Lightweight port integrated energy system distributed optimization method

Abstract

The invention relates to a lightweight port integrated energy system distributed optimization method, which comprises the following steps of: constructing and training a port logistics predictor through historical operation data, and modeling a port logistics-energy system distributed optimization problem; and using the output of the port logistics predictor obtained by training as an initial solution of a logistics system local sub-problem in the optimization problem model, and iteratively solving the port logistics-energy system distributed optimization problem model through an ADMM algorithm to obtain a globally optimal joint scheduling scheme. According to the method, the initial solution of the logistics system sub-problem is predicted by using the convolutional neural network, the solution time of the MIQP sub-problem is shortened, the convergence of the global optimal solution of the system is ensured, the scheduling efficiency of the port energy and the logistics system can be remarkably improved, the lightweight and intelligent solution process is realized, and the method is suitable for popularization and application. And the high-efficiency management requirement of a large-scale port comprehensive energy system is met.

Description

Technical Field

[0001] The present invention relates to a technology in the field of port control, specifically a lightweight distributed optimization method for a port integrated energy system. Background Art

[0002] In the existing port operation scheduling, the energy system and the logistics system are usually managed independently, lacking the interaction of data and decision-making. This separate management method leads to resource waste, increased operating costs and carbon emissions. As the demand for integrated management increases, the scheduling problem of the energy-logistics system is transformed into a large-scale mixed integer programming problem, which is complex and time-consuming to solve, affecting the efficiency of actual application. Summary of the invention

[0003] In order to solve the problem of long calculation time in the existing port integrated energy-logistics system scheduling optimization, the present invention proposes a lightweight distributed optimization method for the port integrated energy system. By using a convolutional neural network to predict the initial solution of the logistics system subproblem, the solution time of the MIQP subproblem is reduced, and the convergence of the system's global optimal solution is guaranteed. This method can significantly improve the efficiency of the port energy and logistics system scheduling, realize a lightweight and intelligent solution process, and meet the efficient management needs of large-scale port integrated energy systems.

[0004] The present invention is achieved through the following technical solutions:

[0005] The present invention relates to a lightweight distributed optimization method for a port integrated energy system. After constructing and training a port logistics predictor through historical operation data, a distributed optimization problem of a port logistics-energy system is modeled, and then the output of the trained port logistics predictor is used as the initial solution of a local sub-problem of the logistics system in the optimization problem model. The distributed optimization problem model of the port logistics-energy system is iteratively solved through an ADMM algorithm to obtain a globally optimal joint scheduling solution. Technical Effects

[0006] The present invention embeds a learning optimization predictor to quickly predict the initial solution of the logistics system in the early iteration stage of ADMM, bypassing the cumbersome solution of the mixed integer programming problem, using a predictor based on a convolutional neural network to accelerate the solution in the early stage, and switching to the existing optimization method in the later stage to ensure the global optimal solution. Compared with the prior art, the present invention can use the predictor model based on the convolutional neural network proposed by the present invention to quickly predict the initial solution of the logistics system in the early iteration stage of ADMM, bypassing the cumbersome solution of the mixed integer programming problem, and through the hybrid strategy proposed by the present invention, flexibly switch the solution method according to the different iteration stages. This strategy makes the method highly adaptable and can dynamically adjust the solution process according to actual needs, thereby optimizing the utilization of computing resources, significantly reducing the calculation time, and significantly improving the solution efficiency, which is particularly suitable for real-time scheduling of large-scale port systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 It is a flow chart of the present invention;

[0008] Figure 2 and Figure 3 It is a schematic diagram of the effect of the embodiment. DETAILED DESCRIPTION

[0009] like Figure 1 As shown, this embodiment relates to a lightweight distributed optimization method for a port integrated energy system, including:

[0010] Step 1: Train the port logistics predictor, including:

[0011] 1.1 Collect information on ship arrival plans and logistics equipment dispatch at the port.

[0012] 1.2 Based on the collected historical operation data, the interactive variables of the energy system and the logistics system are integrated to construct a complete training data set.

[0013] 1.3 Construct a port logistics predictor using convolutional neural network (CNN) as the core structure and train it using the training data set obtained in step 1.2.

