Self-adaptive hybrid power energy management method based on demand power prediction

Abstract

The invention discloses a self-adaptive hybrid power energy management method based on demand power prediction, and belongs to the field of hybrid electric vehicle energy management, and the method comprises the steps: 1, based on the historical information of a fixed route and the positioning information of a navigation system, utilizing an improved LSTM (Long-Short-Term Memory) to carry out power demand prediction on the basis of the historical information of the fixed route and the positioning information of the navigation system; carrying out certain time domain prediction on the vehicle speed and the required power of the fixed commuting vehicle through an LSTM (Long Short Term Memory) algorithm; 2, establishing an energy consumption minimum solving problem in a short time domain by estimating a required power range of a certain time domain, and solving an optimal basic equivalent factor sequence by using a particle swarm algorithm of adaptive weight; 3, equivalent factors under different required powers are corrected by considering the SOC and the temperature of the battery, an equivalent factor correction MAP is constructed through offline simulation and real vehicle calibration, and the equivalent factors are adjusted online in real time; 4, after the determined equivalent factors are obtained, an ECMS strategy is executed to determine the optimal output torque of the engine and the optimal output torque of the motor; according to the invention, the energy management control is close to global optimum, and the economic performance and working condition adaptability of the vehicle can be improved.

Description

Technical Field

[0001] This application relates to the energy management of plug-in hybrid electric vehicles, and in particular to an adaptive plug-in hybrid electric vehicle energy management method based on demand power prediction. Background Art

[0002] With the growing global energy crisis, energy conservation and emission reduction technologies are a critical need for the automotive industry. PHEVs not only reduce fuel consumption but also alleviate drivers' range anxiety. PHEV systems are complex, nonlinear, multivariable, and time-varying systems. Developing energy management control strategies is a key technology for plug-in hybrid systems.

[0003] The energy management strategy of plug-in hybrid electric vehicles (PHEVs) significantly impacts vehicle performance. Hybrid electric vehicles (HEVs) utilize multiple power sources. Developing a rational control strategy to determine the output power of the engine and motor based on power demand significantly impacts vehicle dynamics and economic performance. Accurate power demand prediction can significantly enhance vehicle energy efficiency. This power demand prediction provides the foundation for adaptive adjustment of the equivalent factor. EMCS strategies with short-term predictions of vehicle speed and power demand offer significant advantages over traditional EMCS and rule-based strategies in terms of real-time performance and adaptability to operating conditions, meeting practical application requirements and delivering superior control results.

[0004] How to accurately predict the power demand of the entire vehicle and reasonably distribute the power during vehicle driving is a problem that needs to be solved in the energy management of hybrid vehicles. Summary of the Invention

[0005] In order to overcome the shortcomings of existing vehicle speed prediction, such as low vehicle speed prediction accuracy, general operating adaptability of hybrid energy management control strategies, and failure to consider the impact of system component temperature on energy management control, the present invention proposes an adaptive hybrid energy management method based on demand power prediction. Since the traditional equivalent fuel consumption minimum energy management strategy has poor operating adaptability and general optimization effect of energy management control, the energy management control and thermal management control of most current hybrid vehicles belong to two relatively independent control systems, and the impact of system component temperature on energy management control, especially the impact of battery temperature on energy management control, is rarely considered.

[0006] The technical solution adopted by the present invention to solve the technical problem is to provide an adaptive hybrid energy management method based on demand power prediction, including the following steps:

[0007] Step 1: By collecting vehicle driving data and status information in highway, suburban, and urban areas, including noise reduction and normalization, the vehicle driving status characteristic parameters are extracted from the historical vehicle data on fixed routes. The vehicle status characteristic parameters mainly include vehicle speed, average speed, maximum speed, minimum speed, acceleration ratio, idle ratio, speed variance, and acceleration variance.

