Hybrid bus energy-saving optimization control method and system and storage medium

Abstract

The invention discloses a hybrid bus energy-saving optimization control method and system and a storage medium, and the method comprises the steps: building an HEB traffic digraph based on the impact priority of traffic element distance on HEB of a hybrid bus, and constructing a multi-dimensional tensor composed of a vehicle state and traffic information; reducing the dimension of an input multi-dimensional matrix into a one-dimensional vector for the traffic directed graph through an edge-enhanced graph convolutional neural network EGCN to obtain spatial characteristics of data; eGCN embedded reinforcement learning is combined with a deep neural network DNN to form an ecological driving strategy; and the output of the ecological driving strategy is used as the input state of the energy management strategy to establish an information bridge, so that HEB speed planning at the traffic level and energy management at the power system level are collaboratively optimized, and HEB energy consumption and life decline of the power system are reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy management, and particularly relates to an energy-saving optimization control method and system for a hybrid electric bus, and a storage medium. Background Art

[0002] Transportation has always accounted for an important proportion in the energy consumption and emissions of human society. Facing the increasing pressure of energy and environmental problems, the transportation industry urgently needs effective energy-saving and emission-reduction solutions. HEB has the advantages of low or even zero emissions, renewable energy, and the traffic attribute of public transportation mode, and is an effective way to achieve green transportation. However, the unsatisfactory economy is still one of the problems restricting the development of HEB. The main ways to improve the economy of HEB are eco-driving and energy management. Therefore, governments and international organizations around the world are actively promoting research on eco-driving and energy management in order to improve energy utilization efficiency and reduce energy consumption.

[0003] With the development of network connection technology and big data, the interconnection between members of the traffic environment has become possible, which means that vehicles can receive and process more traffic information. The improvement of vehicle economy is not limited to the historical data and status of the vehicle itself, but extends to the traffic environment outside the vehicle. How to reasonably explore and utilize the role of multi-source traffic information on vehicle working conditions is the key factor in improving eco-driving and energy management of HEB. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an energy-saving optimization control method and system for a hybrid electric bus, and a storage medium, to improve the collaborative optimization strategy of eco-driving and energy management of MARL, analyze the interaction between HEB and each member in the traffic environment, design an EGCN to process the interaction between the bus and multi-source traffic information, quantify the interaction between nodes, and use it as an eco-driving intelligent agent for reinforcement learning to form a collaborative management strategy with energy management to reduce the economic cost in the process of eco-driving and energy management.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] An energy-saving optimization control method for a hybrid electric bus, comprising:

[0007] Step 1, obtain a traffic directed graph of a hybrid electric bus HEB;

[0008] Step 2, reduce the dimensionality of the input multi-dimensional matrix to a one-dimensional vector through an edge-enhanced graph convolutional neural network EGCN of the traffic directed graph to obtain the spatial characteristics of the data;

[0009] Step 3, embed EGCN into reinforcement learning and combine it with a deep neural network DNN to form an eco-driving strategy;

[0010] Step 4: The output of the eco-driving strategy is used as the input state of the energy management strategy to establish an information bridge, realizing the coordinated optimization of HEB speed planning at the traffic level and energy management at the power system level.

[0011] Preferably, in step S1, based on the priority of the impact of traffic element distance on HEB, a HEB traffic directed graph is established, and a multidimensional tensor consisting of vehicle status and traffic information is constructed.

[0012] Preferably, in step S3, an edge-enhanced graph convolutional neural network (EGCN) that quantifies and fuses multi-source traffic information around the HEB is used as an ecological driving reinforcement learning agent and is trained collaboratively with an energy management reinforcement learning agent.

