New energy vehicle energy intelligent management method and system based on space-time variation of road network conditions

Abstract

The application provides a new energy vehicle energy intelligent management method and system based on road network condition space-time change, comprising: searching q paths with distance satisfying a preset condition for a given start and end point pair; constructing a hierarchical generalized regression neural network, and predicting the theoretical optimal energy consumption of each path in the q paths by using the constructed hierarchical generalized regression neural network; selecting the path with the lowest theoretical optimal energy consumption based on the theoretical optimal energy consumption of each path in the q paths, and taking the path with the lowest theoretical optimal energy consumption as the driving path of the new energy vehicle; constructing a bidirectional long short-term memory neural network, and predicting the global optimal SOC trajectory facing the current new energy vehicle driving path by using the constructed bidirectional long short-term memory neural network; in the vehicle journey, following the predicted global optimal SOC trajectory by using the adaptive equivalent fuel consumption minimum method, so that the actual SOC trajectory and the predicted global optimal SOC trajectory have the minimum deviation, and thus the optimal control instruction is output.

Description

Technical Field

[0001] This invention relates to the field of automotive energy-saving control technology, specifically to a method and system for intelligent energy management of new energy vehicles based on spatiotemporal changes in road network conditions, and more specifically to a method and system for predicting the most energy-efficient path for the entire road network and the globally optimal SOC trajectory for the entire path for new energy vehicles. Background Technology

[0002] With the increasing number of cars on the road, energy consumption in the transportation sector is rising, leading to serious environmental and energy shortage problems. Therefore, in order to reduce energy consumption in the transportation sector and further conserve energy and reduce emissions, developing new energy vehicles to replace traditional vehicles has become a key technological approach and has garnered widespread attention from academia and industry.

[0003] Vehicle energy management (EMS) is a key focus for further optimizing the energy consumption of new energy vehicles. New energy vehicles achieve different energy consumption levels under different operating conditions when different energy management strategies are applied. Theoretically, only an EMS that optimizes based on the entire operating condition—a global EMS—can achieve globally optimal energy consumption. However, global EMS, such as dynamic programming (DP) and the Pontryagin Minimum Principle (PMP), requires prior knowledge of the global operating conditions before the vehicle departs, which is impossible in real-world traffic scenarios and therefore cannot be applied in real time. Therefore, current online EMS applications are mostly local EMS aimed at optimizing instantaneous or short-range energy consumption. In particular, to improve the utilization of unknown operating conditions and enhance the adaptability of EMS, Model Predictive Control (MPC-EMS) is widely used due to its short-range operating condition prediction capabilities. However, due to limitations in current operating condition prediction methods, it is impossible to predict the operating conditions every second of the entire journey, so MPC-EMS cannot achieve theoretically optimal energy consumption. Meanwhile, current operating condition prediction methods are limited to utilizing operating condition characteristics, such as historical speed trajectories, average acceleration, and driving intentions of new energy vehicles. However, in urban traffic environments, influenced by numerous traffic facilities and large traffic volumes, the spatiotemporal dynamic changes in road conditions, i.e., macroscopic traffic parameters (MTP), play a more significant role in influencing vehicle operating conditions. Current EMS (Energy Management Systems) neglect the application of MTP, thus their energy-saving effects need improvement.

[0004] Furthermore, existing EMS (Energy Management Systems) are limited to developing optimal EMS for specific driving routes, i.e., optimal EMS for the entire route. However, in urban traffic networks, for the same pair of origin and destination points, different routes have different road conditions, which affects the vehicle's driving conditions and thus leads to different theoretically optimal energy consumption for different routes. Therefore, if a theoretically energy-optimal route for the entire road network can be found before the car departs, and the car travels on this route, the potential of the road network to reduce the energy consumption of new energy vehicles can be further explored. Compared to EMS that only targets a specific route, the energy consumption of new energy vehicles will be further reduced.

[0005] In summary, existing EMS (Energy Management Systems) cannot enable new energy vehicles to achieve optimal energy consumption at both the entire road network and the entire route scale. Furthermore, existing EMS neglects the impact of road network traffic parameters on the energy efficiency of new energy vehicles. Summary of the Invention

[0006] In view of the deficiencies in the existing technology, the purpose of this invention is to provide a method and system for intelligent energy management of new energy vehicles based on the spatiotemporal changes of road network conditions.

[0007] A new energy vehicle energy intelligent management method based on spatiotemporal changes in road network conditions, provided by the present invention, includes:

[0008] Step S1: Search for q paths whose distances from the given origin and destination satisfy a preset condition;

[0009] Step S2: Construct a hierarchical generalized regression neural network and use the constructed hierarchical generalized regression neural network to predict the theoretical optimal energy consumption of each of the q paths;

[0010] Step S3: Based on the theoretical optimal energy consumption of each of the q paths, select the path with the lowest theoretical optimal energy consumption, and use the path with the lowest theoretical optimal energy consumption as the driving path of the new energy vehicle.

[0011] Step S4: Construct a bidirectional long short-term memory neural network and use the constructed bidirectional long short-term memory neural network to predict the globally optimal SOC trajectory for the current driving path of the new energy vehicle.

[0012] Step S5: During the vehicle's journey, the adaptive equivalent fuel consumption minimization method is used to follow the predicted global optimal SOC trajectory, minimizing the deviation between the actual SOC trajectory and the predicted global optimal SOC trajectory, thereby outputting the optimal control command.

[0013] The hierarchical generalized regression neural network predicts the travel time of a road segment through an upper-layer generalized regression neural network (GRNN), and feeds back the predicted travel time and macroscopic traffic parameters obtained based on the travel time to the lower-layer generalized regression neural network, thereby achieving the prediction of the theoretical optimal energy consumption for the entire process.

[0014] The bidirectional long short-term memory neural network performs forward and backward calculations through two layers of long short-term memory neural network, thereby enabling the prediction of the entire SOC trajectory.

[0015] Preferably, step S2 employs:

[0016] Step S2.1: Construct an upper-layer generalized regression neural network to predict the travel time of each segment on q paths in turn, and obtain the macroscopic traffic parameters of each segment on the path in turn based on the travel time;

[0017] Step S2.2: Calculate the macro-traffic parameters of the route based on the macro-traffic parameters of the road segment;

[0018] Step S2.3: Construct a lower-level generalized regression neural network to predict the theoretical optimal energy consumption of q paths based on the macroscopic traffic parameters of the paths;

[0019] The macroscopic traffic parameters for the road segment include: remaining red light time r i,t Green light remaining time g i,t Distance between road sections l i Traffic density ρ of road segment i,t and average speed of the road section Where i represents the i-th segment of the current path, and t represents time t;

[0020] The macroscopic traffic parameters of the route include: average route speed. n represents the number of road segments; the path speed difference is... Path speed standard deviation Average traffic density and path length L n .

[0021] Preferably, step S2.1 employs the following:

[0022] Step S2.1.1: Record the time η when arriving at the current road segment i. i ;

[0023] Step S2.1.2: Based on the arrival time η of the current road segment i Obtain the macroscopic traffic parameters of the current road segment;

[0024] Step S2.1.3: Based on the macroscopic traffic parameters of the current road segment, use the upper-level generalized regression neural network to predict the travel time Δt of the current road segment. i ;

[0025] Step S2.1.4: Based on the arrival time η of the current road segment iThe time η for arriving at the next road segment is obtained by combining the predicted travel time of the current road segment with the predicted travel time of the current road segment. i+1 =η i +Δt i Repeat steps S2.1.1 to S2.1.4 until the travel time and macroscopic traffic parameters for each segment of the entire path are obtained.

[0026] Preferably, step 2.1.3 includes:

[0027] Step 2.1.3.1: Macroscopic traffic parameter vector of road segment i at time t The total number of input samples is Q t This indicates that the subscript t represents the target prediction time; Q is... t Each sample is processed into a matrix X, where each column of matrix X represents a sample.