[0008] Step 2: Without prior information about future power requirements, the hybrid system will struggle to ensure optimal vehicle economic performance. This method uses an improved LSTM to combine historical route information with navigation system positioning information to predict the future speed series of a commuting vehicle. At time t, the LSTM inputs are: the current network input value X(t), the previous LSTM output value h(t-1), and the previous cell state C(t-1). The LSTM outputs are: the current LSTM output value h(t) and the current cell state C(t).

[0009] Step 3: Based on the predicted vehicle speed sequence and the slope angle calculated by the VCU module, the slope angle generally does not have a large sudden change in a short time domain. The vehicle longitudinal dynamics equation is used to solve the vehicle's required power, and the predicted required power has an estimated range.

[0010] Step 4: During the driving process, the vehicle obtains the required power sequence and estimated range of the next period according to step 3 as the prior information of the ECMS control module based on the historical data of vehicle driving and the current real-time data. The problem of minimum equivalent energy consumption in the prediction time domain is established. With reference to the predicted required power and vehicle speed, the particle swarm algorithm with adaptive weights is used to solve the basic optimal equivalent factor in the prediction time domain. The fitness function constructed is as follows:

[0011]

[0012] It can be transformed into the corresponding fitness function as follows

[0013]

[0014] In the formula is the equivalent fuel consumption, Q lhv is the lower calorific value of fuel, η dis is the discharge power of the battery, η char is the charging power of the battery; when k=1, it indicates that the battery is in the discharging state, and k=0, it indicates that the battery is in the charging state; P req is the predicted power demand, P e The power allocated to the engine, η engis the working efficiency of the engine, x represents the equivalent factor s0(t), and y represents the power allocated to the engine. The range of the basic equivalent factor s0(t) is set to [1, 5].

[0015] The calculation method of the inertia weight of the adaptive particle swarm algorithm is as follows:

[0016]

[0017] Where ω min With ω max is the preset minimum inertia weight coefficient and maximum inertia weight coefficient, ω max Set to 0.8, ω min Set to 0.4, f(x i d ) is the fitness of the ith particle in the dth iteration, is the average fitness of all particles at the dth iteration, i = 1, 2, 3..n, n is the size of the particle swarm, is the minimum fitness of all particles at the dth iteration. The particle swarm algorithm with adaptive weights is used to obtain the optimal basic equivalent factor s0(t) so that the fitness function is minimized.

[0018] Step 5: A penalty factor is introduced based on the battery SOC and temperature to modify the equivalent factor, update the equivalent fuel consumption calculation formula, and optimize the calibration parameters through offline simulation to construct an optimal equivalent factor correction map. Considering the battery target SOC value, the SOC must be within a reasonable and safe range. The penalty function r(SOC) is introduced to modify the equivalent factor, and the equivalent fuel consumption is updated using the following formula.

[0019]

[0020] In the above formula, r(SOC) is used as a correction function to convert the current SOC(t) and the target SOC tar The numerical difference is an important basis for online evaluation of battery SOC stability. tar The SOC value of the equilibrium point is set. The reference value is a constant value and is set by the user according to the needs. The equilibrium point range is set between 25% and 70%. The penalty function is calculated using the following formula.

[0021]

[0022] Where SOC max is the upper limit of battery SOC, SOC min is the lower limit of SOC, and ε(t) is the correction coefficient based on battery temperature.

[0023] Under the same battery SOC, different r(soc) values ​​result in different penalty levels. tar value, the penalty function value will be less than 1, the battery cost is low, the vehicle state tends to discharge the battery, and when the SOC is higher than the target SOC tar When the value is , the penalty function value is greater than 1, the battery usage cost is high, and the vehicle state tends to be charging.

