Method and system for regulating comprehensive energy system of traffic logistics facility

Abstract

The application provides a comprehensive energy system regulation method and system considering traffic logistics facilities, and belongs to the technical field of comprehensive energy system regulation. The method comprises the following steps: encoding the driving and charging states of an electric truck, setting minimum duration constraints and state mutual exclusion constraints of the driving state and the charging state based on the encoding result to determine the working state of the electric truck, determining the actual power consumption of the electric truck to construct an average field game theory model, converting the theoretical model into a neural network parameterized representation of a federal learning algorithm, determining global model parameters through local training and parameter aggregation, training an electric truck distributed collaborative scheduling model based on federal average field game through a deep average field Actor-Critic algorithm; inputting real-time state observation data into the model to generate optimal scheduling decisions, including optimal route selection and optimal charging power. The application improves the real-time performance, accuracy and efficiency of comprehensive energy system scheduling.

Description

Technical Field

[0001] This invention belongs to the field of integrated energy system regulation technology, and more specifically, relates to an integrated energy system regulation method and system that takes into account transportation and logistics facilities. Background Technology

[0002] Photovoltaic power generation is significantly affected by weather conditions, while the energy demand of transportation and logistics facilities is often closely related to business volume and exhibits obvious fluctuations. In integrated energy systems, increased load forecasting errors will lead to higher cooling system energy consumption and decreased gas turbine efficiency, directly driving up annual operating costs. For critical transportation hubs such as airports and ports, this uncertainty may result in energy supply and demand imbalances, affecting normal operations.

[0003] The energy consumption of the logistics system primarily stems from the charging activities of electric trucks. Changes in electric truck scheduling lead to variations in electricity load, thereby altering the operating costs of the integrated energy logistics system in the port. Furthermore, factors such as electricity prices and renewable energy production levels affect the unit charging cost at different times, also prompting adjustments to electric truck charging schedules.

[0004] In integrated energy systems, electric trucks require varying times to complete container transport tasks on different routes. If the integrated energy logistics system schedules tasks based on time intervals within the energy system, there can be significant gaps between the completion of an electric truck's task and its next scheduling time. This leads to inefficient waiting times for electric trucks. Furthermore, frequent changes in electric truck routes can cause port congestion, increasing the complexity of the scheduling model. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a comprehensive energy system control method and system considering transportation and logistics facilities. Based on the actual power consumption of electric trucks, a mean-field game theory model is constructed. This model is then transformed into a neural network parameterized representation using the FL (Federated Learning) algorithm. Global model parameters are determined through local training and parameter aggregation. Based on these global model parameters, a distributed cooperative scheduling model for electric trucks based on federated mean-field game theory is trained using the DMF-AC (Deep Mean Field Actor-Critic) algorithm. The mean-field Nash equilibrium condition includes the HJB (Hamilton Jacobi Bellman Equation) and the FP (Fokker-Planck Equation). This invention improves the collaborative scheduling capability of various devices within an integrated energy system by constructing an average-field game model that includes a state space, action space, state dynamics equations, and a long-term expected cost function. This reduces resource waste caused by discrete scheduling and lowers the operating costs of the integrated energy system. Furthermore, by parameterizing the average-field game model into a neural network and training the model parameters through federated learning, this invention resolves the contradictions in collaborative learning of multi-regional data in ports, improving the real-time performance of large-scale electric truck scheduling within the integrated energy system and enhancing the accuracy and scientific rigor of the system's scheduling. Finally, by using pre-trained global parameters to train and schedule an Actor-Critic network in real time, this invention addresses the problem of fixed-ratio allocation of transferable loads and lack of dynamic adjustment capability in existing technologies. This improves the flexibility of load transfer in the integrated energy system, dynamically adjusting according to real-time conditions, reducing efficiency losses caused by fixed-ratio allocation in the integrated energy system's regulation, and increasing the energy utilization rate of the integrated energy system.

[0006] The present invention adopts the following technical solution.

[0007] A first aspect of the present invention provides a method for regulating an integrated energy system that takes into account transportation and logistics facilities, comprising:

[0008] The driving and charging states of electric trucks are encoded, and the minimum duration constraints and state mutual exclusion constraints of driving and charging states are set based on the encoding results to determine the working state of electric trucks.

[0009] The actual power consumption of the electric truck is determined based on the driving time and average power consumption of the electric truck on different routes under different working conditions.

[0010] A mean-field game theory model is constructed based on the actual power consumption of electric trucks. The theoretical model is transformed into a neural network parameterized representation of the federated learning algorithm. The global model parameters are determined through local training and parameter aggregation. Based on the global model parameters, a distributed cooperative scheduling model for electric trucks based on federated mean-field game theory is trained using the deep mean-field Actor-Critic algorithm.

[0011] Real-time status observation data is input into a distributed cooperative scheduling model for electric trucks based on federated mean field games to generate optimal scheduling decisions, including optimal route selection and optimal charging power.

[0012] Preferably, the minimum duration constraints for driving state and charging state include:

[0013] The minimum duration constraint for driving state is expressed by the following formula:

[0014]

[0015] In the formula, This indicates that the b-th electric truck is in the... Driving status over a period of time, Indicates the minimum number of duration segments. Indicates the running index for a continuous time period. , Indicates the start time period number. , This represents the total number of time periods in the scheduling cycle.

