [0.08, 0.21,...] represents future 20-day life decay rate.

[0090] Input short-term prediction feature set into LSTM network to predict 72-hour price fluctuation curve, and input long-term prediction feature set into Prophet model to predict 30-day energy storage cycle life decay trajectory, and generate dynamic coupling atlas.

[0091] Short-Term Price Prediction (LSTM):

[0092] Network Structure: Three-layer LSTM (hidden layer unit number 128) + fully connected layer (output dimension 24);

[0093] Input Data: Short-term feature set (24-dimensional) + historical price (lag 24 time points);

[0094] Training Mechanism: Use Seq2Seq architecture, encoder input 72-hour data, decoder output future 72-hour prediction;

[0095] Output: 24 time points of price curve (such as [0.45, 0.52,..., 0.38] yuan / kWh).

[0096] Long-Term Life Prediction (Prophet):

[0097] Model Configuration:

[0098] Growth Trend: Piecewise Linear (turning point automatically detected);

[0099] Seasonal Term: Fourier Series (order = 5, period = 30 days);

[0100] Input Data: Long-term feature set (24-dimensional) + historical capacity decay rate (such as 0.5% per month);

[0101] Output: 30-day decay trajectory (such as [100%, 99.7%,..., 98.2%]).

[0102] Dynamic Coupling Atlas Generation:

[0103] Time Axis Alignment: Map 72-hour price curve and 30-day life trajectory to a unified time axis (1-hour resolution);

[0104] Coupling relationship visualization:

[0105] X-axis: Time (0~720 hours); Left Y-axis: Electricity price (yuan / kWh); Right Y-axis: Energy storage capacity (%); Correlation line: Label the causal relationship between the electricity price peak and the capacity drop (e.g., when the electricity price is >0.5 yuan, the battery DoD deepens 5%).

[0106] Example atlas: The electricity price peak at hour 120 (5th day) is 0.58 yuan, corresponding to a battery capacity drop from 99.1% to 98.9%.

[0107] By integrating key parameters of each link of the energy system, including the operation cost of the power generation side, the transmission efficiency of the power grid, the user load demand, and the life degradation characteristics of the energy storage device, a spatio-temporal correlation model between electricity price fluctuations and energy storage life is established through time series analysis technology. This model can capture both short-term electricity price responses and long-term energy storage performance degradation dynamics, forming a coupling atlas that reflects the changes in the system's life cycle cost, and establishing a spatio-temporal framework for system-level cost analysis. The traditional isolated short-term scheduling and long-term planning are organically combined. The generation of dynamic coupling atlas not only reveals the interactive influence mechanism between electricity price and energy storage life, but also provides quantitative basis for subsequent multi-time scale optimization, avoiding decision bias caused by time dimension fragmentation.

[0108] S202, based on the dynamic coupling atlas, combining meteorological prediction data and historical load fluctuation characteristics, the influence factors of weather change and load fluctuation are quantified through dynamic uncertainty atlas generation technology, and uncertainty quantification parameters containing probability distribution are output;

[0109] Specifically, according to the long-term energy storage life trajectory in the dynamic coupling atlas, meteorological sensitive features can be extracted to generate meteorological influence coding vectors;

[0110] Meteorological sensitive feature extraction focuses on the correlation analysis between energy storage life and meteorological parameters. First, a multi-factor response model of energy storage battery life degradation is established, and three types of core meteorological indicators, temperature, humidity, and irradiance, are selected as input dimensions. For example:

[0111] Temperature sensitive feature: A piecewise linearization model is used, when the environmental temperature exceeds 25℃, the lithium battery capacity degradation rate increases exponentially with the temperature rise. The temperature-degradation coefficient mapping table is determined through historical data statistics (e.g., the degradation coefficient is 1.15 at 30℃ and 1.32 at 35℃).

[0112] Humidity sensitive feature: Based on the electrochemical corrosion model, when the humidity is >70%, the battery internal resistance growth rate increases to 1.8 times that in dry environment.

[0113] Irradiance sensitive feature: For photovoltaic energy storage, high irradiance (>800W / m2 ) leads to an increase in charge-discharge frequency, defined as a charge-discharge cycle acceleration factor (e.g. +0.2 for each 100 W / m 2 of irradiance increase).

[0114] The feature extraction process is as follows:

[0115] Data alignment: Align the time axis of meteorological station data (sampling frequency 5 minutes) and energy storage life monitoring data (daily capacity detection) through cubic spline interpolation.

[0116] Feature calculation:

[0117] Calculate the daily average temperature fluctuation range (DTTR), which is formalized as the difference between the daily maximum and minimum temperature;

[0118] Calculate the high humidity duration (HHD), defined as the sum of continuous periods with humidity > 70%;

[0119] Integrate the equivalent irradiance dose (EID) by performing trapezoidal numerical integration on the irradiance curve.

[0120] Feature encoding:

[0121] Temperature feature encoding as a 3-dimensional vector: [daily average temperature, DTTR, duration ratio of > 30°C];

[0122] Humidity feature encoding as a 2-dimensional vector: [daily average humidity, HHD];

[0123] Irradiance feature encoding as a 2-dimensional vector: [peak irradiance, EID].

[0124] Finally generate a 7-dimensional weather influence encoding vector, for example: [28.5°C, 12.3°C, 15%, 65%, 4.2h, 850W / m 2 , 5.6kWh / m 2 ].

[0125] Input the weather influence encoding vector into the dynamic Bayesian network, combine with historical load fluctuation features, model the weather-load joint probability distribution, and output the uncertainty propagation diagram;

[0126] The dynamic Bayesian network (DBN) adopts a three-layer spatio-temporal topology:

[0127] Input layer: weather influence encoding vector (7 nodes);

[0128] Hidden layers: 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);

[0129] Output layer: System cost fluctuation (3 nodes), including: generation cost coefficient of variation; network loss fluctuation amplitude; energy storage life decay acceleration factor.

[0130] Network construction process:

[0131] Conditional probability table (CPT) training: Use 5 years of historical data (total 43800 hours of records) for parameter learning; use the expectation maximization (EM) algorithm to iteratively optimize CPT parameters, set the convergence threshold Δ<0.001; for example: when the temperature is >30℃ and the humidity is >75%, the load surge probability is increased to 1.5 times the baseline value.

[0132] Spacetime correlation modeling: Industrial load is negatively correlated with temperature (correlation coefficient -0.7 when air conditioning load proportion increases); commercial load is positively correlated with irradiance (correlation coefficient +0.6 when shopping mall passenger flow increases on sunny days).

