Carbon electricity operation and investment collaborative decision-making method based on alliance chain

Abstract

This invention relates to the field of resource allocation and decision-making data processing, specifically to a collaborative decision-making method for carbon power operation and investment based on a consortium blockchain. This method constructs an integrated decision-making framework encompassing investment planning, carbon power trading, and physical operation, mapping three decision-making sub-models—investment planning, carbon power trading, and physical operation—to consortium blockchain nodes. For each sub-model, a partially observable Markov decision process is constructed. Collaborative multi-center federated learning is performed through parameter interaction between consortium blockchain nodes to update the policy network parameters of each sub-model. Based on the updated policy network parameters, a medium- to long-term collaborative operational decision for the power distribution company is output. This invention achieves dynamic interaction and parameter coordination among investment planning, carbon power trading, and physical operation, solving the decision mismatch problem caused by the separation of these three elements.

Description

Technical Field

[0001] This invention relates to the field of resource allocation and decision-making data processing, specifically to a collaborative decision-making method for carbon-electricity transportation and investment based on consortium blockchain. Background Technology

[0002] In the context of energy transition, power distribution companies need to participate in carbon-electricity trading to reduce carbon emissions and meet user demand. Existing conventional technical solutions mainly focus on short-term optimization strategies for carbon-electricity coupled trading. These solutions typically construct optimization models aimed at minimizing electricity purchase and sale costs or maximizing carbon trading revenue, considering the constraints of renewable energy output and carbon quotas, and obtaining short-term trading curves by solving mathematical programming problems. In this approach, power distribution companies act only as trading entities in the market, allocating electricity and carbon quotas based on the existing physical grid structure and fixed low-carbon resource capacity, without incorporating the expansion of the physical grid and the renewal of low-carbon resources into their decision-making.

[0003] With the opening of the electricity retail market, some technical solutions have begun to focus on the investment planning stage of distribution and retail companies, treating distribution network expansion and low-carbon resource allocation as independent optimization targets. These solutions typically employ deterministic planning models, aiming to minimize the sum of investment and operating costs, and pre-setting fixed operating scenarios and trading boundaries during the planning phase. In practice, investment planning decisions are separated from carbon electricity trading and physical operation. The planning phase does not consider dynamic game theory and risk feedback in market trading, and the operation and trading phases do not transmit grid flexibility and market supply and demand information to the planning phase. The decommissioning of carbon-intensive equipment and the construction of renewable energy projects lack linkage and verification with spinning reserves and flexibility indicators in physical operation.

[0004] Existing technical solutions isolate the investment planning, carbon electricity trading, and physical operation of power distribution companies, lacking a unified modeling and collaborative optimization mechanism. This separation leads to investment planning failing to adapt to the dynamic risk feedback of market transactions and the flexibility verification of physical operation. Carbon electricity trading lacks the capacity support and market space provided by investment planning, and physical operation cannot conversely constrain the action space of investment planning and affect the competitive operational level of trading. The renewable energy construction and decommissioning of carbon-intensive equipment in investment planning are not linked to the reserve needs of physical operation, and trading decisions are not linked to the social welfare of multi-stakeholder games. This results in core technical problems such as decision-making mismatch, resource misallocation, and low overall operational efficiency for power distribution companies in the medium and long term. Summary of the Invention

[0005] The purpose of this invention is to provide a collaborative decision-making method for carbon-electricity transportation and investment based on consortium blockchains, which can solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A collaborative decision-making method for carbon power operation and investment based on consortium blockchain includes the following steps: Constructing an integrated decision-making framework for investment planning, carbon power trading, and physical operation. This framework includes an investment planning decision-making sub-model, a carbon power trading decision-making sub-model, and a physical operation decision-making sub-model. The investment planning decision-making sub-model provides capacity conditions and market space for the carbon power trading decision-making sub-model. The operational benefits and risks of the carbon power trading decision-making sub-model are fed back to the investment planning decision-making sub-model. The physical operation decision-making sub-model verifies the feasibility of the investment planning decision-making sub-model and influences the competitive operational level of the carbon power trading decision-making sub-model. The method then addresses the investment planning decision-making sub-model and its implications. The strategy sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model are constructed based on the state space, action space, and transition probabilities of a partially observable Markov decision process. The investment planning decision sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model are mapped to consortium blockchain nodes. Through parameter interaction between the consortium blockchain nodes, collaborative multi-center federated learning based on the consortium blockchain is performed to update the strategy network parameters of the investment planning decision sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model. Based on the updated strategy network parameters, the medium- and long-term operational collaborative decision of the power distribution company is output, completing the consortium blockchain-based carbon electricity operation and investment collaborative decision-making method.

[0007] Preferably, in the step of constructing an integrated decision-making framework for investment planning, carbon electricity trading, and physical operation, the investment planning decision-making sub-model includes distribution network investment planning and low-carbon resource planning. The distribution network investment planning generates node investment prices, which are allocated according to the occupancy levels of each market participant in the distribution network. The low-carbon resource planning targets the market participants for distributed renewable energy planning and carbon-intensive power generation equipment retirement plans. The carbon electricity trading decision-making sub-model conducts differentiated combined carbon electricity trading with the market participants based on the node investment prices and the output of the low-carbon resource planning. The physical operation decision-making sub-model is executed by a distribution network operator independent of market trading. The distribution network operator optimizes the allocation of dispatchable resources based on the differentiated combined carbon electricity trading and feeds back the flexibility index of the optimized allocation of dispatchable resources to the investment planning decision-making sub-model.

[0008] Preferably, in the step of constructing the state space, action space, and transition probability based on a partially observable Markov decision process for the investment planning decision sub-model, the state space includes the distribution network topology state, low-carbon resource capacity state, and load demand state; the action space includes distribution network line expansion actions, distributed renewable energy construction actions, and carbon-intensive power generation equipment decommissioning actions; the transition probability represents the probability distribution of the distribution network topology state and the low-carbon resource capacity state transitioning to the next time series under the distribution network line expansion action, the distributed renewable energy construction action, and the carbon-intensive power generation equipment decommissioning action; and the reward function of the investment planning decision sub-model includes the difference between the discounted investment cost and the node investment price return.

