Multi-stage stochastic low-carbon power system planning method considering climate uncertainty

CN122779474APending Publication Date: 2026-09-18SICHUAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610918772.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-06-24
Publication Date
2026-09-18

AI Technical Summary

Technical Problem

[0008]针对上述问题,本发明的目的在于提供一种考虑气候不确定性的多阶段随机低碳电力系统规划方法,通过将气候演变轨迹嵌入规划视野,解决现有规划框架忽视气候适应且难以桥接气候模型与电力模型的问题,达到了提升电力系统在气候变化下的适应性与成本效益的效果

Benefits of technology

[0015] 1) This invention proposes a novel multi-stage stochastic programming framework that explicitly incorporates the long-term evolution of renewable energy resource allocation driven by climate change. This framework supports phased optimization to adapt to changing climate trajectories. By embedding climate evolution into the investment decision-making process, the model captures the co-evolution between climate change and power system transformation, enabling adaptive programming that integrates adaptation and mitigation strategies under long-term climate uncertainty.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122779474A_ABST
    Figure CN122779474A_ABST
Patent Text Reader

Abstract

The application relates to the technical field of low-carbon transformation of power systems, and discloses a multi-stage stochastic low-carbon power system planning method considering climate uncertainty. Based on shared socio-economic pathways, representative concentration pathway scenarios and multi-global climate model outputs, a representative scenario tree is constructed through stage-by-stage clustering to depict double uncertainties of scenarios and climate responses; coarse resolution climate data is converted into renewable resource availability factors through climate-consistent time downscaling; a multi-stage stochastic planning model is established with the minimum net present value of investment and operation costs as the target, power generation and energy storage capacity expansion and operation scheduling optimization are carried out according to the climate realization of scenario nodes at each stage; and SDDP is directly used for solving on the scenario tree. Through stage-by-stage adaptive decision-making, the application can reduce the total cost of long-term planning under the guarantee of a low-carbon target and improve the robustness of a planning scheme to future climate change.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of low-carbon transformation technology for power systems, specifically a multi-stage stochastic low-carbon power system planning method that takes into account climate uncertainty. Background Technology

[0002] The low-carbon transition of the power system is a key strategy for mitigating climate change. Existing power system planning studies mainly focus on a medium- to long-term perspective, aiming to develop low-carbon or 100% renewable energy systems as part of a strategic decarbonization pathway consistent with global climate goals. However, most existing studies rely on static or scenario-specific climate assumptions, primarily focusing on climate mitigation strategies aimed at curbing global warming, while failing to recognize that climate change is a dynamic, exogenous, and deeply uncertain risk driver for long-term power system development.

[0003] Existing research on low-carbon planning for power systems that considers uncertainty generally employs two-stage or finite-stage stochastic programming models, decomposing the problem into early-stage capacity investment decisions and later-stage operation and scheduling optimization to handle short-term fluctuations in renewable energy output and load demand. These approaches simplify the sources of uncertainty, primarily focusing on hourly or daily-scale random disturbances, while rarely considering climate change as a long-term, dynamic exogenous driver. Some studies have introduced scenario tree methods to characterize uncertainty, but these are mostly geared towards short-term operational scenarios (such as islanded operation and biomass availability) and do not perform hierarchical modeling of climate evolution trajectories. Regarding solution algorithms, Stochastic Dual Dynamic Programming (SDDP) has some application foundation, but it relies on traditional sampling methods to generate scenarios and approximates the value function through random sampling during backtracking, resulting in slow convergence and high computational cost.

[0004] Existing technological solutions typically output static or deterministic capacity expansion schemes adjusted only according to pre-defined scenarios, lacking the ability to adaptively adapt to the gradual unfolding of climate change. Most existing research relies on static or scenario-specific climate assumptions and remains primarily focused on carbon reduction strategies aimed at curbing global warming, failing to recognize that climate change is a dynamic, exogenous, and highly uncertain long-term risk driver for power system development. Therefore, existing approaches are inadequate in addressing climate adaptation, with limited attention paid to planning frameworks that integrate mitigation and adaptation goals. This gap weakens the power system's ability to cope with evolving climate-related uncertainties. Specifically, three main limitations can be identified:

[0005] (i) Inadequate Inclusion of Climate Evolution in Planning: As power systems become increasingly reliant on weather-dependent renewable energy sources, climate change will become a major external driver reshaping their long-term development paths. Under climate change conditions, photovoltaic resources exhibit significant spatiotemporal heterogeneity. These changes alter the spatial and temporal distribution of renewable energy generation, complicating efforts to maintain long-term supply-demand balance. Research on how climate change impacts power system planning remains scarce. Climate change has not been effectively incorporated into optimization frameworks that proactively adapt to these changing conditions during the planning phase. This limitation may increase investment risk and potentially undermine the feasibility of long-term decarbonization pathways.

[0006] (ii) The Challenges of Bridging Climate Science and Electrical Engineering: Climate modeling and power system planning differ significantly in their objectives, temporal and spatial scales, and data requirements. Climate models typically predict large-scale trends over centuries at coarse resolutions (e.g., monthly or annual time steps), focusing on macroscopic Earth system dynamics. In contrast, power system planning requires high-resolution generation and load data, often at sub-hourly time scales, and spatial granularity consistent with grid topology to support investment decisions spanning decades. This fundamental mismatch in spatiotemporal resolution and data structure poses a significant challenge to integrating climate model outputs into power system models. Consequently, outputs from Global Climate Models (GCMs), such as those included in the Coupled Model Intercomparison Project Phase 6 (CMIP6), often fail to meet the detailed spatiotemporal information requirements of power system planning, becoming a key bottleneck for effective climate-aware power system planning.

[0007] (iii) Inadequate Characterization of Climate Uncertainty: As mentioned earlier, most existing power system planning models not only fail to adequately account for the impacts of climate change but also largely ignore the associated uncertainties. In particular, they rarely consider two key dimensions of uncertainty: (a) scenario uncertainty, stemming from different socioeconomic development and emission pathways, typically represented by the Shared Social Economic Path and Representative Concentration Path (SSP-RCP scenario); and (b) climate response uncertainty, reflecting the differences in the sensitivity of climate models under the same radiative forcing conditions. Consequently, the variability in the predicted availability of renewable energy resources is often underestimated or ignored. This deficiency hinders the effective integration of climate uncertainty into power system planning, ultimately leading to suboptimal investment strategies and increased exposure to unforeseen operational and financial risks. Summary of the Invention

[0008] To address the aforementioned problems, the present invention aims to provide a multi-stage stochastic low-carbon power system planning method that considers climate uncertainty. By embedding climate evolution trajectories into the planning framework, it solves the problems of existing planning frameworks neglecting climate adaptation and struggling to bridge climate and power models, thereby improving the adaptability and cost-effectiveness of the power system under climate change. The technical solution is as follows:

[0009] A multi-stage stochastic low-carbon power system planning method considering climate uncertainty includes the following steps:

[0010] Step 1: Based on shared socio-economic paths and representative concentration path scenarios, as well as the outputs of multiple global climate models, a representative scenario tree is constructed through stage-by-stage clustering to generate climate evolution paths that characterize scenario uncertainty and climate response uncertainty.

[0011] Step 2: Using a climate-consistent time downscaling method, the coarse-resolution climate prediction data of each stage in the representative scenario tree are converted into hourly-resolution time series data to generate renewable resource availability factors for each stage.

[0012] Step 3: With the goal of minimizing the sum of net present value of investment costs and operating costs during the planning period, a multi-stage stochastic low-carbon power system planning model is established based on the representative scenario tree and the renewable resource availability factors at each stage. At each stage, capacity expansion decisions for power generation assets and energy storage assets and operation scheduling optimization are made according to the climate of the corresponding scenario nodes.

[0013] Step 4: A customized stochastic dual dynamic programming algorithm is directly applied to the representative scenario tree to perform forward policy evaluation and backward value function approximation, solving the multi-stage stochastic low-carbon power system planning model to generate the optimal investment and operation decision strategies.