[0133] Joint probability distribution generation: Simulate 100,000 state transitions by Gibbs Sampling; output three-dimensional joint probability distribution: [P(cost fluctuation|weather, load), P(network loss|load characteristics), P(life decay|meteorology)].

[0134] The final uncertainty propagation diagram contains: node topology structure (15-node directed graph); conditional probability matrix (15x15 sparse matrix); joint distribution heat map (resolution 0.1 probability unit).

[0135] Based on the uncertainty propagation diagram, embed the Monte Carlo sampling of extreme weather scenarios in the dynamic coupling graph to generate a risk scenario set;

[0136] Monte Carlo sampling (MCS) is used to generate small probability, high risk extreme weather scenarios to evaluate their impact on the power system. The specific process is as follows:

[0137] ‌Define extreme weather event criteria‌:

[0138] ‌High temperature event‌: Continuous 3-day daily maximum temperature exceeding 95% historical quantile (e.g. 40°C).

[0139] ‌Low temperature event‌: Continuous 2-day daily minimum temperature below 5% historical quantile (e.g. -10°C).

[0140] ‌High wind event: instantaneous wind speed exceeds 20 m / s and lasts more than 6 hours.

[0141] ‌Latin Hypercube Sampling (LHS)

[0142] Generate 1000 sets of weather parameter samples (SAMPLES=1000) within the extreme event time window (e.g., days 15-17). Each set of samples contains temperature perturbation values (TEMPERATURE_SHIFT), humidity perturbation (HUMIDITY_SHIFT), and wind speed perturbation (WIND_SHIFT). For example, the high temperature perturbation for the 50th sample is +2°C (TEMPERATURE_SAMPLE_50 = TEMPERATURE_PREDICTED + 2°C).

[0143] The perturbation range is set according to historical extreme values. For example, the temperature perturbation range is [-3°C, +5°C], and the wind speed perturbation range is [-5 m / s, +10 m / s].

[0144] ‌Scenario embedding and correction:

[0145] According to the perturbed weather parameters, correct the energy storage life trajectory and load prediction in the dynamic coupling atlas: in the high temperature scenario, the energy storage capacity decay rate is increased to 1.5 times the baseline value (e.g., from 0.8% per month to 1.2%). The load demand is adjusted according to the DBN inference result. For example, for every 1°C increase in temperature, the air conditioning load increases by 5%.

[0146] Generate a set of risk scenarios (RISK_SCENARIOS), each containing:

[0147] Scenario ID (e.g., SCENARIO_100), weather parameters, energy storage decay correction coefficient, load correction coefficient.

[0148] According to the risk scenario set, use the kernel density estimation algorithm to quantify the joint influence of weather and load fluctuations on system cost, outputting a multi-dimensional probability density function;

[0149] Kernel density estimation (KDE) is used to fit the joint probability distribution of weather-load-cost, with the following specific steps:

[0150] ‌Cost simulation and sample collection:

[0151] Calculate the cost for the 1000 samples in the risk scenario set, and the total cost (TOTAL_COST) includes:

[0152] Source-side generation cost (GEN_COST): fuel cost of coal-fired units + start-stop cost of gas-fired units.

[0153] Grid-side transmission loss (LOSS_COST): economic cost corresponding to line resistance loss.

[0154] Storage-side life depreciation (STORAGE_COST): equipment replacement cost caused by capacity attenuation.

[0155] Load-side deviation penalty (LOAD_PENALTY): deviation penalty of actual load from predicted value.

[0156] For example, in a certain scenario, GEN_COST = 500,000 yuan, LOSS_COST = 50,000 yuan, STORAGE_COST = 100,000 yuan, and LOAD_PENALTY = 20,000 yuan. The total cost TOTAL_COST = 670,000 yuan.

[0157] ‌Multidimensional Kernel Density Estimation‌:

[0158] The Gaussian Kernel is used 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 h = 1.06 × σ × n^(-1 / 5), where σ is the sample standard deviation and n = 1000.

[0159] To reduce computational complexity, Hoeffding Projection Pursuit is used to decompose the four-dimensional space into two two-dimensional subspaces (e.g., GEN_COST-LOSS_COST and STORAGE_COST-LOAD_PENALTY), and the joint distribution is estimated and synthesized.

[0160] ‌Generating Multidimensional Probability Density Function‌:

[0161] The four-dimensional joint probability density function JPDF(gen_cost, loss_cost, storage_cost, load_penalty) is output, which can query the occurrence probability of any cost combination. For example, the probability of total cost exceeding 80,000 yuan is 5%, mainly contributed by high generation cost and high storage depreciation in high temperature scenarios.

[0162] Align the multidimensional probability density function with the spatiotemporal characteristics of the dynamic coupling graph to generate an uncertainty quantification parameter matrix with confidence intervals.

[0163] Temporal alignment ensures that the probability density function is consistent with the time resolution (hourly) and spatial partition (grid topology) of the dynamic coupling atlas. The implementation is as follows:

[0164] ‌Temporal alignment‌:

[0165] The monthly JPDF (30 days) is decomposed into hourly slices using the Dynamic Time Warping (DTW) algorithm. For example, 30 days are divided into 6 phases (5 days per phase), and the JPDF parameters (mean μ, standard deviation σ) of each phase are mapped to the corresponding weekly interval of the dynamic coupling atlas.

[0166] The hourly cost confidence interval is calculated by Bootstrap resampling. For example, the total cost 95% confidence interval for hour 500 (5 pm on a certain day) is [0.55 million, 0.65 million].

[0167] ‌Spatial alignment‌:

[0168] The grid is divided into multiple regions (e.g., ZONE_01~ZONE_10), each associated with nearby weather station data. The Inverse Distance Weighting (IDW) algorithm is used to calculate the regional weather influence weight. For example, the weight of region ZONE_03 is determined by the three closest weather stations (distances d1=5km, d2=8km, d3=10km), and the weight formula is w_i = 1 / (d_i^2).

[0169] The regionalized weather influence weight is combined with the JPDF to generate the regional cost probability distribution. For example, ZONE_05 has a wider load cost confidence interval ([0.5 million, 0.7 million]) due to its proximity to an industrial area.