[0016] The minimum duration constraint for the state of charge is expressed by the following formula:

[0017]

[0018] In the formula, This indicates that the b-th electric truck is in the... The charging status over a period of time.

[0019] Preferably, the state mutual exclusion constraint is expressed by the following formula:

[0020]

[0021] In the formula, This represents the time period number, t=1, 2, ... ;

[0022] exist and The time is determined to be in a driving state; and It is determined to be in charging state at this time; and The system is then judged to be in an idling state.

[0023] Preferably, the actual power consumption of the electric truck is expressed by the following formula:

[0024]

[0025] In the formula, This represents the actual power consumption of the b-th electric truck in the t-th time period. This represents the charging power consumption of the b-th electric truck in the t-th time period. This indicates the power consumption at idle. For electric trucks on the driving route Average power consumption.

[0026] Preferably, constructing a mean-field game theory model includes:

[0027] The state space of an electric truck is constructed based on its battery charge status, location coordinates, number of outbound container transport tasks, driving status, charging status, and driving speed; the action space is constructed based on route selection variables and charging power.

[0028] Based on the state space, action space, and actual power consumption of a single electric truck, a state dynamics equation for a single electric truck is established using stochastic differential equations; based on the state space, action space, travel time, and electricity price of a single electric truck, a long-term expected cost function for a single electric truck is established.

[0029] Based on the state dynamics equation and long-term expected cost function of a single electric truck, theoretical mean-field distribution parameters and mean-field Nash equilibrium conditions are constructed to obtain a mean-field game theory model.

[0030] Preferably, the long-term expected cost function of a single electric truck is expressed by the following formula:

[0031]

[0032] In the formula, Indicates the b-th electric truck under a given strategy and mean field distribution The long-term expected cost, Represents the expectation operator. Indicates the instantaneous operating cost. Let b represent the state space of the b-th electric truck in the t-th time interval. Indicates the terminal time The cost.

[0033] Preferably, The instantaneous operating cost is expressed by the following formula:

[0034]

[0035] In the formula, , , , , and Indicates the weighting coefficient. This represents the electricity price for the t-th time period. This indicates the number of containers that need to be transferred along route s. This represents the location congestion penalty function. Indicates the position in the t-th time period Truck density at the location Indicates the lower limit of battery capacity. This indicates the maximum charging power.

[0036] Preferably, Indicates the terminal time The cost is expressed by the following formula:

[0037]

[0038] In the formula, , This represents the terminal cost weighting coefficient. This indicates that the b-th electric truck is at the terminal time. The amount of electricity, Indicates the target power level. Indicates an indicator function, hour ,otherwise .

[0039] Preferably, determining the global model parameters includes:

[0040] The mean-field game theory model is mapped to a federated learning model, which includes a global model and a local regional model.

[0041] Based on the Fock-Planck equation, the mean field distribution is parameterized by a neural network and Fourier representation is used to represent the parameterized mean field distribution.

[0042] A local training loss function is constructed based on the Hamilton-Jacobi-Bellman equation, and local parameters are updated by gradient descent based on the local training loss function.

[0043] Aggregate local parameters into updated global model parameters;

[0044] Training stops when the maximum number of rounds is reached, and the trained global model parameters are obtained; otherwise, the number of rounds is incremented, and new global model parameters are issued for retraining.

[0045] A second aspect of the present invention provides an integrated energy system control system considering transportation and logistics facilities, and operates the integrated energy system control method considering transportation and logistics facilities described in the first aspect of the present invention, comprising:

[0046] The working state determination module is used to encode the driving and charging states of the electric truck, and set minimum duration constraints and state mutual exclusion constraints for the driving and charging states based on the encoding results to determine the working state of the electric truck.

[0047] The power consumption determination module is used to determine the actual power consumption of the electric truck based on the driving time and average power consumption of the electric truck on different routes under different operating conditions.

[0048] The model solving module is used to construct a mean-field game theory model based on the actual power consumption of electric trucks, transform the theoretical model into a neural network parameterized representation of the federated learning algorithm, determine the global model parameters through local training and parameter aggregation, and train a distributed cooperative scheduling model for electric trucks based on federated mean-field game using the deep mean-field Actor-Critic algorithm based on the global model parameters.

[0049] The output module is used to input real-time status observation data into the distributed cooperative scheduling model of electric trucks based on federated mean field game to generate optimal scheduling decisions, including optimal route selection and optimal charging power.

[0050] Compared with the prior art, the beneficial effects of the present invention include at least the following:

[0051] This invention reduces scheduling deviations caused by energy consumption estimation errors and energy conversion losses during charging by constraining the minimum duration of driving and charging states, thereby reducing energy waste in the overall energy system regulation.

[0052] This invention calculates actual power consumption by combining factors such as driving status, charging status, and route selection, thereby improving the coordination between electric truck dispatching and energy system dispatching and reducing energy waste caused by the lack of coordination in the overall energy system dispatching.

[0053] This invention improves the coordinated scheduling capability of various devices in an integrated energy system by constructing an average-field game model that includes a state space, action space, state dynamics equations, and a long-term expected cost function, thereby reducing resource waste caused by discrete scheduling and lowering the operating cost of the integrated energy system.