[0009] Preferably, in the step of constructing the state space, action space, and transition probability based on a partially observable Markov decision process for the carbon electricity trading decision sub-model, the state space includes the market electricity price state, the carbon quota price state, and the carbon electricity demand state of the market participants; the action space includes the differentiated combination of trading electricity and carbon emission rights trading quota allocated to the market participants; the transition probability represents the probability distribution of the market electricity price state and the carbon quota price state transitioning to the next time series under the differentiated combination of trading electricity and the carbon emission rights trading quota; and the reward function of the carbon electricity trading decision sub-model includes the difference between carbon electricity trading revenue and the carbon electricity trading default risk penalty value.

[0010] Preferably, in the step of constructing the state space, action space, and transition probability based on a partially observable Markov decision process for the physical operation decision sub-model, the state space includes the schedulable resource output state, branch power flow state, and node voltage state; the action space includes distributed energy storage charging and discharging regulation actions and distributed renewable energy output reduction actions; the transition probability represents the probability distribution of the branch power flow state and the node voltage state transitioning to the next time series under the distributed energy storage charging and discharging regulation actions and the distributed renewable energy output reduction actions; and the reward function of the physical operation decision sub-model includes the sum of physical operation scheduling cost and physical over-limit penalty value.

[0011] Preferably, in the step of performing collaborative multi-center federated learning based on consortium blockchain through parameter interaction between the investment planning decision sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model, the investment planning decision sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model calculate the policy network gradient locally, homomorphically encrypt the policy network gradient, and broadcast it to the consortium blockchain node. The policy network gradient verified by consensus is aggregated in the consortium blockchain node, and the policy network parameters of the investment planning decision sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model are updated based on the aggregated global gradient.

[0012] Preferably, in the step of constructing the state space, action space, and transition probability based on a partially observable Markov decision process for the investment planning decision sub-model, when executing the distributed renewable energy construction action and the carbon-intensive power generation equipment decommissioning action within the action space, an output intermittency characterization parameter and an operating reserve matching constraint are introduced. The output intermittency characterization parameter quantifies the volatility brought about by the distributed renewable energy construction action, and the operating reserve matching constraint limits the spinning reserve reduced by the carbon-intensive power generation equipment decommissioning action to be compensated by the newly added distributed energy storage capacity. The operating reserve matching constraint is used as the action boundary condition of the investment planning decision sub-model, and distributed renewable energy construction actions and carbon-intensive power generation equipment decommissioning actions that do not meet the action boundary condition are filtered out.

[0013] Preferably, in the step of constructing the state space, action space, and transition probability based on a partially observable Markov decision process for the carbon electricity trading decision sub-model, when generating the differentiated combined trading volume and the carbon emission rights trading quota within the action space, a carbon-electricity coupled price formation mechanism is established. This mechanism dynamically adjusts the settlement rate of the differentiated combined trading volume based on the carbon emission intensity and electricity elasticity coefficient of the market participants, and calculates the social welfare function of the market participants in conjunction with a multi-participant game interaction process. Maximizing the social welfare function is used as the action optimization constraint of the carbon electricity trading decision sub-model to guide the allocation of the differentiated combined trading volume and the carbon emission rights trading quota.

[0014] Preferably, in the step of constructing the state space, action space, and transition probability based on a partially observable Markov decision process for the physical operation decision sub-model, for the physical limit penalty value of the physical operation decision sub-model, the branch power flow limit index and node voltage limit index corresponding to the physical limit penalty value are extracted. An operation flexibility evaluation matrix is ​​constructed based on the branch power flow limit index and the node voltage limit index. The operation flexibility evaluation matrix is ​​mapped to the investment planning decision sub-model. When the operation flexibility evaluation matrix is ​​lower than the operation flexibility benchmark, the distribution network line expansion action and the distributed renewable energy investment action in the action space of the investment planning decision sub-model are dynamically pruned to eliminate action combinations that cause the operation flexibility evaluation matrix to be lower than the operation flexibility benchmark.

[0015] Preferably, in the step of performing collaborative multi-center federated learning based on the consortium blockchain through parameter interaction between the consortium blockchain nodes, a multi-timescale alignment mechanism is introduced for the long-term decision variables of the investment planning decision sub-model and the short-term decision variables of the carbon electricity trading decision sub-model and the physical operation decision sub-model. A timestamp identifier is added to the strategy network gradient. The consortium blockchain nodes asynchronously aggregate the strategy network gradients corresponding to the long-term decision variables and the short-term decision variables according to the timestamp identifier. The consistency of the timestamp identifier is verified by a cross-chain parameter verification node, thereby completing the cross-timescale collaborative update of strategy network parameters between the investment planning decision sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs an integrated decision-making framework for investment planning, carbon electricity trading, and physical operation. It maps three decision-making sub-models (investment planning, carbon electricity trading, and physical operation) to consortium blockchain nodes and executes collaborative multi-center federated learning. The investment planning sub-model provides capacity conditions and market space for carbon electricity trading. The operational benefits and risks of the carbon electricity trading sub-model are fed back to the investment planning sub-model. The physical operation sub-model verifies the feasibility of the investment plan and influences the competitive operation level of carbon electricity trading. These three elements form a dynamic interactive closed loop. Based on a partially observable Markov decision process, each decision-making sub-model is modeled. The collaborative updating of strategy network parameters is achieved through parameter interaction between consortium blockchain nodes, overcoming the decision mismatch problem caused by the separation of the three elements. This enables integrated collaborative decision-making for medium- and long-term operations of power distribution companies, driven by cost, carbon electricity, and market information flow.