[0170] ‌Generate uncertainty quantification parameter matrix‌:

[0171] The matrix fields include timestamp (TIMESTAMP), region ID (ZONE_ID), cost mean (COST_MEAN), standard deviation (COST_STD), and confidence interval (COST_CI). For example:

[0172] TIMESTAMP: 2024-07-01T14:00;

[0173] ZONE_ID: ZONE_07;

[0174] COST_MEAN: 0.58 million;

[0175] COST_STD: 0.06 million;

[0176] COST CI: [52 million, 64 million].

[0177] The matrix supports the dynamic adjustment strategy of the scheduling system. For example, when the confidence interval width of a certain area exceeds a threshold (such as 20 million), the standby power supply capacity needs to be increased to hedge risks.

[0178] On the basis of the dynamic coupling graph, meteorological forecasts and historical load data are introduced as external disturbance variables. The joint influence of extreme weather events and load random fluctuations on system cost is analyzed using a probabilistic statistical method. Uncertain factors are converted into quantifiable probability distribution parameters, a mathematical representation of risk transmission is established, and explicit modeling of uncertain factors is achieved, extending traditional deterministic optimization to the field of probabilistic optimization. By quantifying the risk contribution of weather and load fluctuations, the robustness of the scheduling strategy is ensured, enhancing the system's ability to respond to extreme events.

[0179] S203, according to the uncertainty quantification parameter, design a multi-time scale coupling scheduling framework, use a multi-scale coupling factor decomposition model to decompose the scheduling task into hour-level short-term optimization and month-level long-term optimization sub-problems, and output a collaborative scheduling framework containing space-time correlation constraints;

[0180] Specifically, according to the uncertainty quantification parameter matrix, an hour-level scheduling objective function can be constructed, and dynamic constraint conditions of generation cost, transmission loss, and load deviation are defined;

[0181] The hour-level scheduling objective function aims to minimize the total operating cost from the current time to 72 hours in the future while meeting the real-time operating constraints of the power grid. The objective function consists of three parts: source-side generation cost (GEN_COST), network-side transmission loss (LOSS_COST), and load-side load deviation penalty (LOAD_PENALTY).

[0182] Objective function construction:

[0183] ‌Generation cost (GEN_COST): The generation cost of source-side devices such as coal-fired units, gas units, and photovoltaic power stations is calculated based on output (POWER_GEN). For example, the cost of a coal-fired unit is a linear function:

[0184] GEN_COST = a * POWER_COAL + b, where a = 200 yuan / MWh (unit generation cost) and b = 5000 yuan (start-stop fixed cost).

[0185] ‌Transmission loss (LOSS_COST): The loss power (LOSS = I 2 ^2 * R_LINE) is calculated based on line resistance (R_LINE) and current square (I 2R_LINE), converted to economic cost. For example, a line R_LINE = 0.1Ω, current I = 100A, then the hourly loss cost is LOSS_COST = (100 2 0.1) / 1000 * electricity price (assuming electricity price is 0.5 yuan / kWh) = 5 yuan.

[0186] ‌Load deviation penalty (LOAD_PENALTY): When the deviation between actual load (LOAD_ACTUAL) and predicted value (LOAD_PREDICTED) exceeds the threshold (such as ±5%), a penalty is imposed according to the overrun ratio. For example, when the deviation is 10%, the penalty coefficient is 50 yuan / MWh.

[0187] ‌Dynamic constraint condition definition‌:

[0188] ‌Power generation constraint: The output of each unit needs to be within the technical output range. For example, the lower limit of the output of a coal-fired unit is 50MW, and the upper limit is 300MW, i.e. 50MW ≤ POWER_COAL ≤ 300MW.

[0189] ‌Energy storage charging and discharging constraint: The state of charge (SOC) of the energy storage system (such as lithium battery) needs to be maintained within the safe interval (such as 20% ≤ SOC ≤ 90%), and the charging and discharging power is limited by the equipment rating (such as the maximum charging and discharging power is 50MW).

[0190] ‌Load balancing constraint: The total amount of power generation plus the amount of energy storage discharge needs to equal the load demand plus transmission loss, i.e. SUM(POWER_GEN) + POWER_STORAGE_DISCHARGE = LOAD_DEMAND + LOSS.

[0191] ‌Parameter correlation and optimization weight‌:

[0192] Convert the confidence interval in the uncertainty quantification parameter matrix (for example, the cost fluctuation range [55 million yuan, 65 million yuan]) to the robustness weight of the objective function. For example, the power generation cost weight of the period with high uncertainty is increased by 20% to prefer stable output solutions.

[0193] Based on the long-term energy storage decay trajectory in the dynamic coupling graph, a monthly optimization objective function is established to define the energy storage life balance and return on investment (ROI) constraints;

[0194] The monthly optimization objective function focuses on the life decay balance and return on investment (ROI) of the energy storage system within 30 days, and needs to coordinate the charging and discharging strategy to avoid local excessive wear and tear.

[0195] ‌Objective function construction‌:

[0196] ‌LIFE_BALANCE: Balance the capacity decay rate of each energy storage unit by minimizing the difference. For example, define the difference index as: LIFE_BALANCE = MAX(CAPACITY_DECAY_RATE_i) - MIN(CAPACITY_DECAY_RATE_i), where i is the unit number.

[0197] ‌ROI: Calculate the ratio of net income to initial investment of the energy storage system. For example, initial investment COST_INVEST = 10 million yuan, monthly income includes arbitrage profit (PROFIT_ARBITRAGE = 2 million yuan) and ancillary service profit (PROFIT_ANC = 0.5 million yuan), then ROI = (200 + 50) / 1000 = 25%.

[0198] ‌Constraint definition‌:

[0199] ‌Cycle constraint: The monthly charge-discharge cycle number of the energy storage unit cannot exceed the rated value. For example, the rated cycle number of lithium battery is 5000 times, and the monthly upper limit is 200 times.

[0200] ‌DOD limit: The single charge-discharge depth (such as DOD = 80%) needs to meet the manufacturer's requirements, and the monthly average DOD does not exceed 60%.

[0201] ‌Lower limit of ROI: Requires ROI ≥ 15%, otherwise it is considered an economically unfeasible solution.

[0202] ‌Long-term energy storage decay trajectory fusion‌:

[0203] Extract the energy storage capacity decay prediction curve (for example, capacity decay 0.8% per month) from the dynamic coupling graph, and discretize it into 30-day daily decay amount (such as DAY_DECAY = 0.0267%), as a reference benchmark for life balance optimization.

[0204] Associate the decay rate with the charge-discharge strategy. For example, if the number of charge-discharge times increases by 10% on a certain day, the decay rate for that day is corrected to 0.0293%.