[0054] This invention resolves the contradiction of collaborative learning of multi-regional data in ports by parameterizing the average field game model into a neural network and training the model parameters through federated learning. This improves the real-time performance of the integrated energy system for scheduling large-scale electric trucks, enhances the accuracy and scientific nature of the integrated energy system's scheduling, and reduces the resource consumption of the integrated energy system.

[0055] This invention achieves the training and real-time scheduling of the Actor-Critic network based on pre-trained global parameters, solving the problem of fixed-ratio allocation of transferable loads and lack of dynamic adjustment capability in existing technologies. It improves the flexibility of load transfer in integrated energy systems, dynamically adjusts according to real-time status, reduces efficiency losses caused by fixed-ratio allocation in integrated energy system regulation, and improves the energy utilization rate of integrated energy systems. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the integrated energy system control method for transportation and logistics facilities provided in accordance with an embodiment of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0058] like Figure 1 As shown, Embodiment 1 of the present invention provides a method for regulating an integrated energy system that takes into account transportation and logistics facilities, comprising the following steps:

[0059] Step 1: Encode the driving and charging states of the electric truck. Based on the encoding results, set minimum duration constraints and state mutual exclusion constraints for the driving and charging states to determine the working state of the electric truck.

[0060] Step 1.1: Encode the driving and charging status of the electric truck, preferably using binary encoding.

[0061] Because electric trucks need to be recharged when their batteries are low, they cannot always participate in container transport tasks. Therefore, it is necessary to distinguish the operating states of electric trucks. Two binary variables are defined. , , representing the driving and charging status of the electric truck, respectively, are expressed by the following formula:

[0062] (1)

[0063] In the formula, This represents the driving status of the b-th electric truck in the t-th time period, where 1 indicates driving and 0 indicates not driving. This represents the charging status of the b-th electric truck in the t-th time period, where 1 indicates charging and 0 indicates not charging, and b represents the port. Electric truck serial number, , Total number of electric trucks This represents the time period number, t=1, 2, ... , This represents the total number of time periods in the scheduling cycle.

[0064] Step 1.2: Based on the encoding results, set minimum duration constraints for driving and charging states to avoid frequent switching of working states. To avoid frequent switching of the electric truck's working state within the scheduling cycle, constraints are set on the coded driving and charging states of the electric truck according to the number of minimum duration segments for charging and driving, ensuring that driving and charging times last for at least μ time periods, thus obtaining driving and charging states of the electric truck that meet the constraints.

[0065] The minimum duration constraint for driving state is expressed by the following formula:

[0066] (2)

[0067] In the formula, This indicates that the b-th electric truck is in the... Driving status over a period of time, Indicates the minimum duration segment number, when Indicates the running index for a continuous time period. , Indicates the start time period number. .

[0068] The minimum duration constraint for the state of charge is expressed by the following formula:

[0069] (3)

[0070] In the formula, This indicates that the b-th electric truck is in the... The charging status over a period of time.

[0071] Step 1.3: Construct state mutual exclusion constraints. Determine the operating state of the electric truck based on the minimum duration constraints of the driving and charging states, as well as the state mutual exclusion constraints. The state mutual exclusion constraints are expressed by the following formula:

[0072] (4)

[0073] exist and The time is determined to be in a driving state; and It is determined to be in charging state at this time; and The system is then judged to be in an idling state.

[0074] It is worth noting that this invention solves the problems of increased energy consumption and equipment wear caused by frequent start-stop of electric trucks by establishing a minimum duration constraint model for driving and charging states, and solves the problems of equipment aging and efficiency decline caused by frequent start-stop of charging piles. This improves the working efficiency of electric trucks, reduces energy loss of the integrated energy system, improves charging efficiency, and reduces energy conversion loss during the charging process.

[0075] Step 2: Determine the actual power consumption of the electric truck based on the driving time and average power consumption of the electric truck on different routes under different working conditions.

[0076] In a preferred but non-limiting embodiment of the present invention, step 2 includes:

[0077] Step 2.1: Obtain the driving route between the inbound container yard and the outbound container yard at the seaport.

[0078] More preferably, step 2.1 includes:

[0079] Let M represent the number of inbound container yards in the seaport, and N represent the number of outbound container yards. Assume that within a scheduling cycle, the number of inbound containers equals the number of outbound containers. Ignore differences in berth space location, and the routes of electric trucks are determined by their access to the inbound container yards. and outbound container yard The choice is determined by multiplying the number of inbound container yards and the number of outbound container yards at the seaport to obtain the total driving route, which has a total of Available driving routes, the length of each driving route is calculated based on the spatial location between the inbound container yard and the outbound container yard.

[0080] Step 2.2: Determine the empty length and loaded length of the electric truck on the driving route, and the sum of the empty length and the loaded length is the total length of the driving route;

[0081] The product of power consumption and no-load length is the total no-load power consumption, and the product of loaded power consumption and loaded length is the total loaded power. The sum of the total no-load power consumption and the total loaded power, divided by the total length of the travel route, yields the average power consumption. It can be expressed by the following formula:

[0082] (5)

[0083] (6)

[0084] in, For electric trucks on the driving route Average power consumption on For electric trucks on the driving route Travel time on the road This indicates that the speed of the electric truck remains constant during transportation. Indicates no-load power consumption. Indicates the power consumption of the load. , , They represent the driving routes respectively. Total length, unloaded length, and loaded length.