[0014] The port logistics predictor includes: a one-dimensional convolution layer, a pooling layer, a flattening layer, a fully connected layer and an output layer, wherein: the one-dimensional convolution layer extracts local features of the input data and captures key patterns in the interaction data of the energy and logistics systems; the pooling layer downsamples the features to reduce the data dimension and reduce the computational complexity; the flattening layer maps the multidimensional features into a one-dimensional vector to provide input for the subsequent fully connected layer; the fully connected layer extracts complex nonlinear relationships in the data; the output layer generates the predicted values ​​of the interaction variables of the logistics system to obtain the required logistics system predictor.

[0015] Step 2: Modeling of distributed optimization problems of port logistics-energy system, including:

[0016] 2.1 For the logistics system including shore crane equipment, ship scheduling and berth allocation, an equipment management and control model is established to optimize the scheduling and operation plan of logistics equipment.

[0017] The equipment management and control model described above takes minimizing the logistics system cost as the objective function, specifically: Where: c PG,t is the electricity purchase price at time t, e PG 、c CO2 are the carbon emission coefficient and carbon emission price of the power grid, P QC,t , P SR,t is the power consumption of the quay crane and shore power at time t.

[0018] The equipment management and control model meets the requirements of port entry and exit time and equipment operation capacity constraints, wherein the equipment operation capacity constraints include berth allocation restrictions and quay crane task restrictions.

[0019] The berth allocation restrictions include: the total length of the vessel when docked Ships do not overlap when docked Where: h i,j and z i,j is an auxiliary variable. When ship i docks before ship j, then h i,j =1, otherwise h i,j = 0. When ship i is docked on the left side of ship j, then z i,j =1, otherwise z i,j =0. berth , L ship,j is the total length of the pier and the length of ship j, in meters (m). i is the berth allocated to ship i.

[0020] The quay crane task restrictions include: the number of quay crane tasks Each gantry crane can only execute one task at a time. Constraints on the sequential execution of quay crane tasks Power Constraints of Container Terminal Cranes Power Constraints on Vessels Where: C QC,i,j,t is the number of containers of ship i handled by quay crane j at time t, is the rated power of the quay crane, P B,j is the basic power requirement of ship j, in megawatts (MW).

[0021] 2.2 Model the common operation constraint part of the distributed solution problem, that is, coordinate the power interaction between the logistics system and the energy system through Lagrange multipliers, and solve them alternately in each iteration. Specifically, the calculation method of the interaction variables between the logistics system and the energy system in the k+1th iteration is: Where: Φ LS and Φ ES represent the interaction variables of the logistics system and the energy system, L ρ (Φ LS ,Φ ES ,λ) is the augmented Lagrangian function, λ is the Lagrangian multiplier, and ρ is the penalty parameter.

[0022] 2.3 Establish an energy management and control model for photovoltaic power generation, energy storage system, gas turbine, electric boiler and shore power system involved in the port energy system.

[0023] The energy management control model is based on the energy consumption cost Minimize the goal, where: c ESS Represents the total cost of the energy storage system, and Respectively represent the charging and discharging efficiency of the energy storage system at time t, G(Ω s,t ) is the other operating cost of the energy system under scenario s and time t, p s is the occurrence probability of scenario s.

[0024] The energy management control model satisfies energy equipment constraints and power balance constraints.

[0025] The energy equipment constraints include: power output limit of photovoltaic panels Power output and energy capacity limitations of energy storage systems (ESS) Energy state transition equation of ESS Initial energy state of ESS Gas supply restrictions for gas turbines (GT) and boilers (GB) GT's gas conversion equation for electricity and heat GB gas converted into heat Power output limitation of electric boiler (EB) EB's electrical energy is converted into heat energy Where: P PV,t and are the output power and predicted available power of photovoltaic at time t respectively; and are the charging / discharging power of the energy storage system at time t respectively; and are the maximum charge / discharge power of the energy storage system at time t; E ESS,t and are the capacity and maximum capacity of the energy storage system at time t respectively; and is the charge / discharge efficiency of the energy storage system; G GT,t , G GB,t and are the gas consumption and maximum gas input of the gas turbine and gas boiler at time t, both in km 3 / t;P GT,t , P GB,t , P EB,t and are the output power and maximum output power of the gas turbine, gas boiler and electric boiler at time t respectively; H GT,t , H GB,t and H EB,t are the heat outputs of the gas turbine, gas boiler and electric boiler at time t, in megawatts (MW); is the gas turbine power generation efficiency, and are the thermal efficiencies of gas turbines, gas boilers and electric boilers respectively; T is the total dispatching time; the units of the above powers are all megawatts (MW).