[0205] Use a two-level decomposition algorithm to decompose the hourly and monthly targets into independent sub-problems, and realize cross-time scale parameter interaction through coupling factor transfer matrix;

[0206] The two-level decomposition algorithm divides the complex optimization problem into hourly (lower level) and monthly (upper level) sub-problems, and realizes two-way interaction through coupling factor transfer matrix (COUPLING_MATRIX).

[0207] ‌Sub-problem decomposition‌:

[0208] ‌Hourly Subproblem: Solve for hourly generation schedule, storage charge / discharge power, and load allocation within a 72-hour optimization window, aiming to minimize total operational cost.

[0209] ‌Monthly Subproblem: Optimize storage charge / discharge frequency distribution and investment return rate over a 30-day cycle, aiming to minimize lifetime imbalance and maximize ROI.

[0210] ‌Coupling Factor Transfer Matrix Design:

[0211] Rows represent time scales (1-72 hours for hourly, 1-30 days for monthly), and columns represent coupling parameter types (e.g., storage SOC, charge / discharge count).

[0212] ‌Example Matrix Structure:

[0213] COUPLING_MATRIX = [

[0214] [DAY1_HOUR1_SOC, DAY1_HOUR1_DISCHARGE_COUNT,...],

[0215] [DAY1_HOUR2_SOC, DAY1_HOUR2_DISCHARGE_COUNT,...], ...

[0216] [DAY30_HOUR24_SOC, DAY30_HOUR24_DISCHARGE_COUNT,...]

[0217] ]。

[0218] ‌Parameter Transfer Rules:

[0219] Monthly layer transfers monthly balancing targets for storage SOC to the hourly layer (e.g., daily SOC fluctuation not exceeding ±10%).

[0220] Hourly layer feeds actual charge / discharge counts to the monthly layer for updating the lifetime degradation model.

[0221] ‌Iterative Optimization Process:

[0222] ‌Initial Iteration: Monthly layer sets initial charge / discharge plan, and the hourly layer executes the plan and returns actual SOC and cost.

[0223] ‌Parameter Correction: If the actual hourly SOC deviates from the monthly target (e.g., deviation >5%), the monthly layer adjusts the charge / discharge depth limits for subsequent dates.

[0224] Convergence condition: When the rate of change of the objective function is less than 1% for 3 consecutive iterations, it is determined to be converged.

[0225] According to the coupling factor transmission result, a space-time correlation constraint propagation mechanism is introduced to ensure the consistency of short-term scheduling and long-term strategy in key parameters such as energy storage charge and discharge depth, and output a collaborative scheduling framework.

[0226] The space-time correlation constraint propagation mechanism solves the strategy conflict across time scales through rule-based reasoning and dynamic priority adjustment.

[0227] ‌Constraint conflict detection‌:

[0228] ‌Charge and discharge depth conflict‌: For example, the hourly optimization requires a certain energy storage unit to discharge 90% in a single discharge to meet peak load requirements, but the monthly strategy requires the average DOD to be ≤ 60%.

[0229] ‌Life balance conflict‌: A certain energy storage unit has a monthly attenuation rate that exceeds the average value by 20% due to frequent calls, violating the life balance goal.

[0230] ‌Constraint propagation and priority rules‌:

[0231] ‌Hard constraint priority‌: Safety-related constraints (such as SOC ≥ 20%) have unconditional priority over economic constraints.

[0232] ‌Dynamic weight adjustment‌: Apply a penalty term to conflicting parameters (such as DOD). For example, if the DOD of the hourly plan exceeds the limit, add a penalty term PENALTY = 1000 * (DOD - 60%) to the objective function, forcing the algorithm to search for a new feasible solution.

[0233] ‌Rule-based reasoning example‌:

[0234] If the daily predicted load peak exceeds the historical extreme value by 10%, temporarily allow the DOD to increase to 70%, but need to be compensated to 50% in the subsequent day.

[0235] ‌Collaborative framework output‌:

[0236] The final generated scheduling framework includes hourly and monthly optimization models, coupling factor transmission interfaces, and constraint rule libraries.

[0237] ‌Framework running process‌:

[0238] The monthly level generates a charge and discharge plan draft and transmits it to the hourly level;

[0239] The hourly level performs real-time optimization according to the draft and returns the actual parameters;

[0240] The constraint propagation mechanism detects and corrects conflicts;

[0241] Repeat the iteration until all constraints are met.

[0242] ‌Key technology examples and parameter explanations‌:

[0243] ‌GEN_COST‌: Source-side generation cost, including fuel cost, operation and maintenance cost, and start-up and shutdown loss.

[0244] ‌LOSS_COST‌: Economic cost of network-side transmission loss, calculated by multiplying line loss power by real-time electricity price.

[0245] ‌SOC‌ (State of Charge): State of charge of energy storage system, indicating the percentage of current remaining capacity to total capacity.

[0246] ‌DOD‌ (Depth of Discharge): Depth of charge and discharge, indicating the proportion of single discharge capacity to total capacity.

[0247] ‌COUPLING_MATRIX‌: Coupling factor transfer matrix, used for parameter interaction across time scales, matrix elements include SOC, charge and discharge times, etc. Key parameters.

[0248] Based on the uncertainty analysis results, a hierarchical optimization architecture is constructed. Through coupling factor decomposition technology, the interaction parameters between hourly scheduling and monthly planning are identified, a constraint propagation mechanism across time scales is established to ensure the coordination of short-term operation and long-term strategy in key variables such as energy storage charge and discharge depth, the framework solves the dimension disaster problem of multi-time scale optimization, reduces the computational complexity through intelligent decomposition. The introduction of space-time correlation constraints ensures the time consistency of the scheduling scheme, avoiding the overdraft of long-term resources by short-term optimization.

[0249] S204, based on the collaborative scheduling framework, a hybrid algorithm is used for multi-objective optimization solution, where the co-evolution mechanism of genetic algorithm and reinforcement learning is used to process short-term optimization problems, and fuzzy logic decision tree is introduced to optimize long-term strategy, outputting the global optimal scheduling strategy.

[0250] Specifically, according to the hourly target function in the collaborative scheduling framework, the genetic algorithm is used to initialize the population, and the real-time electricity price, load demand and energy storage SOC are coded as chromosome genes to generate an initial short-term solution set;

[0251] The initialization of the genetic algorithm (GA) population is the basis for searching feasible solutions, which requires encoding the hourly scheduling parameters into chromosome structures. Chromosome genes correspond to scheduling decision variables, including real-time price (RTP), load demand (LD), and state of charge (SOC) of energy storage.