[0014] The beneficial effects of this invention are:

[0015] 1) This invention proposes a novel multi-stage stochastic programming framework that explicitly incorporates the long-term evolution of renewable energy resource allocation driven by climate change. This framework supports phased optimization to adapt to changing climate trajectories. By embedding climate evolution into the investment decision-making process, the model captures the co-evolution between climate change and power system transformation, enabling adaptive programming that integrates adaptation and mitigation strategies under long-term climate uncertainty.

[0016] 2) This invention proposes a climate uncertainty modeling method considering dual uncertainties. Scenario uncertainty is captured by selecting a specific future emission trajectory from multiple alternative scenarios, while climate response uncertainty is processed using the outputs of multiple GCMs under a selected SSP-RCP scenario. To effectively characterize the changes and evolution of key climate variables caused by climate response uncertainty, a staged clustering method is applied to construct a representative scenario tree. Furthermore, a climate-consistent timescale downscaling technique is developed to transform long-term climate forecasts into hourly resolution data, ensuring that climate-driven inputs are spatiotemporally consistent with the power system planning model.

[0017] 3) To address the computational challenges posed by high-dimensional, multi-stage stochastic models derived from two-layer climate uncertainty representations, this invention develops a customized stochastic dual dynamic programming (SDDP) algorithm. Unlike traditional sampling-based SDDP methods, the proposed method operates directly on a stage-clustered climate scenario tree, leveraging the tree's compact structure to accelerate convergence. By utilizing the decomposability of the scenario tree and the finite representative nodes of each stage, the algorithm effectively performs deterministic cut generation through a combination of forward policy evaluation and backward value function approximation to approximate the optimal decision policy. Attached Figure Description

[0018] Figure 1 A schematic diagram of a planning framework for climate-adaptive multi-stage stochastic low-carbon power systems.

[0019] Figure 2 Build a flowchart for the climate scenario tree.

[0020] Figure 3 A detailed diagram illustrating the construction of the climate scene tree.

[0021] Figure 4Diagram of the improved IEEE 30-node system.

[0022] Figure 5 This represents the phased planning results for each asset.

[0023] Figure 6 This refers to a phased power generation structure and energy share.

[0024] Figure 7 These are system indicators under different climate uncertainty scenarios.

[0025] Figure 8(a) shows the planning schemes under different climate evolution trajectories, comparing the average capacity factor under LTCT (low total cost trajectory) and HTCT (high total cost trajectory).

[0026] Figure 8(b) shows the planning schemes under different climate evolution trajectories, comparing the installed capacity expansion schemes of LTCT (low total cost trajectory) and HTCT (high total cost trajectory).

[0027] Figure 9 The planning results are presented under scenarios with / without climate uncertainty.

[0028] Figure 10 This represents the total cost under different climate uncertainty scenarios. Detailed Implementation

[0029] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0030] This invention first proposes a systematic method for modeling climate uncertainty, capturing scenario uncertainty and climate response uncertainty by constructing a representative scenario tree. Scenario uncertainty is captured by selecting a specific future emission trajectory from multiple alternative scenarios, while climate response uncertainty is addressed by utilizing the outputs of multiple global climate models under the selected scenario. A stage-wise clustering method is employed to construct the representative scenario tree to effectively characterize the variability and evolution of key climate variables. Secondly, to reconcile the temporal mismatch between coarse-resolution climate forecasts and fine-grained power system planning requirements, a climate-consistent temporal downscaling method is proposed, transforming long-term climate forecasts into hourly-resolution data to ensure that climate-driven inputs are spatiotemporally consistent with the power system planning model. Subsequently, a multi-stage stochastic low-carbon planning model is established, incorporating climate-related uncertainties into the system planning decision-making process, and adapting to the constantly changing climate trajectory through stage-wise optimization. Finally, to address the computational challenges of high-dimensional multi-stage stochastic models arising from the two-layer climate uncertainty representation, a custom stochastic dual dynamic programming algorithm was developed. This algorithm runs directly on the climate scene tree of stage-wise clustering, leveraging the compact structure of the tree to accelerate convergence. Through the integration of forward policy evaluation and backward value function approximation, it efficiently performs deterministic cutting plane generation to approximate the optimal decision policy.

[0031] Explanation of technical effects: This invention, by explicitly modeling climate uncertainty, achieves a 3.8% reduction in total system cost compared to deterministic programming models that do not consider climate uncertainty, and avoids annual load shedding of up to 2992.5 MWh, enhancing the power system's ability to manage climate-induced risks while maintaining cost-effectiveness across a wide range of future climate scenarios. Compared to two-stage stochastic programming models, the multi-stage programming framework of this invention further reduces total system cost and maximum load shedding, validating the value of time flexibility in addressing long-term climate uncertainty. The algorithm has a small and stable convergence iteration count, and the computation time increases approximately linearly with system size, exhibiting good scalability.

[0032] Reference Figures 1 to 2 This invention provides a multi-stage stochastic low-carbon power system planning method that considers climate uncertainty, and the specific implementation is as follows:

[0033] Step 1: Based on shared socioeconomic paths and representative concentration paths, as well as the outputs of multiple global climate models, a representative scenario tree is constructed through stage-by-stage clustering to generate climate evolution paths that characterize scenario uncertainty and climate response uncertainty. In Step 1, a set of climate evolution trajectories is first obtained. It is assumed that N climate evolution trajectories are available, each spanning T planning stages. At each stage t, the climate state is characterized by M climate variables over R spatial regions. The entire dataset is organized as a four-dimensional tensor. At each stage t, the N trajectories are reshaped from matrices into four-dimensional vectors to form a sample matrix. The K-means clustering algorithm is used to cluster this sample matrix to identify K clusters and their centroids. These centroids are used as representative nodes of the scenario tree at that stage. For each trajectory, its assigned cluster label at stage t is recorded, and the conditional transition probability between representative nodes of adjacent stages is estimated using empirical frequency counting. Starting from the root node, all feasible paths are expanded according to probability to construct a complete scenario tree, with each path corresponding to a compressed but typical climate evolution pattern.

[0034] Step 2 employs a climate-consistent time downscaling method to convert long-term monthly climate forecast data into hourly-resolution time series data to generate a high-resolution renewable resource availability factor. In Step 2, since the time granularity of the global climate model output is relatively coarse (usually monthly), historical hourly data of climate variables are first acquired, and the corresponding monthly average is calculated. The predicted monthly average associated with the scene tree nodes is then acquired, and the monthly scaling factor is calculated as the ratio of the predicted monthly average to the historical monthly average. This monthly scaling factor is applied to the historical hourly data to generate an adjusted hourly climate time series. This process preserves the original intra-month hourly variability structure while aligning the long-term monthly average trend of historical data with the predicted climate trajectory, ultimately obtaining an hourly resource characteristic curve that is spatiotemporally consistent with the power system planning model.

[0035] Step 3: Based on representative scenario trees and high-resolution renewable resource availability factors, a multi-stage stochastic low-carbon power system planning model is established. In Step 3, the planning problem is formulated as a multi-stage stochastic programming problem with recourse. Climate-driven uncertainties are represented by the availability factors of wind power and photovoltaic power at each stage. Decision variables are divided into investment decisions and operational decisions. The objective function is to minimize the net present value of investment and operating costs, where investment costs are the present value of the annualized capital costs of new assets, and operating costs are calculated through hourly simulations on typical days, including fixed maintenance costs, variable costs, thermal power unit start-up costs, carbon capture costs, and load shedding penalties. Constraints include stage-level investment constraints, renewable energy resource constraints, operational constraints of coal-fired and gas-fired power plants and their carbon capture units, energy storage system constraints, network structure constraints, and system performance constraints. System performance constraints include capacity adequacy, spinning reserve, carbon emission limits, renewable energy utilization rate, and system inertia constraints. The model is expressed as a Markov decision process, and a nested optimization form is obtained through Bellman equation expansion.