[0028] Step 2.1.3.2: Construct the upper-layer generalized regression neural network, including the input layer, pattern layer, summation layer, and output layer;

[0029] Step 2.1.3.3: Data is passed from the input layer to the pattern layer. In the pattern layer, each neuron is calculated using the following formula to obtain the pattern layer output:

[0030]

[0031] in, It is the output of the j-th neuron in the pattern layer; the superscript t is the predicted target of the upper-layer GRNN: abbreviated as time; dist is the Euclidean distance function; X j δ represents the j-th column of matrix X; δ is the smoothing parameter of the network;

[0032] Step 2.1.3.4: The data processed by the schema layer is passed to the summation layer for calculation; the summation layer can be divided into two parts, and the calculation formula for the first part is as follows:

[0033]

[0034] Among them, S t This is the output of the first part of the summation layer;

[0035] The formula for the second part is as follows:

[0036]

[0037] Among them, y jK This represents the Kth feature of the j-th output sample;

[0038] Step 2.1.3.5: The calculation results from both parts are passed to the output layer, and the prediction result is obtained through the following calculation formula:

[0039]

[0040] in, This represents the predicted travel time of the i-th road segment at time t.

[0041] Preferably, step S2.2 employs the following:

[0042] The macroscopic traffic parameters of the selected route include the average speed of the route. Path speed standard deviation Extremely poor path speed Path length L n Path traffic density

[0043] Where n represents the number of road segments in the current path; the average speed of the path. Path speed standard deviation Extremely poor path speed Path length L n Path traffic density The calculation formulas are as follows:

[0044]

[0045]

[0046]

[0047]

[0048]

[0049] in, Indicates the average speed of vehicles on the road segment; l i Indicates the distance between road segments; ρ i,t Indicates the traffic density of a road segment.

[0050] Preferably, step S2.3 involves: using a lower-level generalized regression neural network and a dynamic programming algorithm (DP) to calculate the true value of the theoretically optimal energy consumption for the entire path.

[0051] Preferably, step S4 employs the following methods:

[0052] Step S4.1: Construct a bidirectional long short-term memory neural network. Based on the macroscopic traffic parameters of the road segments along the driving path of new energy vehicles, use the constructed bidirectional long short-term memory neural network to predict the optimal SOC gradient of each road segment on the path.

[0053] Step S4.2: Calculate the globally optimal SOC trajectory by combining the initial SOC and the segment length;

[0054]

[0055] Among them, SOC i+1 This represents the SOC value at the end of the i-th road segment; SOC init Represents the initial value of SOC; It is the predicted value of the SOC gradient of the optimal road segment.

[0056] Preferably, step S4.1 employs the following:

[0057] Step S4.1.1: Process the macroscopic traffic parameters of the road segment into vectors.

[0058] Step S4.1.2: Construct a bidirectional long short-term memory neural network model, including: an input layer, a fully connected layer, an LSTM layer, a fully connected layer, and an output layer; wherein the LSTM layer includes a forward LSTM layer and a backward LSTM layer;

[0059] Step S4.1.3: Vector The input layer is fed into the fully connected layer, and the fully connected layer is used to extract the dynamic traffic features of the road segment;

[0060]

[0061] Among them, W xi These are the weights of the fully connected layer, b xi It is the bias of the fully connected layer;

[0062] Step S4.1.4: Input the dynamic traffic features into the LSTM layer, and use the forward LSTM layer to learn the influence of the dynamic traffic parameters of the upstream road segment; the data transmission direction of the forward LSTM layer is consistent with the vehicle's driving direction; the working principle of each LSTM unit in the forward LSTM layer is shown in the following formula:

[0063]

[0064] Where i is the i-th LSTM unit. C represents the hidden state output of the current LSTM cell. i This represents the current state of the LSTM cell; σ represents the sigmoid function. It's weight. It is a bias; It will be passed to the next neuron in the feedforward LSTM layer. It will be passed to the next neuron in the feedforward LSTM layer and the fully connected layer;

[0065] Step S4.1.5: The input data also enters the backward LSTM layer, which processes the data using the following formula:

[0066]

[0067] in, This indicates the hidden state output of the current LSTM cell. This represents the current state of the LSTM cell; σ represents the sigmoid function. It's weight. It is a bias; It will be passed to the next neuron in the feedforward LSTM layer. It will be passed to the next neuron in the feedforward LSTM layer and the fully connected layer;

[0068] Step S4.1.5: Output data of the forward LSTM layer and the backward LSTM layer The output is passed to the fully connected layer for extraction.

[0069]

[0070] Among them, W yi b represents the weights of the fully connected layer. yi Indicates the bias of the fully connected layer; This represents the predicted value of the SOC gradient for the optimal road segment.

[0071] Preferably, step S5 employs the following:

[0072] Step S5.1: Set the initial equivalence factor λ0;

[0073] Step S5.2: Specify the control variables as engine torque, motor torque, engine speed, motor speed, and battery power, and the state variable as SOC; the control variables and state variables must meet the following constraints:

[0074]

[0075] Among them, T e (t) represents the engine torque, T m (t) represents the motor torque, ω e (t) represents the engine speed, ω m (t) represents the motor torque, P BP This represents the battery power; the subscripts _max and _min represent the maximum and minimum values ​​of the corresponding items, respectively.

[0076] Step S5.3: Discretize the limit range of the control quantity under each second operating condition to obtain the discrete points of all control variables;

[0077] Step S5.4: For all feasible outcomes consisting of discrete points of all control variables, calculate the equivalent cost using the following formula:

[0078] in,

[0079] J = ∑min{(m eq (t)+a1ΔICE(t)+a2|G trans (t)-G trans (t-1)|)}

[0080] m eq (t)=m f (t)+λ(t)×P BP (t) / F LHV

[0081] Where, m eq (t) represents instantaneous equivalent fuel consumption; m f (t) represents instantaneous fuel consumption; λ(t) represents the equivalent factor; P BP (t) represents instantaneous electric power; F LHV Indicates the lower heating value of fuel; ΔICE(t) represents the change in engine start-stop state; G trans (t) represents the gear; a1 and a2 represent the weights of the costs caused by engine start-stop changes and gear shifting, respectively, and are calibrated by the test bench;

[0082] Step S5.5: Select the feasible result with the smallest J;

[0083] Step S5.6: Calculate the actual SOC corresponding to the feasible result described in step S5.5. Based on the error between the actual SOC and the reference SOC, adjust the equivalent factor using incremental PID control, as shown in the following formula:

[0084] Δλ(t)=k p [e(t)-e(t-1)]+k i e(t)+k d [e(t)+e(t-2)-2e(t-1)]

[0085] e(t) = SOC T (t)-SOC R (t)

[0086] The equivalent factor at the current moment is:

[0087] λ(t) = λ(t-1) + Δλ(t)

[0088] Where, k p k i k d These are the proportional coefficient, integral coefficient, and differential coefficient, respectively; SOCT (t) represents the actual SOC control value, SOC R (t) represents the reference SOC value;

[0089] Repeat steps S5.4 to S5.6 until the difference between the actual SOC and the reference SOC is less than the threshold.

[0090] According to the present invention, a new energy vehicle energy intelligent management system based on spatiotemporal changes in road network conditions is characterized by comprising:

[0091] Module M1: Searches for q paths whose distances from a given origin and destination pair satisfy a preset condition;

[0092] Module M2: Construct a hierarchical generalized regression neural network and use the constructed hierarchical generalized regression neural network to predict the theoretical optimal energy consumption of each of the q paths;

[0093] Module M3: Based on the theoretical optimal energy consumption of each of the q paths, select the path with the lowest theoretical optimal energy consumption and use the path with the lowest theoretical optimal energy consumption as the driving path of the new energy vehicle.

[0094] Module M4: Constructs a bidirectional long short-term memory neural network and uses the constructed bidirectional long short-term memory neural network to predict the globally optimal SOC trajectory for the current driving path of the new energy vehicle;

[0095] Module M5: During the vehicle's journey, it uses the adaptive equivalent fuel consumption minimization method to follow the predicted global optimal SOC trajectory, minimizing the deviation between the actual SOC trajectory and the predicted global optimal SOC trajectory, thereby outputting the optimal control command.

[0096] The hierarchical generalized regression neural network predicts the travel time of a road segment through an upper-layer generalized regression neural network (GRNN), and feeds back the predicted travel time and macroscopic traffic parameters obtained based on the travel time to the lower-layer generalized regression neural network, thereby achieving the prediction of the theoretical optimal energy consumption for the entire process.

[0097] The bidirectional long short-term memory neural network performs forward and backward calculations through two layers of long short-term memory neural network, thereby enabling the prediction of the entire SOC trajectory.