[0024] Battery temperature fluctuations significantly impact both battery life and performance. Significant changes in battery temperature can cause different battery behaviors, affecting charge and discharge performance and impacting the execution of energy management strategies. In addition to considering the current SOC data, the penalty function value must also be adjusted based on battery temperature. At the same SOC, when the battery temperature exceeds 45°C or falls below 10°C, ε(t) is greater than 1, the equivalence factor correction coefficient increases, and the battery's maximum allowable charge or discharge power decreases compared to normal temperature. In this situation, continued use of electricity is not recommended to prevent a reduction in battery life. When the battery temperature is greater than 10°C and less than 45°C, ε(t) is set to 1, and the equivalence factor correction coefficient is not adjusted. ε(t) is obtained using a linear table lookup based on the battery temperature. ε(t) can be used to adjust the charge and discharge strategy based on battery temperature, significantly improving system efficiency while preventing battery overheating or overcooling. The correction coefficient MAP table is determined through offline simulation optimization and on-vehicle calibration. Table 1 shows the correction coefficients based on battery temperature.

[0025]

[0026] Step 6: Based on the equivalent factors determined in Step 5, the equivalent fuel consumption minimization strategy is used to calculate the motor and engine torques that minimize instantaneous fuel consumption under different operating modes. By traversing all feasible solutions, the coupling matrix combination that minimizes total fuel consumption is determined, ultimately achieving the optimal target torques for the engine and motor, thereby improving the fuel economy of the PHEV.

[0027] Compared with the existing technology, the beneficial effects of the present invention are:

[0028] 1. The present invention collects and monitors vehicle driving status data in real time, uses an improved LSTM algorithm combined with historical information of fixed routes and navigation system positioning information to predict the power demand of fixed commuting vehicles in a certain time domain, and then uses an adaptive weighted particle swarm algorithm to solve the basic equivalent factor of the ECMS strategy in the prediction time domain, and corrects the basic equivalent factor according to the battery SOC size and temperature. Then, offline simulation and actual vehicle optimization calibration are used to construct an equivalent factor correction MAP table.

[0029] 2. Based on the resulting equivalent factors, the engine and generator torques that minimize instantaneous fuel consumption are calculated. An ECMS strategy based on demand power prediction and taking into account battery temperature and SOC can approach the global optimal solution while also improving the energy management strategy's adaptability to operating conditions. Compared to traditional methods, this significantly improves vehicle fuel economy. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 Adaptive hybrid energy management method based on demand power prediction

[0031] Figure 2 A schematic diagram of a hybrid vehicle power demand prediction process in the present invention;

[0032] Figure 3 Flowchart of the adaptive energy management strategy based on demand power prediction in the present invention;

[0033] Figure 4 This is a schematic diagram of the ECMS energy management control module;

[0034] Figure 5 and Figure 6 Comparison chart of predicted speed and actual speed

[0035] Figure 7 Equivalent factor three-dimensional MAP diagram DETAILED DESCRIPTION

[0036] In this embodiment, an adaptive hybrid energy management method based on demand power prediction is proposed, which can be applied to the energy management control system of plug-in hybrid electric vehicles. In the specific implementation, based on the equivalent fuel consumption minimum strategy, the improved LSTM prediction model is used to predict the vehicle speed sequence of a fixed commuting vehicle, and the demand power in the next time domain is obtained according to the longitudinal dynamic equation. Based on the predicted demand power range, the particle swarm algorithm with adaptive weights is used to solve the basic optimal equivalent factor. The equivalent factors of different demand powers are corrected and optimized by offline simulation based on the SOC value and temperature of the battery. The ECMS energy management algorithm is executed using the equivalent factors after real-time correction to obtain the optimal torque distribution between the engine and the motor. Figure 1 Diagram of the adaptive hybrid energy management method based on demand power prediction.

[0037] like Figure 2 Vehicle demand power prediction model for plug-in hybrid electric vehicles.

[0038] Step 1: Collect vehicle driving status data and perform filtering and noise reduction on the collected data.

[0039] Step 2: Process the short-stroke motion segments to obtain the vehicle's speed, average speed, maximum speed, minimum speed, acceleration ratio, idle ratio, speed variance, and acceleration variance characteristic parameters, and construct a characteristic parameter matrix. Standardize the characteristic parameters.