[0036] Step 4 employs a customized stochastic dual dynamic programming algorithm to perform forward policy evaluation and backward value function approximation on the representative scenario tree, solving the multi-stage stochastic low-carbon power system planning model to generate the optimal investment and operation decision strategy. In Step 4, during the backward propagation, the stage-level linear programming subproblem is solved for the scenario node of the current stage to obtain the optimal value and dual multipliers. The Benders cut plane is constructed using the dual multipliers to approximate the future cost function, and the new cut plane is added to the value function approximation set of the node. The lower bound of the original problem is obtained by solving the linear problem at the root stage. During the forward propagation, the stage-level optimization problem is solved sequentially along the complete path of the scenario tree, calculating the total path cost. The weighted average of the total path costs is used as the upper bound. The forward and backward propagation processes are executed alternately until the difference between the upper and lower bounds is less than a preset convergence threshold, at which point the optimal decision strategy is output.

[0037] Part 1: Multistage Stochastic Programming Problems under Climate Uncertainty

[0038] This paper first proposes a multi-stage stochastic programming model aimed at achieving fine-grained optimization of power generation portfolios across multiple planning stages, while explicitly considering climate-related uncertainties. Compared to traditional target-year or multi-stage planning methods, which either ignore long-term climate uncertainties or assume fully predictable future climate conditions, this model formulates the planning problem as a recourse-based multi-stage stochastic programming problem. By progressively revealing uncertainties over time, the model enables decisions to evolve adaptively to address both observed outcomes and probabilistic predictions of future climate conditions.

[0039] This model focuses on uncertain parameters driven by climate, denoted as . They indicate the stage The random input. Specifically Capturing the actual availability factors of wind turbines (WTs) and photovoltaics (PVs). These parameters reflect the impact of long-term climate change on the availability of renewable energy resources. Through the phased changes in renewable energy availability factors, climate uncertainty is translated into differences in resource endowments at different planning stages, thus directly affecting investment economics and operational feasibility. In particular, Changes in climate will affect the expected energy output and marginal value of renewable energy investments, as well as operational constraints such as renewable energy generation caps, reserve capacity adequacy, and emissions reduction feasibility relative to a changing baseline. Therefore, climate-driven uncertainty alters the viable technology space and the relative effectiveness of different mitigation options.

[0040] In each stage The decision variables are divided into investment and operation components. Investment decisions are denoted as... Operational decisions are recorded as These decisions are made randomly. The function, for the sake of simplicity, is not explicitly written out, i.e. and , representing the system's capacity expansion and operational response to climate uncertainty, respectively.

[0041] A. Phased Objective Function

[0042] In each stage The objective is to minimize the net present value of investment and operating costs while satisfying asset-level and system-level constraints, as shown in equation (1a). Vector and Indicates cost coefficient. Investment cost. Indicates the stage The present value of the annualized cost of capital for newly installed assets at the outset, as defined in formula (1b). Capital recovery factor. Used to convert capital investment into equivalent annualized value. Operating costs The calculations were performed on a typical day at the hourly level, including fixed operation and maintenance costs (1d), variable costs (1e), start-up costs of thermal power units (1f), carbon capture costs (1g), and penalties for load reduction (1h).

[0043] The objective function of the multi-stage stochastic low-carbon power system planning model is expressed by the following formula:

[0044] (1a);

[0045] (1b);

[0046] (1c);

[0047] (1d);

[0048] (1e);

[0049] (1f);

[0050] (1g);

[0051] (1h);

[0052] in, For the planning phase index, ; This is the collection of all planning phases. For the stage The vector of investment decision variables (new installed capacity of each asset); For the stage The vector of operational decision variables (power generation, reserves, etc.); These are cost coefficient vectors for investment and operational decision variables, respectively; The interest rate (discount rate); The number of years included in each planning phase; For the stage Net present value of investment costs; For the stage Net present value of operating costs; Index of years within a period; For asset indexing (such as WT, PV, CU, ES, etc.); Separate generator sets and energy storage systems; Indexing of power grid nodes (buses); For assets The capital recovery coefficient; For the stage Domestic assets At the node The investment cost reduction coefficient; For assets At the node The unit investment cost; For the stage Assets at the end At the node The cumulative installed capacity; For the previous stage ( The cumulative installed capacity of ) To cover fixed operation and maintenance costs; For assets At the node The unit fixed operation and maintenance cost; Variable operating costs; Typical daily index; For typical days The actual number of days in a year; Index of time (hour) for each typical day; For assets At the node Variable operating cost per unit; For assets In the stage , typical day ,time and nodes contribution; For energy storage systems The charging power; Costs for unit start-up and shutdown; For assets At the node The cost per startup; For assets At any moment Online (already powered on) capacity; For carbon capture costs; A collection of units equipped with carbon capture systems; The unit cost of carbon capture; For the actual carbon capture by the carbon capture system quantity; Penalty cost for loss of load; For nodes The unit loss-of-load penalty cost; For nodes At any moment The power of the load shedding.

[0053] B. Phased Investment Constraints

[0054] Given that this invention primarily aims to capture the long-term evolution dynamics of the power generation portfolio, investment decision variables for various assets are modeled as continuous variables to maintain computational feasibility. The relevant investment constraints are given in equations (2a) and (2b), where node-level and system-level capacity limits are specified by constraints (2a) and (2b), respectively. Note that asset capacity under stage 0 represents the existing installed capacity at the start of the planning period.

[0055] The phased investment constraints are as follows:

[0056] (2a);

[0057] (2b);

[0058] in, For the stage Domestic assets At the node Maximum allowed new capacity (investment limit per node); For assets throughout the system The maximum total allowed installed capacity.

[0059] C. Staged Operation Constraints of Different Units

[0060] 1) Renewable energy resources

[0061] The output of renewable energy units, such as wind turbines (WT) and photovoltaics (PV), is primarily constrained by the availability of their natural resources.

[0062] The constraints on renewable energy generation are shown in the following formula:

[0063] (3);

[0064] in, Contributing to renewable energy (WT / PV); This is the renewable energy resource availability factor (output coefficient, driven by climate data). This refers to the cumulative installed capacity of renewable energy.

[0065] 2) Coal-fired units

[0066] The operating constraints for coal-fired power units (CUs) are as follows: Constraint (4a) ensures that the power generation output plus spinning reserve does not exceed the committed capacity. Constraint (4b) limits the spinning reserve to the maximum ramp rate of the unit. Constraint (4c) constrains the ramp rate of output variation. Constraint (4d) limits the committed capacity to the rated installed capacity. Constraint (4e) defines the initial committed capacity.

[0067] The operating constraints for coal-fired power units are shown in the following formula:

[0068] (4a);

[0069] (4b);

[0070] (4c);

[0071] (4d);

[0072] (4e);

[0073] in, This refers to the minimum technical output ratio of a coal-fired power unit (relative to its online capacity). For coal-fired power units at all times Online capacity; The actual output of the coal-fired power unit; Spinning reserve capacity provided for coal-fired power units; and These are the maximum uphill and downhill gradients for coal-fired power units, respectively. This refers to the cumulative installed capacity of coal-fired power units. For the initial stage ( =0) Initial online capacity ratio of coal-fired units.

[0074] 3) Gas turbine units

[0075] Compared to coal-fired units, gas-fired units (GUs) have rapid start-up and shutdown capabilities and high ramp rates. Therefore, their modeling can be simplified by ignoring the constraints of committed capacity and ramp rate.

[0076] The operating constraints of the gas turbine unit are as follows:

[0077] (5);

[0078] 4) Generator sets equipped with carbon capture systems

[0079] For coal-fired and gas-fired power units (CUCs and GUCs) equipped with carbon capture systems, in addition to basic operating constraints, the following must be considered: Specifically, constraint (6a) defines the amount of carbon emissions generated during unit operation. Constraint (6b) limits the maximum amount of carbon dioxide that the carbon capture system can capture. Constraint (6c) describes the energy consumption of the carbon capture system during the capture process. Net carbon emissions are calculated via (6d).