[0098] Compared with the prior art, the present invention has the following beneficial effects:

[0099] 1. This invention, by predicting the theoretically optimal energy consumption of new energy vehicles, realizes the most energy-efficient path planning for new energy vehicles across the entire road network, providing a theoretically optimal energy consumption path for new energy vehicles;

[0100] 2. This invention achieves prediction of the globally optimal SOC trajectory of new energy vehicles based on macroscopic traffic parameters, breaking through the limitation of current EMS prediction based only on short-range operating conditions.

[0101] 3. This invention combines the above-developed optimal energy-saving path planning for new energy vehicles across the entire road network with the optimal SOC trajectory prediction algorithm for the entire path, thereby achieving global optimal energy management for new energy vehicles and enabling them to achieve optimal energy consumption across the entire road network. Attached Figure Description

[0102] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0103] Figure 1 This is a flowchart of a new energy vehicle energy management method based on the spatiotemporal changes of road network conditions.

[0104] Figure 2 This is a schematic diagram of the optimal energy consumption prediction model based on the hierarchical GRNN path theory.

[0105] Figure 3 This is a schematic diagram illustrating the principle of Bi-LSTM optimal road segment SOC gradient prediction. Detailed Implementation

[0106] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0107] Example 1

[0108] According to the present invention, a method for intelligent energy management of new energy vehicles based on spatiotemporal changes in road network conditions is provided, such as... Figure 1 The above includes:

[0109] Step S1: Search for q paths whose distances from the given origin and destination satisfy a preset condition;

[0110] Step S2: Construct a hierarchical generalized regression neural network (HGRNN) and use the constructed hierarchical generalized regression neural network to predict the theoretical optimal energy consumption of each of the q paths;

[0111] Step S3: Based on the theoretical optimal energy consumption of each of the q paths, select the path with the lowest theoretical optimal energy consumption, and use the path with the lowest theoretical optimal energy consumption as the driving path of the new energy vehicle.

[0112] Step S4: Construct a bidirectional long short-term memory neural network (Bi-LSTM) and use the constructed bidirectional long short-term memory neural network to predict the globally optimal SOC trajectory for the current driving path of the new energy vehicle.

[0113] Step S5: During the vehicle's journey, the Adaptive Equivalent Minimum Fuel Consumption Method (AECMS) is used to follow the predicted global optimal SOC trajectory, minimizing the deviation between the actual SOC trajectory and the predicted global optimal SOC trajectory, thereby outputting the optimal control command.

[0114] The hierarchical generalized regression neural network (GRNN) predicts road segment travel time using an upper-layer GRNN and feeds back the predicted travel time and macroscopic traffic parameters obtained based on the travel time to the lower-layer GRNN, thereby achieving prediction of the theoretically optimal energy consumption for the entire route. The working principle of the hierarchical GRNN is as follows: Figure 2 As shown;

[0115] The bidirectional long short-term memory neural network predicts the entire SOC trajectory by performing forward and backward calculations using two layers of long short-term memory neural networks. The working principle of the bidirectional long short-term memory neural network is as follows: Figure 3 As shown;

[0116] Specifically, step S2 employs the following:

[0117] Step S2.1: Construct an upper-layer generalized regression neural network to predict the travel time of each segment on q paths in turn, and obtain the macroscopic traffic parameters of each segment on the path in turn based on the travel time;

[0118] Step S2.2: Calculate the macro-traffic parameters of the route based on the macro-traffic parameters of the road segment;

[0119] Step S2.3: Construct a lower-level generalized regression neural network to predict the theoretical optimal energy consumption of q paths based on the macroscopic traffic parameters of the paths;

[0120] The road segments on the route are divided according to the topological structure, that is, the intersections are used as the basis for the road segment division;

[0121] Macroscopic traffic parameters are obtained from Intelligent Transportation Systems (ITS) or traffic forecasting methods;

[0122] The macroscopic traffic parameters for the road segment include: remaining red light time r i,t Green light remaining time g i,t Distance between road sections l i Traffic density ρ of road segment i,t and average speed of the road section Where i represents the i-th segment of the current path, and t represents time t;

[0123] The macroscopic traffic parameters of the route include: average route speed. n represents the number of road segments; the path speed difference is... Path speed standard deviation Average traffic density and path length L n .

[0124] Specifically, step S2.1 employs the following:

[0125] Step S2.1.1: Record the time η when arriving at the current road segment i. i ;

[0126] Step S2.1.2: Based on the arrival time η of the current road segment i Obtain the macroscopic traffic parameters of the current road segment;

[0127] Step S2.1.3: Based on the macroscopic traffic parameters of the current road segment, use the upper-level generalized regression neural network to predict the travel time Δt of the current road segment. i ;

[0128] Step S2.1.4: Based on the arrival time η of the current road segment i The time η for arriving at the next road segment is obtained by combining the predicted travel time of the current road segment with the predicted travel time of the current road segment. i+1 =η i +Δt i Repeat steps S2.1.1 to S2.1.4 until the travel time and macroscopic traffic parameters for each segment of the entire path are obtained.

[0129] Specifically, step 2.1.3 includes:

[0130] Step 2.1.3.1: Macroscopic traffic parameter vector of road segment i at time t The total number of input samples is Q t This indicates that the subscript t represents the target prediction time; Q is... t Each sample is processed into a matrix X, where each column of matrix X represents a sample.

[0131] Step 2.1.3.2: Construct the upper-layer generalized regression neural network, including the input layer, pattern layer, summation layer, and output layer;

[0132] Step 2.1.3.3: Data is passed from the input layer to the pattern layer. In the pattern layer, each neuron is calculated using the following formula to obtain the pattern layer output:

[0133]

[0134] in, It is the output of the j-th neuron in the pattern layer; the superscript t is the predicted target of the upper-layer GRNN: abbreviated as time; dist is the Euclidean distance function; X j δ represents the j-th column of matrix X; δ is the smoothing parameter of the network;

[0135] Step 2.1.3.4: The data processed by the schema layer is passed to the summation layer for calculation; the summation layer can be divided into two parts, and the calculation formula for the first part is as follows:

[0136]

[0137] Among them, S t This is the output of the first part of the summation layer;

[0138] The formula for the second part is as follows:

[0139]

[0140] Among them, y jK This represents the Kth feature of the j-th output sample;

[0141] Step 2.1.3.5: The calculation results from both parts are passed to the output layer, and the prediction result is obtained through the following calculation formula:

[0142]

[0143] in, This represents the predicted travel time of the i-th road segment at time t.

[0144] Specifically, step S2.2 employs the following:

[0145] The macroscopic traffic parameters of the selected route include the average speed of the route. Path speed standard deviation Extremely poor path speed Path length L n Path traffic density

[0146] Where n represents the number of road segments in the current path; the average speed of the path. Path speed standard deviation Extremely poor path speed Path length L n Path traffic density The calculation formulas are as follows:

[0147]

[0148]

[0149]

[0150]

[0151]

[0152] in, Indicates the average speed of vehicles on the road segment; l i Indicates the distance between road segments; ρ i,t Indicates the traffic density of a road segment.

[0153] Specifically, step S2.3 employs the following method: based on a lower-layer generalized regression neural network, the true value of the theoretically optimal energy consumption for the entire path is calculated using a dynamic programming algorithm (DP). The prediction principle of the lower-layer GRNN is the same as that of the upper-layer GRNN described in steps 2.1.3.1 to 2.1.3.5; the only difference lies in the input samples. The input samples of the lower-layer GRNN are...

[0154] Specifically, step S4 employs the following:

[0155] Step S4.1: Construct a bidirectional long short-term memory neural network. Based on the macroscopic traffic parameters of the road segments along the driving path of new energy vehicles, use the constructed bidirectional long short-term memory neural network to predict the optimal SOC gradient of each road segment on the path.

[0156] Step S4.2: Calculate the globally optimal SOC trajectory by combining the initial SOC (known value) and the road segment length;

[0157]

[0158] Among them, SOC i+1 This represents the SOC value at the end of the i-th road segment; SOC init Represents the initial value of SOC; This represents the predicted optimal SOC gradient for road segment i.

[0159] Specifically, step S4.1 employs the following:

[0160] Step S4.1.1: Process the macroscopic traffic parameters of the road segment into vectors.

