A green port integrated energy system collaborative optimization scheduling method

Abstract

This invention discloses a collaborative optimization scheduling method for a green port integrated energy system, comprising the following steps: S1: Constructing a multidimensional mathematical model of the green port integrated energy system, including an electric truck load model; introducing a maximum charging delay index into the electric truck load model, and constructing a charging and discharging priority queue based on the maximum charging delay index; S2: Constructing a multi-objective comprehensive utility function, including comprehensive economic cost, environmental cost, system stability cost, and rigid constraint penalty term; S3: Constructing a two-layer collaborative optimization framework, including an upper-layer parameter evolution layer and a lower-layer real-time control layer; This solution completely breaks through the bottleneck of existing port scheduling technology, providing an efficient, robust, economical, and environmentally friendly energy management strategy for green and smart ports.

Description

Technical Field

[0001] This invention relates to the field of port energy management technology, specifically to a collaborative optimization scheduling method for a green port integrated energy system. Background Technology

[0002] With the increasing severity of the global energy crisis and environmental pollution, and the deepening implementation of the national "dual-carbon" strategy, the green transformation of the transportation industry, as a key area of ​​carbon emissions, is urgently needed. Ports, as core hubs of the global logistics supply chain and centers of energy consumption, have their energy structure optimization directly impacting the effectiveness of emission reduction in the transportation sector. According to data from the International Association of Ports and Harbors (IAPH), port carbon emissions account for a large proportion of total shipping emissions, and this proportion is showing an upward trend year by year. To address this challenge, the construction of green and smart ports, centered on "electricity substitution" and "clean energy integration," has become an industry consensus.

[0003] Modern large ports are undergoing a profound energy transformation. On the one hand, port areas have abundant wind and solar energy resources, and large-scale distributed photovoltaic and wind turbine generators are being connected to the port microgrid, transforming the port from a simple energy consumer into a "prosumer." On the other hand, the electrification level of port machinery and equipment is constantly improving, with a large number of quay cranes, yard cranes, and port horizontal transport equipment replacing traditional fuel-powered equipment. For example, Shenzhen Mawan Port has put hundreds of electric container trucks into operation, and the coverage rate of the shore power system has reached more than 90%.

[0004] However, this dual transformation of energy and load has brought unprecedented challenges to the port energy management system. First, the port load has significant "rigidity" and "fluctuation" characteristics. Port operations are greatly affected by ship schedules, and the load curve often exhibits a dramatic "double peak" characteristic, with a huge peak-to-valley difference. Furthermore, the random access of high-power shore power and fast-charging loads for container trucks can cause instantaneous impacts on the power grid. Second, renewable energy generation has inherent intermittency and randomness. Photovoltaic power generation is concentrated at noon, while wind power output is greater at night, which is a serious "temporal and spatial mismatch" with the peak periods of port load.

[0005] This supply-demand mismatch has led to serious consequences: on the source side, due to insufficient absorption capacity, wind and solar power curtailment occurs frequently, resulting in a waste of clean energy; on the grid side, in order to maintain power balance, ports have to purchase electricity from the main grid at high prices during peak hours, leading to a surge in operating costs and increasing the peak-shaving pressure on the upper-level grid; on the load side, the disorderly charging of electric trucks not only exacerbates the peak load of the grid, but may also affect the efficiency of logistics operations due to charging queues; therefore, how to achieve coordinated and optimized scheduling of port "source-grid-load-storage" elements in a multi-energy coupled and highly uncertain environment is a technical problem that urgently needs to be solved.

[0006] Existing optimized dispatch schemes for port integrated energy systems (PIES) or microgrids can be mainly summarized into two major technical routes: The first category consists of model-based optimization methods, such as Mixed Integer Linear Programming (MILP) or Model Predictive Control (MPC). The core idea of ​​this approach is to construct the scheduling problem as a rigorous mathematical programming model, typically using mixed integer linear programming. However, this type of method has the following drawbacks: 1. Computational complexity grows exponentially: Port PIES has a large number of devices and complex constraints. When using the MILP method, a large number of binary variables need to be introduced; as the system scale increases, the computation time required for solving the problem grows exponentially, often making it difficult to obtain an accurate solution within the minute-level scheduling cycle required by the port, which seriously fails to meet the real-time requirements.

[0007] 2. Difficulty in nonlinear modeling: Battery life loss models and nonlinear output curves of wind power generation require extensive linearization approximation in MILP. This approximation introduces model errors, leading to discrepancies between the economic efficiency of scheduling results and actual operating performance.

[0008] 3. Limited Uncertainty Handling: Although MPC can roll over and correct prediction errors, its robustness largely depends on the accuracy of the prediction model. For drastic random fluctuations in wind and solar power output and load, standard deterministic MILP / MPC lacks flexible adaptive capabilities. While some robust optimization methods can handle uncertainty, they typically come at the cost of sacrificing economic efficiency (conservative scheduling).

[0009] The second category consists of intelligent algorithms based on empirical learning, such as traditional heuristic algorithms PSO / GA or standard reinforcement learning DQN / DDPG. These approaches aim to approximate the optimal control policy through search or learning, and are suitable for complex systems where it is difficult to establish accurate analytical models. However, these methods have the following drawbacks: 1. Traditional heuristic algorithms are prone to getting trapped in local optima: Algorithms such as PSO and GA are essentially random search methods based on swarm intelligence. The objective function of a port energy system is a complex, high-dimensional, non-convex, multi-peaked surface. During the search process, these algorithms are easily captured by suboptimal solutions (local optima) on the objective function surface, leading to premature convergence. For example, the algorithm may find a strategy that minimizes electricity purchase costs at a certain time, but at the expense of battery life or due to wind curtailment, failing to find the globally optimal economic-environmental balance.

[0010] 2. Parameter Sensitivity: The performance of deep reinforcement learning algorithms such as DQN and DDPG is highly sensitive to hyperparameters such as initial neural network connection weights, learning rate, and discount factor. The selection of these parameters often relies on manual experience or time-consuming grid search. If the parameters are set improperly, it may lead to divergence or extremely slow convergence during the training process.

[0011] 3. Low cold start efficiency: In the early stages of training, the agent lacks experience and requires a large amount of random trial and error to begin learning. In a large-scale, high-dimensional state space like a port, the initial training cycle is extremely long, lacking the engineering capabilities for rapid deployment and iteration.