[0080] The carbon capture operation and energy consumption constraints are shown in the following equation:

[0081] (6a);

[0082] (6b);

[0083] (6c);

[0084] (6d);

[0085] in, For assets Total carbon emissions generated; For assets carbon emission intensity (tons) / MWh); For assets contribution; The amount of carbon emissions sent to CCS (Carbon Capture and Storage) for processing; The carbon capture efficiency (captivity rate) of CCS. The electrical power consumed by CCS operation; Energy consumption ratio of CCS (MWh / ton) ); This refers to the unit's final net carbon emissions (emissions-capture).

[0086] 5) Energy storage system

[0087] Pumped hydro storage units (PSs) and energy storage systems (ESs) are modeled using a unified approach. Specifically, constraint (7a) limits the charge and discharge power within the rated range, while constraint (7b) limits the energy storage to no more than the rated energy capacity. Constraint (7c) describes the coupling relationship between energy storage and charge / discharge power. Constraint (7d) enforces the consistency of energy storage at the beginning and end of the cycle. Constraint (7e) limits the spinning reserve capacity provided by the energy storage system.

[0088] The operating constraints of the energy storage system are shown in the following formula:

[0089] (7a);

[0090] (7b);

[0091] (7c);

[0092] (7d);

[0093] (7e);

[0094] in, The charging power for energy storage; The discharge power of the stored energy; The rated continuous discharge time (in hours) for energy storage; and These are the charging efficiency and discharging efficiency of energy storage, respectively. Spinning reserve capacity provided for energy storage.

[0095] D. Phase constraints of network structure

[0096] A linearized DC power flow model is used to describe the power exchange between network nodes. Specifically, constraint (8a) establishes the linear power flow relationship between regions, while constraints (8b) and (8c) impose operating limits on the power flow of transmission lines and the node voltage angle, respectively. Node power balance requirements are implemented through (8d).

[0097] The network structure stage constraints are shown in the following equation:

[0098] (8a);

[0099] (8b);

[0100] (8c);

[0101] (8d);

[0102] in, branch road The active power flowing upstream; and branch road The voltage phase angle between the start and end nodes; branch road The reactance; branch road The transmission capacity limit; For nodes The voltage phase angle; To provide power to the generator set; This refers to the energy storage discharge / charge power. Branch power; For node load requirements; This refers to the power loss at the node. Power consumed by CCS.

[0103] E. System performance phase constraints

[0104] 1) Capacity adequacy constraint

[0105] Capacity adequacy constraints require that, in each time period, the total available generating capacity meets or exceeds the total load demand and reserve capacity requirements, thereby ensuring the security and reliability of the power supply.

[0106] (9a);

[0107] in, For assets At the node Available capacity factor (considering forced outages); For assets The cumulative installed capacity; For nodes The required proportion of standby capacity; For the stage Middle node Peak load.

[0108] 2) Rotational spare constraint

[0109] To ensure sufficient reserve capacity in response to sudden generator failures or load fluctuations, a spinning reserve constraint is introduced to ensure that the total rapid response reserve capacity must meet or exceed the specified reserve requirements.

[0110] (9b);

[0111] in, For assets Provided spinning reserve capacity; The proportion of spinning reserve required by the load; For node load requirements; The proportion of spinning reserve demand for renewable energy; To make a real contribution to renewable energy.

[0112] 3) Carbon emission restrictions

[0113] Carbon emission limits restrict the total carbon emissions of various conventional units when carbon capture is taken into account, ensuring that the power system develops along a prescribed carbon reduction path.

[0114] (9c);

[0115] in, For typical days The number of days; Carbon emission intensity; Net carbon emissions; For the stage Maximum permissible total carbon emissions.

[0116] 4) Renewable energy utilization rate constraints

[0117] Renewable energy utilization constraints apply to the ratio of renewable energy generation to available renewable energy resources. Imposing a lower limit will promote the priority use and efficient integration of renewable energy.

[0118] (9d);

[0119] in, Minimum utilization rate of renewable energy; Renewable energy availability factor; Installed capacity for renewable energy.

[0120] 5) System inertia constraints

[0121] System inertia constraints address the challenges to frequency stability posed by high-penetration renewable energy sources by specifying the minimum level of inertia that the system must meet under different operating conditions.

[0122] (9e);

[0123] in, For assets The provided inertial constant; For the installed capacity of the asset (gas turbine unit / gas turbine unit C / ES);

[0124] For the online capacity of CU / CUC; This represents the maximum allowable inertial descent tolerance of the system. This is the reference value for the current system's inertial constant; This is for node load requirements.

[0125] F. Multi-stage stochastic low-carbon power system planning model

[0126] Finally, the multi-stage stochastic programming problem under climate uncertainty is formulated as a Markovian Decision Process (MDP), characterized by states... ,action Transition probability and stage costs describe.

[0127] Specifically, in the stage The system state is defined as follows: ,in This indicates the system configuration inherited from the previous stage. Representation phase Climate uncertainty is realized.

[0128] In a given state Under the conditions, stage The decision (action) is defined as Actions are subject to the following:

[0129] (10);

[0130] (11);

[0131] Among them, constraints It captures the cross-temporal coupling of investment decisions across consecutive stages, while the remaining investment and operational constraints are compactly represented as Vectors and matrices , , , , and This represents the corresponding constraint parameters.

[0132] The state transition from stage ω to stage ω+1 consists of both deterministic and stochastic components. Specifically, given a state-action pair... The state of the next stage is defined as follows: ,in, Determined by the current action, It follows stochastic climate processes. Therefore, the state transition probability can be decomposed into a deterministic part related to the evolution of the investment-driven system and a stochastic part describing the dynamics of climate uncertainty, namely:

[0133] (12);

[0134] award With cost function This indicates that the selected actions reflect the investment and operating costs. The optimal solution of the original MDP-based multi-stage model satisfies the Bellman equation:

[0135] (13);

[0136] in Indicates from stage The minimum expected cumulative cost to the terminal stage. By expanding the Bellman recursion, the MDP can adopt an equivalent nested optimization form during the planning period. As shown in Equation (14), the decision is made in a forward recursive manner, that is, the investment and operation decisions at each stage depend on the realization of current uncertainties and the expected costs of future stages.

[0137] (14);

[0138] in, For the stage The vector of investment decision variables (such as the new / cumulative installed capacity of various assets); For the stage The vector of operational decision variables (such as power generation, reserve capacity, charging / discharging power, etc.); For the stage The set of feasible decisions, which depends on the system configuration left over from the previous stage. And the current stage of climate uncertainty ; The existing system installed capacity (a known constant) is used in the initial planning phase (phase 0). For the stage Climate uncertainties are realized (such as wind speed and the availability factor of solar radiation). For the stage Climate uncertainty The mathematical expectation operator.

[0139] Part Two: Climate Scenario Tree Construction and Multi-Stage Planning Model Reconstruction

[0140] To implement a multi-stage stochastic programming model (14), it is essential to construct a structured and computationally tractable representation of long-term climate uncertainty that can be seamlessly integrated into the optimization framework. Climate predictions from multiple global climate models (GCMs) inherently capture the uncertainty of climate evolution, manifesting as diverse and numerous possible trajectories. However, the sheer number of these uncertain paths leads to excessive computational complexity, making direct integration into the programming model impractical. Furthermore, the nested expectation structure of the models further complicates the feasibility of understanding them. To overcome these challenges, this invention proposes a staged clustering method for constructing a scenario tree that compresses the set of GCM trajectories into a manageable set of representative paths. This scenario tree preserves basic statistical and temporal properties, facilitating a recursive, node-based reformulation of the programming problem.

[0141] A. Constructing a scenario tree for climate uncertainty through phased clustering.

[0142] To effectively characterize the uncertainty of long-term climate variable evolution in a computationally tractable manner, this invention proposes a scenario tree generation framework based on staged clustering. This method transforms a large set of climate evolution trajectories simulated by various Global Climate Change Models (GCMs) under a given SSP scenario into a smaller but statistically representative scenario tree, suitable for multi-stage stochastic optimization models. The entire process is as follows: Figure 3 As shown.