[0040] Step 3: The speed prediction model uses the LSTM algorithm, combining historical information from fixed routes with navigation system positioning information. The LSTM hyperparameters are optimized using an improved sparrow search algorithm. The vehicle's historical speed sequence is used as input for LSTM speed prediction model training, and the output is a speed sequence for a period of time in the future, typically 3 to 5 seconds.

[0041] Step 4. The LSTM prediction model contains three gates: forget gate, input gate, output gate and a memory unit.

[0042] Forget gate neuron f(t)

[0043] f(t)=σ(W f ·[h(t-1),x(t)]+b f )

[0044] Input gate neuron i(t)

[0045] i(t)=σ(W i ·[h(t-1),x(t)]+b i )

[0046] memory gate neurons

[0047]

[0048] Long-term memory C(t) at time t

[0049]

[0050] Output gate neuron o(t)

[0051] o(t)=σ(W o [h(t-1),x(t)]+b o )

[0052] Short-term memory h(t) at time t

[0053] h(t)=o(t)*tanh(C(t))

[0054] Where σ refers to the sigmoid function, t is the time, C(t) is the long-term memory cell, h(t) is the memory state, and x(t) is the input. is the internal memory neuron, h(t) is the short-term memory state, f(t) is the forget gate, i(t) is the update gate, and o(t) is the output gate. f ,b f ,W i ,b i ,W o ,b o ,W c ,b c The speed prediction model is trained using the collected historical speed sequence of the vehicle to obtain a trained speed prediction model, and the output of the model is the future speed sequence.

[0055] Step 5: Use the improved sparrow search algorithm to optimize the number of training cycles and learning rate for each LSTM hidden layer node. The vehicle's historical speed sequence is used as input, and the output is the predicted future speed sequence. Each input and output pattern consists of a fixed-length moving window, expressed as follows.

[0056] [V k+1 ,V k+2 ,…V k+Hn ]=f(V k-Hm+1 ,V k-Hm+2 ,…V k )

[0057] In the above formula, H m is the dimension of the input velocity sequence; where f represents the nonlinear mapping function of the LSTM-based prediction module; H n Indicates the length of the output future velocity sequence, V k is the vehicle speed at time k. The mean square error and root mean square error are selected as evaluation criteria. The mean square error (MSE) measures the quality of a prediction model by calculating the square of the distance between the predicted value and the true value. The closer the predicted data is to the true value, the smaller the MSE value. MSE is calculated using the following formula.

[0058]

[0059] Where n is the number of training data, V i is the actual value of the output vehicle speed, Predicted vehicle speed values. Vehicle characteristic parameters are obtained in real time during vehicle driving and used as input to the speed prediction model to obtain the future vehicle speed sequence online.

[0060] Step 6: Based on the predicted future vehicle speed, the required power sequence in the future time domain can be obtained using the principle of vehicle longitudinal dynamics power balance in the following formula.

[0061]

[0062] Among them, v is the predicted speed; m is the vehicle mass, C D is the drag coefficient, A is the frontal area of ​​the vehicle, δ is the vehicle rotation mass coefficient, a is the slope angle, η t For transmission efficiency.

[0063] The slope angle a generally does not change suddenly. The slope angle value is calculated by the VCU module. The future power demand can be obtained according to step 6. The prediction of the power demand takes into account the influence of errors. The acceptable range of the estimated range deviation is generally around 3 kW.

[0064] Step 7: The value of the bank angle ɑ significantly affects the accuracy of the required power calculation. During normal vehicle operation, ignoring slip, the vehicle speed is approximately equal to the wheel speed. The wheel speed calculation accounts for sensor and road noise, significantly increasing wheel speed errors. When the vehicle accelerates or decelerates, especially when considering the influence of anti-lock braking system control, wheel speed determination can result in larger errors. To reduce the impact of noise, a low-pass filter can be applied to calculate the filtered wheel speed value.