[0161] Step S4.1.2: Construct a bidirectional long short-term memory neural network model, including: an input layer, a fully connected layer, an LSTM layer, and an output layer; wherein, the LSTM layer includes a forward LSTM layer and a backward LSTM layer; the data transmission direction between LSTM units in the forward LSTM layer is the same as the vehicle's driving direction, while that in the backward LSTM layer is the opposite. Therefore, the processing of the forward LSTM layer considers the influence of macroscopic traffic parameters of the upstream road segment on the SOC gradient of the downstream road segment, and the processing of the backward LSTM layer considers the influence of macroscopic traffic parameters of the downstream road segment on the upstream road segment. Considering the influence between road segments improves prediction accuracy.

[0162] Step S4.1.3: Vector The input layer is fed into the fully connected layer, and the fully connected layer is used to extract the dynamic traffic features of the road segment;

[0163]

[0164] Among them, W xi These are the weights of the fully connected layer, b xi It is the bias of the fully connected layer;

[0165] Step S4.1.4: Input the dynamic traffic features into the LSTM layer, and use the forward LSTM layer to learn the influence of the dynamic traffic parameters of the upstream road segment; the data transmission direction of the forward LSTM layer is consistent with the vehicle's driving direction; the working principle of each LSTM unit in the forward LSTM layer is shown in the following formula:

[0166]

[0167] Where i is the i-th LSTM unit. C represents the hidden state output of the current LSTM cell. i This represents the current state of the LSTM cell; σ represents the sigmoid function. It's weight. It is a bias; It will be passed to the next neuron in the feedforward LSTM layer. It will be passed to the next neuron in the feedforward LSTM layer and the fully connected layer;

[0168] Step S4.1.5: The input data also enters the backward LSTM layer, which processes the data using the following formula:

[0169]

[0170] in, This indicates the hidden state output of the current LSTM cell. This represents the current state of the LSTM cell; σ represents the sigmoid function. It's weight. It is a bias; It will be passed to the next neuron in the feedforward LSTM layer. It will be passed to the next neuron in the feedforward LSTM layer and the fully connected layer;

[0171] Step S4.1.5: Output data of the forward LSTM layer and the backward LSTM layer The output is passed to the fully connected layer for extraction.

[0172]

[0173] Among them, W yi b represents the weights of the fully connected layer. yi Indicates the bias of the fully connected layer; This represents the predicted value of the SOC gradient for the optimal road segment.

[0174] Specifically, step S5 employs the following:

[0175] Step S5.1: Set the initial equivalence factor λ0;

[0176] Step S5.2: Specify the control variables as engine torque, motor torque, engine speed, motor speed, and battery power, and the state variable as SOC; the control variables and state variables must meet the following constraints:

[0177]

[0178] Among them, T e (t) represents the engine torque, T m (t) represents the motor torque, ω e (t) represents the engine speed, ω m (t) represents the motor torque, P BP This represents the battery power; the subscripts _max and _min represent the maximum and minimum values ​​of the corresponding items, respectively.

[0179] Step S5.3: Discretize the limit range of the control quantity under each second operating condition to obtain the discrete points of all control variables;

[0180] Step S5.4: For all feasible outcomes consisting of discrete points of all control variables, calculate the equivalent cost using the following formula:

[0181] in,

[0182] J = ∑min{(m eq (t)+a1ΔICE(t)+a2|G trans (t)-G trans(t-1)|)}

[0183] m eq (t)=m f (t)+λ(t)×P BP (t) / F LHV

[0184] Where, m eq (t) represents instantaneous equivalent fuel consumption; m f (t) represents instantaneous fuel consumption; λ(t) represents the equivalent factor; P BP (t) represents instantaneous electric power; F LHV Indicates the lower heating value of fuel; ΔICE(t) represents the change in engine start-stop state; G trans (t) represents the gear; a1 and a2 represent the weights of the costs caused by engine start-stop changes and gear shifting, respectively, and are calibrated by the test bench;

[0185] Step S5.5: Select the feasible result with the smallest J;

[0186] Step S5.6: Calculate the actual SOC corresponding to the feasible result described in step S5.5. Based on the error between the actual SOC and the reference SOC, adjust the equivalent factor using incremental PID control, as shown in the following formula:

[0187] Δλ(t)=k p [e(t)-e(t-1)]+k i e(t)+k d [e(t)+e(t-2)-2e(t-1)]

[0188] e(t) = SOC T (t)-SOC R (t)

[0189] The equivalent factor at the current moment is:

[0190] λ(t) = λ(t-1) + Δλ(t)

[0191] Where, k p k i k d These are the proportional coefficient, integral coefficient, and differential coefficient, respectively; SOC T (t) represents the actual SOC control value, SOC R (t) represents the reference SOC value;

[0192] Repeat steps S5.4 to S5.6 until the difference between the actual SOC and the reference SOC is less than the threshold.

[0193] According to the present invention, a new energy vehicle energy intelligent management system based on spatiotemporal changes in road network conditions includes:

[0194] Module M1: Searches for q paths whose distances from a given origin and destination pair satisfy a preset condition;

[0195] Module M2: Construct a hierarchical generalized regression neural network (HGRNN) and use the constructed hierarchical generalized regression neural network to predict the theoretical optimal energy consumption of each of the q paths;

[0196] Module M3: Based on the theoretical optimal energy consumption of each of the q paths, select the path with the lowest theoretical optimal energy consumption and use the path with the lowest theoretical optimal energy consumption as the driving path of the new energy vehicle.

[0197] Module M4: Constructs a bidirectional long short-term memory neural network (Bi-LSTM) and uses the constructed bidirectional long short-term memory neural network to predict the globally optimal SOC trajectory for the current driving path of the new energy vehicle;

[0198] Module M5: During the vehicle's journey, it uses the Adaptive Equivalent Minimum Fuel Consumption Method (AECMS) to follow the predicted global optimal SOC trajectory, minimizing the deviation between the actual SOC trajectory and the predicted global optimal SOC trajectory, thereby outputting the optimal control command.

[0199] The hierarchical generalized regression neural network (GRNN) predicts road segment travel time using an upper-layer GRNN and feeds back the predicted travel time and macroscopic traffic parameters obtained based on the travel time to the lower-layer GRNN, thereby achieving prediction of the theoretically optimal energy consumption for the entire route. The working principle of the hierarchical GRNN is as follows: Figure 2 As shown;

[0200] The bidirectional long short-term memory neural network predicts the entire SOC trajectory by performing forward and backward calculations using two layers of long short-term memory neural networks. The working principle of the bidirectional long short-term memory neural network is as follows: Figure 3 As shown;

[0201] Specifically, module M2 adopts:

[0202] Module M2.1: Constructs an upper-layer generalized regression neural network to predict the travel time of each segment on q paths in turn, and obtains the macroscopic traffic parameters of each segment on the path in turn based on the travel time;

[0203] Module M2.2: Calculates macroscopic traffic parameters of the route based on macroscopic traffic parameters of road segments;

[0204] Module M2.3: Constructs a lower-level generalized regression neural network to predict the theoretical optimal energy consumption of q paths based on macroscopic traffic parameters of the paths;

[0205] The road segments on the route are divided according to the topological structure, that is, the intersections are used as the basis for the road segment division;

[0206] Macroscopic traffic parameters are obtained from Intelligent Transportation Systems (ITS) or traffic forecasting methods;

[0207] The macroscopic traffic parameters for the road segment include: remaining red light time r i,t Green light remaining time g i,t Distance between road sections l i Traffic density ρ of road segment i,t and average speed of the road section Where i represents the i-th segment of the current path, and t represents time t;

[0208] The macroscopic traffic parameters of the route include: average route speed. n represents the number of road segments; the path speed difference is... Path speed standard deviation Average traffic density and path length L n .

[0209] Specifically, module M2.1 adopts:

[0210] Module M2.1.1: Records the arrival time η of the current road segment i. i ;

[0211] Module M2.1.2: Based on the arrival time η of the current road segment i Obtain the macroscopic traffic parameters of the current road segment;

[0212] Module M2.1.3: Based on the macroscopic traffic parameters of the current road segment, it uses an upper-level generalized regression neural network to predict the travel time Δt of the current road segment. i ;

[0213] Module M2.1.4: Based on the arrival time η of the current road segment i The time η for arriving at the next road segment is obtained by combining the predicted travel time of the current road segment with the predicted travel time of the current road segment. i+1 =η i +Δt i Repeatedly trigger modules M2.1.1 to M2.1.4 until the travel time and macroscopic traffic parameters of each segment on the entire path are obtained.