[0252] ‌Chromosome Encoding Design‌:

[0253] ‌Gene Structure‌: Each chromosome represents a scheduling plan for the next 72 hours, containing 216 genes (3 parameters x 72 hours). For example, the first hour gene group is [RTP_1, LD_1, SOC_1], and the 72nd hour is [RTP_72, LD_72, SOC_72].

[0254] ‌Encoding Format‌: Real-Coding is used. For example, RTP ranges from 0.1 yuan / kWh to 1.0 yuan / kWh, LD ranges from 100 MW to 300 MW, and SOC ranges from 20% to 90%.

[0255] ‌Population Size‌: The initial population contains 100 individuals (chromosomes) randomly generated. For example, individual 1 has RTP_1 = 0.5 yuan / kWh, LD_1 = 150 MW, and SOC_1 = 60%.

[0256] ‌Fitness Function Definition‌:

[0257] The fitness value reflects the quality of the solution, and the calculation formula is the inverse of the total cost (Fitness = 1 / TOTAL_COST), aiming to minimize the cost.

[0258] The total cost (TOTAL_COST) includes generation cost (GEN_COST), transmission loss (LOSS_COST), and load deviation penalty (LOAD_PENALTY). For example, if the total cost of a certain individual is 1.2 million yuan, the fitness value is 1 / 120 = 0.0083.

[0259] ‌Initial Solution Set Generation and Selection‌:

[0260] After randomly generating 100 individuals, calculate the fitness value of each individual, eliminate individuals with fitness values lower than the average (e.g., average value 0.005), and retain the top 50 high-quality individuals.

[0261] An elitism strategy is applied to the remaining individuals, which are directly copied to the next generation, ensuring that excellent genes are not lost.

[0262] Based on the initial short-term solution set, a deep reinforcement learning agent is embedded to dynamically adjust the crossover and mutation probabilities, and a Q-learning algorithm is used to optimize the convergence speed of the solution set, outputting the short-term candidate solutions on the Pareto front.

[0263] 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, accelerating the convergence to the Pareto front.

[0264] State-action space definition:

[0265] State (State): Current population diversity indicators include fitness variance (FV) and gene similarity (GS). For example, FV=0.002 indicates that the population has small fitness differences, and the mutation probability needs to be increased.

[0266] Action (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].

[0267] Reward (Reward): Based on the population evolution speed, if the optimal fitness of the new generation improves by 10%, the reward is +10; if it degrades, it is punished by -5.

[0268] Q-learning algorithm flow:

[0269] Q-table initialization: Construct a state-action value table, with initial Q values of 0.

[0270] Exploration and utilization: Use the ε-greedy strategy, with a probability of ε=0.3 for random exploration of actions in the first 50 generations, and gradually reduce it to ε=0.1 thereafter.

[0271] Q-value update: Update the Q table based on the reward. For example, if the fitness improves after performing the action "CR+0.1" in a certain state, update the Q value as 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.

[0272] ‌Pareto front candidate solution generation‌:

[0273] After 100 iterations, non-dominated solutions are selected from the final population, i.e., no other solution is better than this solution in all objectives.

[0274] For example, solution A has a total cost of 1 million yuan and a life balance index of 5%; solution B has a cost of 105 million yuan and a balance of 3%. If solution A is better than solution B in cost, and solution B is better than solution A in balance, both belong to the Pareto front candidate solution.

[0275] Input the short-term candidate solution of the Pareto front into the fuzzy logic decision tree, combine the uncertainty parameters in the monthly optimization target, and generate a long-term strategy fuzzy rule base;

[0276] Fuzzy Logic Decision Tree (FLDT) maps short-term candidate solutions to long-term strategy rules, and needs to define fuzzy sets and inference rules.

[0277] ‌Fuzzy set definition‌:

[0278] ‌Input variables‌:

[0279] Load demand (LD): fuzzified as "low (Low)", "medium (Medium)", "high (High)", with intervals [100MW, 150MW], [150MW, 250MW], [250MW, 300MW].

[0280] Real-time electricity price (RTP): fuzzified as "low price (Low)", "medium price (Mid)", "high price (High)", with intervals [0.1 yuan, 0.4 yuan], [0.4 yuan, 0.7 yuan], [0.7 yuan, 1.0 yuan].

[0281] Energy storage SOC: fuzzified as "low energy storage (Low)", "medium energy storage (Mid)", "high energy storage (High)", corresponding to [20%, 50%], [50%, 80%], [80%, 90%].

[0282] ‌Output variables: long-term charging and discharging strategy, including "aggressive charging (Aggressive Charge)", "conservative charging and discharging (Conservative)", "aggressive discharging (Aggressive Discharge)".

[0283] ‌Fuzzy rule base construction‌:

[0284] Rules are generated based on expert experience and historical data. For example:

[0285] Rule 1: If LD=High and RTP=Low, then Strategy=Aggressive Charge (charge at low price period to cope with high load demand).

[0286] Rule 2: If SOC=High and RTP=High, then Strategy=Aggressive Discharge (discharge for profit at high price period).

[0287] The rule base contains 27 rules (3 input variables x 3 fuzzy levels^3).

[0288] Uncertainty parameter fusion:

[0289] Combine the uncertainty quantification parameters in monthly optimization (such as weather impact probability) to correct the rule confidence. 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.

[0290] Based on the long-term strategy fuzzy rule base, the long-term strategy is iteratively optimized by dynamic programming algorithm, and the monthly charging and discharging plan is output, which is decoupled from the Pareto front short-term candidate solution;

[0291] Dynamic programming (DP) algorithm decomposes the 30-day scheduling problem into daily decision stages to minimize long-term cost and meet life balance constraints.

[0292] State variables and decision variables:

[0293] State variables (State):

[0294] Energy storage SOC (range 20%~90%), discretized into 10% intervals (such as 20%, 30%, …, 90%).

[0295] Energy storage decay rate (DR), discretized into three levels of 0.5%, 0.8%, and 1.0% per month.

[0296] Decision variables (Action): Daily charging and discharging strategy, including charging power (CP) and discharging power (DP).

[0297] State transition equation:

[0298] SOC update: SOC_{t+1} = SOC_t + (CP - DP) / CAPACITY, where CAPACITY is the total capacity of the energy storage (such as 100 MWh).