[0017] 2. This invention constructs a refined interactive constraint and feedback mechanism within the decision-making sub-model. In investment planning, an operational reserve matching constraint is introduced, limiting the rotational reserves for the retirement and reduction of carbon-intensive power generation equipment to be compensated by newly added distributed energy storage capacity, ensuring physical operational safety. In carbon electricity trading, settlement rates are dynamically adjusted based on carbon emission intensity and electricity elasticity coefficient, with the maximization of the social welfare function as the optimization constraint, preventing the market from falling into suboptimal solutions. Exceeding limits in physical operation indicators are extracted to construct an operational flexibility assessment matrix, which is mapped to the investment planning decision-making sub-model to dynamically tailor the action space, avoiding investment actions that do not meet flexibility benchmarks. A multi-timescale alignment mechanism is introduced, asynchronously aggregating and verifying the gradients of the strategy network between long-term investment planning and short-term trading operations, achieving cross-timescale alignment and collaborative updating of decision variables. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the overall implementation process of the collaborative decision-making for carbon power transportation and investment based on consortium blockchains in this invention. Figure 2 This is a flowchart illustrating the closed-loop data interaction between the three sub-models of the integrated decision-making framework of this invention. Figure 3 The flowchart for modeling and constraint execution of the investment planning decision-making sub-model POMDP (Partially Observable Markov Decision Process) of this invention is shown below. Figure 4 This is a flowchart illustrating the modeling and pricing coupling process of the carbon electricity trading decision sub-model POMDP of this invention. Figure 5 This is a flowchart illustrating the modeling and flexibility feedback process of the Physical Operation Decision Sub-model (POMDP) ​​of this invention. Figure 6 This is a flowchart of the consortium blockchain collaborative federated learning and multi-timescale alignment process of the present invention. Detailed Implementation

[0019] refer to Figure 1In one embodiment, an integrated decision-making framework for investment planning, carbon electricity trading, and physical operation is constructed. This integrated decision-making framework comprises three core components: an investment planning decision-making sub-model, a carbon electricity trading decision-making sub-model, and a physical operation decision-making sub-model. The investment planning decision-making sub-model outputs distribution network expansion plans, distributed renewable energy investment plans, and carbon-intensive power generation equipment decommissioning plans, providing capacity conditions and market space for the carbon electricity trading decision-making sub-model. Based on the capacity conditions provided by the investment planning decision-making sub-model, the carbon electricity trading decision-making sub-model generates differentiated combined carbon electricity trading plans for different market participants. Its operational revenue and risk data are fed back to the investment planning decision-making sub-model for adjusting subsequent investment planning strategies. The physical operation decision-making sub-model receives the trading execution plan generated by the carbon electricity trading decision-making sub-model, optimizes the allocation of schedulable resources, verifies the physical feasibility of the plans output by the investment planning decision-making sub-model, and feeds back the flexibility index of the optimized allocation of schedulable resources to the investment planning decision-making sub-model, while also influencing the competitive operation level of the carbon electricity trading decision-making sub-model.

[0020] For the investment planning decision-making sub-model, carbon electricity trading decision-making sub-model, and physical operation decision-making sub-model, respectively, we construct the state space, action space, and transition probabilities based on a partially observable Markov decision process. The partially observable Markov decision process consists of a quintuple. Define, where Representing the state space, Represents the action space, This represents the state transition probability function. Represents the reward function, The observation space is defined. For each decision sub-model, a corresponding quintuple of parameters is defined based on its decision objective and constraints. The observation space of the investment planning decision sub-model includes observations of distribution network topology, low-carbon resource capacity, and load demand. The observation space of the carbon electricity trading decision sub-model includes observations of market electricity prices, carbon allowance prices, and carbon electricity demand from market participants. The observation space of the physical operation decision sub-model includes observations of dispatchable resource output, branch power flow, and node voltage.

[0021] The investment planning decision-making sub-model, carbon electricity trading decision-making sub-model, and physical operation decision-making sub-model are mapped to consortium blockchain nodes. The investment planning decision-making sub-model is mapped to an investment planning node, the carbon electricity trading decision-making sub-model to a carbon electricity trading node, and the physical operation decision-making sub-model to a physical operation node. These three nodes together form a consortium blockchain network, executing collaborative multi-center federated learning based on the consortium blockchain through parameter interaction between the nodes. During the collaborative multi-center federated learning process, each node calculates the policy network gradient locally based on its own partially observable Markov decision process model, encrypts the policy network gradient, and broadcasts it to other nodes in the consortium blockchain network. All nodes receive the encrypted gradients broadcast by other nodes, verify them through consensus, aggregate the gradients, and update their local policy network parameters based on the aggregated global gradient.

[0022] The updated strategy network parameters are used to output medium- and long-term operational coordination decisions for power distribution companies. These decisions include a 5-10 year distribution network investment plan, an annual carbon electricity trading plan, and a monthly physical operation scheduling plan. The distribution network investment plan includes the timeline, line parameters, and investment budget for distribution network line expansion; the type, capacity, and location of distributed renewable energy projects; and the timeline and capacity for retiring carbon-intensive power generation equipment. The annual carbon electricity trading plan includes differentiated combinations of trading volumes and carbon emission rights trading quotas for different market participants, as well as trading prices and settlement methods. The monthly physical operation scheduling plan includes charging and discharging plans for distributed energy storage and output reduction plans for distributed renewable energy.

[0023] This embodiment constructs an integrated decision-making framework for investment planning, carbon electricity trading, and physical operation. It maps the three decision-making sub-models to consortium blockchain nodes and executes collaborative multi-center federated learning, achieving a dynamic interactive closed loop among the three. By modeling each decision-making sub-model based on a partially observable Markov decision process, it can handle the uncertainty and partial observability in the decision-making process. Through parameter interaction between consortium blockchain nodes, it achieves collaborative updates of policy network parameters, overcoming the decision mismatch problem caused by the separation of the three components.

[0024] In this embodiment, the interaction parameters of each decision sub-model in the integrated decision-making framework of investment planning-carbon electricity trading-physical operation are shown in Table 1.

[0025] Table 1. Interaction Parameters of Each Decision Sub-model in the Integrated Decision-Making Framework of Investment Planning, Carbon Electricity Trading, and Physical Operation

[0026] refer to Figure 2In one embodiment, an integrated decision-making framework for investment planning, carbon electricity trading, and physical operation is constructed. The investment planning decision-making sub-model includes a distribution network investment planning module and a low-carbon resource planning module. The distribution network investment planning module generates distribution network line expansion plans and node investment prices based on the existing distribution network topology, load growth forecasts, and operational flexibility requirements. Node investment prices are allocated according to the occupancy levels of each market participant in the distribution network. The occupancy levels of market participants are quantified using their maximum electricity load, electricity consumption time distribution, and access node locations. The low-carbon resource planning module provides distributed renewable energy planning and carbon-intensive power generation equipment retirement plans for market participants. Distributed renewable energy planning includes the construction capacity, site selection, and timelines for distributed power sources such as photovoltaics and wind power. Carbon-intensive power generation equipment retirement plans include the retirement capacity and timelines for carbon-intensive power generation equipment such as coal-fired and gas-fired power plants.