[0214] Specifically, step 2.1.3 includes:

[0215] Step 2.1.3.1: Macroscopic traffic parameter vector of road segment i at time t The total number of input samples is Q t This indicates that the subscript t represents the target prediction time; Q is... t Each sample is processed into a matrix X, where each column of matrix X represents a sample.

[0216] Step 2.1.3.2: Construct the upper-layer generalized regression neural network, including the input layer, pattern layer, summation layer, and output layer;

[0217] Step 2.1.3.3: Data is passed from the input layer to the pattern layer. In the pattern layer, each neuron is calculated using the following formula to obtain the pattern layer output:

[0218]

[0219] in, It is the output of the j-th neuron in the pattern layer; the superscript t is the predicted target of the upper-layer GRNN: abbreviated as time; dist is the Euclidean distance function; X j δ represents the j-th column of matrix X; δ is the smoothing parameter of the network;

[0220] Step 2.1.3.4: The data processed by the schema layer is passed to the summation layer for calculation; the summation layer can be divided into two parts, and the calculation formula for the first part is as follows:

[0221]

[0222] Among them, S t This is the output of the first part of the summation layer;

[0223] The formula for the second part is as follows:

[0224]

[0225] Among them, y jK This represents the Kth feature of the j-th output sample;

[0226] Step 2.1.3.5: The calculation results from both parts are passed to the output layer, and the prediction result is obtained through the following calculation formula:

[0227]

[0228] in, This represents the predicted travel time of the i-th road segment at time t.

[0229] Specifically, module M2.2 adopts:

[0230] The macroscopic traffic parameters of the selected route include the average speed of the route. Path speed standard deviation Extremely poor path speed Path length L n Path traffic density

[0231] Where n represents the number of road segments in the current path; the average speed of the path. Path speed standard deviation Extremely poor path speed Path length L n Path traffic density The calculation formulas are as follows:

[0232]

[0233]

[0234]

[0235]

[0236]

[0237] in, Indicates the average speed of vehicles on the road segment; l i Indicates the distance between road segments; ρ i,t Indicates the traffic density of a road segment.

[0238] Specifically, module M2.3 employs a dynamic programming algorithm (DP) based on a lower-layer generalized regression neural network (GRNN) to calculate the true value of the theoretically optimal energy consumption for the entire path. The prediction principle of the lower-layer GRNN is the same as that of the upper-layer GRNN described in steps 2.1.3.1 to 2.1.3.5; the only difference lies in the input samples. The input samples for the lower-layer GRNN are...

[0239] Specifically, module M4 adopts:

[0240] Module M4.1: Construct a bidirectional long short-term memory neural network to predict the optimal SOC gradient of each road segment based on the macroscopic traffic parameters of the driving path of new energy vehicles.

[0241] Module M4.2: Calculates the globally optimal SOC trajectory by combining the initial SOC (known value) and the road segment length;

[0242]

[0243] Among them, SOC i+1 This represents the SOC value at the end of the i-th road segment; SOC initRepresents the initial value of SOC; This represents the predicted optimal SOC gradient for road segment i.

[0244] Specifically, module M4.1 adopts:

[0245] Module M4.1.1: Processes macroscopic traffic parameters of road segments into vectors.

[0246] Module M4.1.2: Constructs a bidirectional long short-term memory neural network model, including: an input layer, a fully connected layer, an LSTM layer, and an output layer; wherein, the LSTM layer includes a forward LSTM layer and a backward LSTM layer; the data transmission direction between LSTM units in the forward LSTM layer is the same as the vehicle's driving direction, while that in the backward LSTM layer is the opposite. Therefore, the processing of the forward LSTM layer considers the influence of macroscopic traffic parameters of the upstream road segment on the SOC gradient of the downstream road segment, and the processing of the backward LSTM layer considers the influence of macroscopic traffic parameters of the downstream road segment on the upstream road segment. Considering the influence between road segments improves prediction accuracy.

[0247] Module M4.1.3: Vectors The input layer is fed into the fully connected layer, and the fully connected layer is used to extract the dynamic traffic features of the road segment;

[0248]

[0249] Among them, W xi These are the weights of the fully connected layer, b xi It is the bias of the fully connected layer;

[0250] Module M4.1.4: Inputs dynamic traffic features into the LSTM layer, utilizing the forward LSTM layer to learn the influence of dynamic traffic parameters from the upstream road segment; the data transmission direction of the forward LSTM layer is consistent with the vehicle's travel direction; the working principle of each LSTM unit in the forward LSTM layer is shown in the following formula:

[0251]

[0252] Where i is the i-th LSTM unit. C represents the hidden state output of the current LSTM cell. i This represents the current state of the LSTM cell; σ represents the sigmoid function. It's weight. It is a bias; It will be passed to the next neuron in the feedforward LSTM layer. It will be passed to the next neuron in the feedforward LSTM layer and the fully connected layer;

[0253] Module M4.1.5: Input data also enters the backward LSTM layer, which processes the data using the following formula:

[0254]

[0255] in, This indicates the hidden state output of the current LSTM cell. This represents the current state of the LSTM cell; σ represents the sigmoid function. It's weight. It is a bias; It will be passed to the next neuron in the feedforward LSTM layer. It will be passed to the next neuron in the feedforward LSTM layer and the fully connected layer;

[0256] Module M4.1.5: Output data of the forward LSTM layer and the backward LSTM layer The output is passed to the fully connected layer for extraction.

[0257]

[0258] Among them, W yi b represents the weights of the fully connected layer. yi Indicates the bias of the fully connected layer; This represents the predicted value of the SOC gradient for the optimal road segment.

[0259] Specifically, module M5 adopts:

[0260] Module M5.1: Set the initial equivalence factor λ0;

[0261] Module M5.2: The specified control variables are engine torque, motor torque, engine speed, motor speed, and battery power; the state variable is SOC. The control variables and state variables must meet the following constraints:

[0262]

[0263] Among them, T e (t) represents the engine torque, T m (t) represents the motor torque, ω e (t) represents the engine speed, ω m (t) represents the motor torque, P BP This represents the battery power; the subscripts _max and _min represent the maximum and minimum values ​​of the corresponding items, respectively.

[0264] Module M5.3: Discretizes the limit range of the control quantity under each second operating condition to obtain the discrete points of all control variables;

[0265] Module M5.4: For all feasible outcomes consisting of discrete points of all control variables, calculate the equivalent cost using the following formula:

[0266] in,

[0267] J = ∑min{(m eq (t)+a1ΔICE(t)+a2|G trans (t)-G trans (t-1)|)}

[0268] m eq (t)=m f (t)+λ(t)×P BP (t) / F LHV

[0269] Where, m eq (t) represents instantaneous equivalent fuel consumption; m f (t) represents instantaneous fuel consumption; λ(t) represents the equivalent factor; P BP (t) represents instantaneous electric power; F LHV Indicates the lower heating value of fuel; ΔICE(t) represents the change in engine start-stop state; G trans (t) represents the gear; a1 and a2 represent the weights of the costs caused by engine start-stop changes and gear shifting, respectively, and are calibrated by the test bench;

[0270] Module M5.5: Select the option with the smallest J among all feasible results;

[0271] Module M5.6: Calculates the actual SOC corresponding to the feasible result described in module M5.5. Based on the error between the actual SOC and the reference SOC, it uses incremental PID adjustment of the equivalent factor, as shown in the following formula:

[0272] Δλ(t)=k p [e(t)-e(t-1)]+k i e(t)+k d [e(t)+e(t-2)-2e(t-1)]

[0273] e(t) = SOC T (t)-SOC R (t)

[0274] The equivalent factor at the current moment is:

[0275] λ(t) = λ(t-1) + Δλ(t)

[0276] Where, k p k i k d These are the proportional coefficient, integral coefficient, and differential coefficient, respectively; SOCT (t) represents the actual SOC control value, SOC R (t) represents the reference SOC value;

[0277] Repeat modules M5.4 through M5.6 until the difference between the actual SOC and the reference SOC is less than the threshold.