[0065] V fit (k)=ωV(k)+(1-ω)V fit (k-1)

[0066] In the above formula, V fit (k) represents the filtered output value of the wheel speed at time k; V fit (k-1) represents the filtered value of the wheel speed over the k-th time step; V(k) represents the sampled wheel speed at time k; and w is the filter coefficient. Because the braking force and slip ratio are relatively small, the vehicle speed can be directly derived from the wheel speed.

[0067] Step 8: Calculate the acceleration Ac in the X-axis (longitudinal) direction of the vehicle by differentiating the vehicle speed.

[0068] Calculate the X-axis acceleration Ac using the following formula.

[0069]

[0070] Step 9. After the above calculation, the X-axis acceleration needs to be filtered again (taking the 10-cycle average value) with a Δt of 0.1s. Use the accelerometer to obtain the X-axis acceleration measurement value: X-axis acceleration measurement value = X-axis acceleration - X-axis acceleration offset. After filtering, take the 10-cycle average value and use the following formula to calculate the slope angle and slope (%).

[0071]

[0072] In the above formula, i is the slope (%), Ax is the acceleration measurement value in the X direction, (m / s2), A c is the calculated value of acceleration in the X direction; g is the acceleration due to gravity, which is 9.8m / s2. The slope is positive or negative with A x -A c The positive and negative are consistent. Taking the front of the vehicle as a reference, when the vehicle is going uphill, the slope is positive, and when the vehicle is going downhill, the slope is negative.

[0073] Step 10: Calculate the vehicle's driving force, rolling resistance, acceleration resistance, and air resistance using the vehicle's basic parameters and speed, based on the vehicle longitudinal dynamics formula. The identification method strategy call cycle is 10ms. During this cycle, the slope angle signal value typically does not experience sudden changes, and the probability of data mutation is low.

[0074] like Figure 3 Shown is a roadmap for adaptive energy management based on demand power prediction.

[0075] Step 1: The vehicle data acquisition and processing module collects vehicle information data in real time, filters and reduces noise on the data, and extracts vehicle driving status characteristic parameters through a large amount of real historical data.

[0076] Step 2: According to step 1, the vehicle driving state characteristic parameters are obtained as input to the vehicle driving speed and required power prediction module.

[0077] Step 3: Based on the vehicle speed and required power in the time domain predicted in step 2, Figure 5 and Figure 6 The figure below is a comparison of the predicted speed and the actual speed. The adaptive weighted particle swarm algorithm is used to solve the equivalent factor of the initial basis optimization, and the fitness function is constructed as follows:

[0078]

[0079] The formula can be transformed into the corresponding fitness function as follows

[0080]

[0081] Where Q lhv is the lower calorific value of fuel, η dis is the discharge power of the battery, η char is the charging power of the battery; when k=1, it indicates that the battery is in the discharging state, and k=0, it indicates that the battery is in the charging state; P req is the predicted required power, Pe is the power allocated to the engine, η eng is the engine's efficiency, x represents the equivalent factor s0(t), and y represents the power allocated to the engine. The inertia weight calculation method of the adaptive particle swarm algorithm is as follows.

[0082]

[0083] In the above formula, ω min With ω max is the preset minimum inertia weight coefficient and maximum inertia weight coefficient, ω max Set to 0.8, ω min Set to 0.4, is the average fitness of all particles at the dth iteration, i = 1, 2, 3..n, n is the size of the particle swarm, is the minimum fitness of all particles at the dth iteration. The adaptively weighted particle swarm optimization algorithm is used to obtain the optimal initial equivalent factor s*0(t) to minimize the fitness function. The adaptively weighted particle swarm optimization algorithm (PSO) is used to solve the optimal equivalent factor sequence with the minimum equivalent fuel consumption within the prediction time domain. The first element of the prediction sequence is used as the currently determined basic equivalent factor, and rolling prediction and optimization are repeated.

[0084] Step 4: The SOC value is a key parameter affecting the equivalent factor. Therefore, the battery SOC limit must be considered to ensure that the SOC is within a reasonable and safe range. Therefore, a penalty function r(SOC) is introduced to appropriately modify the equivalent factor and update the equivalent fuel consumption formula.