[0012] 4. Action Space and Continuity Issues: DQN is only applicable to discrete actions: Energy storage charging and discharging power, and interaction power with the main grid are continuous variables. DQN must discretize the action space, which leads to reduced control accuracy and makes it impossible to achieve fine-grained power regulation; Poor convergence stability of continuous action algorithms (DDPG / A2C): Although algorithms such as DDPG can handle continuous actions, their training process is often more difficult to converge, and the design requirements for the reward function are extremely high. Once there is a slight flaw in the design of the reward function, it may cause the control action output by the Actor network to oscillate or diverge.

[0013] Therefore, the application of existing technologies in port energy systems faces a triple dilemma of "long solution time, low scheduling accuracy, and poor robustness," and there is an urgent need for a hybrid intelligent algorithm that can combine the advantages of global optimization and online control.

[0014] Furthermore, existing technologies for port load management lack a mechanism to incorporate flexible loads as adjustable resources into active scheduling. The main technical solutions focus on meeting rigid operational demands, employing rule-driven or simple time-of-use pricing. Prioritizing load demand, they neglect the load's adjustment potential and the real-time interaction needs of the power grid. This rule-based or passive response strategy fails to achieve deep coupling and synergistic optimization between the logistics and energy systems. Due to its passivity and rule-based nature, it introduces significant technical and economic drawbacks. These drawbacks are the fundamental reason for proposing the "maximum charging delay index" and the two-layer collaborative framework in this invention. The main drawbacks are: triggering a "peak drift" effect, idle flexible resource potential, separation of the logistics and energy systems, and difficulty in adapting to fluctuations in new energy sources.

[0015] 1. Triggering the "Peak Load Shifting" Effect: While a simple time-of-use (TOU)-based pricing strategy can avoid charging during peak electricity prices, it can lead to all electric trucks simultaneously charging at the start of a price trough (e.g., 11 PM). This large-scale synchronous charging creates a new artificial load peak, known as "peak load shifting." This shifted peak may exceed the capacity of substations or distribution network branches, causing transformer overload, reduced equipment lifespan, and even triggering voltage drops or faults in the regional power grid, severely impacting the safe and stable operation of the power grid.

[0016] 2. Idle Flexible Resource Potential: Electric trucks, as mobile energy storage units, possess V2G (vehicle-to-grid) and delayed charging potential. Current technology treats trucks solely as energy consumers, neglecting their ability to absorb energy as a "sponge" during periods of wind and solar power surplus, and their capacity to discharge as a "backup power source" during sudden system power outages. This passive management approach causes ports to miss opportunities to utilize truck fleets as virtual power plants (VPPs) to participate in grid ancillary services and peak shaving, directly wasting significant economic and regulatory potential.

[0017] 3. The disconnect between logistics and energy systems: Current dispatching only focuses on the EV's State of Charge (SOC) and the grid's static electricity price, failing to incorporate logistics efficiency into dispatching decisions. This may lead to: 4. Overcharging: Highly flexible EVs occupy charging resources, but remain at charging stations for extended periods after being fully charged, reducing the utilization rate of charging stations.

[0018] 5. Operational risks: EVs with low flexibility (such as vehicles about to begin high-intensity operations) may fail to reach the target SOC within the specified time due to resource occupation, directly affecting the service level (QoS) of port operations.

[0019] 6. Difficulty adapting to renewable energy fluctuations: The lack of an active load control mechanism means that when renewable energy (wind and solar) output changes abruptly, the system can only rely on expensive energy storage systems or quickly purchase electricity from the main grid to balance power. If electric trucks could adjust power based on real-time renewable energy forecasts (e.g., increasing charging power during peak solar output), the renewable energy absorption rate could be significantly improved, and curtailment losses reduced.

[0020] In summary, there is an urgent need for a port energy system optimization and scheduling method that can balance global optimization and real-time control, possess multi-objective collaborative capabilities, and fully tap the potential of flexible load adjustment. Summary of the Invention

[0021] To address the shortcomings of existing technologies, this invention provides a collaborative optimization scheduling method for a green port integrated energy system. The method aims to achieve real-time optimal coordination of all elements of the port, including "source-grid-load-storage," in terms of minimizing economic costs, maximizing environmental benefits, and stabilizing system operation. Its core lies in constructing a hierarchical intelligent decision-making architecture, which solves the inherent defects of traditional single algorithms in dealing with high-dimensional uncertainties in port integrated energy systems.

[0022] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A collaborative optimization scheduling method for a green port integrated energy system is provided, which includes the following steps: S1: Construct a multi-dimensional mathematical model of a green port integrated energy system, which includes an electric truck load model; introduce a maximum charging delay index into the electric truck load model to quantify the charging time margin of each electric truck, and construct a charging and discharging priority queue based on the maximum charging delay index. S2: Construct a multi-objective comprehensive utility function, which includes comprehensive economic cost, environmental cost, system stability cost, and rigid constraint penalty term; S3: Construct a two-layer collaborative optimization framework, which includes an upper parameter evolution layer and a lower real-time control layer. The upper parameter evolution layer uses the biologically enhanced dung beetle optimization algorithm to perform offline global optimization, optimizing the hyperparameters of the flexible motion evaluation algorithm and the multi-objective weights of the multi-objective comprehensive utility function in the lower real-time control layer. The lower real-time control layer uses the flexible motion evaluation algorithm and takes the multi-objective comprehensive utility function as the optimization objective to perform online rolling time-domain control, and finally outputs real-time power scheduling commands.

[0023] Furthermore, the formula for calculating the maximum charging delay index in step S1 is as follows: ;

[0024] in, For the first i electric vehicles t The maximum charging delay at any given moment indicates the amount of time an electric vehicle can be charged before it reaches the SOC (State of Charge) requirement for battery departure. , The first i The registration time of electric vehicles and the owner's desired deregistration time; T i For the first i The necessary charging time for an electric vehicle; For the first i Off-grid SOC (State of Charge) index for electric vehicles; For the first iThe SOC index of an electric vehicle at time t; Rated charging efficiency; For the first i The rated charging power of an electric vehicle; Battery capacity; The method for constructing the discharge priority queue is as follows: S11: Based on the maximum charging delay index for each electric vehicle The positive and negative signs, and magnitudes, are used to generate an ascending queue for charging priority and a descending queue for discharging priority; the method is as follows: ... Electric vehicles are placed in an ascending priority charging queue, and according to... Arrange in ascending order; Electric vehicles are assigned to a descending discharge priority queue, and according to... Arrange them in descending order; S12: By calculating the uniformity of charge index, the same For identical electric vehicles, a secondary ranking is performed, and the formula for calculating the uniformity of charge index is:

[0025]

[0026] in, , These are the charge uniformity indices of electric vehicles to be corrected in the charging and discharging priority queues, respectively. , These are the highest and lowest SOC (State of Charge) indicators for electric vehicles; electric vehicles with higher uniform charge indicators are prioritized; when uniform charge values ​​are equal, then... Smaller electric vehicles are given priority.