[0085] Step 2.3: Calculate the actual power consumption of the electric truck based on its operating status, expressed by the following formula:

[0086] (7)

[0087] In the formula, This represents the actual power consumption of the b-th electric truck in the t-th time period. This represents the charging power consumption of the b-th electric truck in the t-th time period. This indicates the power consumption at idle speed.

[0088] It is worth noting that by combining factors such as driving status, charging status, and route selection to calculate actual power consumption, the problem of insufficient accuracy in existing technologies, such as the disconnect between electric truck scheduling and energy system scheduling and the failure of traditional energy consumption models to consider actual factors such as route gradient and driving speed, is solved. This reduces the scheduling deviation of the integrated energy system caused by energy consumption estimation errors, improves the coordination between electric truck scheduling and energy system scheduling, and reduces energy waste caused by the lack of coordination in the scheduling of the integrated energy system.

[0089] Step 3: Based on the actual power consumption of the electric trucks obtained in Step 2, construct a mean-field game theory model, transform the theoretical model into a neural network parameterized representation of the federated learning algorithm, obtain global model parameters through local training and parameter aggregation, and train a distributed cooperative scheduling model for electric trucks based on federated mean-field game using the deep mean-field Actor-Critic algorithm based on the global model parameters.

[0090] Step 3.1: Construct an average-field game theory model based on the actual power consumption of electric trucks.

[0091] More preferably, step 3.1 includes:

[0092] Step 3.1.1: Construct the state space of a single electric truck based on its battery charge status, location coordinates, number of outbound container transport tasks, driving status, charging status, and driving speed; construct the action space based on route selection variables and charging power.

[0093] More preferably, step 3.1.1 includes:

[0094] The state space of a single electric truck is represented by the following formula:

[0095] (8)

[0096] In the formula, Let b represent the state space of the b-th electric truck in the t-th time interval. This represents the state of charge of the battery of the b-th electric truck in the t-th time period. This represents the position coordinates of the b-th electric truck in the t-th time interval. This represents the number of inbound container transport tasks being performed by the b-th electric truck during the t-th time period. This represents the speed of the b-th electric truck during the t-th time period.

[0097] The action space is represented by the following formula:

[0098] (9)

[0099] In the formula, This represents the motion space of the b-th electric truck during the t-th time interval. This indicates whether the b-th truck chooses route s in the t-th time period.

[0100] Step 3.1.2: Based on the state space, action space, and actual power consumption of a single electric truck, establish the state dynamics equation of a single electric truck through stochastic differential equations; based on the state space, action space, travel time, and electricity price of a single electric truck, establish the long-term expected cost function of a single electric truck.

[0101] More preferably, step 3.1.2 includes:

[0102] Based on the state space and action space of a single electric truck, the state-dynamic equations of a single electric truck are established through stochastic differential equations, expressed as follows:

[0103] (10)

[0104] In the formula, Represents the state vector of the b-th electric truck. The differential, The deterministic drift term is represented by the following formula:

[0105] (11)

[0106] In the formula, Indicates selection based on route Calculated driving power consumption S represents the total number of routes. Indicates charging efficiency. Indicates selection based on route Calculate the direction vector. Indicates the direction of route s. Represents the mean field interaction intensity coefficient. Indicates truck density In position gradient at, Indicates the position in the t-th time period Truck density at the location This represents the task completion rate coefficient. Indicates an indicator function, The value is 1 if the condition is met, and 0 otherwise. Indicates the basic acceleration. The coefficient representing the sensitivity of speed to congestion. This represents a vector consisting entirely of 1s.

[0107] The diffusion term is represented by the following formula:

[0108] (12)

[0109] In the formula, Represents a diagonal matrix. The diffusion coefficient represents the state of charge of a battery. The diffusion coefficient represents the position coordinates. The diffusion coefficient represents the amount of task. The diffusion coefficient represents the velocity.

[0110] Driving status and charging status The diffusion coefficient is 0 because it is a discrete state. This represents the differential of the Wiener process.

[0111] Based on the state space, action space, travel time, and electricity price of a single electric truck, the long-term expected cost function of a single electric truck is established, expressed by the following formula:

[0112] (13)

[0113] In the formula, Indicates the b-th electric truck under a given strategy and mean field distribution The long-term expected cost, Represents the expectation operator. The instantaneous operating cost is expressed by the following formula:

[0114] (14)

[0115] In the formula, , , , , and Indicates the weighting coefficient. This represents the electricity price for the t-th time period. This indicates the number of containers that need to be transferred along route s. The location congestion penalty function is expressed by the following formula:

[0116] (15)

[0117] In the formula, This represents the congestion sensitivity coefficient. Indicates the lower limit of battery capacity. This indicates the maximum charging power.

[0118] Indicates the terminal time The cost is expressed by the following formula:

[0119] (16)

[0120] In the formula, , This represents the terminal cost weighting coefficient. This indicates that the b-th electric truck is at the terminal time. The amount of electricity, Indicates the target power level. Indicates an indicator function, hour ,otherwise .