[0278] Example 2

[0279] Example 2 is a preferred example of Example 1.

[0280] Appendix Figure 1 This is a flowchart illustrating the new energy vehicle energy management method based on spatiotemporal changes in road network conditions, as described in this invention. Figure 1 As shown, the new energy vehicle energy management method based on the spatiotemporal changes of road network conditions provided by this invention includes the following steps:

[0281] Step S1: Given a starting point and an ending point, use a navigation system or ITS to obtain the path and path length between the starting point and the ending point pair, and select the q paths with the shortest distance.

[0282] Step S2: Based on macroscopic traffic parameters, construct a hierarchical GRNN (HGRNN) model to predict the theoretical optimal energy consumption of these q paths.

[0283] Appendix Figure 2 The schematic diagram of the hierarchical GRNN (HGRNN) described in step S2 includes:

[0284] Step S201: Based on the macroscopic traffic parameters of the road segments, construct an upper-level GRNN model to predict the travel time of each road segment in these q paths.

[0285] Step S201.1: Record the time η of arrival at the current road segment. i ;

[0286] Where i represents the i-th segment of a certain path;

[0287] Step S201.2: Combine the current time η of the current road segment i Using ITS or traffic forecasting methods, macroscopic traffic parameters of the current road segment are obtained, including the remaining red light time r. i,t Green light remaining time g i,t Distance between road sections l i Traffic density ρ of road segment i,t Average speed of vehicles on the road section

[0288] Where t represents the current time as time t; ρ i,t =α i,t / l i αi,t The number of vehicles boarding at the i-th road segment; k represents the kth vehicle, v k (t) represents the speed of the k-th vehicle at time t. The road segment is divided by a topological structure, i.e., intersections.

[0289] If the current road segment is the first road segment, then the remaining red light time r i,t Green light remaining time g i,t Distance between road sections l i Traffic density ρ of road segment i,t Average speed of vehicles on the road section Obtain from ITS;

[0290] If the current road segment is not the first road segment, then the remaining red light time r i,t Green light remaining time g i,t Distance between road sections l i Traffic density ρ of road segment obtained from ITS i,t Average speed of vehicles on the road section Obtained through traffic forecasting methods;

[0291] Step S201.3: Based on the macroscopic traffic parameters of the current road segment, i.e., the remaining red light time r i,t Green light remaining time g i,t Distance between road sections l i Traffic density ρ of road segment i,t Average speed of vehicles on the road section The travel time Δτ for the current road segment is predicted using a generalized regressive neural network (GRNN). i,t ;

[0292] To achieve step S201.3, the GRNN model needs to be trained first, and then the trained model is used to predict the travel time of the road segment. The training steps are as follows:

[0293] Step S201.31: Obtain the remaining red light time r from the ITS. i,t Green light remaining time g i,t Distance between road sections l i Traffic density ρ of road segment i,t Average speed of vehicles on the road section Travel time Δτ i,t And perform data processing, including filtering and max-min normalization;

[0294] Step S201.32: In MATLAB, construct the upper-level GRNN, which has four layers. The first layer is the input layer with 5 neurons; the second layer is the pattern layer with the same number of neurons as the number of input samples; the third layer is the summary layer; and the fourth layer is the output layer, which outputs the predicted road segment travel time.

[0295] Step S201.33: Save the trained upper-layer GRNN.

[0296] After training, perform the following steps:

[0297] Step S201.4: Combine the current arrival time η of the road i And the travel time Δτ of the current road segment predicted from the upper-level GRNN model. i,t The arrival time η of the next road segment is obtained. i+1 =η i +Δτ i,t ;

[0298] Step S201.5: Repeat steps S202 to S204 until the travel time Δτ for each segment of the entire path is calculated. i,t Macro-level traffic parameters, namely, the remaining red light time r. i,t Green light remaining time g i,t Distance between road sections l i Traffic density ρ of road segment i,t Average speed of vehicles on the road section They all got it.

[0299] Step S202: Based on the macroscopic traffic parameters of the path, construct a lower-level GRNN to predict the theoretical optimal energy consumption of these q paths.

[0300] To achieve step S202, the lower-level GRNN model needs to be trained first, and then the trained model needs to be used to predict the theoretically optimal energy consumption. The training steps are as follows:

[0301] Step S202.1: Obtain road segment distance l offline from ITS i Traffic density ρ of road segment i,t Average speed of vehicles on the road section Calculate the path average speed Path speed standard deviation Extremely poor path speed Path length L n Path traffic density As input features of the model; the operating conditions are obtained offline from the vehicle CAN communication, and the true value of the theoretical optimal energy consumption is calculated by combining DP with the full operating conditions. This value is used as the target value for model prediction, and data processing is performed, including filtering and maximum-minimum normalization.

[0302] Step S202.2: In MATLAB, construct the lower-level GRNN, which has four layers. The first layer is the input layer with 5 neurons; the second layer is the pattern layer with the same number of neurons as the number of input samples; the third layer is the summary layer; and the fourth layer is the output layer, which outputs the theoretically optimal energy consumption of the predicted path.

[0303] Step S202.3: Save the trained lower-level GRNN.

[0304] Then, using the trained lower-level GRNN, online prediction of the theoretically optimal energy consumption is performed, specifically including:

[0305] Step S202.4: Receive the macroscopic traffic parameters of each road segment in the q paths obtained from the upper-layer GRNN, and calculate the average speed of each path. Path speed standard deviation Extremely poor path speed Path length L n Path traffic density

[0306] Where n represents the number of road segments; the average speed of the path Path speed standard deviation Extremely poor path speed Path length L n Path traffic density The calculation formulas are as follows:

[0307]

[0308]

[0309]

[0310]

[0311]

[0312] Step S202.5: Input the macroscopic traffic parameters of the path calculated above into the lower-level GRNN model trained in steps S202.1 to S202.3 to obtain the theoretical optimal energy consumption of the q paths.

[0313] Step S3: From these q paths, select the theoretically optimal path with the lowest energy consumption as the most energy-efficient path planned for the new energy vehicle, and make it travel on this path.

[0314] Step S4: Based on the macroscopic traffic parameters of the road segment, construct a Bi-LSTM model to predict the optimal SOC gradient for each road segment on the path.

[0315] Appendix Figure 3 The Bi-LSTM optimal road segment SOC gradient prediction model described in step S4 includes:

[0316] To achieve step S4, a Bi-LSTM model needs to be trained first, and then the trained model needs to be used to predict the optimal road segment SOC gradient. The training steps are as follows:

[0317] Step S401: Obtain the remaining red light time r from the ITS offline. i,t Green light remaining time g i,t Distance between road sections l i Traffic density ρ of road segment i,t Average speed of vehicles on the road section The data is then processed, including filtering and min-max normalization, using the samples as input and converting each sample into a vector.

[0318] Step S402: Obtain the global operating conditions offline from CAN and use DP to calculate the true value of the optimal SOC gradient for each road segment throughout the entire process, which is used as the prediction target of Bi-LSTM.

[0319] Step S402: In MATLAB, construct a bidirectional LSTM neural network (Bi-LSTM) with 5 layers. The first layer is the input layer, which contains macroscopic traffic parameters of the road segment; the second layer is a fully connected layer; the third layer is the Bi-LSTM layer; the fourth layer is a fully connected layer; and the fifth layer is the output layer.

[0320] In step S402, the computation and training steps of the Bi-LSTM model are as follows:

[0321] Step S402.1: Extract the dynamic traffic features of the road segment using the fully connected layer. The calculation formula is as follows:

[0322]

[0323] Among them W xi These are the weights of the fully connected layer, b xi These are the biases of the fully connected layer, and they are all parameters that need to be trained.

[0324] Step S402.2: Use the forward LSTM layer to learn the impact of dynamic traffic parameters of the upstream road segment. The calculation formula is as follows:

[0325]

[0326] Where f f This represents the principle of LSTM neuron computation.

[0327] Step S402.3: Use the backward LSTM layer to learn the impact of dynamic traffic parameters of the downstream road segment. The calculation formula is as follows:

[0328]

[0329] Step S402.4: Extract the output result using a fully connected layer. The calculation formula is as follows:

[0330]

[0331] Among them, W yi These are the weights of the fully connected layer, b yi These are the biases of the fully connected layer, and they are all parameters that need to be trained; It is the predicted value of the SOC gradient of the optimal road segment.