[0013] Preferably, in step S4,

[0014] The EMS is constructed based on the DDPG framework of the DNN agent. The state space of the EMS includes the current vehicle speed, acceleration, power battery SOC, and fuel cell power, which can be expressed as:

[0015] s EMS =[v FCB ,a FCB ,SOC,P fc ]

[0016] Among them, a FCB is the output action from the eco-driving strategy;

[0017] EMS controls the power increment ΔP of the fuel cell fc To achieve the purpose of energy management, the action space is set to a EMS ∈[-4,4]kW, limiting the loss caused by the instantaneous power change of the fuel cell in the decision space, the state transition of the fuel cell power can be expressed as:

[0018] P fc (t) = P fc (t-1)+ΔP fc (t)

[0019] EMS adds SOC shaping reward items on the basis of energy consumption and life loss rewards to measure the current SOC and reference SOC (SOC ref ) distance, guiding the Agent to make the battery SOC close to SOC ref Decision making, the reward function can be expressed as:

[0020]

[0021] Among them, β1 and β2 are the weights of the economic reward function and the SOC shaping reward function respectively.

[0022] The present invention also provides an energy-saving optimization control system for a hybrid electric bus, including:

[0023] An acquisition module, configured to acquire a HEB traffic directed graph;

[0024] A first processing module, configured to reduce the dimensionality of the input multi-dimensional matrix to a one-dimensional vector through an EGCN for the traffic directed graph to obtain the spatial characteristics of the data;

[0025] A second processing module, configured to combine the EGCN embedded reinforcement learning with DNN to form an eco-driving strategy;

[0026] An optimization control module, configured to establish an information bridge with the output of the eco-driving strategy as the input state of the energy management strategy, and realize the collaborative optimization of the HEB speed planning at the traffic level and the energy management at the power system level.

[0027] Preferably, the acquisition module establishes a HEB traffic directed graph based on the influence priority of traffic elements on the HEB, and constructs a multi-dimensional tensor composed of vehicle states and traffic information.

[0028] Preferably, the second processing module is used for the edge-enhanced graph convolutional neural network EGCN that quantifies and fuses multi-source traffic information around the HEB, serves as an eco-driving reinforcement learning agent, and collaboratively trains with the energy management reinforcement learning agent.

[0029] Preferably, the calculation process of the optimization control module is as follows: an EMS is constructed based on the DDPG framework of the DNN agent, and the state space of the EMS includes the current vehicle speed, acceleration, power battery SOC, and fuel cell power, which is expressed as:

[0030] s EMS =[v FCB ,a FCB ,SOC,P fc

[0031] where a FCB is the output action from the eco-driving strategy;

[0032] The EMS realizes the energy management purpose by controlling the power increment ΔP fc of the fuel cell. The action space is set as a EMS ∈[-4,4]kW, and the loss caused by the instantaneous power change of the fuel cell is restricted in the decision space. The state transition of the fuel cell power can be expressed as:

[0033] P fc (t)=P fc (t - 1)+ΔP fc (t) ​

[0034] The EMS adds a SOC shaping reward term on the basis of energy consumption and life loss rewards to measure the distance between the current SOC and the reference SOC (SOC ref ), guiding the Agent to make decisions that bring the battery SOC closer to SOC ref . The reward function can be expressed as:

[0035]

[0036] where β1 and β2 are the weights of the economic reward function and the SOC shaping reward function respectively.

[0037] An embodiment of the present invention also provides a storage medium, on which a computer program is stored, and the computer program executes the energy-saving optimization control method for a hybrid electric bus when running.

[0038] Aiming at the key bottlenecks that hinder the improvement of the energy conversion and management efficiency of HEB in the connected environment, which come from the mining of traffic information, as well as the problem of the timeliness of the energy management strategy and the optimization of the balance between energies, the present invention adopts a collaborative optimization strategy of multi-agent reinforcement learning for HEB based on an improved agent to reduce the economic cost in the process of eco-driving and energy management. The main steps are as follows: Design an edge-enhanced graph convolutional neural network that quantifies and fuses multi-source traffic information around the HEB, as an eco-driving reinforcement learning agent, and co-train with the energy management reinforcement learning agent to realize the collaborative optimization of the HEB speed planning at the traffic level and the energy management at the powertrain level, reducing the energy consumption of the HEB and the life decline of the powertrain.