[0121] It is worth noting that this invention solves the problem of delayed scheduling response at a fixed time scale in the prior art by constructing state dynamics equations and long-term expected cost functions, thereby improving the response speed of integrated energy system scheduling and reducing the efficiency loss caused by the delay in integrated energy system scheduling response.

[0122] Step 3.1.3: Based on the state dynamics equation and long-term expected cost function of a single electric truck, construct the theoretical mean field distribution parameters and mean field Nash equilibrium conditions. The mean field Nash equilibrium conditions include the HJB equation, the FP equation, and the theoretical optimal action, thus obtaining the mean field game theory model.

[0123] The theoretical mean field distribution is constructed as follows:

[0124] (17)

[0125] In the formula, This is the theoretical average field distribution, used to describe the density distribution of electric trucks in state x during the t-th time period. Let B represent the Dirac delta function, B represent the total number of electric trucks in the port, and x represent the state vector of the electric trucks.

[0126] The normalization condition for the theoretical mean field distribution is expressed by the following formula:

[0127] (18)

[0128] In the formula, Representing the state space, This represents the integral element in the state space.

[0129] The theoretical value function is set based on the long-term expected cost function, and is expressed by the following formula:

[0130] (19)

[0131] In the formula, Let be the theoretical value function, representing the minimum expected cost starting from state x in the t-th time period.

[0132] The Hamilton-Jacobi-Bellman equations are constructed based on the theoretical value function and the state dynamics equations, and are expressed as follows:

[0133] (20)

[0134] In the formula, Represents the theoretical value function The partial derivative with respect to the t-th time interval, Let V(t,x) represent the gradient of the theoretical value function V(t,x) with respect to the state x. The Hamiltonian is expressed by the following formula:

[0135] (twenty one)

[0136] In the formula, Represents the action space, This represents the action in action space A.

[0137] The Fock-Planck equation is constructed based on the theoretical mean field distribution and Hamiltonian, and is expressed as follows:

[0138] (twenty two)

[0139] In the formula, This represents the partial derivative of the theoretical mean field distribution m(t,x) with respect to time t. Denotes the divergence operator for state x. The partial derivative of the Hamiltonian H with respect to the gradient p is obtained according to the envelope theorem. , Let represent the theoretically optimal action in time interval t, state x. The theoretically optimal action is determined according to the Hamilton-Jacobi-Bellman equation and is expressed by the following formula:

[0140] (twenty three)

[0141] The boundary conditions for the Hamilton-Jacobi-Bellman equations and the Fock-Planck equations are constructed as follows:

[0142] (twenty four)

[0143] In the formula, For the initial theoretical mean field distribution, This is the terminal cost function.

[0144] It is worth noting that this invention solves the conflict between individual decision-making and group optimality in large-scale electric truck scheduling by constructing the mean-field Nash equilibrium condition, improves the optimality of group decision-making in the integrated energy system, achieves optimal scheduling under Nash equilibrium, and reduces congestion and efficiency decline caused by group conflict in the integrated energy system.

[0145] Step 3.2: Based on the mean-field game theory model obtained in Step 3.1, parameterize the mean-field game theory model into a neural network, and determine the global model parameters through local training and parameter aggregation.

[0146] Step 3.2.1: Map the mean-field game theory model obtained in Step 3.1 to a federated learning model. The federated learning model includes a global model and a regional local model.

[0147] The solutions to the HJB and FP equations in the mean-field game theory model obtained in step 3.1 are mapped to global model parameters, as expressed by the following formula:

[0148] (25)

[0149] In the formula, Represents the global model parameter set. Represents the global value function network parameters, corresponding to the solution of the HJB equation: theoretical value function ; Represents the global mean field distribution network parameters, corresponding to the solution of the FP equation: theoretical mean field distribution ; Represents the global policy network parameters, corresponding to the theoretically optimal action. , g represents the global context.

[0150] The local model for the region is expressed by the following formula:

[0151] (26)

[0152] In the formula, This represents the set of local model parameters for the k-th port region. This represents the local value function network parameters for the k-th port region. This represents the local mean field distribution network parameters of the k-th port area. This represents the local policy network parameters for the k-th port region, where k represents the region number of the port division (k=1, 2, ..., K), and K represents the total number of port divisions. Each region has its own local data and local model, and l represents local.

[0153] It is worth noting that this invention solves the problem of complex partial differential equations being difficult to solve directly by mapping theoretical models to federated learning, thereby improving the model solving efficiency of integrated energy systems, reducing the computational resource consumption of integrated energy systems, and supporting large-scale real-time scheduling of integrated energy systems.

[0154] Step 3.2.2: Based on the FP equation, parameterize the mean field distribution using a neural network, and use Fourier representation to represent the parameterized mean field distribution.

[0155] Use parameters as The neural network approximates the value function as the theoretical value function. ; Use parameters as The neural network approximates the average field distribution as the theoretical average field distribution. ; Use parameters as The neural network with the optimal action approximates the theoretical optimal action. .

[0156] The parameters are The neural network with the average field distribution is represented by the following formula:

[0157] (27)

[0158] In the formula, The parameter is The average field distribution, Represents the state feature mapping function. Represents a time feature mapping function. Represents the learnable weight matrix. , This represents the transpose symbol.