[0299] Decay rate update: If the number of daily charge-discharge cycles exceeds 2, increase the DR by 0.1%.

[0300] Value function and iterative optimization:

[0301] Value Function: V(s) = Minimum total cost (monthly GEN_COST + STORAGE_COST) + Life balance penalty (LIFE_BALANCE_PENALTY).

[0302] 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.

[0303] Backward iteration: Calculate the optimal strategy from the 30th day to the 1st day. For example, the 30th day SOC needs to reach 50% to meet the monthly balance target.

[0304] Input the Pareto frontier short-term candidate solution and monthly charge-discharge plan into the non-dominated sorting genetic algorithm, and calculate the comprehensive cost and risk score by combining the entropy weight method to screen the global optimal scheduling strategy.

[0305] Non-Dominated Sorting Genetic Algorithm (NSGA-II) is used to integrate short-term and long-term solution sets, and the Entropy Weight Method is used to quantify multi-objective weights to select the optimal solution.

[0306] Non-dominated sorting and crowding distance calculation:

[0307] Non-dominated sorting: The solution set is layered according to the dominance relationship. The first layer is the Pareto frontier solution, the second layer is the solution dominated by the first layer, and so on.

[0308] Crowding distance: Calculate the density of the solution in the target space, and prefer to keep the solution in the sparse area to maintain diversity. For example, the distance between the adjacent solutions of a solution on the cost axis is 100,000 yuan, and the distance on the balance axis is 2%, then the crowding distance = 10 + 2 = 12.

[0309] Entropy weight method weight distribution:

[0310] Data standardization: Normalize the cost (unit: ten thousand yuan) and risk score (unit: %) to [0, 1].

[0311] Information entropy calculation:

[0312] Cost Entropy E_COST = -Σ(p_i * ln p_i), where p_i is the cost proportion of the i-th solution.

[0313] Similarly, calculate Risk Entropy E_RISK.

[0314] ‌Weight Distribution: Weight W_COST = (1 - E_COST) / [(1 - E_COST) + (1 - E_RISK)]. For example, if E_COST=0.2, E_RISK=0.5, then W_COST=0.8 / 1.3≈61.5%.

[0315] ‌Global Optimal Solution Screening:

[0316] Calculate the comprehensive score of each solution: SCORE = W_COST * normalized cost + W_RISK * normalized risk.

[0317] Select the solution with the highest comprehensive score. For example, if the normalized cost of solution X is 0.3 and the risk is 0.2, then SCORE=0.6150.3 + 0.3850.2=0.254, and if it is the highest value, it is selected.

[0318] ‌Key Technology Examples and Parameter Descriptions:

[0319] ‌SOC (State of Charge): The remaining capacity of the energy storage system as a percentage of the total capacity, which is the core state variable of the charging and discharging strategy.

[0320] ‌NSGA-II: Non-dominated Sorting Genetic Algorithm, used to handle multi-objective optimization problems, ensuring the diversity and convergence of the solution set through hierarchical sorting and congestion calculation.

[0321] ‌Entropy Weight Method: An objective weight distribution method based on information entropy, avoiding the influence of subjective preference on decision-making.

[0322] For different time scale optimization problems, combine the advantages of intelligent algorithms: genetic algorithm handles discrete decision variables, reinforcement learning dynamically adjusts search direction to accelerate short-term optimization convergence; fuzzy logic handles language rules in long-term strategies, converting expert experience into calculable decision trees, hybrid algorithms fully utilize the complementarity of each technology, ensuring solution accuracy while improving computational efficiency. Global strategy realizes the multi-objective balance of economy, reliability and equipment life, providing optimal decision support for source-grid-load-storage collaborative operation.

[0323] It can be seen that, according to the source-side power generation cost, the network-side transmission loss, the load-side load demand and the storage-side life attenuation data of the energy system, a dynamic coupling graph containing short-term and long-term cost evolution paths is output; based on the dynamic coupling graph, uncertainty quantization parameters containing probability distribution are output; according to the uncertainty quantization parameters, a collaborative scheduling framework containing space-time correlation constraints is output; based on the collaborative scheduling framework, a multi-objective optimization solution is solved by using a hybrid algorithm, and a globally optimal scheduling strategy is output, so that a globally optimal decision can be made to balance system economy, safety and equipment life.

[0324] Another embodiment of the application provides a source-network-load-storage collaborative scheduling system based on multi-time scale cost optimization, referring to Figure 3 , the system can include:

[0325] The analysis module 301 is configured to perform space-time correlation analysis on electricity price fluctuations and energy storage cycle life by using a time series prediction algorithm according to the source-side power generation cost, the network-side transmission loss, the load-side load demand and the storage-side life attenuation data of the energy system, and output a dynamic coupling graph containing short-term and long-term cost evolution paths.

[0326] The generation module 302 is configured to generate an impact factor of weather changes and load fluctuations by using a dynamic uncertainty graph generation technology based on the dynamic coupling graph, in combination with meteorological prediction data and historical load fluctuation characteristics, and output uncertainty quantization parameters containing probability distribution.

[0327] The decomposition module 303 is configured to design a multi-time scale coupled scheduling framework according to the uncertainty quantization parameters, and decompose the scheduling task into hour-level short-term optimization and month-level long-term optimization sub-problems by using a multi-scale coupling factor decomposition model, and output a collaborative scheduling framework containing space-time correlation constraints.

[0328] The output module 304 is configured to solve a multi-objective optimization by using a hybrid algorithm based on the collaborative scheduling framework, wherein a short-term optimization problem is processed by using a synergistic evolution mechanism of a genetic algorithm and reinforcement learning, and a fuzzy logic decision tree is introduced to optimize a long-term strategy, and a globally optimal scheduling strategy is output.

[0329] It can be seen that, according to the source-side power generation cost, the network-side transmission loss, the load-side load demand and the storage-side life attenuation data of the energy system, a dynamic coupling graph containing short-term and long-term cost evolution paths is output; based on the dynamic coupling graph, uncertainty quantization parameters containing probability distribution are output; according to the uncertainty quantization parameters, a collaborative scheduling framework containing space-time correlation constraints is output; based on the collaborative scheduling framework, a multi-objective optimization solution is solved by using a hybrid algorithm, and a globally optimal scheduling strategy is output, so that a globally optimal decision can be made to balance system economy, safety and equipment life.