[0027] The carbon electricity trading decision-making sub-model, based on the outputs of nodal investment prices and low-carbon resource planning, conducts differentiated combined carbon electricity trading with market participants. This differentiated combined carbon electricity trading includes a combination of electricity trading and carbon emission rights trading. For market participants with lower carbon emission intensity, lower electricity prices and higher carbon emission rights trading quotas are offered; for market participants with higher carbon emission intensity, higher electricity prices and lower carbon emission rights trading quotas are offered. The physical operation decision-making sub-model is executed by the distribution network operator, independent of market transactions. Based on the execution plan of the differentiated combined carbon electricity trading, the distribution network operator optimizes the allocation of dispatchable resources. Dispatchable resources include distributed energy storage, adjustable loads, and distributed renewable energy. The distribution network operator feeds back the flexibility indicators of the optimized allocation of dispatchable resources to the investment planning decision-making sub-model. Operational flexibility indicators include the distribution network's voltage regulation capability, power flow regulation capability, and fault recovery capability.

[0028] refer to Figure 3 This paper constructs a state space, action space, and transition probabilities based on a partially observable Markov decision process for an investment planning decision-making sub-model. The state space of the investment planning decision-making sub-model includes the distribution network topology state, low-carbon resource capacity state, and load demand state. The distribution network topology state consists of the on / off status and capacity parameters of all lines in the distribution network. The low-carbon resource capacity state consists of the total installed capacity of distributed renewable energy, the remaining capacity of carbon-intensive power generation equipment, and the total capacity of distributed energy storage. The load demand state consists of the maximum load, average load, and load curve characteristics of all nodes in the distribution network.

[0029] The action space of the investment planning decision-making sub-model includes distribution network line expansion, distributed renewable energy investment and construction, and carbon-intensive power generation equipment decommissioning. Distribution network line expansion includes the number of the lines to be expanded, the expansion capacity, and the expansion time. Distributed renewable energy investment includes the type, capacity, location, and construction time of the distributed power sources to be invested in. Carbon-intensive power generation equipment decommissioning includes the number of the power generation equipment to be decommissioned, the decommissioning capacity, and the decommissioning time.

[0030] The transition probability of the investment planning decision-making sub-model represents the probability distribution of the distribution network topology state and low-carbon resource capacity state transitioning to the next time series under the actions of distribution network line expansion, distributed renewable energy construction, and carbon-intensive power generation equipment decommissioning. The transition probabilities are calculated through historical data statistics and Monte Carlo simulation. For distribution network line expansion, the transition probability is 1, indicating that the distribution network topology state will be deterministically updated to include the newly expanded line after the expansion. For distributed renewable energy construction, the transition probability is determined by the construction cycle and success rate of the distributed power source. For carbon-intensive power generation equipment decommissioning, the transition probability is 1, indicating that the low-carbon resource capacity state will be deterministically updated to a state minus the capacity of the decommissioned equipment after the equipment decommissioning.

[0031] The reward function of the investment planning decision-making sub-model includes the difference between the discounted value of the investment cost and the return on the investment price at each node. The mathematical expression for the reward function is:

[0032] in, Indicates at time In state And perform the action Instant rewards at any time Indicates the time span of the decision-making process. Indicates at time The return on investment in nodes, Indicates at time Execute action The investment cost, This represents the discount rate.

[0033] When executing distributed renewable energy construction and decommissioning of carbon-intensive power generation equipment within the action space, a power output intermittency characterization parameter and operational reserve matching constraints are introduced. The power output intermittency characterization parameter quantifies the volatility introduced by distributed renewable energy construction. The mathematical expression for the power output intermittency characterization parameter is:

[0034] in, The parameter representing the intermittency of output is This indicates the number of historical data sampling points. Indicates the first Actual output of distributed renewable energy at each sampling point This represents the average output of distributed renewable energy sources.

[0035] The operational reserve matching constraint stipulates that the spinning reserves reduced by the retirement of carbon-intensive power generation equipment must be compensated by the increase in distributed energy storage capacity. The mathematical expression for the operational reserve matching constraint is:

[0036] in, This indicates the newly added distributed energy storage capacity. This indicates the capacity of carbon-intensive power generation equipment to be decommissioned. This represents the rotating reserve compensation coefficient. The operational reserve matching constraint is used as the action boundary condition of the investment planning decision sub-model, filtering out distributed renewable energy construction actions and carbon-intensive power generation equipment decommissioning actions that do not meet the action boundary conditions.

[0037] This embodiment details the implementation of the investment planning decision-making sub-model, including the specific content of distribution network investment planning and low-carbon resource planning, as well as the construction method of the state space, action space, and transition probability based on a partially observable Markov decision process. By introducing output intermittency characterization parameters and operational reserve matching constraints, physical operational safety can be ensured, avoiding the problem of insufficient spinning reserves caused by the retirement of carbon-intensive power generation equipment.

[0038] In this embodiment, the mapping relationship between the state space and action space of the investment planning decision sub-model is shown in Table 2.

[0039] Table 2. Mapping of State Space and Action Space of Investment Planning Decision Submodel

[0040] refer to Figure 4 In one embodiment, a state space, action space, and transition probabilities based on a partially observable Markov decision process are constructed for the carbon electricity trading decision sub-model. The state space of the carbon electricity trading decision sub-model includes the market electricity price state, the carbon allowance price state, and the carbon electricity demand state of market participants. The market electricity price state consists of the day-ahead market electricity price, the real-time market electricity price, and the medium- and long-term contract electricity price. The carbon allowance price state consists of the national carbon market allowance price and the regional carbon market allowance price. The carbon electricity demand state of market participants consists of each market participant's monthly electricity consumption demand, monthly carbon emission demand, and electricity elasticity coefficient.

[0041] The action space of the carbon electricity trading decision-making sub-model includes differentiated combinations of traded electricity volume and carbon emission trading quotas allocated to market participants. Differentiated combinations of traded electricity volume include different proportions of base load, peak load, and off-peak load. Carbon emission trading quotas include different proportions of quotas allocated free of charge and quotas purchased for a fee.