[0143] Specifically, assuming it is available A series of climate evolution trajectories, each covering Each planning phase. In each phase... Climate status from spatial regions Each climate variable (e.g., monthly average wind speed and solar radiation intensity) describes the dataset. Therefore, the entire dataset can be organized as a four-dimensional tensor. At each stage ,Will The implementation of the trajectory from Matrix reshaping 3D vectors form a sample matrix Subsequently, K-means clustering was applied to the matrix to identify... Each cluster has a centroid. It is used as the representative node of stage ω in the scene tree.

[0144] For each trajectory In the stage The assigned cluster label is denoted as The conditional transition probability between representative nodes in adjacent stages is estimated using empirical frequency counting.

[0145] (15);

[0146] in, For the planning phase index, ; Number of phases in the overall planning; To be from the stage Representative node Transition to Stage Representative node The conditional probability; For the first A climate trajectory in stages Cluster tags; For counting symbols; For the stage The number of clusters (representing nodes) is determined by parameters preset during K-means clustering.

[0147] This generates a series of stage transition matrices that collectively capture the transtemporal evolution and conditional dependencies representing the climate state. Starting from the root node (stage 1), a complete scenario tree is constructed by probabilistically expanding all feasible paths, with each path corresponding to a compressed but typical climate evolution pattern. Let the first... The scenario path is The probability of its occurrence is equal to the product of the transition probabilities along the path. Each node Representing a path Previous stage This enables the comprehensive realization of climate, which can then be integrated into multi-stage investment decision-making models.

[0148] B. Climate-consistent hourly resolution time downsampling

[0149] Because the temporal granularity of GCM outputs is relatively coarse (typically monthly), a statistically consistent temporal downsampling method is further proposed to generate hourly renewable energy profiles while preserving historical variability and future climate trends. It is important to note that spatial downsampling is generally unnecessary for transmission-level planning, as the original resolution (≈25 km) of modern GCM datasets is sufficient. The process includes:

[0150] Step 1: Set For the region Climate variables Hourly historical data. The historical monthly average for the corresponding month o is defined as:

[0151] (16);

[0152] in, For a set of regions; A set of climate variables; For months Total number of hours.

[0153] Step 2: Set Represents the relationship between scene tree nodes The predicted monthly averages of the relevant climate variables. The monthly scaling factor is defined as:

[0154] (17);

[0155] Step 3: Apply the monthly scaling factor to generate the adjusted hourly climate time series, as follows:

[0156] (18);

[0157] This adjustment method aligns the long-term monthly average trend of historical data with the predicted climate trajectory while preserving the original intra-month hourly variation structure. Therefore, for each scenario path... This allows us to obtain a set of climate-consistent and temporally detailed hourly resource curves. Subsequently, these adjusted curves can be processed by a renewable energy simulation model to generate the final phased wind and solar resource realization. These results will serve as inputs to the multi-stage decision-making process.

[0158] Please note that the proposed method focuses on preserving climate-driven trends at monthly and quarterly scales, which is most relevant to long-term capacity expansion planning, while retaining realistic intra-diurnal variability for operational representation and flexibility assessment by anchoring future scenarios to historical hourly data. Under future climate conditions, variations in intra-diurnal characteristics and potential changes in transvariable physical couplings are difficult to robustly describe and primarily affect short-term operational behavior and the utilization of flexibility resources. Their impact on overall energy availability and optimal capacity expansion outcomes is relatively limited when assessed over multi-year planning spans. Nevertheless, more advanced temporal downsampling techniques remain valuable for improving data accuracy and may further enhance the precision and applicability of long-term planning results.

[0159] C. Reconstruction of Multi-Stage Planning Problems Based on Scenario Trees

[0160] Given that long-term uncertainty is modeled through a scenario tree, with each node corresponding to a unique implementation path, problem (14) can be equivalently restated in a node-based manner, as defined by equations (19a)-(19c). Let and Representing nodes respectively The parent and child nodes. Cost arrival function. Aggregated in implementation The immediate cost of the node and the expected future costs of all its child nodes. In the final stage Only the specific costs of this stage are retained, since there are no further transfers.

[0161] (19a);

[0162] (19b)

[0163] (19c);

[0164] in, The cost-to-reach function (the expected total cost from this node to the end point); To move to child nodes The probability of.

[0165] Part Three: Solution Theory

[0166] The multi-stage planning problem (19) is built on a climate-driven scenario tree, where each stage corresponds to the realization of uncertainty, followed by a decision. As the number of stages and the uncertainty space increase, the scenario tree grows exponentially, leading to a significant increase in decision nodes and constraints. This exponential growth makes direct enumeration or fully analytical methods computationally infeasible.

[0167] To address this challenge, a custom stochastic dual dynamic programming (SDDP) algorithm is employed. Unlike traditional sampling-based implementations, the proposed method operates directly on a staged clustering climate scenario tree composed of a finite number of representative scenarios, preserving the diversity of climate uncertainty while significantly reducing the computational burden. In this algorithm, the expected future cost function... It is approximated by a piecewise linear representation constructed from Benders cutting planes, which are iteratively generated from the dual solution during the backtracking phase. By decomposing the multi-stage problem into staged subproblems and refining the cost-to-approximation at these representative scenario nodes, the custom SDDP algorithm leverages controlled scenario scale to achieve accelerated convergence and high scalability for large-scale, climate-driven multi-stage stochastic programming problems.

[0168] 1) Backtracking Phase (Value Function Approximation)

[0169] Consideration stage Scenario nodes Its characteristics are caused by uncertainty The implementation and the system state inherited from the previous stage The determined value function can be approximated by solving the following linear programming problem:

[0170] (20a);

[0171] in, For nodes Investment decisions; Auxiliary variables used to replicate the state from the previous stage; For nodes Operational decisions; For nodes Climate realization; As an approximation variable for future costs; The state value passed in from the previous stage; For dual multipliers of equality constraints; Investment coupling constraints; For phase-specific constraints; This is a constant term that approximates future costs; The gradient (dual multiplier) approximates the future cost. Investment decisions at reference points;

[0172] (20b);

[0173] (20c);

[0174] Solving the above linear programming problem, that is, the phased optimal programming subproblem under the given climate uncertainty and the system configuration inherited from the previous stage, yields the optimal value. and the corresponding dual multipliers These are used to construct new Benders cuts, thereby refining the value function. The piecewise linear approximation. The newly generated cut will be added to the node. The associated value functions are approximated in the set and will be invoked during the subsequent forward pass.

[0175] After the uplink process is completed, through the root phase =1 can solve the following linear problem to obtain an effective lower bound (LB) for the original multi-stage stochastic problem:

[0176] (twenty one);

[0177] Note 1: For a given planning phase The scale of phased sub-problems is primarily determined by the dimensions of investment and operational decisions. It represents a collection of power generation and energy storage assets. This represents a set of representative scenarios under given climate uncertainty. This represents the set of runtime steps in each representative scenario. The number of investment decision variables grows in an O(A) order, while the number of operation and maintenance decision variables grows in an O(A) order. The number of constraints increases exponentially. Therefore, the total number of constraints approximately increases by [percentage missing]. The subproblem size increases linearly with both investment-related constraints and time-coupled operational constraints.

[0178] 2) Forward transmission process (strategy evaluation and cost estimation):

[0179] During the forward propagation, the decision strategy is evaluated along each complete path from the root node to the terminal node of the scene tree. This is for implementations involving a series of uncertainties. Given a defined path, decisions for each stage are computed sequentially by solving a phased optimization problem (20). Subsequent cost functions are approximated by piecewise linear functions constructed using standard Benders cuts, which are iteratively generated using dual information obtained from backpropagation.

[0180] This process generates a set of optimal stage decisions. And calculate the total cost along this path:

[0181] (twenty two);

[0182] 3) Convergence Criteria

[0183] As the iterations proceed, additional Benders cuts are generated, gradually improving the accuracy of the piecewise linear approximation of the value function. Alternating backward and forward traversals continue until the solution converges.