[0026] The power balance constraints include: upper grid power and natural gas supply restrictions Heat balance equation Natural Gas Balance Equation Where: P G,t and They are the electric power purchased from the upper power grid at time t and the maximum value of the purchased electric power, in megawatts (MW); G G,t and is the gas purchase volume from the upper-level gas grid and the maximum supply volume from the upper-level gas grid at time t, is the natural gas load forecast value at time t, in km 3 / t;P SR,t and P QC,t are the shore power consumption and the quay crane power consumption at time t, respectively, in megawatts (MW); and It is the predicted power of electric load and predicted value of thermal load at time t, in megawatts (MW).

[0027] Step 3: Use the output of the port logistics predictor trained in step 1 As the initial solution to the local sub-problem of the logistics system, the distributed optimization problem model of the port logistics-energy system established in step 2 is iteratively solved by the ADMM algorithm, which specifically includes:

[0028] 3.1 Obtain the ship arrival and departure time and load information from the ship arrival plan, and collect scenario prediction data to provide input for the optimization solution of the logistics system and energy system.

[0029] 3.2 Local energy management control problem of energy system using GUROBI solver Solve and optimize the operation plan of the energy system.

[0030] 3.3 In the initial stage of distributed optimization, determine whether the current number of iterations is less than the set threshold (i.e., whether it is in the early iteration stage). If it is greater than the threshold, use the commercial solver to directly solve the local equipment management and control problem of the logistics system to obtain the initial solution. If it is less than the threshold, call the logistics port logistics predictor trained in step 1, and use the predictor to quickly solve the local equipment management and control problem of the logistics system, thereby accelerating the iteration process.

[0031] 3.4 After each iteration, determine whether the iteration termination condition is met, that is, whether the original residual and the dual residual are both less than the given value. If the condition is met, the iteration is terminated to form the final logistics-energy system joint scheduling plan; if the condition is not met, repeat steps 3.2 and 3.3 until the condition is met, specifically: the original residual Dual residual Among them: When calculating the dual residual, the number of iterations k ≥ 2 should be ensured.

[0032] 3.5 When the residual obtained in step 3.4 is less than the given value, the iteration ends. Integrate the scheduling results of the logistics system and the energy system to form a global optimal joint scheduling solution.

[0033] After specific practical experiments, taking a low-carbon port as an example, through the coordinated scheduling of the energy system and the logistics system, the port ship plan is shown in Table 1, and the port parameter settings in the simulation are shown in Table 2.

[0034] Table 1

[0035] Table 2

[0036] Learning optimization predictor training set generation and ADMM parameter setting: In order to generate a dataset that can be used to train the learning optimization predictor, a series of energy system net interaction powers and ship arrival schedules are randomly set as follows: Figure 2 As shown, the corresponding net interaction power of the logistics system is pre-solved using the Gurobi optimizer. Then a training set and a test set are created. The solution process can be executed with a mixed integer programming gap lower than the actual distributed scheduling algorithm. Although the process of generating a single sample is time-consuming, these samples can be generated in parallel through multi-process management, thereby reducing the time required to generate a dataset. The convolutional neural network is trained using the PyTorch framework and 600 pairs of samples are generated, with a ratio of 6:4 between the training set and the test set. The convergence threshold ∈ of ADMM is set to 0.01 and ρ is set to 5.

[0037] like Figure 3 The figure shows the error distribution of the convolutional neural network learning optimization predictor on the test set, where the error is defined as the difference between the operating cost predicted by the logistics system and the operating cost calculated by the actual net interaction power. It can be seen that most of the errors of the well-trained convolutional neural network are concentrated around 0, and the deviation between the predicted logistics system operating cost and the actual cost is mainly within $100, indicating the effectiveness of the convolutional neural network as a learning optimization predictor.

[0038] like Figure 3 As shown in the figure, the original and dual residual convergence speed curves of different methods are compared. It can be seen that the initial step of the existing ADMM iteration is usually time-consuming, but the convergence speed is very fast. The CNN-ADMM solution method effectively bypasses the process of solving the logistics scheduling MIQP problem in the initial iteration, while ensuring the convergence speed and reducing the overall time required for distributed optimization.