[0042] The transition probability of the carbon electricity trading decision-making sub-model represents the probability distribution of the transition from the market electricity price state and the carbon quota price state to the next time series under differentiated combination of traded electricity volume and carbon emission rights trading quota. The transition probability is calculated through time series analysis and machine learning methods. For the market electricity price state, the transition probability is affected by factors such as market supply and demand, fuel prices, and renewable energy output. For the carbon quota price state, the transition probability is affected by factors such as the total amount of carbon quotas, carbon emission intensity targets, and emission reduction costs for market participants.

[0043] The reward function of the carbon electricity trading decision-making sub-model includes the difference between carbon electricity trading revenue and the penalty value for carbon electricity trading default risk. The mathematical expression of the reward function is:

[0044] in, Indicates at time In state And perform the action Instant rewards at any time Indicates the number of market entities. Indicates to the first The price at which individual market entities sell electricity. Indicates to the first The amount of electricity sold by each market entity Indicates to the first The price at which individual market entities sell carbon emission rights Indicates to the first Carbon emission allowances sold by individual market entities Indicates the risk penalty coefficient. This indicates the default risk value in carbon electricity trading.

[0045] When generating differentiated combined trading volumes and carbon emission rights trading quotas within the action space, a carbon-electricity coupling price formation mechanism is established. This mechanism dynamically adjusts the settlement rate for differentiated combined trading volumes based on the market participants' carbon emission intensity and electricity elasticity coefficient. The mathematical expression for the carbon-electricity coupling price is:

[0046] in, Indicates the first The price of carbon-electric coupling for individual market participants Indicates the benchmark electricity price. Indicates the carbon emission intensity adjustment coefficient. Indicates the first Carbon emission intensity of each market entity This indicates the electrical elasticity adjustment coefficient. Indicates the first Electricity elasticity coefficient of each market entity.

[0047] The social welfare function of market participants is calculated by combining the interaction process of a multi-agent game. The social welfare function includes two components: consumer surplus and producer surplus. The mathematical expression of the social welfare function is:

[0048] in, Indicates social welfare. Indicates the first Electricity demand function of each market participant Indicates the first The power supply function of a power generator. Indicates the number of distributors. Indicates the first The power supply of a power generator. Indicates the first The electricity price for individual power generators.

[0049] Maximizing the social welfare function serves as the action optimization constraint for the carbon electricity trading decision sub-model, guiding the allocation of differentiated combinations of traded electricity and carbon emission rights trading quotas. During the action optimization process, all possible combinations of differentiated traded electricity and carbon emission rights trading quotas are traversed, the social welfare function value corresponding to each combination is calculated, and the combination with the largest social welfare function value is selected as the optimal action.

[0050] This embodiment details the implementation of the carbon electricity trading decision-making sub-model, including the method for constructing the state space, action space, and transition probabilities based on a partially observable Markov decision process. By establishing a carbon-electricity coupled price formation mechanism, settlement rates can be dynamically adjusted based on the carbon emission intensity and electricity elasticity coefficient of market participants. By using the maximization of the social welfare function as a constraint for action optimization, the market can be prevented from falling into suboptimal solutions.

[0051] In this embodiment, the differentiated combination parameters for carbon electricity trading for different market participants are shown in Table 3.

[0052] Table 3. Differentiated Combination Parameters for Carbon Electricity Trading for Different Market Participants

[0053] refer to Figure 5In one embodiment, a state space, action space, and transition probabilities based on a partially observable Markov decision process are constructed for the physical operation decision sub-model. The state space of the physical operation decision sub-model includes the dispatchable resource output state, branch power flow state, and node voltage state. The dispatchable resource output state consists of the charging and discharging power of distributed energy storage, the actual output of distributed renewable energy, and the regulating power of adjustable loads. The branch power flow state consists of the active and reactive power of all branches in the distribution network. The node voltage state consists of the voltage amplitude and phase angle of all nodes in the distribution network.

[0054] The action space of the physical operation decision-making sub-model includes distributed energy storage charging and discharging regulation actions and distributed renewable energy output reduction actions. Distributed energy storage charging and discharging regulation actions include the charging and discharging power and time for each distributed energy storage unit. Distributed renewable energy output reduction actions include the output reduction ratio and reduction time for each distributed renewable energy unit.

[0055] The transition probabilities of the physical operation decision sub-model represent the probability distribution of the transition of branch power flow state and node voltage state to the next time series under distributed energy storage charging and discharging regulation actions and distributed renewable energy output reduction actions. The transition probabilities are calculated using power system power flow calculations and Monte Carlo simulations. For distributed energy storage charging and discharging regulation actions, the transition probabilities are determined by the energy storage's charging and discharging efficiency and state of charge. For distributed renewable energy output reduction actions, the transition probabilities are determined by the renewable energy output prediction accuracy and reduction response speed.

[0056] The reward function of the physical operation decision sub-model consists of the sum of the physical operation scheduling cost and the physical overrun penalty value. The mathematical expression of the reward function is:

[0057] in, Indicates at time In state And perform the action Instant rewards at any time Indicates the physical operation and scheduling cost. This represents the penalty coefficient for exceeding the limit. This represents the physical limit penalty value.

[0058] For the physical limit violation penalty value of the physical operation decision sub-model, the corresponding branch power flow limit violation index and node voltage limit violation index are extracted. The mathematical expression of the branch power flow limit violation index is as follows:

[0059] in, Indicates the first The power flow of the branch road exceeds the limit indicator. Indicates the first The actual active power of the branch circuit Indicates the first The maximum permissible active power of the branch circuit.

[0060] The mathematical expression for the node voltage over-limit index is:

[0061] in, Indicates the first Voltage over-limit indicators for each node Indicates the first The actual voltage amplitude of each node Indicates the first The reference voltage amplitude of each node, This indicates the maximum permissible voltage deviation.

[0062] An operational flexibility assessment matrix is ​​constructed based on branch power flow over-limit indicators and node voltage over-limit indicators. The mathematical expression of the operational flexibility assessment matrix is:

[0063] in, This represents the operational flexibility assessment matrix. Indicates the number of branches in the distribution network. Indicates the number of nodes in the distribution network. They represent the first The power flow of the branch road exceeds the limit indicator. They represent the first Voltage over-limit indicators for each node.