[0159] The parameter is Average field distribution Expanding this into a Fourier series yields the Fourier mean field. It is expressed by the following formula;

[0160] (28)

[0161] In the formula, Represents the imaginary unit. Represents the Fourier series index, n= N, N+1, ..., N 1, N, where N represents the truncation order of the Fourier series. Represents the vector dot product. This represents the nth Fourier coefficient.

[0162] Will Substitute into the FP equation to determine the Fourier coefficients. The dynamic equation is expressed by the following formula:

[0163] (29)

[0164] In the formula, Represents the set of Fourier coefficients The dynamic function used to describe the nth Fourier coefficient of the mean field distribution m(t,x) The rate of change with time t is expressed by the following formula:

[0165] (30)

[0166] In the formula, Represents the set of Fourier coefficients. , This indicates that the conditions are met simultaneously. Summing all combinations of (p,q), where p and q represent Fourier pattern indices of the same type as n. = N, N+1, ..., N 1, N, This represents the gradient operator with respect to state x. Represents the mean field interaction intensity coefficient. The Fourier coefficients represent the deterministic drift term. , Let q represent the q-th Fourier coefficient of the Fourier transform.

[0167] It is worth noting that this invention solves the problem of high computational complexity in solving FP equations using traditional numerical methods by parameterizing the average field distribution through neural networks, thereby improving the scheduling efficiency of integrated energy systems.

[0168] Step 3.2.3: Construct a local training loss function based on the HJB equation, and update the local parameters using gradient descent based on the local training loss function.

[0169] More preferably, step 3.2.3 includes:

[0170] Based on the HJB equations, a local training loss function is constructed, expressed as follows:

[0171] (31)

[0172] In the formula, This represents the local loss function for the k-th port region. Represents the local mean field distribution in the k-th port area The expectations below This represents the gradient of the local value function in the k-th region. Represents the gradient of the global value function. Denotes the square of the L2 norm. This represents the regularization coefficient, used to balance gradient matching and distribution consistency. The KL divergence term represents the local mean field distribution that approximates the global mean field distribution, ensuring the consistency of distribution across regions. This is a condition for mean field game equilibrium, preventing regional distributions from deviating excessively from the global distribution. It is expressed by the following formula:

[0173] (32)

[0174] Gradient matching term To make the gradient of the local value function approximate the gradient of the global value function, if the gradients of all port regions approximate the global gradient, then the global gradient approximately satisfies the HJB equation. Calculate the expectation under the current distribution, focusing on the high-probability region of the current distribution.

[0175] Local parameter updates are expressed by the following formula:

[0176] (33)

[0177] In the formula, Represents the local parameters of the k-th port area, including , Indicates the learning rate. Represents the loss function For local parameters The gradient.

[0178] It is worth noting that this invention solves the problem of high scheduling costs caused by random scenario generation in the prior art through local training and parameter aggregation, improves the robustness of the scheduling scheme, adapts to a variety of uncertain scenarios, reduces the scheduling cost of the integrated energy system caused by scenario assumption bias, and reduces the cost fluctuation of the integrated energy system.

[0179] Step 3.2.4: Aggregate the local parameters into the updated global model parameters. . Include .

[0180] Training stops when the maximum number of rounds is reached, and the trained global model parameters are obtained; otherwise, the number of rounds is incremented, and new global model parameters are issued for retraining.

[0181] Step 3.3: Based on the global model parameters, a distributed cooperative scheduling model for electric trucks based on federated mean field game is trained using the deep mean field Actor-Critic algorithm.

[0182] Step 3.3.1, the trained global model parameters Constructing a mean field distribution network .

[0183] Step 3.3.2, the trained global model parameters Set the parameters for the Critic network to construct the state value function network and action value function network of the Critic network.

[0184] State-valued function networks are represented by the following formula:

[0185] (34)

[0186] In the formula, Represents the state value function, The parameter is Multilayer perceptron, Representing state x and mean-field pooling characteristics splicing, The mean-field pooling characteristic is represented by the following formula:

[0187] (35)

[0188] In the formula, Indicates the mean field distribution The expectation of the next state x, Represents the mean field distribution The variance of the next state x Represents the mean field distribution The skewness of the lower state x.

[0189] Action-value function networks are represented by the following formula:

[0190] (36)

[0191] In the formula, Represents the action value function. Indicates an immediate reward. Indicates the discount factor. Indicates the target network parameters. Indicates the state at the next moment. Indicates the mean field distribution at the next moment. Indicates a time interval.

[0192] The training loss function for the Critic network is expressed by the following formula:

[0193] (37)

[0194] In the formula, This represents the loss function of the Critic network. This represents the expectation in the experience playback buffer D.

[0195] Step 3.3.3: Use the trained global model parameters as the parameters of the Actor network to construct the policy network.

[0196] The output of the policy network is expressed by the following formula:

[0197] (38)

[0198] In the formula, This represents the action probability distribution output by the policy network. Indicates the exploration level control parameter, Represents the normalization function. The raw output of the policy network is expressed by the following formula:

[0199] (39)

[0200] In the formula, Represents the weights of the neural network. This represents the bias of the neural network. This represents the activation function. Represents state x, action a, and mean-field pooling feature Pool( () splicing.