[0330] The embodiment of the present application also provides a storage medium, wherein the storage medium stores a computer program, and the computer program is arranged to execute the steps in any one of the method embodiments.

[0331] Specifically, in the embodiment, the storage medium can be arranged to store a computer program for executing the following steps:

[0332] S201, according to source-side power generation cost, network-side transmission loss, load-side load demand and storage-side life attenuation data of an energy system, time series prediction algorithm is adopted to perform space-time correlation analysis on electricity price fluctuation and energy storage cycle life, and a dynamic coupling graph containing short-term and long-term cost evolution paths is outputted;

[0333] S202, based on the dynamic coupling graph, meteorological prediction data and historical load fluctuation characteristics are combined, and a dynamic uncertainty graph generation technology is used to quantize influence factors of weather change and load fluctuation, and an uncertainty quantization parameter containing a probability distribution is outputted;

[0334] S203, according to the uncertainty quantization parameter, a multi-time scale coupled scheduling framework is designed, a multi-scale coupling factor decomposition model is adopted to decompose the scheduling task into hour-level short-term optimization and month-level long-term optimization sub-problems, and a collaborative scheduling framework containing space-time correlation constraints is outputted;

[0335] S204, based on the collaborative scheduling framework, a hybrid algorithm is used for multi-objective optimization solution, wherein a genetic algorithm and a reinforcement learning co-evolution mechanism are used to process the short-term optimization problem, and a fuzzy logic decision tree is introduced to optimize the long-term strategy, and a globally optimal scheduling strategy is outputted.

[0336] It can be seen that according to source-side power generation cost, network-side transmission loss, load-side load demand and storage-side life attenuation data of an energy system, a dynamic coupling graph containing short-term and long-term cost evolution paths is outputted; based on the dynamic coupling graph, an uncertainty quantization parameter containing a probability distribution is outputted; according to the uncertainty quantization parameter, a collaborative scheduling framework containing space-time correlation constraints is outputted; based on the collaborative scheduling framework, a hybrid algorithm is used for multi-objective optimization solution, and a globally optimal scheduling strategy is outputted, so that global optimal decision can be realized to balance system economy, safety and equipment life.

[0337] The embodiment of the present application also provides an electronic device, comprising a memory and a processor, the memory stores a computer program, and the processor is arranged to execute the computer program to execute the steps in any one of the method embodiments.

[0338] Specifically, the electronic device can further comprise a transmission device and an input and output device, wherein the transmission device is connected with the processor, and the input and output device is connected with the processor.

[0339] Specifically, in the present embodiment, the above processor can be configured to execute the following steps by means of a computer program:

[0340] S201, according to the source side power generation cost, the network side transmission loss, the load side load demand and the storage side life attenuation data of the energy system, time series prediction algorithm is adopted to carry out space-time correlation analysis on electricity price fluctuation and energy storage cycle life, and dynamic coupling atlas containing short-term and long-term cost evolution path is outputted;

[0341] S202, based on the dynamic coupling atlas, combining meteorological prediction data and historical load fluctuation characteristics, the influence factors of weather change and load fluctuation are quantified by dynamic uncertainty atlas generation technology, and uncertainty quantization parameters containing probability distribution are outputted;

[0342] S203, according to the uncertainty quantization parameters, a multi-time scale coupled scheduling framework is designed, and a multi-scale coupling factor decomposition model is used to decompose the scheduling task into hour-level short-term optimization and month-level long-term optimization sub-problems, and a collaborative scheduling framework containing space-time correlation constraints is outputted;

[0343] S204, based on the collaborative scheduling framework, a hybrid algorithm is used for multi-objective optimization solution, wherein the short-term optimization problem is processed by the synergistic evolution mechanism of genetic algorithm and reinforcement learning, and fuzzy logic decision tree is introduced to optimize long-term strategy, and global optimal scheduling strategy is outputted.

[0344] It can be seen that according to the source side power generation cost, the network side transmission loss, the load side load demand and the storage side life attenuation data of the energy system, the dynamic coupling atlas containing short-term and long-term cost evolution path is outputted; based on the dynamic coupling atlas, the uncertainty quantization parameters containing probability distribution are outputted; according to the uncertainty quantization parameters, the collaborative scheduling framework containing space-time correlation constraints is outputted; based on the collaborative scheduling framework, the hybrid algorithm is used for multi-objective optimization solution, and the global optimal scheduling strategy is outputted, so as to realize the global optimal decision to balance the system economy, safety and equipment life.

[0345] The above describes the structure, features and effects of the present application in detail according to the embodiments shown in the drawings. The above description is only the preferred embodiments of the present application, but the present application is not limited to the embodiments shown in the drawings. Any changes or modifications made in accordance with the concept of the present application, or equivalent embodiments with equivalent changes, shall be within the scope of protection of the present application.

Claims

1. A source-grid-load-storage collaborative scheduling method based on multi-time-scale cost optimization, characterized in that, The method includes: Based on the energy system's source-side power generation cost, grid-side transmission loss, load-side load demand, and energy storage lifespan decay data, a time series prediction algorithm is used to conduct a spatiotemporal correlation analysis of electricity price fluctuations and energy storage cycle lifespan, outputting a dynamic coupling map that includes 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 influence 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-timescale coupled scheduling framework is designed. A multi-scale coupling factor decomposition model is used to decompose the scheduling task into hourly short-term optimization and monthly long-term optimization sub-problems, outputting a collaborative scheduling framework with spatiotemporal constraints. Specifically, based on the uncertainty quantification parameter matrix, an hourly scheduling objective function is constructed, defining dynamic constraints for generation cost, transmission loss, and load deviation. Based on the long-term energy storage decay trajectory in the dynamic coupling graph, a monthly optimization objective function is established, defining constraints for energy storage lifetime equilibrium and return on investment. A two-level decomposition algorithm is used to decompose the hourly and monthly objectives into independent sub-problems, achieving cross-timescale parameter interaction through a coupling factor transfer matrix. Based on the coupling factor transfer results, a spatiotemporal constraint propagation mechanism is introduced to ensure consistency between short-term and long-term strategies on key parameters, outputting a collaborative scheduling framework. These key parameters include charge / discharge depth. Based on the aforementioned collaborative scheduling framework, a hybrid algorithm is used to solve multi-objective optimization problems. Specifically, a co-evolutionary mechanism of genetic algorithm and reinforcement learning is used to handle short-term optimization problems, and a fuzzy logic decision tree is introduced to optimize long-term strategies, outputting the globally optimal scheduling strategy.