[0064] The operational flexibility assessment matrix is ​​mapped to the investment planning decision-making sub-model. When the operational flexibility assessment matrix is ​​lower than the operational flexibility benchmark, the distribution network line expansion and distributed renewable energy investment actions in the action space of the investment planning decision-making sub-model are dynamically pruned, eliminating action combinations that cause the operational flexibility assessment matrix to fall below the operational flexibility benchmark. The operational flexibility benchmark is determined through historical operational data statistics and expert experience.

[0065] refer to Figure 6The investment planning decision-making sub-model, carbon electricity trading decision-making sub-model, and physical operation decision-making sub-model are mapped to consortium blockchain nodes, respectively. Collaborative multi-center federated learning based on the consortium blockchain is executed through parameter interaction between these nodes. The gradients of the policy network are computed locally for each of the three sub-models. The policy network employs a deep neural network structure, comprising an input layer, hidden layers, and an output layer. The input layer receives observations of the observable Markov decision process, the hidden layers extract and transform features from the observations, and the output layer outputs the probability distribution of the actions.

[0066] The mathematical expression for the gradient of the policy network is:

[0067] in, Indicates policy network parameters gradient, Representation strategy The state distribution under the following conditions Indicates the state Select action The probability, Indicates the state Next action The action value function.

[0068] The policy network gradient is homomorphically encrypted and then broadcast to the consortium blockchain nodes. The homomorphic encryption uses the Paillier encryption algorithm. The public key for the Paillier encryption algorithm is... The private key is The mathematical expression for the encryption process is:

[0069] in, Indicates ciphertext, Indicates plain text, Represents a random number.

[0070] The policy network gradient verified through consensus is aggregated within the consortium blockchain nodes. The consensus mechanism employs a practical Byzantine fault-tolerant algorithm. The mathematical expression for gradient aggregation is:

[0071] in, Represents the global gradient. Indicates the number of nodes in the consortium blockchain. Indicates the first The local gradient of each node.

[0072] The strategy network parameters of the investment planning decision sub-model, carbon electricity trading decision sub-model, and physical operation decision sub-model are updated based on the aggregated global gradient. The parameter update employs the stochastic gradient descent algorithm. The mathematical expression for parameter update is:

[0073] in, This represents the updated policy network parameters. This represents the policy network parameters before the update. This represents the learning rate.

[0074] To address the long-term decision variables of the investment planning decision-making sub-model and the short-term decision variables of the carbon electricity trading decision-making sub-model and the physical operation decision-making sub-model, a multi-timescale alignment mechanism is introduced. A timestamp is added to the policy network gradient. The timestamp contains the start and end times of the decision cycle. Consortium blockchain nodes asynchronously aggregate the policy network gradients corresponding to the long-term and short-term decision variables based on the timestamp.

[0075] The mathematical expression for asynchronous gradient aggregation is:

[0076] in, Indicates time global gradient, Represents long-period gradient weights. Indicates time Gradient of long-period decision variables, Indicates the number of short-cycle decisions. Indicates the first The gradient of sub-short cycle decision-making.

[0077] The cross-chain parameter verification node verifies the consistency of timestamp identifiers. Independent of other nodes in the consortium blockchain network, the cross-chain parameter verification node is responsible for verifying the validity of timestamp identifiers across different time scales. When timestamp identifiers are inconsistent, gradient aggregation is rejected, and an anomaly alert is sent to the consortium blockchain network. This facilitates cross-timescale collaborative updates of strategy network parameters between the investment planning decision sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model.

[0078] This embodiment details the implementation of the physical operation decision sub-model and the consortium blockchain collaborative multi-center federated learning. By constructing an operational flexibility evaluation matrix and mapping it to the investment planning decision sub-model, investment actions that do not meet the flexibility benchmark can be avoided. By introducing a multi-timescale alignment mechanism, the decision variables for long-term investment planning and short-term transaction operations can be aligned and collaboratively updated.

[0079] In this embodiment, the consortium blockchain node types and functional permissions are shown in Table 4.

[0080] Table 4. Consortium Blockchain Node Types and Functional Permissions

[0081] In one embodiment, an integrated decision-making framework for investment planning, carbon electricity trading, and physical operation is constructed. The investment planning decision-making sub-model operates on a 5-year decision-making cycle, updating the investment plan annually. The carbon electricity trading decision-making sub-model operates on a 1-year decision-making cycle, updating the trading plan monthly. The physical operation decision-making sub-model operates on a 1-day decision-making cycle, updating the scheduling plan hourly. The three sub-models interact with each other through a unified data interface. The data interface employs a standardized communication protocol to ensure the real-time performance and reliability of data transmission.

[0082] State spaces, action spaces, and transition probabilities based on partially observable Markov decision-making processes are constructed for the investment planning decision-making sub-model, the carbon power trading decision-making sub-model, and the physical operation decision-making sub-model, respectively. The investment planning decision-making sub-model has a state space containing 100 discrete states and an action space containing 50 discrete actions. The carbon power trading decision-making sub-model has a state space containing 200 discrete states and an action space containing 100 discrete actions. The physical operation decision-making sub-model has a state space containing 500 discrete states and an action space containing 200 discrete actions. The transition probability matrices are trained using historical data, including distribution network operation data, carbon power trading data, and market data from the past 10 years.

[0083] The investment planning decision-making sub-model, carbon electricity trading decision-making sub-model, and physical operation decision-making sub-model are each mapped to a consortium blockchain node. The consortium blockchain network adopts a permissioned blockchain architecture, allowing only authorized nodes to join the network. Each node has a unique digital identity certificate for authentication and data encryption. Communication between nodes is peer-to-peer, ensuring the security and privacy of data transmission.

[0084] Collaborative multi-center federated learning based on the consortium blockchain is executed through parameter interaction between consortium blockchain nodes. Each node trains its own policy network model locally, sharing only encrypted model gradients without sharing the original data. The training process uses batch gradient descent with a batch size of 32. The training consists of 1000 rounds, with gradient aggregation and parameter updates performed after each round.