[0027] Furthermore, the multidimensional mathematical model also includes a power balance constraint model, a renewable energy output model, an energy storage system model, a shore power system load model, and a port crane load model.

[0028] Furthermore, the expression for the multi-objective comprehensive utility function in step S2 is:

[0029] in, , , These are the multi-objective weight coefficients optimized by the upper-level parameter evolution layer. Considering the overall economic cost; For environmental costs; Costs related to system stability; This is a rigid constraint penalty term; x( t () represents the system state vector; u ( t ) represents the system action vector.

[0030] Furthermore, the overall economic cost This includes the cost of purchasing electricity from the grid, the revenue from selling electricity to the grid, the loss costs of aging energy storage systems, and the operation and maintenance costs of renewable energy. The calculation formula is as follows:

[0031] in, These represent the power purchased from and sold to the power grid, respectively. These are real-time time-of-use electricity prices; Costs associated with the aging and depletion of energy storage batteries; The operation and maintenance costs of renewable energy; Environmental costs The carbon trading mechanism comprises the carbon emission costs incurred from purchasing electricity from the grid and the carbon emission reduction benefits from absorbing renewable energy. Its calculation formula is as follows:

[0032] in, The price to pay for carbon emissions; The price of carbon emission reduction benefits; Carbon emission factors for supplying electricity to the power grid; This serves as the port's baseline carbon emission allowance; purchase costs only arise when the allowance is exceeded. ; This refers to the total amount of new energy power generation; System stability cost It is constructed using the volatility of the power exchange between the port and the power grid, and its calculation formula is as follows:

[0033] Furthermore, rigid constraint penalty terms The expression is:

[0034] Where M is the penalty factor used to amplify the impact of constraint violation on the objective function; Because the electric vehicle failed to reach the departure time Service quality breach; This indicates the amount of violation of the state of charge constraints of the energy storage system; This indicates the amount of violation of the charging and discharging power constraints of the energy storage system; This indicates the amount of violation of service quality constraints for electric vehicles.

[0035] Furthermore, in step S3, the upper parameter evolution layer passes the parameters of each iteration to the lower real-time control layer during the offline global optimization process, and performs rolling time-domain pre-training simulation on the historical scene dataset to obtain the total running cost in the scene. This cost is then passed to the upper parameter evolution layer as the fitness value of the iteration parameters, and the upper parameter evolution layer uses the fitness value as the basis for selecting the iteration parameters. The expression for fitness value is:

[0036] in, This represents the set of parameters to be optimized in the upper layer. T sim This is the total time step of the scheduling simulation, i.e., the discrete time length of the rolling time-domain optimization; t This represents the current discrete-time step.

[0037] Furthermore, the biologically enhanced dung beetle optimization algorithm in step S3 optimizes parameters through a parabolic adaptation strategy, a Gaussian distribution parameter replacement strategy, and a Levy flight escape mechanism.

[0038] Furthermore, the flexible action evaluation algorithm approximates the optimal control strategy through the Actor-Critic dual network structure based on the current system state at each scheduling moment and the optimized parameters of the upper parameter evolution layer, outputs and executes real-time power scheduling commands, and at the same time re-collects system state and outputs commands at the next moment.

[0039] Furthermore, the flexible motion evaluation algorithm is trained using an Actor-Critic dual-network structure; The Critic network is used to receive the current system state. x ( t ) and the actions output by the Actor network u ( t It predicts the cumulative cost from the current moment to the end of the scheduling cycle, and updates the network weights by minimizing the time difference error, which is constructed based on the Bellman error. Actor networks are used to approximate the optimal control strategy; they receive the current system state. x ( t ), directly output the optimal action u ( t The network weights are updated along the gradient direction of the cumulative cost estimated by the Critic network, so that the control policy evolves in the direction of minimizing the long-term cost.

[0040] The beneficial effects of this invention are as follows: 1. This scheme improves the performance and robustness of the algorithm: by using the upper-level biological enhancement dung beetle optimization algorithm to globally optimize the hyperparameters of the lower-level flexible action evaluation algorithm, the cold start time of the algorithm is significantly shortened and the convergence speed is significantly improved. This avoids the problems of slow convergence and easy divergence of traditional deep reinforcement learning algorithms in high-dimensional dynamic systems, and enhances the generalization ability and robustness of the algorithm.

[0041] 2. This solution enables multi-objective collaborative optimization: By dynamically adjusting the weights of multiple objectives such as economic cost, environmental benefits, system stability, and equipment lifespan, this solution achieves real-time dynamic balance among multiple objectives. In typical port scenarios, operating costs are significantly reduced, energy storage battery lifespan is extended, and carbon emissions are significantly reduced, thereby significantly improving the overall performance of the scheduling strategy.

[0042] 3. This solution improves load flexibility and scheduling efficiency: It introduces a maximum charging delay index to accurately quantify the charging time margin of electric trucks, constructs a dynamic charging and discharging priority queue, realizes active scheduling of flexible loads, effectively avoids the "peak drift" phenomenon, improves the renewable energy consumption rate, significantly improves the charging efficiency of electric trucks, and at the same time ensures the level of port logistics services.

[0043] 4. This solution can optimize energy management and control carbon emissions: It constructs an environmental cost model based on a carbon trading mechanism, accurately calculates the carbon emissions from electricity purchases and the emission reduction benefits from renewable energy consumption, guides the system to prioritize the consumption of clean energy, reduces dependence on the main power grid, thereby reducing carbon emissions and improving the cleanliness and sustainability of the port's energy system.

[0044] 5. This solution can improve the stability and real-time performance of the system: By adopting a rolling time-domain control mechanism and combining it with optimized control parameters at the upper level, a robust real-time response to minute-level high-frequency fluctuations in the port energy system is achieved, significantly improving scheduling accuracy and significantly reducing power fluctuations between the port and the main power grid, thereby enhancing the system's safe and stable operation capability. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. The above and other objects, features, and advantages of the present invention will become clearer through the accompanying drawings. The same reference numerals indicate the same parts in all the drawings. The drawings are not intentionally drawn to scale to actual dimensions; the focus is on illustrating the main points of the invention.