[0184] The algorithm terminates once the optimal gap (defined as the difference between the upper bound (UB) and the lower bound (LB)) falls below a preset threshold, as shown in Equation (23). In this invention, the tolerance ε is set to a small value (e.g., 10⁻⁴), following conventional methods to balance solution accuracy and computational efficiency in long-term planning problems. LB is calculated according to Equation (21), while UB is estimated by weighted averaging of the target values ​​across all scenario paths.

[0185] (twenty three);

[0186] (twenty four);

[0187] Part Four: Case Studies

[0188] The multi-stage stochastic programming method proposed in this invention was validated on an improved IEEE 30-node system. Various assets were allocated in stages to address climate uncertainty.

[0189] A. Test System Description

[0190] Figure 4 An improved IEEE 30-node system is presented. The entire planning period is 40 years, divided into four phases of 10 years each. An annual load growth rate of 4% is assumed throughout the planning period, with a phased 20% carbon reduction target to guide the system's decarbonization path. At the start of the planning period, the system comprises 370 MW of wind turbines (WTs), 330 MW of photovoltaics (PVs), 1100 MW of gas turbines (CUs), and 400 MW of gas turbine units (GUs). Technical and economic parameters for all candidate assets are listed in Table 1.

[0191] Figure 4The improved IEEE 30-node system comprises three regions, and the monthly variations in wind speed and solar radiation in each region relative to Phase 1 are recorded.

[0192] Table 1 Candidate Asset Parameters

[0193] .

[0194] To capture long-term uncertainties associated with climate change, scenario trees were constructed based on predictions from 15 global climate models (GCMs), including ACCESS-CM2, ACCESS-ESM1-5, and other models participating in CMIP6, covering the period from 2020 to 2060, using the SSP2-4.5 path. Climate data from three adjacent provinces in China were selected and mapped to three corresponding regions within the IEEE 30-node system. Figure 4 The monthly variations in wind speed and solar radiation for each region are presented as phase-wise changes relative to Phase 1. To ensure computational feasibility, the generated wind and solar energy time series are clustered into twelve representative scenarios (typical days) for annual operational simulation.

[0195] B. Multi-stage planning outcomes under climate uncertainty

[0196] The proposed planning model is used to determine the optimal allocation of these assets at different planning stages. Table 2 summarizes the weighted installed capacity of different assets in each region at the final stage and the cost over the entire planning period. Assets include ES (Energy Storage), WT (Wind Turbine), PV (Photovoltaic), CU (Coal-fired Unit), CUC (Coal-fired Unit with Cogeneration), GU (Gas Turbine), and GUC (Gas Turbine Unit with Cogeneration).

[0197] 1) Climate Mitigation Planning Strategy Analysis: Driven by climate mitigation goals, it is clear that by the end of the planning period, asset deployment strategies in various regions showed a high dependence on renewable energy. Simultaneously, these regions also introduced thermal power units equipped with carbon capture technology to further enhance climate mitigation efforts.

[0198] Table 2 Results of Power System Planning

[0199] .

[0200] IC, OC, and TC represent the net present value of investment costs, operating costs, and total costs over the entire planning period, respectively.

[0201] Figure 5 This illustrates the evolution of cumulative installed capacity across different planning phases. It is directly observable that, driven by climate change mitigation goals and increased renewable energy penetration, the capacity of wind (WT) and solar (PV) units has gradually expanded to varying degrees in each phase. With increasing load demand, total installed capacity shows a steady upward trend. Notably, renewable energy capacity grew rapidly in phases 3 and 4.

[0202] Figure 6 The data shows a declining trend in the proportion of thermal power units, which decreased by 29.5% in the power generation structure by the end of the planning period compared to the initial stage. This trend reflects the gradual replacement of thermal power by wind and solar energy as the main power supply. The proportion of renewable energy steadily increased throughout the planning period, eventually reaching approximately 70%. Meanwhile, the proportion of energy storage rose to 7.8% by the end of the planning period, indicating increased integration of energy storage systems to enhance their climate resilience.

[0203] 2) Validation of the effectiveness of climate adaptation planning strategies: Figure 7 Key system indicators were compared under multiple climate uncertainty scenarios at the end of the planning period. The results clearly demonstrate that changes in climate variables have a significant impact on the deployment of renewable energy capacity. Specifically, the share of renewable energy in total installed capacity ranged from 68.5% to 71.9%, reflecting the spatial and temporal heterogeneity of wind and solar resource availability. These changes affect investment economics, thus shaping capacity expansion decisions at different planning stages. Notably, climate uncertainty has a more significant impact on renewable energy output than on installed capacity. The share of renewable energy generation exhibits greater fluctuations under different scenarios, highlighting that changes in climate conditions directly affect system performance through generation uncertainty.

[0204] Furthermore, analysis of flexible resources, particularly energy storage systems (ES), reveals that climate-driven uncertainties also alter system flexibility requirements. The largest variation in ES deployment reached 3.8%, highlighting the moderating role of ES in adapting to uncertain climate trajectories. Overall, these findings validate the effectiveness of the proposed multi-stage planning framework in addressing climate-related uncertainties. By supporting iterative adjustments to the power generation mix, this framework contributes to more cost-effective and resilient system development under diverse climate futures.

[0205] Climate uncertainty is represented by a set of different climate development trajectories. To study their impact on long-term planning outcomes, two extreme trajectories were selected for comparison: the Low Total Cost Trajectory (LTCT) and the High Total Cost Trajectory (HTCT), corresponding to the minimum and maximum cumulative system costs over the entire planning period, respectively. Figure 8(a) shows the planning outcomes for each stage under LTCT and HTCT. Due to the heterogeneity of climate realization, the average capacity factor for renewable energy sources (RES) varies significantly across stages. LTCT shows a consistently high stage-average capacity factor, indicating more favorable renewable energy resource conditions and therefore higher unit investment efficiency for RES. In contrast, HTCT is associated with relatively poorer resource availability.

[0206] These differences are further reflected in the installed capacity required to ensure energy balance and system reliability. As shown in Figures 8(a) and 8(b), under the unfavorable resource realization conditions represented by the HTCT, a greater quantity and higher proportion of renewable energy capacity must be deployed, while other power generation technologies are adjusted accordingly. Notably, a complementarity of resource conditions is observed between different stages. For the HTCT, although resource conditions improve from stage 2 to stage 3, a significant amount of renewable energy capacity has already been installed in the early stages, resulting in a moderate increase in additional investment. Therefore, the increase in the renewable energy share from stage 2 to stage 3 is approximately 2.2% lower than that under the Long Term Carbon Reduction Scenario (LTCT). These results demonstrate that the climate evolution trajectory directly influences system-level planning decisions, including the timing, scale, and composition of power generation capacity investments, by impacting the availability of renewable resources at each stage. This confirms the proposed framework's ability to capture trajectory-dependent resource evolution and corresponding system-level responses.

[0207] C. The value of climate uncertainty modeling

[0208] To further illustrate the necessity of explicitly modeling climate uncertainty in long-term power system planning, this study compares three representative planning paradigms:

[0209] (i) Deterministic programming model that does not consider climate uncertainty (wo CU-DT).

[0210] (ii) Two-stage stochastic programming model considering climate uncertainty (wCU-TS).

[0211] (iii) The proposed multi-stage planning model (wCU-MS) that takes into account climate uncertainty.

[0212] Through detailed system operation simulations, the planning strategies derived from these models are evaluated under climate-affected operating conditions, aiming to assess their economic performance and technological adaptability under real-world uncertainties.

[0213] As shown in Table 3, ignoring climate uncertainty leads to a 3.8% increase in total system cost during actual operation. This cost increase is primarily due to insufficient representation of wind and solar resource curves, resulting in a load reduction of up to 2992 MWh. Introducing climate uncertainty through two-stage stochastic programming partially mitigates this issue. Compared to the deterministic baseline, the wCU-TS scenario achieves a 0.53% reduction in total cost and a 29.8% reduction in maximum annual load loss. However, a considerable residual maximum annual load loss (2102 MWh) remains. Furthermore, the total system cost under the two-stage model is still 3.2% higher than the proposed wCU-MS scenario, further validating the value of time flexibility in addressing long-term climate uncertainty.