[0201] The training loss function for the Actor network is expressed by the following formula:

[0202] (40)

[0203] In the formula, This represents the loss function of the Actor network. Indicates the current strategy The expectations below Represents the entropy regularization coefficient. This represents the action value function output by the Critic network. The entropy representing the action probability distribution output by the policy network is expressed by the following formula:

[0204] (41)

[0205] Based on the training loss functions of the Critic network and the Actor network, the global model parameters are updated using gradient descent. This process continues until a set number of iterations is reached, resulting in a distributed collaborative scheduling model for electric trucks based on federated mean-field game theory.

[0206] It is worth noting that this invention solves the real-time conflict between electric truck charging and transportation tasks by implementing an Actor-Critic network, improves the task completion rate of the integrated energy system, completes the scheduling tasks of the integrated energy system while ensuring charging needs, reduces the delay penalty caused by task conflicts, and reduces the additional cost of integrated energy system scheduling.

[0207] Step 4: Input the real-time status observation data into the distributed cooperative scheduling model of electric trucks based on federated mean field game obtained in Step 3 to generate the optimal scheduling decision and output the charging power plan and route selection scheme.

[0208] The optimal scheduling decision is expressed by the following formula:

[0209] (42)

[0210] In the formula, Let represent the final optimal action of the b-th truck at time t. This represents the trained Actor network. This represents real-time status observation data. This indicates the optimal route selection. This indicates the optimal charging power.

[0211] Embodiment 2 of the present invention provides an integrated energy system control system considering transportation and logistics facilities, which operates the integrated energy system control method considering transportation and logistics facilities described in Embodiment 1, including:

[0212] The working state determination module is used to encode the driving and charging states of the electric truck, and set minimum duration constraints and state mutual exclusion constraints for the driving and charging states based on the encoding results to determine the working state of the electric truck.

[0213] The power consumption determination module is used to determine the actual power consumption of the electric truck based on the driving time and average power consumption of the electric truck on different routes under different operating conditions.

[0214] The model solving module is used to construct a mean-field game theory model based on the actual power consumption of electric trucks, transform the theoretical model into a neural network parameterized representation of the federated learning algorithm, determine the global model parameters through local training and parameter aggregation, and train a distributed cooperative scheduling model for electric trucks based on federated mean-field game using the deep mean-field Actor-Critic algorithm based on the global model parameters.

[0215] The output module is used to input real-time status observation data into the distributed cooperative scheduling model of electric trucks based on federated mean field game to generate optimal scheduling decisions, including optimal route selection and optimal charging power.

[0216] Compared with the prior art, the beneficial effects of the present invention include at least the following:

[0217] This invention reduces scheduling deviations caused by energy consumption estimation errors and energy conversion losses during charging by constraining the minimum duration of driving and charging states, thereby reducing energy waste in the overall energy system regulation.

[0218] This invention calculates actual power consumption by combining factors such as driving status, charging status, and route selection, thereby improving the coordination between electric truck dispatching and energy system dispatching and reducing energy waste caused by the lack of coordination in the overall energy system dispatching.

[0219] This invention improves the coordinated scheduling capability of various devices in an integrated energy system by constructing an average-field game model that includes a state space, action space, state dynamics equations, and a long-term expected cost function, thereby reducing resource waste caused by discrete scheduling and lowering the operating cost of the integrated energy system.

[0220] This invention resolves the contradiction of collaborative learning of multi-regional data in ports by parameterizing the average field game model into a neural network and training the model parameters through federated learning. This improves the real-time performance of the integrated energy system for scheduling large-scale electric trucks, enhances the accuracy and scientific nature of the integrated energy system's scheduling, and reduces the resource consumption of the integrated energy system.

[0221] This invention achieves the training and real-time scheduling of the Actor-Critic network based on pre-trained global parameters, solving the problem of fixed-ratio allocation of transferable loads and lack of dynamic adjustment capability in existing technologies. It improves the flexibility of load transfer in integrated energy systems, dynamically adjusts according to real-time status, reduces efficiency losses caused by fixed-ratio allocation in integrated energy system regulation, and improves the energy utilization rate of integrated energy systems.

[0222] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0223] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for regulating an integrated energy system considering transportation and logistics facilities, characterized in that, include: The driving and charging states of electric trucks are encoded, and the minimum duration constraints and state mutual exclusion constraints of driving and charging states are set based on the encoding results to determine the working state of electric trucks. The actual power consumption of the electric truck is determined based on the driving time and average power consumption of the electric truck on different routes under different working conditions. A mean-field game theory model is constructed based on the actual power consumption of electric trucks. The theoretical model is transformed into a neural network parameterized representation of the federated learning algorithm. The global model parameters are determined through local training and parameter aggregation. Based on the global model parameters, a distributed cooperative scheduling model for electric trucks based on federated mean-field game theory is trained using the deep mean-field Actor-Critic algorithm. Real-time status observation data is input into a distributed cooperative scheduling model for electric trucks based on federated mean field games to generate optimal scheduling decisions, including optimal route selection and optimal charging power.