[0039] Compared with the prior art, the present invention has the following technical effects:

[0040] 1. The present invention proposes the concept of a spatio-temporal graph of the interaction between HEB and nearby traffic elements, as well as a traffic element priority influence mechanism, quantifying and fusing the influence of multi-source traffic information on the target vehicle.

[0041] 2. The present invention proposes an edge-enhanced graph convolutional neural network (EGCN) that quantifies and fuses multi-source traffic information around the HEB. By considering the node features of vehicle states and traffic information and the relative features of other nodes to the HEB node, it realizes the full utilization of multi-source information in the energy-saving strategy of buses.

[0042] 3. The present invention designs a multi-agent deep reinforcement learning strategy that embeds EGCN into eco-driving and powertrain energy management of reinforcement learning, designs an objective function considering energy consumption and powertrain cost, realizes the collaborative optimization of eco-driving and energy management, and reduces the energy consumption of the HEB and the life decline of the powertrain. Description of the Drawings

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the provided drawings.

[0044] Figure 1 It is a flowchart of the energy-saving optimization control method for a hybrid bus in an embodiment of the present invention;

[0045] Figure 2 It is the structure of the EGCN module in the present invention and its roadmap for extracting the characteristics of multi-source traffic information. Detailed implementation manners

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0047] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific implementation manners.

[0048] Embodiment 1:

[0049] As Figure 1 shown, an energy-saving optimization control method for a hybrid bus in an embodiment of the present invention describes a collaborative optimization strategy for the ecological driving and energy management of a fuel cell bus in a connected traffic environment. An EGCN is designed to handle the interaction between the bus and multi-source traffic information, quantify the interaction between nodes, and embed the EGCN into a new reinforcement learning algorithm. The intelligent agent can more efficiently capture the spatio-temporal dependence relationship of the target vehicle in the traffic flow during ecological driving. By establishing a multi-agent reinforcement learning strategy for ecological driving and energy management, an objective function considering energy consumption and powertrain cost is designed to achieve the collaborative optimization of ecological driving and energy management, including:

[0050] Step 1: Based on the influence priority of traffic elements' distance on the hybrid electric bus (HEB), establish a directed graph of HEB traffic and construct a multi-dimensional tensor composed of vehicle states and traffic information.

[0051] By identifying the relative positions of bus stops and traffic lights, the priority of their impact on bus operation is determined when quantifying and integrating multi-source traffic information. Specifically, in terms of ecological driving, the intelligent agent alternately processes the impact of bus stops and traffic lights on the target vehicle, which is reflected in four scenarios: whether there is a preceding vehicle during the entry process and whether there is a preceding vehicle at the intersection.

[0052] Establish a graph agent based on EGCN: G = (X, E), X represents node features, E represents edge features. During the HEB driving process, there are HEB and neighboring vehicles, and its node feature matrix is represented by X FCB and X nei , the edge feature matrix is represented as E FCB and E nei Node characteristics and vehicle status and traffic information include: own vehicle acceleration, own vehicle speed, own vehicle horizontal coordinate position, own vehicle vertical coordinate position, surrounding vehicle acceleration, surrounding vehicle speed, surrounding vehicle horizontal coordinate position, and surrounding vehicle vertical coordinate position.

[0053] Among them, the characteristic matrix of the ego vehicle and the surrounding vehicle can be expressed as X = [X FCB X nei ]:

[0054]

[0055] In order to fully express the impact of traffic information on HEB, the matrix extracts the information of the vehicle and a maximum of 6 surrounding vehicles, where a represents the node acceleration, v represents the node speed, and p represents the node position information.

[0056] E represents the interaction between node features, which are the relative acceleration, relative velocity, relative horizontal coordinate position and relative vertical coordinate position between other nodes and HEB node. Therefore, the edge features between nodes can be expressed as the edge feature matrix E = [E FCB E nei ]:

[0057]

[0058] In order to fully express the impact of traffic information on HEB, the matrix extracts the information of the ego vehicle and the surrounding vehicles with a maximum number of 6, where the first element is zero, indicating the impact of the ego vehicle on the ego vehicle. FCB represents the relative influence of the ego vehicle on the ego vehicle, Δa nei Indicates that Δv nei represents the relative speed between the vehicle and the surrounding vehicle, Δp nei Indicates the relative position of the self-vehicle and the surrounding vehicle.