[0046] Figure 1A flowchart illustrating the collaborative optimization scheduling method for the integrated energy system of green ports.

[0047] Figure 2 A schematic diagram illustrating the execution logic of the collaborative optimization scheduling method for the integrated energy system of green ports. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0049] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0050] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0051] Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0052] like Figure 1 and Figure 2 As shown, this scheme provides a collaborative optimization scheduling method for a green port integrated energy system, aiming to achieve real-time optimal coordination of all elements of the port, including "source-grid-load-storage", in terms of minimizing economic costs, maximizing environmental benefits, and stabilizing system operation. The core of this scheme lies in constructing a hierarchical intelligent decision-making architecture, which solves the inherent defects of traditional single algorithms in dealing with PIES high-dimensional uncertainty problems.

[0053] In practical implementation, this technical solution includes the following key and interlocking steps: Step 1: Construction of a multidimensional mathematical model for the Green Port Integrated Energy System (PIES): This step aims to establish a mathematical foundation model that accurately reflects the characteristics of energy conversion, transmission, storage, and load within the port, particularly the Maximum Charging Delay Index (MCDI) model on the flexible load side.

[0054] (1) Port system topology and power balance constraints: The port PIES is considered as a hybrid AC / DC microgrid system. At any scheduling time t (scheduling time step Δt), the system must satisfy a strict power balance constraint. This is the primary prerequisite for all scheduling decisions.

[0055] Power purchased from / sold from the main power grid (mutually exclusive). Real-time output power of photovoltaic and wind power generation. : The discharge and charging power of the energy storage system (ESS) (mutually exclusive). Port baseline load (quay cranes, storage yards, etc.) and shore power load. Total charging power of all electric trucks (EVs).

[0056] (2) Renewable Energy Output Model: In the near-water area of ​​the green port cluster, wind and solar resources are naturally abundant, and wind power generation can complement photovoltaic power generation to a certain extent. Therefore, renewable energy power generation systems represented by wind power and photovoltaic power generation can serve as important power generation equipment for the port's micro-energy system. The new energy power generation model adopted is as follows:

[0057] In the formula, for wind power generation systems, for The output power of the wind turbine at all times. for Predicted power of the wind turbine at any given time. This refers to the air density under the current atmospheric conditions. The correlation coefficient for wind energy conversion. for Wind speed at all times For wind power generation equipment, the swept cross-sectional area is used; for photovoltaic power generation systems, for Photovoltaic power output at all times for Predicted power output of photovoltaic systems at any given time. Temperature under standard conditions. PN represents the illuminance under standard conditions, and PN represents the rated power of the photovoltaic panel during operation. The temperature conversion relationship of photovoltaic panels, Photovoltaic panels Real-time operating temperature and irradiance.

[0058] (3) Energy storage system model: The power supplied by the port to the ship through the shore power system is constrained by the characteristics of the ship itself, and the charging and discharging constraints of its energy storage battery are as follows: V

[0059]

[0060]

[0061]

[0062] In the formula, is The state of charge (SOC) of the energy storage at any given moment. and These are the lower and upper limits of the energy storage state of charge, respectively; for Energy stored in a constant-time energy storage system; This refers to the rated energy of the energy storage system. for The charging and discharging power of energy stored at all times; and These refer to the charging and discharging efficiencies of the energy storage system, respectively. and These are the maximum charging power and maximum discharging power of the energy storage, respectively. (4) Port-specific conformity model, which specifically includes: (41) Shore power system load: The shore power system provides electrical energy to berthed vessels, and its load is determined by the number of berths, vessel type, and operating power.

[0063] in This represents the current number of ships docked. Let be the shore power demand of the i-th ship.

[0064] (42) Port cranes and cargo yards: The power of port quay cranes for loading and unloading is shown in (x)

[21] , which is the main form of ship berthing service.

[0065]

[0066] In the formula, The rate at which goods are loaded and unloaded; For the quality of the cargo on board; The service time of a vessel in port.

[0067] (43) Electric freight trucks: In the port power grid, orderly management of electric freight trucks is a prerequisite for electric vehicles to participate in the port power grid tracking and dispatching plan. First, the charging requirements of electric vehicles must be prioritized. Therefore, considering the characteristics of electric vehicle entry SOC, exit SOC requirements, electric vehicle battery charging and discharging power and efficiency, battery capacity, etc., a dynamic electric vehicle charging and discharging priority queue based on the maximum charging delay index is established to orderly manage the charging and discharging of electric vehicles. Then, based on this, the charging and discharging task indicators of electric vehicle batteries and grid energy storage batteries are arranged in the Energy Storage System (ESS) according to the upper-level dispatching instructions, and the indicators are fed back to the upper-level power grid. After receiving the feedback from the ESS, the port power grid calculates the updated dispatching plan in the prediction time domain through the prediction model.

[0068] (431) Maximum charging delay index for electric vehicles: Define the individual electric vehicles accessing the port The necessary charging time is:

[0069] In the formula: For off-grid SOC; for The SOC state at time t; Rated charging efficiency; for Rated charging power; For battery capacity. According to Necessary charging time, Individual network access time Off-grid time as expected by car owners Then you can get each vehicle The maximum charging delay of an individual at time t is

[0070] Maximum charging delay express The amount of time an individual battery can be charged before reaching the SOC (State of Charge) target required for user-off-site battery operation. The larger Compared to other electric vehicles, they have more time to charge. hour, The indicated charging time is sufficient for discharging tasks, and such electric vehicles should be prioritized for discharging tasks; otherwise, the charging time is insufficient. smaller Compared to other electric vehicles, it has a shorter charging time and is prioritized lower in the electric vehicle discharging task; when When, it means It needs to be charged immediately, lacks the ability to perform discharging tasks, and smaller The more urgent the charging need, the higher the charging priority. (432) Construction of electric vehicle charging and discharging priority queue: When an electric vehicle connects to the park power grid at time t, a charging and discharging priority queue for electric vehicles at time t can be established based on the maximum charging delay index mentioned above. The specific steps are as follows: Establish a charging / discharging priority queue. Calculate the charging / discharging priority of all electric vehicles connected to the park's power grid at time t. ,according to The positive and negative signs and the size are used to generate an ascending priority queue for charging. and discharge priority descending queue . The charging and discharging priority queue has been revised. To avoid frequent scheduling of some electric vehicles, a charge uniformity index is adopted, prioritizing vehicles with the same charge level. The electric vehicles were reordered:

[0071]

[0072] In the formula: , These are the priorities to be corrected in the charging and discharging priority queues, respectively. Uniform charge; , They are respectively The highest and lowest SOC. The greater the uniformity of charge, the higher the priority for action; when the uniformity of charge is equal, then... smaller Priority actions. After the above steps, the actual electric vehicle charging and discharging priority queue can be determined. , .