[0085] The updated strategy network parameters are used to output medium- and long-term operational coordination decisions for power distribution companies. These decisions include a 5-year power distribution network investment plan, a 1-year carbon electricity trading plan, and a 1-month physical operation and scheduling plan. The decision outputs are presented in a structured document format, containing detailed parameter descriptions and implementation guidelines.

[0086] This embodiment provides specific parameter settings and implementation procedures for an integrated decision-making framework encompassing investment planning, carbon electricity trading, and physical operation, which can be directly applied to actual power distribution company operational decision-making scenarios. By employing a permissioned blockchain architecture and encrypted communication technology, data security and privacy protection are ensured. Through batch gradient descent algorithms and multi-round training, the accuracy and stability of the policy network model are improved.

[0087] In one embodiment, an integrated decision-making framework for investment planning, carbon electricity trading, and physical operation is constructed. The investment planning decision-making sub-model includes a distribution network investment planning module, a low-carbon resource planning module, and an investment benefit assessment module. The distribution network investment planning module uses a mixed-integer linear programming method to solve for the optimal line expansion scheme. The low-carbon resource planning module uses a multi-objective optimization method to solve for the optimal distributed renewable energy construction and carbon-intensive power generation equipment decommissioning schemes. The investment benefit assessment module uses the net present value method and the internal rate of return method to evaluate the economic benefits of the investment schemes.

[0088] The carbon electricity trading decision-making sub-model includes a market forecasting module, a trading strategy generation module, and a risk assessment module. The market forecasting module uses a Long Short-Term Memory (LSTM) network to predict future market electricity prices and carbon allowance prices. The trading strategy generation module uses reinforcement learning to generate optimal differentiated carbon electricity trading schemes. The risk assessment module uses the Conditional Value at Risk (VaR) method to evaluate the risk level of the trading schemes.

[0089] The physical operation decision-making sub-model comprises a power flow calculation module, a schedulable resource optimization module, and an operation status monitoring module. The power flow calculation module uses the Newton-Raphson method to calculate the power flow distribution of the distribution network. The schedulable resource optimization module uses model predictive control to solve for the optimal schedulable resource allocation scheme. The operation status monitoring module monitors the operation status of the distribution network in real time, promptly detecting and handling abnormal situations.

[0090] For the investment planning decision-making sub-model, carbon electricity trading decision-making sub-model, and physical operation decision-making sub-model, state space, action space, and transition probabilities based on partially observable Markov decision processes are constructed respectively. The state space adopts continuous state representation and is modeled using a Gaussian process. The action space adopts continuous action representation and is solved using a deterministic policy gradient algorithm. The transition probabilities are fitted using a neural network and trained using a supervised learning method.

[0091] The investment planning decision-making sub-model, carbon electricity trading decision-making sub-model, and physical operation decision-making sub-model are mapped to consortium blockchain nodes, respectively. The consortium blockchain network employs sharding technology to improve transaction processing speed and system throughput. Each shard is responsible for handling gradient aggregation and parameter update tasks for a portion of the nodes. Shards interact and collaborate with each other through a cross-shard communication protocol.

[0092] Collaborative multi-center federated learning based on consortium blockchain is executed through parameter interaction between consortium blockchain nodes. A federated averaging algorithm is used for gradient aggregation, and differential privacy technology is employed to protect the privacy of model gradients. An adaptive learning rate adjustment mechanism is introduced during training to dynamically adjust the learning rate based on model convergence, thereby improving training efficiency and model accuracy.

[0093] The updated strategy network parameters are used to output medium- and long-term operational collaborative decisions for power distribution and sales companies. The decision output includes multiple alternatives, each with corresponding economic, environmental, and risk level assessments. Power distribution and sales companies can choose the most suitable decision based on their own circumstances and development goals.

[0094] This embodiment provides a modular implementation method for an integrated decision-making framework encompassing investment planning, carbon electricity trading, and physical operation. Each module can be developed and upgraded independently. By employing continuous state and action representations, the uncertainties and dynamics of the decision-making process can be described more accurately. The use of sharding and differential privacy techniques improves system performance and security.

Claims

1. A method for carbon power operation and investment coordination decision based on a consortium chain, characterized in that, Includes the following steps: An integrated decision-making framework for investment planning, carbon power trading, and physical operation is constructed. This framework includes an investment planning decision-making sub-model, a carbon power trading decision-making sub-model, and a physical operation decision-making sub-model. The investment planning decision-making sub-model provides capacity conditions and market space for the carbon power trading decision-making sub-model. The operating benefits and risks of the carbon power trading decision-making sub-model are fed back to the investment planning decision-making sub-model. The physical operation decision-making sub-model verifies the feasibility of the investment planning decision-making sub-model and influences the competitive operation level of the carbon power trading decision-making sub-model. For the investment planning decision sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model, respectively, a state space, action space, and transition probability based on a partially observable Markov decision process are constructed; The investment planning decision sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model are respectively mapped to consortium blockchain nodes. Through parameter interaction between the consortium blockchain nodes, collaborative multi-center federated learning based on the consortium blockchain is performed to update the strategy network parameters of the investment planning decision sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model. Based on the updated strategy network parameters, output the medium- and long-term operational collaborative decision-making of power distribution and sales companies, and complete the carbon power operation and investment collaborative decision-making method based on consortium blockchain.

2. The carbon electricity operation and investment coordination decision-making method based on the alliance chain according to claim 1, characterized in that, In the steps of constructing an integrated decision-making framework for investment planning, carbon electricity trading, and physical operation, the investment planning decision-making sub-model includes distribution network investment planning and low-carbon resource planning. The distribution network investment planning generates node investment prices, which are allocated according to the occupancy of the distribution network by each market participant. The low-carbon resource planning targets the market participants for distributed renewable energy planning and carbon-intensive power generation equipment retirement plans. The carbon electricity trading decision-making sub-model conducts differentiated combined carbon electricity trading with the market participants based on the node investment prices and the output of the low-carbon resource planning. The physical operation decision-making sub-model is executed by a distribution network operator independent of market trading. The distribution network operator optimizes the allocation of dispatchable resources based on the differentiated combined carbon electricity trading and feeds back the flexibility index of the optimized allocation of dispatchable resources to the investment planning decision-making sub-model.