[0059] The edge matrix needs to be normalized before filtering the feature matrix to avoid data differences across orders of magnitude. The edge matrix is normalized to:

[0060]

[0061] Among them, i represents the HEB and other traffic elements, and k represents the node features, including v, p, h, and t.

[0062] Similarly, the normalization of the node feature matrix needs to be performed separately for each type of feature, and the process is as follows:

[0063]

[0064] The state transition process of EGCN is as follows:

[0065]

[0066] Among them, among them, W l is the current weight matrix, X l-1 is the input state of the previous step length, represents the normalized edge matrix under the current feature.

[0067] Step 2: Reduce the dimensionality of the input multi-dimensional matrix to a one-dimensional vector through the edge-enhanced graph convolutional neural network EGCN for the traffic directed graph established in Step 1 to obtain the spatial characteristics of the data, enabling the intelligent agent to more efficiently capture the spatio-temporal dependence relationship of the target vehicle in the traffic flow during eco-driving.

[0068] EGCN is used to learn the spatial characteristics of the target vehicle. The spatial features include the node features of the vehicle state and traffic information and the relative features of other nodes and the HEB node, and perform dimensionality reduction and fusion on the node feature matrix to extract the effective spatial information in the graph structure.

[0069] Generally, the propagation rule of the GCN network is:

[0070]

[0071] Among them, represents the preprocessing step, is the matrix with self-connection added, I N is the identity matrix; is the degree matrix H (l) is the l-th layer, θ (l) contains the parameters of this layer.

[0072] Among them, A is the adjacency matrix, representing the connection between roads; X is the feature matrix; represents the preprocessing step; σ represents the sigmoid function of the non-linear model; W0 ∈ R P×H represents the weight matrix from the input to the hidden layer, P is the length of the feature matrix, and H is the number of hidden units.

[0073] In the present invention, EGCN enhances the processing of edge information on the basis of GCN, adds extra weights to the edges, strengthens the role of the edges in feature propagation, better captures node information, encodes the topological structure of the road network and the attributes on the road, obtains spatial dependence, extracts richer road information of the HEB ego vehicle and its surroundings, combines the relative positions of bus stops and traffic lights to prioritize the impact on the HEB, quantifies and fuses multi-source traffic information, and uses one layer of EGCN to process features, which is specifically embodied as follows:

[0074] f(X,A) = σ(EXW0)

[0075] s = s(f,Bni,Bei)

[0076] Among them, E is the HEB edge feature matrix, representing the relative position relationship between the surrounding vehicle and the HEB; Bni is the node feature of the bus stop or traffic light; Bei is the edge feature between the bus stop or traffic light and the HEB.

[0077] Through the collaborative optimization of the eco-driving strategy and the energy management strategy as Figure 2 shown, the specific implementation plan is as follows:

[0078] Step 3: Embed EGCN into the combination of reinforcement learning and the deep neural network DNN to form an eco-driving strategy, namely DEGN, in order to better capture the relationship between the HEB and traffic elements during the eco-driving decision-making process.

[0079] Combine the eco-driving map agent established in Step 2 with deep Q-learning (DQL) to design a deep learning framework based on EGCN, in order to better capture the relationship between the HEB and traffic elements during the eco-driving decision-making process.