[0085]

[0086] In the above formula, r(SOC) is used as a correction function to convert the current SOC(t) and the target SOC tar The gap between the two is an important basis for online estimation of battery SOC reliability. tar The reference value is the set balance point SOC value. The reference value can be considered as a constant value. The user can set it according to the needs. The general balance point range is 25% to 70%.

[0087]

[0088] In the above formula, SOC max is the upper limit of battery SOC, SOC min is the lower limit of SOC, ε(t) is the correction coefficient based on battery temperature, and the α value is 3.

[0089] Battery temperature changes have a significant impact on battery life and battery performance. Significant changes in battery temperature will affect the performance of the PHEV energy management strategy. In addition to considering the current SOC size, the penalty function value also needs to be corrected with reference to the battery temperature. Under the same SOC, when the battery temperature exceeds 45°C or is below 10°C, ε(t) is greater than 1, the equivalent factor correction coefficient will increase, and the maximum allowable charging or discharging power of the battery will become smaller than that under normal temperature conditions. In this case, it is not recommended to continue using electricity. When the battery temperature is greater than 10°C and less than 45°C, ε(t) is equal to 1, and the equivalent factor correction coefficient is not corrected. ε(t) is obtained by linear table lookup based on the battery temperature. ε(t) can adjust the charge and discharge strategy according to the battery temperature, which can greatly improve system efficiency, avoid battery overheating or overcooling, and extend the battery life. The correction coefficient is calibrated through offline simulation optimization and the correction coefficient MAP table is determined. For the equivalent factor correction coefficient ε(t) based on battery temperature, the battery temperature correction coefficient MAP table is as follows.

[0090]

[0091] Under the same battery SOC, different r(soc) values ​​have different corresponding penalty levels; when the SOC is lower than the target SOC tar value, the penalty function value will be less than 1, the battery cost is low, the vehicle state tends to discharge the battery, and when the SOC is higher than the target SOC tar When the value is , the value of the penalty function is greater than 1; the cost of using the battery is high, and the vehicle state tends to be charging.

[0092] Step 5: Construct a modified MAP table of equivalent factors according to step 4 to obtain the optimal equivalent factor s1(t) that matches the current driving state.

[0093] Step 6: Execute the ECMS control strategy to reasonably distribute the vehicle power in real time, determine the optimal torque output of the engine and drive motor, minimize the equivalent fuel consumption data, and achieve optimal vehicle economy.

[0094] like Figure 4 The figure shows the ECMS energy management control process of PHEV.

[0095] Step 1: According to the driver's required power and the current SOC state of the power battery.

[0096] Step 2: Determine the control range of battery power as [P batt,min (t),P batt,max (t)].

[0097] Step 3: Discretize the battery power requirement range as the candidate value of the battery power requirement, [P batt,min (t),…P batt,mid (t),…P batt,max (t)].

[0098] Step 4: At the current time t, the required power of the engine is

[0099] P e (t) = P req (t)-P batt (t)

[0100] Step 5: According to the battery SOC and temperature, the corrected equivalent fuel factor s1(t) is determined by looking up the table. The instantaneous equivalent fuel consumption calculation formula is as follows:

[0101]

[0102] In the formula is the engine fuel consumption (g / s), is the equivalent fuel consumption of the battery (g / s), is the total instantaneous fuel consumption corresponding to time t (g / s).

[0103] Step 6: Calculate the equivalent fuel consumption of the battery using the following formula.

[0104]

[0105] In the above formula, Q lhv is the lower calorific value of fuel, η dis is the discharge power of the battery; η char is the charging power of the battery; when k=1, it indicates that the battery is in the discharging state, and k=0, it indicates that the battery is in the charging state.

[0106] Step 7: Based on steps 5 and 6, convert the instantaneous equivalent fuel consumption calculation formula into the following formula.