[0073] Step 2: Construction and optimization of the multi-objective comprehensive utility function: This invention transforms the multi-objective optimization problem into a single-step utility function of the lower-level ADHDP to guide its learning and decision-making; minimizing the single-step utility function represents pursuing the global optimal solution, and its expression is:

[0074] in, , , These are the weighting coefficients optimized by the upper-level BDBO; This represents the state variables of the system at time t, including the operating status information of each subsystem in the port integrated energy system; The control decision variables at time t include energy storage charging and discharging power, grid power purchase and sale, output of electric-hydrogen coupling equipment, and electric vehicle charging strategies.

[0075] Overall economic cost The aim is to minimize net operating expenses and introduce the cost of energy storage lifetime depletion to guide the rational use of energy storage; its expression is:

[0076] Real-time time-of-use electricity pricing; The aging and loss cost of energy storage batteries. This cost has a non-linear relationship with the depth of charge and discharge (DOD). This invention uses a linearized model to approximate the loss, indirectly extending battery life by penalizing high-power charge and discharge or deep cycling in the objective function. The operation and maintenance costs of renewable energy (usually a fixed value).

[0077] Environmental Costs (Carbon Trading Model) ; Based on the carbon trading mechanism, electricity is linked to carbon emissions, guiding the system to prioritize the consumption of clean energy. Its expression is:

[0078] in, Carbon emission purchase price; Carbon emission reduction benefit price; Carbon emission factors of grid-connected power supply; Port benchmark carbon emission allowance. Purchase costs only arise when the allowance is exceeded. ); The total amount of new energy power generation is directly linked to the benefits of emission reduction.

[0079] System stability cost ; Aimed at minimizing fluctuations in the interaction power between the port and the main power grid, and ensuring the safety and friendliness of the power grid, its expression is:

[0080] This penalty penalizes the rate of change of tie-line power, encouraging the ADHDP strategy to output smooth power purchase and sales curves.

[0081] Rigid constraint penalty term ; Used to handle rigid constraints that must be satisfied; when a scheduling action violates a constraint, a very large penalty value M is applied, the expression of which is:

[0082] in, The rigid constraint penalty term at time t is used to quantify the degree of violation of various constraints by the system during operation; M is the penalty factor, which is a sufficiently large positive number used to amplify the impact of constraint violation on the objective function, so as to ensure that rigid constraints are satisfied first during the optimization process. This indicates the amount of violation of the State of Charge (SOC) constraint of the energy storage system. When the SOC of the energy storage exceeds its allowable upper and lower limits, this value is greater than 0, otherwise it is 0. This indicates the amount of violation of the charging and discharging power constraints of the energy storage system. When the charging or discharging power of the energy storage exceeds its rated maximum power limit, this value is greater than 0; otherwise, it is 0. This represents the violation of Quality of Service (QoS) constraints for electric vehicles, used to characterize the failure of an electric vehicle to reach the target state of charge before its departure time. The degree of deviation is usually defined as the shortfall between the actual SOC and the target SOC.

[0083] Step 3: Design and interaction mechanism of the two-layer DBDO-SAC collaborative optimization framework: This invention employs a two-layer collaborative optimization framework to decouple the complex global optimization problem from the real-time control problem, overcoming the shortcomings of traditional single-layer optimization in terms of stability and real-time performance.

[0084] (1) Upper-level parameter evolution layer (BDBO), whose function is to perform offline global optimization; decision variables The optimization goal of BDBO is to find an optimal set of hyperparameters. This set includes the initial weights and learning rate of the SAC network. Discount Factor (Impact on long-term planning capabilities), and the weights of multiple objectives in the objective function. (Affecting the balance between the economy, environmental protection, and stability).

[0085] (2) The lower real-time control layer (SAC) is responsible for performing online rolling time-domain control; decision action u(t): SAC outputs the optimal control action in real time based on the current system state x(t). .

[0086]

[0087] SAC executes u(t) only at time t, and then reprograms based on the new state x(t+1) at time t+1.

[0088] (3) Inter-layer collaboration and fitness evaluation mechanism: The fitness evaluation process of BDBO depends on the simulation performance of ADHDP; Parameter transfer: In each iteration of BDBO, the population individuals are passed down to ADHDP; Pre-training and evaluation: SAC uses this set of parameters Ω to perform rolling time-domain pre-training simulation on a pre-prepared typical historical scenario dataset; Fitness feedback: After the simulation, SACT returns its total running cost (the reciprocal of the cumulative reward) in that scenario as the fitness feedback. Fitness value of individual parameters For BDBO, its expression is:

[0089] in, This represents an individual in the upper-level parameter evolution layer. The fitness function value is used to measure the overall performance of this parameter combination throughout the entire scheduling cycle; This represents the set of parameters to be optimized in the upper layer, including but not limited to policy network parameters, value function parameters, or weight coefficients related to the control policy. T sim This represents the total time step of the scheduling simulation, i.e., the discrete time length of the rolling time-domain optimization; t This represents the current discrete-time step.

[0090] BDBO uses this fitness value as the basis for evolutionary selection, constructing a global optimization closed loop with the goals of algorithm performance and long-term cost.

[0091] Step 4: Improvement and Implementation of the Upper-Level Biological Enhanced Dung Beetle Optimization Algorithm (BDBO); To ensure the globality of parameter optimization and the stability of convergence, this invention makes three key enhancements to the standard dung beetle optimization algorithm (DBO): (1) Parabolic adaptation strategy: Problem: The rolling ball behavior of standard DBO usually uses a linear shrinkage factor, which can easily lead to over-concentration of the search in the early stage or slow convergence in the later stage; Improvement mechanism: Introduce a parabolic shrinkage factor. and To balance exploration and development.

[0092]

[0093] In the early stages of iteration Larger, later It decreases rapidly, and Following the opposite trend, this non-linear adjustment ensures that the algorithm explores a wide range in the early stage and develops local areas with high precision in the later stage, thus improving the efficiency and accuracy of parameter search.