[0039] Compared with the existing technology, this method embeds the convolutional neural network learning optimization predictor into the ADMM optimization framework to ensure the accuracy of the understanding and the convergence of the algorithm. Compared with the existing technology, this strategy improves the computational efficiency without sacrificing convergence and is suitable for large-scale port energy-logistics systems.

[0040] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principle and purpose of the present invention. The protection scope of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. Each implementation scheme within its scope shall be subject to the constraints of the present invention.

Claims

1. A lightweight distributed optimization method for a port integrated energy system, characterized in that: After constructing and training the port logistics predictor through historical operation data, the distributed optimization problem of the port logistics-energy system is modeled. The output of the trained port logistics predictor is used as the initial solution to the local sub-problem of the logistics system in the optimization problem model. The distributed optimization problem model of the port logistics-energy system is iteratively solved through the ADMM algorithm to obtain the globally optimal joint scheduling solution.

2. The lightweight distributed optimization method for port integrated energy system according to claim 1 is characterized in that: The port logistics predictor is obtained by: 1.1 Collect information on the ship arrival plan and logistics equipment dispatch at the port; 1.2 Based on the collected historical operation data, the interactive variables of the energy system and the logistics system are integrated to construct a complete training data set; 1.3 Construct a port logistics predictor using convolutional neural network (CNN) as the core structure and train it using the training data set obtained in step 1.

2.

3. The lightweight distributed optimization method for port integrated energy system according to claim 1 or 2 is characterized in that: The port logistics predictor includes: a one-dimensional convolution layer, a pooling layer, a flattening layer, a fully connected layer and an output layer, wherein: the one-dimensional convolution layer extracts local features of the input data and captures key patterns in the interaction data of the energy and logistics systems; the pooling layer downsamples the features to reduce the data dimension and reduce the computational complexity; the flattening layer maps the multidimensional features into a one-dimensional vector to provide input for the subsequent fully connected layer; the fully connected layer extracts complex nonlinear relationships in the data; the output layer generates the predicted values ​​of the interaction variables of the logistics system to obtain the required logistics system predictor.

4. The lightweight distributed optimization method for port integrated energy system according to claim 1 is characterized in that: The modeling of the distributed optimization problem of the port logistics-energy system specifically includes: 2.1 Establish an equipment management and control model for the logistics system, including shore crane equipment, ship scheduling and berth allocation, to optimize the scheduling and operation plan of logistics equipment to minimize the cost of the logistics system. LS (Φ LS ) as the goal, while meeting the constraints of ship arrival plan and equipment operation capacity; 2.2 Model the common operation constraint part of the distributed solution problem, that is, coordinate the power interaction between the logistics system and the energy system through Lagrange multipliers, and solve them alternately in each iteration. Specifically, the calculation method of the interaction variables between the logistics system and the energy system in the k+1th iteration is: Where: Φ LS and Φ ES represent the interaction variables of the logistics system and the energy system, L ρ (Φ LS ,Φ ES ,λ) is the augmented Lagrangian function, λ is the Lagrangian multiplier, and ρ is the penalty parameter; 2.3 Establish an energy management and control model for the photovoltaic power generation, energy storage system, gas turbine, electric boiler and shore power system involved in the port energy system. The model is based on the energy consumption cost F ES (Φ ES ) is minimized, while ensuring the feasibility of energy equipment operation and the dynamic balance between the port's energy demand.

5. The lightweight distributed optimization method for port integrated energy system according to claim 4 is characterized in that: The equipment management and control model described above takes minimizing the logistics system cost as the objective function, specifically: Where: c PG,t is the electricity purchase price at time t, e PG 、c CO2 are the carbon emission coefficient and carbon emission price of the power grid, P QC,t , P SR,t is the power consumption of the quay crane and shore power at time t.

6. The lightweight distributed optimization method for port integrated energy system according to claim 4 or 5 is characterized in that: The equipment management and control model meets the requirements of port entry and exit time and equipment operation capacity constraints, wherein the equipment operation capacity constraints include berth allocation restrictions and quay crane task restrictions; The berth allocation restrictions include: the total length of the vessel when docked Ships do not overlap when docked Where: h i,j and z i,j is an auxiliary variable. When ship i docks before ship j, then h i,j =1, otherwise h i,j = 0, when ship i is docked on the left side of ship j, then z i,j =1, otherwise z i,j =0,L berth , L ship,j is the total length of the wharf and the length of ship j, in meters (m), b i is the berth allocated to ship i; The quay crane task restrictions include: the number of quay crane tasks Each gantry crane can only execute one task at a time. Constraints on the sequential execution of quay crane tasks Power Constraints of Container Terminal Cranes Power Constraints on Vessels Where: C QC,i,j,t is the number of containers of ship i handled by quay crane j at time t, is the rated power of the quay crane, P B,j is the basic power requirement of ship j, in megawatts (MW).