[0214] Table 3 Results of Power System Planning

[0215] .

[0216] IC, OC, and TC represent the net present value of investment costs, operating costs, and total costs over the entire planning period, respectively. YCE represents annual carbon emissions, and MYLL represents the maximum annual load shedding.

[0217] right Figure 9 A comparative analysis of capacity expansion decisions across different planning phases reveals that the non-CU-DT strategy underinvested in wind and solar capacity in the second phase, while overinvesting in subsequent phases (the third and fourth phases). This bias primarily stems from failing to account for the declining long-term availability of renewable resources, thus distorting the timing and scale of investment decisions. Although the CU-TS scenario exhibits an expansion pattern closer to the proposed multi-phase strategy in some phases, it generally shares the same intertemporal investment trends as the non-CU-DT scenario, reflecting the limitations of a single-response structure in capturing evolving climate uncertainties.

[0218] In addition, such as Figure 10The simulations across multiple scenarios, without the CU assumption, demonstrate that the phased fixed planning decisions derived under these assumptions have limited adaptability to different climate conditions. In the worst-case scenario, the total system cost is 8.34% higher than the adaptive optimal strategy. Compared to the baseline without CU-DT, the scenario with CU-TS shows improvement in most scenarios; however, it still suffers from insufficient adaptability in extreme cases. In particular, its worst-case total system cost is still 7.2% higher than the proposed multi-stage strategy. These findings highlight the importance of integrating climate-related uncertainties into a multi-stage planning framework to enhance system resilience and economic efficiency under long-term climate change.

[0219] In summary, this invention provides a climate-adaptive multi-stage stochastic low-carbon power system planning method that considers climate uncertainty. This method includes constructing a scenario tree based on dual climate uncertainty, downscaling the time scale for climate consistency, establishing a multi-stage stochastic programming model, and solving it using a customized stochastic dual dynamic programming algorithm. By embedding the climate evolution trajectory into the planning perspective, it solves the problems of existing planning frameworks neglecting climate adaptation and struggling to bridge climate and power models, thereby improving the adaptability and cost-effectiveness of the power system under climate change.

Claims

1. A multi-stage stochastic low-carbon power system planning method considering climate uncertainty, characterized in that, Includes the following steps: Step 1: Based on shared socio-economic paths and representative concentration path scenarios, as well as the outputs of multiple global climate models, a representative scenario tree is constructed through stage-by-stage clustering to generate climate evolution paths that characterize scenario uncertainty and climate response uncertainty. Step 2: Using a climate-consistent time downscaling method, the coarse-resolution climate prediction data of each stage in the representative scenario tree are converted into hourly-resolution time series data to generate renewable resource availability factors for each stage. Step 3: With the goal of minimizing the sum of net present value of investment costs and operating costs during the planning period, a multi-stage stochastic low-carbon power system planning model is established based on the representative scenario tree and the renewable resource availability factors at each stage. At each stage, capacity expansion decisions for power generation assets and energy storage assets and operation scheduling optimization are made according to the climate of the corresponding scenario nodes. Step 4: A customized stochastic dual dynamic programming algorithm is directly applied to the representative scenario tree to perform forward policy evaluation and backward value function approximation, solving the multi-stage stochastic low-carbon power system planning model to generate the optimal investment and operation decision strategies.

2. The multi-stage stochastic low-carbon power system planning method considering climate uncertainty according to claim 1, characterized in that, In step 1, the stage-by-stage clustering to construct the representative scene tree includes: Obtain a set of climate evolution trajectories, assuming there are N climate evolution trajectories available, each trajectory spanning... Each planning phase; For each stage The current climate trajectory data is clustered into a limited number of representative nodes, and each representative node corresponds to a group of trajectories with similar climate characteristics. With cluster centroid Representative nodes are formed for each stage, and the conditional transition probabilities between nodes are calculated using the number of trajectories as weights. A representative scenario tree connecting the representative nodes of each stage is then constructed. For each trajectory... In the stage The assigned cluster label is denoted as Estimating representative nodes in adjacent stages using empirical frequency counting and conditional transition probabilities : ; in, For the planning phase index, ; For the set of all planning stages, This represents the total number of planning phases. To be from the stage Representative node Transition to Stage Representative node The conditional probability; For the first A climate trajectory in stages Cluster tags; For counting symbols; For the stage The number of clusters, which represents the number of nodes.

3. The multi-stage stochastic low-carbon power system planning method considering climate uncertainty according to claim 1, characterized in that, In step 2, the climate-consistent time downscaling method includes: Obtain historical hourly data of climate variables, calculate the corresponding monthly averages, and determine the hourly distribution patterns for each month; Obtain the monthly forecast data output by the global climate model. Using the monthly forecast data as a benchmark, apply the hourly distribution model to downscale the monthly forecast data to hourly resolution. The specific process is as follows: Step 2.1: Let For the region Climate variables Hourly historical data; corresponding monthly average data for the month of o. Defined as: ; in, For a set of regions; A set of climate variables; For months Total hours; Step 2.2: Let Represents the relationship between scene tree nodes Forecast monthly averages of relevant climate variables; define monthly scaling factors. for: ; Step 2.3: Apply monthly scaling factors to generate adjusted hourly climate time series. The details are as follows: ; in, The month to which the hour belongs Monthly scaling factor.

4. The multi-stage stochastic low-carbon power system planning method considering climate uncertainty according to claim 1, characterized in that, In step 3, the objective function of the multi-stage stochastic low-carbon power system planning model is expressed by the following formula: ; ; ; ; ; ; ; ; in, For the planning phase index, ; For the set of all planning stages, This represents the total number of planning phases. For the stage The vector of investment decision variables; For the stage The vector of operational decision variables; These are cost coefficient vectors for investment and operational decision variables, respectively; For interest rates; The number of years included in each planning phase; For the stage Net present value of investment costs; For the stage Net present value of operating costs; Index of years within a period; For asset indexing; Separate generator sets and energy storage systems; Index for power grid nodes; For assets The capital recovery coefficient; For the stage Domestic assets At the node The investment cost reduction coefficient; For assets At the node The unit investment cost; For the stage Assets at the end At the node The cumulative installed capacity; For the previous stage ( The cumulative installed capacity of ) To cover fixed operation and maintenance costs; For assets At the node The unit fixed operation and maintenance cost; Variable operating costs; Typical daily index; For typical days The actual number of days in a year; A time index for each typical day; For assets At the node Variable operating cost per unit; For assets In the stage , typical day ,time and nodes contribution; The charging power for the energy storage system; Costs for unit start-up and shutdown; For assets At the node The cost per startup; For assets At any moment Online capacity; For carbon capture costs; A collection of units equipped with carbon capture systems; The unit cost of carbon capture; For the actual carbon capture by the carbon capture system quantity; Penalty cost for loss of load; For nodes The unit loss-of-load penalty cost; For nodes At any moment The power of the load shedding.

5. The multi-stage stochastic low-carbon power system planning method considering climate uncertainty according to claim 4, characterized in that, The phased investment constraints of the multi-stage stochastic low-carbon power system planning model are as follows: ; ; in, For the stage Domestic assets At the node Maximum allowed new capacity; For assets throughout the system The maximum total allowed installed capacity.