[0080] DQL is a classic DRL algorithm. The agent selects an action a at the state s at time t according to the policy π(a|s) t and enters the next state s′ through the state transition matrix t in the environment, and obtains a reward r(s,a) from the environment. Based on the characteristics of the Bellman formula, design a reward function and continuously iteratively optimize a state-action value function until convergence. The reward function is as shown in the formula: until convergence. The reward function is as shown in the formula:

[0081]

[0082] Among them, θ is a parameter, γ ∈ (0,1) is a discount factor, is the state space dimension, is the action space dimension, and Q in the formula θ is represented by a network with parameter θ, Indicates that the parameter is θ - The target network, which is the current Q θ is a lagged version. During DQL training, exploration is based on the ε-greedy optimization strategy, and actions are selected with a probability of 1 - ε and action decisions are made with a probability of ε ∈ (0, 1].

[0083] DEGN uses the EGCN model to estimate the expected reward for taking actions in a given state, that is, EGCN is used instead of Q θ (s t ,a t ), so the Q-value function can be expressed as:

[0084]

[0085] The input state information of DEGN needs to consider the ego-vehicle state, traffic element states, and the influence of traffic elements on the ego-vehicle, and can be expressed as:

[0086]

[0087] Through EGCN, the X and E features are convolutionally dimension-reduced and fused with S B 、S T into a one-dimensional feature as the state input for eco-driving.

[0088] Among them, S X represents the vehicle state and multi-source traffic information, S E represents the influence of multi-source traffic information on the ego-vehicle, S B represents the relative features of the bus stop with respect to the HEB, and S T represents the relative features of the signalized intersection with respect to the HEB.

[0089] The action value of DEGN is the HEB acceleration (m / s 2 ), and based on the actual collected HEB operating conditions, the upper and lower limits of acceleration and the acceleration increment are designed to design the discrete action space:

[0090] a eco =[-1.5, -1.0, -0.5, 0, 0.5, 1.0, 1.5]

[0091] The main purpose of the eco-driving strategy is to optimize the economic losses caused by energy consumption and the degradation of the powertrain life at the operating condition level. Therefore, the reward function can be expressed as:

[0092]

[0093] Among them, is the unit price of hydrogen, c fc is the unit cost of fuel cell system life loss, c batis the unit price of the life loss cost of the power battery.

[0094] In eco-driving, to ensure vehicle economy, the vehicle's power demand needs to be met. The change in the vehicle's position is an obvious feature reflecting the dynamic performance. Therefore, the dynamic performance reward can be expressed as:

[0095]

[0096] During vehicle driving, safety is an element that the intelligent agent must consider. In longitudinal eco-driving decision-making, it is necessary to ensure a safe distance between the host vehicle and the vehicle ahead, specifically reflected in that the relative distance between the host vehicle and the vehicle ahead shall not be lower than the safe distance d safe , that is, the deceleration action made by the intelligent agent cannot cause the at the next moment. Specifically, it is as follows. Therefore, the safety reward function can be expressed as:

[0097]

[0098] Therefore, the eco-driving function can be expressed as:

[0099]

[0100] Among them, α1, α2, and α3 respectively represent the weights of the sub-reward functions.

[0101] Step 4: Use the output of the eco-driving strategy in Step 3 as the input state of the energy management strategy to establish an information bridge, realizing the collaborative optimization of the HEB speed planning at the traffic level and the energy management at the power system level, and reducing the HEB energy consumption and the power system life degradation.

[0102] The energy management strategy is responsible for optimizing the power system energy distribution while meeting the power demand. Given that the power distribution of the power system does not have a spatial attribute, an EMS is constructed based on the DDPG framework of the DNN intelligent agent to handle the continuous action space and refine the energy distribution decision.

[0103] The state space of the EMS includes the current vehicle speed, acceleration, power battery SOC, and fuel cell power, and can be expressed as:

[0104] s EMS =[v FCB ,a FCB ,SOC,P fc

[0105] Among them, a FCB is the output action from the eco-driving strategy.