7. The lightweight distributed optimization method for port integrated energy system according to claim 4 is characterized in that: The energy management control model is based on the energy consumption cost Minimize the goal, where: c ESS Represents the total cost of the energy storage system, and Respectively represent the charging and discharging efficiency of the energy storage system at time t, G(Ω s,t ) is the other operating cost of the energy system under scenario s and time t, p s is the occurrence probability of scenario s.

8. The lightweight distributed optimization method for port integrated energy system according to claim 4 or 7, characterized in that: The energy management control model satisfies energy equipment constraints and power balance constraints; The energy equipment constraints include: power output limit of photovoltaic panels Power output and energy capacity limitations of energy storage systems (ESS) Energy state transition equation of ESS Initial energy state of ESS Gas supply restrictions for gas turbines (GT) and boilers (GB) GT's gas conversion equation for electricity and heat GB gas converted into heat Power output limitation of electric boiler (EB) EB's electrical energy is converted into heat energy Where: P PV,t and are the output power and predicted available power of photovoltaic at time t respectively; and are the charging / discharging power of the energy storage system at time t respectively; and are the maximum charge / discharge power of the energy storage system at time t; E ESS,t and are the capacity and maximum capacity of the energy storage system at time t respectively; and is the charge / discharge efficiency of the energy storage system; G GT,t , G GB,t and are the gas consumption and maximum gas input of the gas turbine and gas boiler at time t, both in km 3 / t;P GT,t , P GB,t , P EB,t and are the output power and maximum output power of the gas turbine, gas boiler and electric boiler at time t respectively; H GT,t , H GB,t and H EB,t are the heat outputs of the gas turbine, gas boiler and electric boiler at time t, in megawatts (MW); is the gas turbine power generation efficiency, and are the thermal efficiencies of gas turbines, gas boilers and electric boilers respectively; T is the total dispatching time; the above power units are all megawatts (MW); The power balance constraints include: upper grid power and natural gas supply restrictions Heat balance equation Natural Gas Balance Equation , where: P G,t and They are the electric power purchased from the upper power grid at time t and the maximum value of the purchased electric power, in megawatts (MW); G G,t and is the gas purchase volume from the upper-level gas grid and the maximum supply volume from the upper-level gas grid at time t, is the natural gas load forecast value at time t, in km 3 / t;P SR,t and P QC,t are the shore power consumption and the quay crane power consumption at time t, respectively, in megawatts (MW); and It is the predicted power of electric load and predicted value of thermal load at time t, in megawatts (MW).

9. The lightweight distributed optimization method for port integrated energy system according to any one of claims 1 to 4, characterized in that: The globally optimal joint scheduling scheme is obtained in the following way: 3.1 Obtain ship arrival and departure time and load information from the ship arrival plan, and collect scenario prediction data to provide input for the optimization of logistics and energy systems; 3.2 Local energy management control problem of energy system using GUROBI solver Solve and optimize the operation plan of the energy system; 3.3 In the initial stage of distributed optimization, determine whether the current number of iterations is less than the set threshold, that is, whether it is in the early iteration stage. If it is greater than the threshold, use the commercial solver to directly solve the local equipment management and control problem of the logistics system to obtain the initial solution. If it is less than the threshold, call the logistics port logistics predictor trained in step 1 to quickly solve the local equipment management and control problem of the logistics system through the predictor, thereby accelerating the iteration process; 3.4 After each iteration, determine whether the iteration termination condition is met, that is, whether the original residual and the dual residual are both less than the given value. If the condition is met, the iteration is terminated to form the final logistics-energy system joint scheduling plan; if the condition is not met, repeat steps 3.2 and 3.3 until the condition is met, specifically: the original residual Dual residual Among them: When calculating the dual residual, the number of iterations k ≥ 2 should be ensured; 3.5 When the residual obtained in step 3.4 is less than the given value, the iteration ends, and the scheduling results of the logistics system and the energy system are integrated to form a globally optimal joint scheduling plan.

Images