[0094] (2) Gaussian distribution parameter replacement strategy: Problem: The brooding and foraging behaviors in DBO mainly rely on uniformly distributed random numbers for mutation, which lacks specificity; Improvement mechanism: In the position update of brooding and foraging dung beetles, the random perturbation term is replaced with a Gaussian distribution. The random numbers. More importantly, the variance of the Gaussian distribution. It decreases linearly or exponentially with the number of iterations t.

[0095]

[0096] This mechanism enables the algorithm to shrink the step size and range of mutations in the later stages of iteration, achieving fine-tuning of parameter optimization and helping to escape subtle saddle points near local optima.

[0097] (3) Lévy flight escape mechanism: Problem: When the population gets stuck in a flat region or a local optimum trap, the algorithm will stagnate for a long time; Improvement mechanism: Set a stagnation counter. If the global optimal fitness value is... If there is no improvement after N consecutive generations, the Lévy flight jump maneuver will be forcibly performed on some individual dung beetles.

[0098]

[0099] Lévy's long-tailed flight pattern can produce occasional large-step jumps, allowing the population to quickly escape the current local optimum and improving the algorithm's robustness.

[0100] Step 5: Real-time control and rolling time-domain implementation of the lower-level SAC algorithm: The lower-level SAC is an online, real-time, model-free controller responsible for translating the optimized parameters from the upper layer into actual power scheduling commands. It is based on the principle of adaptive dynamic programming (ADP) and approximates the optimal control strategy through an Actor-Critic dual-network structure.

[0101] (1) Definition of state and action space: The state vector x(t) must contain all the information needed for PIES to make accurate decisions;

[0102] For the future Predicted values ​​in the time domain; These are the most pressing and most flexible metrics for EVs in the charging and discharging queues, respectively.

[0103] Action vector u(t): The control actions of the system are all continuous variables, which conform to the characteristics of DDPG / ADHDP.

[0104]

[0105] Ratio determines the proportion of total power used to charge or discharge the EV at the current moment.

[0106] (2) Actor-Critic dual network structure and training: The Critic network (evaluation network) approximates the optimal value function J(x); it receives the state x(t) and the output action u(t) of the Actor network, predicts the cumulative cost J(x(t)) from the current time to the end of the scheduling cycle; its update objective is to minimize the time difference (TD) error, i.e., the Bellman error.

[0107] Actor Network (Execution Network): Approximates the optimal policy π(x). It receives the state x(t) and directly outputs the optimal action u(t).

[0108]

[0109] The weights of the Actor network Update along the gradient direction of the Q function to ensure that the policy evolves in the direction of minimizing long-term cost.

[0110] (3) Rolling Time Control (RHC) mechanism: SAC uses the RHC mechanism for online deployment; Time domain setting: a finite prediction time domain is set. (For example, 1 hour, time step Δt = 15 minutes); Decision process: At each scheduling time t, ADHDP: Collect the latest real-time system state x(t). Utilize predictive models such as LSTM to obtain future... The predictive data within the network. The Actor network uses real-time inference based on x(t) to output the optimal action. The system only executes The first action in, namely , Wait until time t+1, then recollect data, repeat the entire process, and correct the prediction error from the previous time step. This mechanism combines the adaptability of reinforcement learning to achieve a robust real-time response to PIES minute-level high-frequency fluctuations.

[0111] The beneficial effects of this solution are explained in detail below in response to the problems of existing technologies: This invention aims to provide an innovative and robust solution to address the three core technical deficiencies in the optimization and scheduling of existing Green Port Integrated Energy Systems (PIES): algorithm performance, multi-objective coordination, and load flexibility mining.

[0112] (1) Overcoming and solving the performance bottleneck of the algorithm: This invention aims to solve the problems of slow convergence, parameter sensitivity, and local optima faced by traditional reinforcement learning algorithms (especially value function-based DQN or policy-based DDPG / ADP) in high-dimensional, continuous, and dynamic systems such as ports.

[0113] (11) Solving the problems of slow convergence speed and cold start: Traditional adaptive dynamic programming (ADP) or deep reinforcement learning (DRL) algorithms require a lot of random trial and error in the early stage of training ("cold start") to accumulate enough experience. When the port PIES state space (containing dozens of dimensions such as SOC, power, price, weather forecast, etc. of multiple devices) is extremely large, this process is very time-consuming, usually requiring tens of thousands or even hundreds of thousands of time steps for training, which is completely inconsistent with the requirements of rapid iteration and deployment of port systems.

[0114] The improvement objective of this invention is to introduce a two-layer collaborative framework, where the upper layer employs the Bio-Enhanced Dung Beetle Optimization (BDBO) algorithm to optimize the key hyperparameters of the lower-layer SAC (including the initial weights and learning rate of the neural network). Discount Factor Global optimization is performed. This mechanism is equivalent to adjusting the network parameters to a near-optimal initial configuration state through the global search capability of the evolutionary algorithm before the algorithm begins learning. This allows SAC to avoid starting from a completely random state, significantly shortening the cold start time and enabling the algorithm to quickly enter the effective learning phase. Its convergence speed is improved by at least 40% compared to the traditional ADP algorithm.

[0115] (12) Solving the problem of insufficient parameter sensitivity and generalization ability: Existing algorithms rely on manual experience to tune parameters. The parameters need to be readjusted for different seasons (such as high photovoltaic power in summer and low wind power in winter) and different operating conditions (such as dense shipping schedules and drastic changes in electricity prices). When the hyperparameters of the algorithm (such as the learning rate) are not set properly, it is easy to cause oscillation or divergence in the training process, resulting in poor policy generalization ability.

[0116] The improvement objective of this invention is to use the BDBO algorithm as a parameter optimization layer, incorporating the weight coefficients Ω (which affect multi-objective balance) in the objective function and the training parameters of the SAC as optimization variables. Through cross-validation and fitness evaluation on training datasets from multiple typical scenarios, BDBO ultimately outputs a globally optimal set of parameters with the highest robustness and adaptability to cross-seasonal changes. This enables the SAC controller to achieve cross-scenario adaptability and generalization capabilities, maintaining efficient operation under different working conditions without manual intervention.

[0117] (13) Solving the problem of getting trapped in local optima: When dealing with high-dimensional, multi-peak non-convex optimization problems, traditional swarm intelligence algorithms and ADP algorithms are easily attracted by local optimal solutions on the surface of the objective function, resulting in the final strategy not being globally optimal.