[0332] Step S402.5: Calculate the error between the predicted value and the actual value, using the following formula:

[0333]

[0334] Where Q is the sample size; y i The true value of the optimal SOC gradient for the i-th road segment;

[0335] Step S402.6: Repeat steps S402.2 to S402.5 until the error loss is less than the threshold.

[0336] Step S403: Save the trained Bi-LSTM.

[0337] Step S404: Input the macroscopic traffic parameters of the road segments planned in Step S3 into the trained Bi-LSTM model to obtain the optimal SOC gradient y for each road segment along the entire path. i .

[0338] Step S405: Calculate the globally optimal SOC trajectory, using the following formula:

[0339]

[0340] Among them, SOC i+1 SOC represents the SOC value at the end of the i-th road segment. init Represents the initial value of SOC;

[0341] Step S5: During the vehicle's journey, the Adaptive Equivalent Minimum Fuel Consumption Method (AECMS) is used to follow the predicted global optimal SOC trajectory, minimizing the deviation between the actual SOC trajectory and the predicted global optimal SOC trajectory, thereby outputting the optimal control command.

[0342] Step S501: Convert electricity consumption into fuel consumption based on the equivalent factor. The calculation formula is as follows:

[0343] m eq (t)=m f (t)+λ(t)×P BP (t) / F LHV

[0344] Where, m eq (t) represents instantaneous equivalent fuel consumption; m f (t) represents instantaneous fuel consumption; λ(t) is the equivalent factor; P BP (t) represents the instantaneous electric power; F LHV It has a low calorific value.

[0345] Step S502: To ensure that the actual SOC follows the reference SOC in real time, an incremental PID controller is used for adjustment, and the output is Δλ(t), which is the change of the equivalent factor, as shown in the following formula:

[0346] Δλ(t)=k p [e(t)-e(t-1)]+k i e(t)+k d [e(t)+e(t-2)-2e(t-1)]

[0347] e(t) = SOC T (t)-SOC R (t)

[0348] The equivalent factor at the current moment is:

[0349] λ(t) = λ(t-1) + Δλ(t)

[0350] Where, k p k i k d These are the proportional coefficient, integral coefficient, and differential coefficient, respectively. SOC T (t) represents the actual SOC control value, SOC R (t) is the reference SOC value.

[0351] Step S503: Considering the losses caused by gear shifting and engine start-stop, the objective function of the control strategy is defined as follows:

[0352] J = ∑min{(m eq (t)+a1ΔICE(t)+a2|G trans (t)-G trans (t-1)|)}

[0353] Where ΔICE(t) represents the change in engine start-stop state; G trans (t) represents the gear; a1 and a2 represent the weights of the costs caused by engine start-stop changes and gear shifting, respectively;

[0354] Step S504: Select engine torque T as the control variable e (t), motor torque T m(t), State variable selection SOC

[0355] Step S505: The control strategy must ensure that the variables meet the following constraints:

[0356]

[0357] Where, ω e (t) represents the engine speed, ω m (t) represents the motor torque; the subscripts _max and _min represent the maximum and minimum values ​​of the corresponding terms, respectively.

[0358] This invention enables the prediction of the theoretically optimal energy consumption of new energy vehicles, thereby achieving the most energy-efficient path planning for new energy vehicles across the entire road network and providing a theoretically optimal energy-consuming path. Furthermore, this invention predicts the globally optimal State of Charge (SOC) trajectory of new energy vehicles based on macroscopic traffic parameters, fully considering the impact of the spatiotemporal dynamic changes in road network conditions on the energy conservation of new energy vehicles, overcoming the limitations of current EMS (Energy Management System) predictions based only on short-range operating conditions. In summary, this invention simultaneously achieves the most energy-efficient path planning for new energy vehicles across the entire road network and real-time globally optimal energy management for the entire path, featuring high energy efficiency and good real-time performance.

[0359] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0360] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for intelligent energy management of new energy vehicles based on spatiotemporal changes in road network conditions, characterized in that, include: Step S1: Search for pairs of points whose distances from the given start and end points satisfy a preset condition. Path; Step S2: Construct a hierarchical generalized regression neural network and use the constructed hierarchical generalized regression neural network to predict... The theoretical optimal energy consumption for each path in the path; Step S3: Based on The path with the lowest theoretical optimal energy consumption among the paths is selected and used as the driving path for new energy vehicles. Step S4: Construct a bidirectional long short-term memory neural network and use the constructed bidirectional long short-term memory neural network to predict the globally optimal SOC trajectory for the current driving path of the new energy vehicle. Step S5: During the vehicle's journey, the adaptive equivalent fuel consumption minimization method is used to follow the predicted global optimal SOC trajectory, minimizing the deviation between the actual SOC trajectory and the predicted global optimal SOC trajectory, thereby outputting the optimal control command. The hierarchical generalized regression neural network predicts the travel time of a road segment through the upper-layer generalized regression neural network, and feeds back the predicted travel time and the macroscopic traffic parameters obtained based on the travel time to the lower-layer generalized regression neural network, thereby achieving the prediction of the theoretical optimal energy consumption for the entire process. The bidirectional long short-term memory neural network performs forward and backward calculations through two layers of long short-term memory neural network, thereby enabling the prediction of the entire SOC trajectory. Step S5 employs the following: Step S5.1: Set the initial equivalence factor ; Step S5.2: Specify the control variables as engine torque, motor torque, engine speed, motor speed, and battery power, and the state variable as SOC; the control variables and state variables must meet the following constraints: in, For engine torque, This is the motor torque. Engine speed, This represents the battery power; the subscripts _max and _min represent the maximum and minimum values ​​of the corresponding items, respectively. Step S5.3: Discretize the limit range of the control quantity under each second operating condition to obtain the discrete points of all control variables; Step S5.4: For all feasible outcomes consisting of discrete points of all control variables, calculate the equivalent cost using the following formula: in, in, Indicates instantaneous equivalent fuel consumption; Indicates instantaneous fuel consumption; Indicates the equivalent factor; Indicates instantaneous electrical power; This indicates the low calorific value of the fuel. Indicates changes in engine start / stop status; Indicates gear position; and The weights representing the costs caused by engine start-stop changes and gear shifting are respectively calibrated by the test bench. Step S5.5: Select the feasible result with the smallest J; Step S5.6: Calculate the actual SOC corresponding to the feasible result described in step S5.

5. Based on the error between the actual SOC and the reference SOC, adjust the equivalent factor using incremental PID control, as shown in the following formula: The equivalent factor at the current moment is: in, , , These are the proportional coefficient, integral coefficient, and differential coefficient, respectively. This is the actual SOC control value. For reference SOC value; Repeat steps S5.4 to S5.6 until the difference between the actual SOC and the reference SOC is less than the threshold.

2. The intelligent energy management method for new energy vehicles based on spatiotemporal changes in road network conditions according to claim 1, characterized in that, Step S2 employs the following: Step S2.1: Construct the upper-layer generalized regression neural network to predict sequentially. The travel time of each segment on the route is calculated, and the macroscopic traffic parameters of each segment on the route are obtained sequentially based on the travel time. Step S2.2: Calculate the macro-traffic parameters of the route based on the macro-traffic parameters of the road segment; Step S2.3: Construct a lower-level generalized regression neural network for path-based macroscopic traffic parameter prediction. The theoretical optimal energy consumption for this path; The macroscopic traffic parameters for the aforementioned road segment include: remaining red light time. Remaining time for the green light Distance between road sections Traffic density of road sections and average speed of the road section ;in, Indicates the first position of the current path Each section of road, represent time; The macroscopic traffic parameters of the route include: average route speed. , This means that the path has Each road segment; extremely poor path speed. Path speed standard deviation Average traffic density and path length .