[0106] The EMS realizes the energy management purpose by controlling the power increment ΔP fc of the fuel cell, and the action space is set as a EMS ​∈[-4,4] kW, the loss caused by the instantaneous power change of the fuel cell is restricted in the decision space, and the state transition of the fuel cell power can be expressed as:

[0107] P fc (t) = P fc (t - 1)+ΔP fc (t)

[0108] The EMS adds a SOC shaping reward term on the basis of the energy consumption and life loss rewards to measure the distance between the current SOC and the reference SOC (SOC ref ), and guides the Agent to make decisions that make the battery SOC close to SOC ref , avoiding overcharging / overdischarging of the power battery. The reward function can be expressed as:

[0109]

[0110] where β1 and β2 are the weights of the economic reward function and the SOC shaping reward function respectively.

[0111] Example 2: [[ID=2,7]]

[0112] The embodiment of the present invention also provides an energy-saving optimization control system for a hybrid electric bus, including:

[0113] An acquisition module, configured to acquire the HEB traffic directed graph;

[0114] A first processing module, configured to reduce the dimension of the input multi-dimensional matrix to a one-dimensional vector through the EGCN for the traffic directed graph to obtain the spatial characteristics of the data;

[0115] A second processing module, configured to combine the EGCN embedding reinforcement learning and the DNN to form an eco-driving strategy;

[0116] An optimization control module, configured to establish an information bridge with the output of the eco-driving strategy as the input state of the energy management strategy, and realize the collaborative optimization of the HEB speed planning at the traffic level and the energy management at the power system level.

[0117] As an implementation manner of the embodiment of the present invention, the acquisition module establishes the HEB traffic directed graph based on the influence priority of traffic elements on the HEB, and constructs a multi-dimensional tensor composed of vehicle states and traffic information.

[0118] As an implementation manner of the embodiment of the present invention, the second processing module is used to quantify and fuse the edge-enhanced graph convolutional neural network EGCN of multi-source traffic information around the HEB, as an eco-driving reinforcement learning agent, and co-train with the energy management reinforcement learning agent.

[0119] As an implementation manner of the embodiment of the present invention, the calculation process of the optimization control module is as follows: An EMS is constructed based on the DDPG framework of the DNN agent. The state space of the EMS includes the current vehicle speed, acceleration, power battery SOC, and fuel cell power, which is expressed as:

[0120] s EMS =[v FCB ,a FCB ,SOC,P fc

[0121] where a FCB is the output action from the eco-driving strategy;

[0122] The EMS realizes the energy management purpose by controlling the power increment ΔP fc of the fuel cell. The action space is set as a EMS ∈[-4,4]kW, and the loss caused by the instantaneous power change of the fuel cell is restricted in the decision space. The state transition of the fuel cell power can be expressed as:

[0123] P fc (t)=P fc (t - 1)+ΔP fc (t)

[0124] The EMS adds a SOC shaping reward term on the basis of the energy consumption and life loss rewards to measure the distance between the current SOC and the reference SOC (SOC ref ), and guides the Agent to make a decision to make the battery SOC close to SOC ref . The reward function can be expressed as:

[0125]

[0126] where β1 and β2 are the weights of the economic reward function and the SOC shaping reward function respectively.

[0127] Embodiment 3:

[0128] The embodiment of the present invention also provides a storage medium, on which a computer program is stored, and the computer program executes the energy-saving optimization control method for a hybrid bus when running.

[0129] The above embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.​

Claims

1. A method for optimizing energy saving control of a hybrid bus, characterized in that, Including: Step 1: Obtain the traffic directed graph of the hybrid electric bus (HEB); Step 2: Use the edge-enhanced graph convolutional neural network (EGCN) to reduce the dimensionality of the input multi-dimensional matrix of the traffic directed graph to a one-dimensional vector to obtain the spatial characteristics of the data; Step 3: Embed EGCN into reinforcement learning and combine it with the deep neural network (DNN) to form an eco-driving strategy; Step 4: Use the output of the eco-driving strategy as the input state of the energy management strategy to establish an information bridge, and achieve the collaborative optimization of the speed planning of the HEB at the traffic level and the energy management at the power system level.

2. The energy-saving optimization control method for a hybrid bus according to claim 1, wherein In Step S1, based on the influence priority of traffic elements' distances on the HEB, establish the traffic directed graph of the HEB, and construct a multi-dimensional tensor composed of vehicle states and traffic information.