[0118] The improvement objective of this invention is to introduce targeted improvement strategies into the upper-level BDBO algorithm, such as the parabolic adaptation strategy and the Lévy flight escape mechanism. The parabolic adaptation strategy maintains the exploration breadth of the population in early iterations, avoiding premature concentration; the Lévy flight mechanism can monitor whether the algorithm has stagnated, and once stagnation is detected, it triggers random long-distance jumps to help the population escape local optima. This enhancement mechanism ensures the global optimization accuracy of the selected parameter set for the complex PIES objective function.

[0119] (2) Balancing and resolving difficulties in multi-objective coordination: This invention aims to solve the technical problem of dynamically and optimally balancing four conflicting objectives in port scheduling: economic cost, environmental benefits, system stability, and equipment lifespan.

[0120] (21) Resolving the dynamic trade-off between conflicting objectives: Minimizing economic costs usually requires discharging from the energy storage system or selling electricity to the grid during peak electricity price periods, but this accelerates battery aging (shortens equipment lifespan) and may cause fluctuations in tie-line power. Minimizing carbon emissions requires maximizing the absorption of new energy sources, but this may lead to wind and solar curtailment during periods of overcapacity, increasing operating costs. Traditional single-objective optimization or fixed-weight multi-objective optimization cannot dynamically adjust objective priorities when operating conditions change in real time.

[0121] The improvement objective of this invention is to transform a multi-objective optimization problem into a single-objective constrained optimization problem, and to dynamically measure the multiple objectives using the single-step utility function U(x, u) of the SAC (Single-Objective Optimization). Simultaneously, the objective weights are... , , This is incorporated into the optimization scope of the upper-level BDBO. This means that BDBO can find an optimal weight combination offline, enabling these four objectives to achieve the best global balance throughout the entire scheduling cycle. SAC learns online precisely how to maximize the long-term overall benefit under these weight combinations.

[0122] Introducing the quantification of equipment lifespan and stability: quantifying the aging and depreciation costs of energy storage batteries. Explicitly introduce the objective function In addition, the power fluctuation variance of the tie line is introduced. As a penalty, SAC will proactively avoid control actions that, while economically viable in the short term, could damage equipment lifespan or disrupt the power grid in the long run.

[0123] (22) Solving the problem of accurate quantification of environmental benefits and economic benefits: The calculation of carbon emissions requires complex models to track the carbon footprint brought about by electricity purchase, fuel consumption (shore power) and new energy consumption. Existing models are often too simplistic and difficult to accurately quantify the real environmental costs of ports in the carbon trading market.

[0124] The improvement objective of this invention is to employ a precise environmental cost model based on a carbon trading mechanism. This model not only calculates the carbon emissions introduced from purchasing electricity from the large power grid (based on the grid carbon emission factor), but also... The model also explicitly includes the carbon emission reduction benefits generated from absorbing wind and solar renewable energy. This precise quantitative model can guide the SAC agent to prioritize scheduling strategies with the highest environmental benefits while meeting economic requirements.

[0125] (3) Breakthroughs and solutions to insufficient load flexibility exploitation: The present invention aims to completely break through the limitations of the existing technology that regards flexible loads such as electric trucks (EVs) as passive consumers, and upgrade them to controllable active adjustment resources in PIES.

[0126] (31) Solving the problem of accurate quantification and scheduling of load flexibility: Existing technologies lack a simple and universal index to evaluate the time window in which each EV can delay charging or participate in V2G. Simple SOC thresholds or time window rules cannot adapt to the randomness of truck operations.

[0127] The objective of this invention is to introduce a maximum charging delay index (MCDI). This indicator is calculated by taking the departure time of the EV. Time required for rigid charging The difference between them is used to precisely quantify the time margin. The larger the size, the greater the flexibility in vehicle dispatching.

[0128] Decision-making mechanism: based on The size of the queue determines the size of the dynamic charging priority queue. Smaller priority charging queues and discharge priority queues ( Large priority discharge / V2G). This mechanism is the first to integrate the logistics task information of the truck ( Deeply coupled with energy dispatch commands (charging and discharging power), it ensures that the port's logistics service level (QoS) remains unaffected while optimizing the energy system.

[0129] (32) Solving the problems of "peak drift" and curtailment of wind and solar power: Disordered charging or simple TOU rules can create new peak values, while preventing the system from utilizing the electricity generated during peak periods of photovoltaic and wind power generation.

[0130] The improvement objective of this invention is to fully integrate the orderly scheduling of EVs with the real-time decision-making of the SAC (Solar Energy Controller). When assessing the state of the EV group, the SAC agent references peak grid load, real-time electricity prices, and wind and solar forecasts. It instructs the EV group to charge at a distributed and dynamically adjusted power level during periods of low electricity prices and high renewable energy output (such as midday solar power). This not only smooths the load curve and eliminates "peak drift," but also maximizes the conversion of renewable energy into valuable energy storage, significantly improving the renewable energy integration rate.

[0131] In summary, this solution constructs a high-fidelity PIES mathematical model and an MCDI flexible load model. Secondly, it achieves collaborative optimization of algorithm parameters and scheduling strategies through a two-layer framework. Finally, guided by the optimized parameters of the upper layer, the lower-layer ADHDP outputs stable, economical, and environmentally friendly optimal power allocation commands online based on a comprehensive objective function. This realizes collaborative optimization scheduling of PIES and completely breaks through the bottleneck of existing port scheduling technologies, providing an efficient, robust, economical, and environmentally friendly energy management strategy for green and smart ports.

[0132] Although the specific embodiments of the invention have been described in detail with reference to the accompanying drawings, this should not be construed as limiting the scope of protection of this patent; various modifications and variations that can be made by a person skilled in the art without inventive effort within the scope described in the claims are still within the scope of protection of this patent.