3. The intelligent energy management method for new energy vehicles based on spatiotemporal changes in road network conditions according to claim 2, characterized in that, Step S2.1 adopts the following: Step S2.1.1: Record arrival at the current road segment i The moment ; Step S2.1.2: Based on the arrival time of the current road segment Obtain the macroscopic traffic parameters of the current road segment; Step S2.1.3: Based on the macroscopic traffic parameters of the current road segment, predict the travel time of the current road segment using the upper-level generalized regression neural network. ; Step S2.1.4: Based on the arrival time of the current road segment The time to reach the next road segment is obtained by combining the predicted travel time of the current road segment with the predicted travel time of the current road segment. ; Repeat steps S2.1.1 to S2.1.4 until the travel time and macroscopic traffic parameters for each segment of the entire path are obtained.

4. The intelligent energy management method for new energy vehicles based on spatiotemporal changes in road network conditions according to claim 3, characterized in that, Step 2.1.3 includes: Step 2.1.3.1: Road Section exist Macroscopic traffic parameter vector at time The total number of input samples is Indicates subscript t It is to predict the target time; Each sample is processed into a matrix. ,matrix Each column represents a sample; Step 2.1.3.2: Construct the upper-layer generalized regression neural network, including the input layer, pattern layer, summation layer, and output layer; Step 2.1.3.3: Data is passed from the input layer to the pattern layer. In the pattern layer, each neuron is calculated using the following formula to obtain the pattern layer output: in, It is the first of the pattern layers. The output of each neuron; superscript It is the prediction target of the upper-level generalized regression neural network: time is an abbreviation; It is the Euclidean distance function; Representative matrix The List; It is the smoothing parameter of the network; Step 2.1.3.4: The data processed by the schema layer is passed to the summation layer for calculation; the summation layer can be divided into two parts, and the calculation formula for the first part is as follows: in, This is the output of the first part of the summation layer; The formula for the second part is as follows: in, Representing the The output sample of the first... One feature; Step 2.1.3.5: The calculation results from both parts are passed to the output layer, and the prediction result is obtained through the following calculation formula: in, This represents the predicted travel time of the i-th road segment at time t.

5. The intelligent energy management method for new energy vehicles based on spatiotemporal changes in road network conditions according to claim 2, characterized in that, Step S2.2 adopts the following: The macroscopic traffic parameters of the selected route include the average speed of the route. Path speed standard deviation Extremely poor path speed Path length Path traffic density in, Indicates that the current path has Each road segment; average path speed Path speed standard deviation Extremely poor path speed Path length Path traffic density The calculation formulas are as follows: in, Indicates the average speed of vehicles on the road segment; Indicates the distance of a road segment; Indicates the traffic density of a road segment.

6. The intelligent energy management method for new energy vehicles based on spatiotemporal changes in road network conditions according to claim 2, characterized in that, Step S2.3 employs the following method: based on a lower-level generalized regression neural network, the dynamic programming algorithm (DP) is used to calculate the true value of the theoretically optimal energy consumption for the entire path.

7. The intelligent energy management method for new energy vehicles based on spatiotemporal changes in road network conditions according to claim 1, characterized in that, Step S4 employs the following: Step S4.1: Construct a bidirectional long short-term memory neural network. Based on the macroscopic traffic parameters of the road segments along the driving path of new energy vehicles, use the constructed bidirectional long short-term memory neural network to predict the optimal SOC gradient of each road segment on the path. Step S4.2: Calculate the globally optimal SOC trajectory by combining the initial SOC and the segment length; in, Indicates the first SOC value at the end of each road segment; Represents the initial value of SOC; It is the predicted value of the SOC gradient of the optimal road segment.

8. The intelligent energy management method for new energy vehicles based on spatiotemporal changes in road network conditions according to claim 7, characterized in that, Step S4.1 adopts the following: Step S4.1.1: Process the macroscopic traffic parameters of the road segment into vectors. ; Step S4.1.2: Construct a bidirectional long short-term memory neural network model, including: an input layer, a fully connected layer, an LSTM layer, a fully connected layer, and an output layer; wherein the LSTM layer includes a forward LSTM layer and a backward LSTM layer; Step S4.1.3: Vector The input layer is fed into the fully connected layer, and the fully connected layer is used to extract the dynamic traffic features of the road segment; in, These are the weights of the fully connected layer. It is the bias of the fully connected layer; Step S4.1.4: Input the dynamic traffic features into the LSTM layer, and use the forward LSTM layer to learn the influence of the dynamic traffic parameters of the upstream road segment; the data transmission direction of the forward LSTM layer is consistent with the vehicle's driving direction; the working principle of each LSTM unit in the forward LSTM layer is shown in the following formula: in, For the i-th LSTM unit, This represents the hidden state output of the feedforward LSTM layer. Indicates the state of the forward LSTM layer; Represents the sigmoid function; , , , It's weight. , , , It is a bias; It will be passed to the next neuron in the feedforward LSTM layer. It will be passed to the next neuron in the feedforward LSTM layer and the fully connected layer; Step S4.1.5: The input data also enters the backward LSTM layer, which processes the data using the following formula: in, This represents the hidden state output of the backward LSTM layer. Indicates the state of the backward LSTM layer; Represents the sigmoid function; , , , It's weight. , , , It is a bias; It will be passed to the next neuron in the feedforward LSTM layer. It will be passed to the next neuron in the feedforward LSTM layer and the fully connected layer; Step S4.1.5: Output data of the forward LSTM layer and the backward LSTM layer , The output is passed to the fully connected layer for extraction. in, Indicates the weights of the fully connected layer. Indicates the bias of the fully connected layer; This represents the predicted value of the SOC gradient for the optimal road segment.

9. A new energy vehicle energy intelligent management system based on spatiotemporal changes in road network conditions, characterized in that, include: Module M1: Searches for locations where the distance between a given start and end point satisfies a preset condition. Path; Module M2: Constructs a hierarchical generalized regression neural network and uses the constructed hierarchical generalized regression neural network to predict... The theoretical optimal energy consumption for each path in the path; Module M3: Based on The path with the lowest theoretical optimal energy consumption among the paths is selected and used as the driving path for new energy vehicles. Module M4: Constructs a bidirectional long short-term memory neural network and uses the constructed bidirectional long short-term memory neural network to predict the globally optimal SOC trajectory for the current driving path of the new energy vehicle; Module M5: During the vehicle's journey, it uses the adaptive equivalent fuel consumption minimization method to follow the predicted global optimal SOC trajectory, minimizing the deviation between the actual SOC trajectory and the predicted global optimal SOC trajectory, thereby outputting the optimal control command. The hierarchical generalized regression neural network predicts the travel time of a road segment through the upper-layer generalized regression neural network, and feeds back the predicted travel time and the macroscopic traffic parameters obtained based on the travel time to the lower-layer generalized regression neural network, thereby achieving the prediction of the theoretical optimal energy consumption for the entire process. The bidirectional long short-term memory neural network performs forward and backward calculations through two layers of long short-term memory neural network, thereby enabling the prediction of the entire SOC trajectory. The module M5 adopts: Module M5.1: Set the initial equivalence factor ; Module M5.2: The specified control variables are engine torque, motor torque, engine speed, motor speed, and battery power; the state variable is SOC. The control variables and state variables must meet the following constraints: in, For engine torque, This is the motor torque. Engine speed, This represents the battery power; the subscripts _max and _min represent the maximum and minimum values ​​of the corresponding items, respectively. Module M5.3: Discretizes the limit range of the control quantity under each second operating condition to obtain the discrete points of all control variables; Module M5.4: For all feasible outcomes consisting of discrete points of all control variables, calculate the equivalent cost using the following formula: in, in, Indicates instantaneous equivalent fuel consumption; Indicates instantaneous fuel consumption; Indicates the equivalent factor; Indicates instantaneous electrical power; This indicates the low calorific value of the fuel. Indicates changes in engine start / stop status; Indicates gear position; and The weights representing the costs caused by engine start-stop changes and gear shifting are respectively calibrated by the test bench. Module M5.5: Select the option with the smallest J among all feasible results; Module M5.6: Calculates the actual SOC corresponding to the feasible result described in module M5.

5. Based on the error between the actual SOC and the reference SOC, it uses incremental PID adjustment of the equivalent factor, as shown in the following formula: The equivalent factor at the current moment is: in, , , These are the proportional coefficient, integral coefficient, and differential coefficient, respectively. This is the actual SOC control value. For reference SOC value; Repeat modules M5.4 through M5.6 until the difference between the actual SOC and the reference SOC is less than the threshold.

Images