2. The integrated energy system control method considering transportation and logistics facilities according to claim 1, characterized in that: The minimum duration constraints for driving and charging states include: The minimum duration constraint for driving state is expressed by the following formula: In the formula, This indicates that the b-th electric truck is in the... Driving status over a period of time, Indicates the minimum number of segments in duration. Indicates the running index for a continuous time period. , Indicates the start time period number. , This represents the total number of time periods in the scheduling cycle. The minimum duration constraint for the state of charge is expressed by the following formula: In the formula, This indicates that the b-th electric truck is in the... The charging status over a period of time.

3. The integrated energy system control method considering transportation and logistics facilities according to claim 2, characterized in that: A mutual exclusion constraint is expressed by the following formula: In the formula, This represents the time period number, t=1, 2, ... ; exist and The time is determined to be in a driving state; and It is determined to be in charging state at this time; and The system is then judged to be in an idling state.

4. The integrated energy system control method considering transportation and logistics facilities according to claim 3, characterized in that: The actual power consumption of an electric truck is expressed by the following formula: In the formula, This represents the actual power consumption of the b-th electric truck in the t-th time period. This represents the charging power consumption of the b-th electric truck in the t-th time period. This indicates the power consumption at idle. For electric trucks on the driving route Average power consumption.

5. The integrated energy system control method considering transportation and logistics facilities according to claim 4, characterized in that: The construction of the mean-field game theory model includes: The state space of an electric truck is constructed based on its battery state of charge, location coordinates, number of outbound container transport tasks, driving status, charging status, and driving speed; the action space is constructed based on route selection variables and charging power. Based on the state space, action space, and actual power consumption of a single electric truck, a state dynamics equation for a single electric truck is established using stochastic differential equations; based on the state space, action space, travel time, and electricity price of a single electric truck, a long-term expected cost function for a single electric truck is established. Based on the state dynamics equation and long-term expected cost function of a single electric truck, theoretical mean-field distribution parameters and mean-field Nash equilibrium conditions are constructed to obtain a mean-field game theory model; the mean-field Nash equilibrium conditions include the Hamilton-Jacobi-Bellman equation and the Fock-Planck equation.

6. The integrated energy system control method considering transportation and logistics facilities according to claim 5, characterized in that: The long-run expected cost function of a single electric truck is expressed by the following formula: In the formula, Indicates the b-th electric truck under a given strategy and mean field distribution The long-term expected cost, Represents the expectation operator. Indicates the instantaneous operating cost. Let b represent the state space of the b-th electric truck in the t-th time interval. Indicates the terminal time The cost.

7. The integrated energy system control method considering transportation and logistics facilities according to claim 6, characterized in that: The instantaneous operating cost is expressed by the following formula: In the formula, , , , , and Indicates the weighting coefficient. This represents the electricity price for the t-th time period. This indicates the number of containers that need to be transferred along route s. This represents the location congestion penalty function. Indicates the position in the t-th time period Truck density at the location Indicates the lower limit of battery capacity. Indicates the maximum charging power. This indicates whether the b-th truck chooses route s in the t-th time period. For electric trucks on the driving route Travel time on the road This indicates that the b-th electric truck is at the terminal time. The amount of electricity.

8. The integrated energy system control method considering transportation and logistics facilities according to claim 7, characterized in that: Indicates the terminal time The cost is expressed by the following formula: In the formula, , This represents the terminal cost weighting coefficient. This indicates that the b-th electric truck is at the terminal time. The amount of electricity, Indicates the target power level. Indicates an indicator function, hour ,otherwise .

9. The integrated energy system control method considering transportation and logistics facilities according to claim 5, characterized in that: Determining global model parameters includes: The mean-field game theory model is mapped to a federated learning model, which includes a global model and a local regional model. Based on the Fock-Planck equation, the mean field distribution is parameterized by a neural network and Fourier representation is used to represent the parameterized mean field distribution. A local training loss function is constructed based on the Hamilton-Jacobi-Bellman equation, and local parameters are updated by gradient descent based on the local training loss function. Aggregate local parameters into updated global model parameters; Training stops when the maximum number of rounds is reached, and the trained global model parameters are obtained; otherwise, the number of rounds is incremented, and new global model parameters are issued for retraining.

10. A comprehensive energy system incorporating transportation and logistics facilities, comprising the control method for the comprehensive energy system incorporating transportation and logistics facilities as described in any one of claims 1-9, characterized in that, include: The working state determination module is used to encode the driving and charging states of the electric truck, and set minimum duration constraints and state mutual exclusion constraints for the driving and charging states based on the encoding results to determine the working state of the electric truck. The power consumption determination module is used to determine the actual power consumption of the electric truck based on the driving time and average power consumption of the electric truck on different routes under different operating conditions. The model solving module is used to construct a mean-field game theory model based on the actual power consumption of electric trucks, transform the theoretical model into a neural network parameterized representation of the federated learning algorithm, determine the global model parameters through local training and parameter aggregation, and train a distributed cooperative scheduling model for electric trucks based on federated mean-field game using the deep mean-field Actor-Critic algorithm based on the global model parameters. The output module is used to input real-time status observation data into the distributed cooperative scheduling model of electric trucks based on federated mean field game to generate optimal scheduling decisions, including optimal route selection and optimal charging power.

Images