3. The energy-saving optimization control method for a hybrid bus according to claim 2, wherein In Step S3, the edge-enhanced graph convolutional neural network (EGCN) that quantifies and fuses multi-source traffic information around the HEB is used as an eco-driving reinforcement learning agent and co-trained with the energy management reinforcement learning agent.

4. The energy-saving optimization control method for a hybrid bus according to claim 3, characterized in that In Step S4, Build an energy management system (EMS) based on the DDPG framework of the DNN agent. The state space of the EMS includes the current vehicle speed, acceleration, state of charge (SOC) of the power battery, and fuel cell power, expressed as: s EMS = [v FCB , a FCB , SOC, P fc ​ where a FCB is the output action from the eco-driving strategy; The EMS controls the power increment ΔP of the fuel cell fc to achieve the purpose of energy management, and the action space is set as a EMS ∈[-4,4]kW. The loss caused by the instantaneous power change of the fuel cell is restricted in the decision space, and the state transition of the fuel cell power can be expressed as: P fc P(t) = fc P(t - 1)+ΔP fc (t) EMS adds an SOC shaping reward term on the basis of energy consumption and life loss rewards to measure the distance between the current SOC and the reference SOC (SOC ref ), guiding the Agent to make decisions that bring the battery SOC closer to SOC ref . The reward function can be expressed as: where β1 and β2 are the weights of the economic reward function and the SOC shaping reward function respectively.

5. A hybrid bus energy-saving optimization control system, characterized in that, Including: An acquisition module for obtaining the traffic directed graph of the HEB; A first processing module for using the EGCN to reduce the dimensionality of the input multi-dimensional matrix of the traffic directed graph to a one-dimensional vector to obtain the spatial characteristics of the data; A second processing module for embedding EGCN into reinforcement learning and combining it with the DNN to form an eco-driving strategy; An optimization control module for using the output of the eco-driving strategy as the input state of the energy management strategy to establish an information bridge, and achieving the collaborative optimization of the speed planning of the HEB at the traffic level and the energy management at the power system level.

6. The energy-saving optimization control system for a hybrid bus according to claim 4, characterized in that, The acquisition module, based on the influence priority of traffic elements' distances on the HEB, establishes the traffic directed graph of the HEB and constructs a multi-dimensional tensor composed of vehicle states and traffic information.

7. The energy-saving optimization control system for a hybrid bus according to claim 5, wherein The second processing module is used to quantify and fuse the edge-enhanced graph convolutional neural network (EGCN) of multi-source traffic information around the HEB, as an eco-driving reinforcement learning agent, and co-train with the energy management reinforcement learning agent.

8. The energy-saving optimization control system for a hybrid bus according to claim 7, characterized in that, The calculation process of the optimization control module is: build an EMS based on the DDPG framework of the DNN agent. The state space of the EMS includes the current vehicle speed, acceleration, state of charge (SOC) of the power battery, and fuel cell power, expressed as: s EMS = [v FCB , a FCB , SOC, P fc ​ where a FCB is the output action from the eco-driving strategy; The EMS controls the power increment ΔP of the fuel cell fc to achieve the purpose of energy management, and the action space is set as a EMS ∈[-4,4]kW, and the loss caused by the instantaneous power change of the fuel cell is restricted in the decision space. The state transition of the fuel cell power can be expressed as: P fc P(t) = fc P(t - 1)+ΔP fc (t) EMS adds an SOC shaping reward term on the basis of energy consumption and life loss rewards to measure the distance between the current SOC and the reference SOC (SOC ref ), guiding the Agent to make decisions that bring the battery SOC closer to SOC ref . The reward function can be expressed as: where β1 and β2 are the weights of the economic reward function and the SOC shaping reward function respectively.

9. A storage medium, characterized in that, A computer program is stored on the storage medium, and the computer program, when running, executes the energy-saving optimization control method for the hybrid electric bus as described in any one of claims 1 to 4.

Images