6. The multi-stage stochastic low-carbon power system planning method considering climate uncertainty according to claim 4, characterized in that, The multi-stage stochastic low-carbon power system planning model also includes phased operational constraints, which include: 1) Renewable energy generation constraints, as shown in the following formula: ; in, Contributing to renewable energy; Availability factor for renewable energy resources; This refers to the cumulative installed capacity of renewable energy. For wind turbine units, For photovoltaics; 2) Operating constraints of coal-fired power units, as shown in the following formula: ; ; ; ; ; in, This refers to the minimum technical output ratio of coal-fired power units. For coal-fired power units at all times Online capacity; The actual output of the coal-fired power unit; Spinning reserve capacity provided for coal-fired power units; and These are the maximum uphill and downhill gradients for coal-fired power units, respectively. This refers to the cumulative installed capacity of coal-fired power units. This represents the initial online capacity ratio of coal-fired power units at the beginning of the phase. 3) Operating constraints of the gas turbine unit are as follows: ; in, This refers to the actual output of the gas turbine unit; Spinning reserve capacity provided for gas turbine units; This refers to the cumulative installed capacity of the gas turbine units; 4) Carbon capture operations and energy consumption constraints, as shown in the following formula: ; ; ; ; in, For assets Total carbon emissions generated; For assets carbon emission intensity; For assets contribution; The amount of carbon emissions fed into the carbon capture system; The carbon capture efficiency of the carbon capture system; The electrical power consumed by the carbon capture system during operation; The energy efficiency ratio of CCS; For the unit's final net carbon emissions; For coal-fired power units and combined heat and power; For gas turbine units and combined heat and power; 5) Operating constraints of the energy storage system are shown in the following formula: ; ; ; ; ; in, The charging power for the energy storage system; The discharge power of the stored energy; The energy stored in energy storage; The rated continuous discharge time of the energy storage system; and These refer to the charging efficiency and discharging efficiency of the energy storage system, respectively. For assets Provided spinning reserve capacity.

7. The multi-stage stochastic low-carbon power system planning method considering climate uncertainty according to claim 4, characterized in that, The multi-stage stochastic low-carbon power system planning model also includes network structure stage constraints, as shown in the following equation: ; ; ; ; in, branch road Active power flowing upstream; and branch road The voltage phase angle between the start and end nodes; branch road The reactance; branch road The transmission capacity limit; For nodes The voltage phase angle; and The discharge and charging power of the energy storage system; For node load requirements; For nodes At any moment The power of the load shedding; The power consumed by carbon capture and storage; Let i be the set of branches starting from node i. Let i be the set of branches that terminate at node i.

8. The multi-stage stochastic low-carbon power system planning method considering climate uncertainty according to claim 4, characterized in that, The multi-stage stochastic low-carbon power system planning model also includes system performance stage constraints, specifically: 1) Capacity adequacy constraint The capacity adequacy constraint requires that, in each time period, the total available generating capacity meets or exceeds the total load demand and reserve capacity requirements, thereby ensuring the security and reliability of the power supply, as shown in the following formula: ; in, For assets At the node Available capacity factor; For nodes The required proportion of standby capacity; For the stage Middle node Peak load; 2) Rotational spare constraint A spinning reserve constraint is introduced to ensure that the total fast response reserve capacity must meet or exceed the specified reserve requirements, as shown in the following formula: ; in, For assets Provided spinning reserve capacity; The proportion of spinning reserve required by the load; For node load requirements; The proportion of spinning reserve demand for renewable energy; To make a real contribution to renewable energy; 3) Carbon emission restrictions Carbon emission limits are imposed on the total carbon emissions of all conventional generating units after considering carbon capture, ensuring that the power system develops along a prescribed carbon reduction path, as shown in the following formula: ; in, For typical days The number of days; Carbon emission intensity; Net carbon emissions; For the stage Maximum permissible total carbon emissions; 4) Renewable energy utilization rate constraints Renewable energy utilization constraints apply to the ratio of renewable energy generation to available renewable energy resources. A lower limit is imposed to promote the priority use and efficient integration of renewable energy, as shown in the following formula: ; in, Minimum utilization rate of renewable energy; Renewable energy availability factor; This refers to the cumulative installed capacity of renewable energy. 5) System inertia constraints System inertia constraints address the challenges to frequency stability posed by high-penetration renewable energy sources by specifying the minimum level of inertia that the system must meet under different operating conditions, as shown in the following equation: ; in, For assets The provided inertial constant; This represents the maximum allowable inertial descent tolerance of the system. This is the reference value for the current system's inertial constant; It is a gas turbine unit.

9. The multi-stage stochastic low-carbon power system planning method considering climate uncertainty according to claim 4, characterized in that, The multi-stage stochastic low-carbon power system planning model establishes the multi-stage stochastic programming problem under climate uncertainty as a Markov decision process; its characteristics are determined by the state... ,action Transition probability and stage costs describe; In the stage The system state is defined as follows: ,in This indicates the system configuration inherited from the previous stage. Representation phase Climate uncertainty is realized; In a given state Under the conditions, stage The decision, or action, is defined as Actions are subject to the following: ; ; in, For the stage The system status, ; For the stage The action, ; Set up as an actionable set; It is a function of the feasible region; The coefficient matrix / vector for investment cross-stage coupling constraints; The coefficient matrix / vector represents the mixed constraints of operation and investment within a given phase. For the stage Climate uncertainty is realized; constraint Capture the cross-period coupling of investment decisions across consecutive stages, and compactly represent the remaining investment and operational constraints as follows: Vectors and matrices , , , , and This represents the corresponding constraint parameters; Given a state-action pair The state of the next stage is defined as follows: ;in, Determined by the current action, Follows random climate processes; The state transition probability is decomposed into a deterministic part related to the evolution of the investment-driven system and a stochastic part describing the dynamics of climate uncertainty, namely: ; in, The state transition probability; award With cost function This indicates the investment and operating costs incurred by the selected action; the optimal solution of the original multi-stage model based on the Markov decision process satisfies the Bellman equation: ; in, Indicates from stage Minimum expected cumulative cost to the final stage; By expanding the Bellman recursion, the Markov decision process adopts an equivalent nested optimization form during the planning period; decisions are made in a forward recursive manner, meaning that the investment and operational decisions at each stage depend on the realization of current uncertainties and the expected costs of future stages, as shown in the following equation: ; in, For the stage The set of feasible decisions, which depends on the system configuration left over from the previous stage. and the current stage of climate uncertainty ; Regarding the stages Climate uncertainty The mathematical expectation operator; The above equation is equivalently restated in a node-based manner; let and Representing nodes respectively Parent and child nodes; cost arrival function Aggregation in implementation The immediate cost of the node and the expected future costs of all its child nodes; in the final stage Only the specific costs of this stage are retained, as shown in the following formula: ; ; ; in, As a cost-to-reach function; To move to child nodes The probability of.

10. The multi-stage stochastic low-carbon power system planning method considering climate uncertainty according to claim 9, characterized in that, In step 4, the customized stochastic dual dynamic programming algorithm is directly run on the representative scene tree, including: Backward value function approximation: The backward propagation process involves solving stage-level linear programming subproblems for each scenario node in the current stage to obtain the optimal value and dual multipliers. The dual multipliers are then used to construct a Benders cut plane to approximate the future cost function. This new cut plane is added to the node's approximate value function set. Finally, the linear problem is solved at the root stage to obtain the lower bound of the original problem. The backward recursion is expressed as follows: ; in, For nodes Investment decisions; Auxiliary variables used to replicate the state from the previous stage; For nodes Operational decisions; For nodes Climate realization; As an approximation variable for future costs; The state value passed in from the previous stage; For dual multipliers of equality constraints; This is a constant term that approximates future costs; This represents the gradient for approximating future costs. Investment decisions at reference points; ; ; After the pass-through is complete, through the root phase =1 solves the following linear problem to obtain an effective lower bound (LB) for the original multi-stage stochastic problem: ; in, For the root stage Decision variables for the next iteration; This approximates future costs. For the first The constant term for slicing; For the first The gradient of the stripe; For the first Investment decisions at the root stage of the next iteration; For the first The reference point for the next iteration; Forward strategy evaluation: The forward pass process evaluates the decision strategy along each complete path from the root node to the terminal node of the scene tree; this process produces a set of optimal stage decisions. And calculate the total cost along this path: ; When the optimal gap, i.e., the difference between the upper bound UB and the lower bound LB, is lower than the preset convergence threshold... If the condition is met, the algorithm terminates; the upper bound UB is estimated by weighted averaging of the target values ​​across all scenario paths, as shown in the following formula: ; ; in, The convergence threshold is M; M is the total number of scenes. For the stage Along the scene The conditional transition probability; For the scene Total cost; The forward and backward propagation processes are executed alternately until the algorithm terminates and outputs the optimal decision strategy.