3. The alliance chain-based carbon electricity operation and investment coordination decision-making method according to claim 1, characterized in that, In the steps of constructing the state space, action space, and transition probabilities based on a partially observable Markov decision process for the investment planning decision sub-model, the state space includes the distribution network topology state, low-carbon resource capacity state, and load demand state; the action space includes distribution network line expansion actions, distributed renewable energy construction actions, and carbon-intensive power generation equipment decommissioning actions; the transition probabilities represent the probability distribution of the distribution network topology state and the low-carbon resource capacity state transitioning to the next time series under the distribution network line expansion action, the distributed renewable energy construction action, and the carbon-intensive power generation equipment decommissioning action; and the reward function of the investment planning decision sub-model includes the difference between the discounted investment cost and the nodal investment price return.

4. The carbon-electricity transportation and investment collaborative decision-making method based on consortium blockchain as described in claim 2, characterized in that, In the steps of constructing the state space, action space, and transition probability based on a partially observable Markov decision process for the carbon electricity trading decision sub-model, the state space includes the market electricity price state, the carbon quota price state, and the carbon electricity demand state of the market participants. The action space includes the differentiated combination of trading electricity and carbon emission rights trading quota allocated to the market participants. The transition probability represents the probability distribution of the market electricity price state and the carbon quota price state transitioning to the next time series under the differentiated combination of trading electricity and the carbon emission rights trading quota. The reward function of the carbon electricity trading decision sub-model includes the difference between carbon electricity trading revenue and the carbon electricity trading default risk penalty value.

5. The carbon-electricity transportation and investment collaborative decision-making method based on consortium blockchain as described in claim 3, characterized in that, In the steps of constructing the state space, action space, and transition probability based on a partially observable Markov decision process for the physical operation decision sub-model, the state space includes the schedulable resource output state, branch power flow state, and node voltage state; the action space includes distributed energy storage charging and discharging regulation actions and distributed renewable energy output reduction actions; the transition probability represents the probability distribution of the branch power flow state and the node voltage state transitioning to the next time series under the distributed energy storage charging and discharging regulation actions and the distributed renewable energy output reduction actions; and the reward function of the physical operation decision sub-model includes the sum of physical operation scheduling cost and physical over-limit penalty value.

6. The carbon-electricity transportation and investment collaborative decision-making method based on consortium blockchain according to claim 1, characterized in that, The investment planning decision sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model are respectively mapped to consortium blockchain nodes. In the step of executing the collaborative multi-center federated learning based on the consortium blockchain through parameter interaction between the consortium blockchain nodes, the investment planning decision sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model calculate the policy network gradient locally. After homomorphically encrypting the policy network gradient, it is broadcast to the consortium blockchain nodes. The policy network gradient verified by consensus is aggregated in the consortium blockchain nodes. The policy network parameters of the investment planning decision sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model are updated based on the aggregated global gradient.

7. The collaborative decision-making method for carbon-based power generation and transportation investment based on consortium blockchain as described in claim 3, characterized in that, In the steps of constructing the state space, action space, and transition probability based on a partially observable Markov decision process for the investment planning decision sub-model, when executing the distributed renewable energy construction action and the carbon-intensive power generation equipment decommissioning action within the action space, an output intermittency characterization parameter and an operating reserve matching constraint are introduced. The output intermittency characterization parameter quantifies the volatility brought about by the distributed renewable energy construction action, and the operating reserve matching constraint limits the spinning reserve reduced by the carbon-intensive power generation equipment decommissioning action to be compensated by the newly added distributed energy storage capacity. The operating reserve matching constraint is used as the action boundary condition of the investment planning decision sub-model, and distributed renewable energy construction actions and carbon-intensive power generation equipment decommissioning actions that do not meet the action boundary condition are filtered out.

8. The collaborative decision-making method for carbon power transportation and investment based on consortium blockchain as described in claim 4, characterized in that, In the step of constructing the state space, action space, and transition probability of the carbon electricity trading decision sub-model based on a partially observable Markov decision process, when generating the differentiated combined trading volume and the carbon emission rights trading quota within the action space, a carbon-electricity coupled price formation mechanism is established. This mechanism dynamically adjusts the settlement rate of the differentiated combined trading volume based on the carbon emission intensity and electricity elasticity coefficient of the market participants, and calculates the social welfare function of the market participants in conjunction with a multi-participant game interaction process. Maximizing the social welfare function is used as the action optimization constraint of the carbon electricity trading decision sub-model to guide the allocation of the differentiated combined trading volume and the carbon emission rights trading quota.

9. The carbon-electricity transportation and investment collaborative decision-making method based on consortium blockchain as described in claim 5, characterized in that, In the step of constructing the state space, action space, and transition probability based on a partially observable Markov decision process for the physical operation decision sub-model, for the physical limit penalty value of the physical operation decision sub-model, the branch power flow limit index and node voltage limit index corresponding to the physical limit penalty value are extracted. Based on the branch power flow limit index and node voltage limit index, an operation flexibility evaluation matrix is ​​constructed. The operation flexibility evaluation matrix is ​​mapped to the investment planning decision sub-model. When the operation flexibility evaluation matrix is ​​lower than the operation flexibility benchmark, the distribution network line expansion action and the distributed renewable energy investment action in the action space of the investment planning decision sub-model are dynamically pruned, and action combinations that cause the operation flexibility evaluation matrix to be lower than the operation flexibility benchmark are eliminated.

10. The collaborative decision-making method for carbon power transportation and investment based on consortium blockchain as described in claim 6, characterized in that, In the step of performing collaborative multi-center federated learning based on consortium blockchain through parameter interaction between consortium blockchain nodes, a multi-timescale alignment mechanism is introduced for the long-term decision variables of the investment planning decision sub-model and the short-term decision variables of the carbon electricity trading decision sub-model and the physical operation decision sub-model. A timestamp identifier is added to the strategy network gradient. The consortium blockchain nodes asynchronously aggregate the strategy network gradients corresponding to the long-term decision variables and the short-term decision variables according to the timestamp identifier. The consistency of the timestamp identifier is verified by a cross-chain parameter verification node, thus completing the cross-timescale collaborative update of strategy network parameters between the investment planning decision sub-model, the carbon electricity trading decision sub-model, and the physical operation decision sub-model.

Images