2. The method according to claim 1, characterized in that, Based on data on energy system source-side power generation costs, grid-side transmission losses, load-side demand, and energy storage lifespan degradation, a time series forecasting algorithm is used to perform spatiotemporal correlation analysis on electricity price fluctuations and energy storage cycle lifespan, outputting a dynamic coupling map containing short-term and long-term cost evolution paths, including: Based on the time series data of power generation cost on the source side and the lifetime decay curve of the storage side, multi-scale wavelet transform is used to extract the fluctuation characteristics of power generation cost and generate a spatiotemporal correlation feature vector. By fusing spatiotemporal correlation feature vectors with topological distribution data of network-side transmission loss, and modeling the source-network-load-storage coupling relationship through a spatiotemporal attention mechanism, a four-dimensional dynamic correlation tensor is output. Based on the four-dimensional dynamic correlation tensor, the tensor decomposition algorithm is used to separate the short-term electricity price sensitivity factor and the long-term energy storage lifetime influencing factor, and generate a dual-time-scale prediction feature set containing short-term prediction feature set and 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 for 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 decay trajectory. The two sets are then 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 quantifies the influencing factors of weather changes and load fluctuations, and outputs uncertainty quantification parameters containing probability distributions, including: Based on the long-term energy storage lifetime 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 a dynamic Bayesian network, and combined with historical load fluctuation characteristics, the weather-load joint probability distribution is modeled, and the uncertainty propagation map is output. Based on the uncertainty propagation graph, extreme weather scenarios sampled in Monte Carlo are embedded in the dynamic coupling graph 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. Align the multidimensional probability density function with the spatiotemporal features of the dynamic coupling spectrum to generate an uncertainty quantization parameter matrix with confidence intervals.

4. The method according to claim 3, characterized in that, Based on the aforementioned collaborative scheduling framework, a hybrid algorithm is used for multi-objective optimization. Specifically, a co-evolutionary mechanism combining genetic algorithms and reinforcement learning is employed to handle short-term optimization problems, while fuzzy logic decision trees are introduced to optimize long-term strategies, outputting a globally optimal scheduling strategy. This includes: Based on the hourly objective function in the collaborative scheduling framework, a genetic algorithm is used to initialize the population, and real-time electricity price, load demand and energy storage SOC are encoded into chromosome genes to generate an 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 probabilities. The Q-learning algorithm is used to optimize the convergence speed of the solution set and output Pareto front 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, a long-term strategy fuzzy rule base is generated. Based on a long-term strategy fuzzy rule base, a dynamic programming algorithm is used to iteratively optimize the long-term strategy and output a monthly charge-discharge plan that is decoupled from the short-term candidate Pareto front. The Pareto front short-term candidate solutions and monthly charge-discharge plans are input into a non-dominated sorting genetic algorithm. The entropy weight method is used to calculate the comprehensive cost and risk score, and the globally optimal scheduling strategy is obtained by screening.

5. A source-grid-load-storage coordinated scheduling system based on multi-time-scale cost optimization, characterized in that, The system includes: The analysis module is used to perform spatiotemporal correlation analysis on 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 energy storage life decay data, using time series prediction algorithms to output a dynamic coupling map containing short-term and long-term cost evolution paths. The generation module is used to quantify the influencing factors of weather changes and load fluctuations based on the dynamic coupling spectrum, combined with meteorological forecast data and historical load fluctuation characteristics, through dynamic uncertainty spectrum generation technology, and output uncertainty quantification parameters containing probability distribution. The decomposition module is used to design a multi-timescale coupled scheduling framework based on the uncertainty quantification parameters. It employs a multi-scale coupling factor decomposition model to decompose the scheduling task into hourly short-term optimization and monthly long-term optimization sub-problems, outputting a collaborative scheduling framework that includes spatiotemporal constraints. Specifically, based on the uncertainty quantification parameter matrix, an hourly scheduling objective function is constructed, defining dynamic constraints for generation cost, transmission loss, and load deviation. Based on the long-term energy storage decay trajectory in the dynamic coupling graph, a monthly optimization objective function is established, defining constraints for energy storage lifetime equilibrium and return on investment. A two-layer decomposition algorithm is used to decompose the hourly and monthly objectives into independent sub-problems, achieving cross-timescale parameter interaction through a coupling factor transfer matrix. Based on the coupling factor transfer results, a spatiotemporal constraint propagation mechanism is introduced to ensure consistency between short-term and long-term strategies on key parameters, outputting a collaborative scheduling framework. These key parameters include charge / discharge depth. The output module is used to solve multi-objective optimization problems based on the cooperative scheduling framework using a hybrid algorithm. In this module, a co-evolutionary mechanism of genetic algorithm and reinforcement learning is used to handle short-term optimization problems, and a fuzzy logic decision tree is introduced to optimize long-term strategies, outputting the globally optimal scheduling strategy.

6. The system according to claim 5, characterized in that, The analysis module is specifically used for: Based on the time series data of power generation cost on the source side and the lifetime decay curve of the storage side, multi-scale wavelet transform is used to extract the fluctuation characteristics of power generation cost and generate a spatiotemporal correlation feature vector. By fusing spatiotemporal correlation feature vectors with topological distribution data of network-side transmission loss, and modeling the source-network-load-storage coupling relationship through a spatiotemporal attention mechanism, a four-dimensional dynamic correlation tensor is output. Based on the four-dimensional dynamic correlation tensor, the tensor decomposition algorithm is used to separate the short-term electricity price sensitivity factor and the long-term energy storage lifetime influencing factor, and generate a dual-time-scale prediction feature set containing short-term prediction feature set and 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 for 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 decay trajectory. The two sets are then fused to generate a dynamic coupling map.

7. The system according to claim 6, characterized in that, The generation module is specifically used for: Based on the long-term energy storage lifetime 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 a dynamic Bayesian network, and combined with historical load fluctuation characteristics, the weather-load joint probability distribution is modeled, and the uncertainty propagation map is output. Based on the uncertainty propagation graph, extreme weather scenarios sampled in Monte Carlo are embedded in the dynamic coupling graph 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. Align the multidimensional probability density function with the spatiotemporal features of the dynamic coupling spectrum to generate an uncertainty quantization parameter matrix with confidence intervals.

8. A storage medium, characterized in that, The storage medium stores a computer program, wherein the computer program is configured to execute the method of any one of claims 1-4 when it is run.

9. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method of any one of claims 1-4.

Images