Claims

1. A method for collaborative optimization and dispatching of green port integrated energy system, characterized in that, Includes the following steps: S1: Construct a multi-dimensional mathematical model of a green port integrated energy system, which includes an electric truck load model; introduce a maximum charging delay index into the electric truck load model to quantify the charging time margin of each electric truck, and construct a charging and discharging priority queue based on the maximum charging delay index. S2: Construct a multi-objective comprehensive utility function, which includes comprehensive economic cost, environmental cost, system stability cost, and rigid constraint penalty term; S3: Construct a two-layer collaborative optimization framework, which includes an upper parameter evolution layer and a lower real-time control layer. The upper parameter evolution layer uses the biologically enhanced dung beetle optimization algorithm to perform offline global optimization, optimizing the hyperparameters of the flexible motion evaluation algorithm and the multi-objective weights of the multi-objective comprehensive utility function in the lower real-time control layer. The lower real-time control layer uses the flexible motion evaluation algorithm and takes the multi-objective comprehensive utility function as the optimization objective to perform online rolling time-domain control, and finally outputs real-time power scheduling commands.

2. The collaborative optimization scheduling method for a green port integrated energy system according to claim 1, characterized in that, The formula for calculating the maximum charging delay in step S1 is as follows: ； in, For the first i electric vehicles t The maximum charging delay at any given moment indicates the amount of time an electric vehicle can be charged before it reaches the SOC (State of Charge) requirement for battery departure. , The first i The registration time of electric vehicles and the owner's desired deregistration time; T i For the first i The necessary charging time for an electric vehicle; For the first i Off-grid SOC (State of Charge) index for electric vehicles; For the first i The SOC index of an electric vehicle at time t; Rated charging efficiency; For the first i The rated charging power of an electric vehicle; Battery capacity; The method for constructing the discharge priority queue is as follows: S11: Based on the maximum charging delay index for each electric vehicle The positive and negative signs, and magnitudes, are used to generate an ascending queue for charging priority and a descending queue for discharging priority; the method is as follows: ... Electric vehicles are placed in an ascending priority charging queue, and according to... Arrange in ascending order; Electric vehicles are assigned to a descending discharge priority queue, and according to... Arrange them in descending order; S12: By calculating the uniformity of charge index, the same For identical electric vehicles, a secondary ranking is performed, and the formula for calculating the uniformity of charge index is: in, , These are the charge uniformity indices of electric vehicles to be corrected in the charging and discharging priority queues, respectively. , These are the highest and lowest SOC (State of Charge) indicators for electric vehicles; electric vehicles with higher uniform charge indicators are prioritized; when uniform charge values ​​are equal, then... Smaller electric vehicles are given priority.

3. The collaborative optimization scheduling method for a green port integrated energy system according to claim 1, characterized in that, The multidimensional mathematical model also includes a power balance constraint model, a renewable energy output model, an energy storage system model, a shore power system load model, and a port crane load model.

4. The collaborative optimization scheduling method for a green port integrated energy system according to claim 1, characterized in that, The expression for the multi-objective comprehensive utility function in step S2 is: in, , , These are the multi-objective weight coefficients optimized by the upper-level parameter evolution layer. Considering the overall economic cost; For environmental costs; Costs related to system stability; This is a rigid constraint penalty term; x ( t () represents the system state vector; u ( t ) represents the system action vector.

5. The collaborative optimization scheduling method for a green port integrated energy system according to claim 4, characterized in that, The comprehensive economic cost This includes the cost of purchasing electricity from the grid, the revenue from selling electricity to the grid, the loss costs of aging energy storage systems, and the operation and maintenance costs of renewable energy. The calculation formula is as follows: in, These represent the power purchased from and sold to the power grid, respectively. These are real-time time-of-use electricity prices; Costs associated with the aging and depletion of energy storage batteries; The operation and maintenance costs of renewable energy; The environmental costs The carbon trading mechanism comprises the carbon emission costs incurred from purchasing electricity from the grid and the carbon emission reduction benefits from absorbing renewable energy. Its calculation formula is as follows: in, The price to pay for carbon emissions; The price of carbon emission reduction benefits; Carbon emission factors for supplying electricity to the power grid; This serves as the port's baseline carbon emission allowance; purchase costs only arise when the allowance is exceeded. ; This refers to the total amount of new energy power generation; The system stability cost It is constructed using the volatility of the power exchange between the port and the power grid, and its calculation formula is as follows: 。 6. The collaborative optimization scheduling method for a green port integrated energy system according to claim 5, characterized in that, The rigid constraint penalty item The expression is: Where M is the penalty factor used to amplify the impact of constraint violation on the objective function; Because the electric vehicle failed to reach the departure time Service quality breach; This indicates the amount of violation of the state of charge constraints of the energy storage system; This indicates the amount of violation of the charging and discharging power constraints of the energy storage system; This indicates the amount of violation of service quality constraints for electric vehicles.

7. The collaborative optimization scheduling method for a green port integrated energy system according to claim 1, characterized in that, In step S3, the upper parameter evolution layer passes the parameters of each iteration to the lower real-time control layer during the offline global optimization process, and performs rolling time-domain pre-training simulation on the historical scene dataset to obtain the total running cost in the scene. This cost is then used as the fitness value of the iteration parameters and transmitted to the upper parameter evolution layer. The upper parameter evolution layer uses the fitness value as the basis for selecting the iteration parameters. The expression for the fitness value is: in, This represents the set of parameters to be optimized in the upper layer. T sim This is the total time step of the scheduling simulation, i.e., the discrete time length of the rolling time-domain optimization; t This represents the current discrete-time step.

8. The collaborative optimization scheduling method for a green port integrated energy system according to claim 1, characterized in that, The biologically enhanced dung beetle optimization algorithm in step S3 optimizes parameters through a parabolic adaptation strategy, a Gaussian distribution parameter replacement strategy, and a Levy flight escape mechanism.

9. The collaborative optimization scheduling method for a green port integrated energy system according to claim 4, characterized in that, The flexible action evaluation algorithm approximates the optimal control strategy through an Actor-Critic dual network structure based on the current system state and the optimized parameters of the upper parameter evolution layer at each scheduling moment. It outputs and executes real-time power scheduling commands, and at the same time, it re-acquires system state and outputs commands at the next moment.

10. The collaborative optimization scheduling method for a green port integrated energy system according to claim 1, characterized in that, The flexible motion evaluation algorithm is trained using an Actor-Critic dual-network structure; The Critic network is used to receive the current system state. x ( t ) and the actions output by the Actor network u ( t It predicts the cumulative cost from the current moment to the end of the scheduling cycle, and updates the network weights by minimizing the time difference error, which is constructed based on the Bellman error. Actor networks are used to approximate the optimal control strategy; they receive the current system state. x ( t ), directly output the optimal action u ( t The network weights are updated along the gradient direction of the cumulative cost estimated by the Critic network, so that the control policy evolves in the direction of minimizing the long-term cost.

Images