[0107]

[0108] u is the ratio of engine power to required power, η eng (t) is the efficiency of the engine at time t.

[0109] Step 8. Optimal control variable u * The mathematical expression of (t) can be written as

[0110]

[0111] Considering the physical property limitations of each component, the main constraints are shown in the following equations.

[0112]

[0113] Where Q SOCmax and Q SOCmin are the upper and lower limits of battery SOC respectively; I max and I min They are the upper and lower limits of current respectively; T mmax and T mmin They are the upper and lower limits of the driving motor torque; T emax and T emin is the maximum and minimum torque of the engine; N emax and N emin N is the upper and lower limits of the engine speed; mmax and N mmin They are the upper and lower limits of the driving motor speed respectively. gmax and T gmin are the upper and lower limits of the generator torque, N gmax and N gmin and are the upper and lower limits of the generator speed respectively.

[0114] By referring to formula (22), the current minimum instantaneous fuel consumption can be calculated, and the corresponding engine power requirement Pe(t) and the current battery power requirement P can be obtained. batt (t).

[0115] Step 9: Based on the determined equivalent factor s1(t), the engine required power Pe(t) and the current battery required power P batt (t), and then obtain the current optimal operating point of the engine and the optimal target operating point of the motor.

Claims

1. An adaptive hybrid energy management method based on demand power prediction, which can be applied to the control of energy management of plug-in hybrid electric vehicles, is characterized by: The following steps are involved: Step 1: By collecting the driving data and status information of the vehicle in the highway, suburbs, urban areas and other working conditions, including noise reduction and normalization of the data information, the vehicle driving status characteristic parameters are extracted from the historical data of the vehicle on the fixed route; The state characteristic parameters are vehicle speed, average speed, maximum speed, minimum speed, acceleration ratio, idle ratio, speed variance and acceleration variance; Step 2: Use the improved LSTM to combine historical information of fixed routes with navigation system positioning information to predict the future speed series of fixed commuting vehicles. At time t, the LSTM inputs are three: the current network input value X(t), the LSTM output value h(t-1) at the previous time, and the cell state C(t-1) at the previous time; the LSTM outputs are two: the current LSTM output value h(t) and the current cell state C(t). Step 3: Based on the vehicle speed sequence predicted in step 2 and the slope angle calculated by the VCU module, the vehicle longitudinal dynamics equation is used to solve the vehicle's required power and predict the required power estimation range; Step 4: Based on step 3, the power demand sequence and estimated range for the next period are obtained as the prior information of the ECMS control module. A model for solving the problem of minimum equivalent energy consumption in the prediction time domain is established. Referring to the predicted power demand and vehicle speed, the particle swarm algorithm with adaptive weights is used to solve the basic optimal equivalent factor in the prediction time domain. The fitness function constructed is as follows: Transformed into the corresponding fitness function as formula (2): in, is the equivalent fuel consumption at time t, Q lhv is the lower calorific value of fuel, η dis is the discharge power of the battery, η char is the charging power of the battery; when k=1, it indicates that the battery is in the discharging state, and k=0, it indicates that the battery is in the charging state; P req is the predicted power demand, P e The power allocated to the engine, η eng is the working efficiency of the engine, x represents the equivalent factor s0(t) at time t, y represents the power allocated to the engine, and the range of s0(t) is set to [1, 5]; The calculation method of the inertia weight of the adaptive particle swarm algorithm is as follows: Among them, ω min With ω max is the preset minimum inertia weight coefficient and maximum inertia weight coefficient, f(x i d ) is the fitness of the ith particle in the dth iteration, is the average fitness of all particles at the dth iteration, i = 1, 2, 3..n, n is the size of the particle swarm, is the minimum fitness of all particles at the dth iteration, and the particle swarm algorithm with adaptive weights is used to obtain the optimal basic equivalent factor s*0(t) to minimize the fitness function; Step 5: According to the battery SOC and battery temperature, a penalty coefficient is introduced to correct the equivalent factor, and the equivalent fuel consumption calculation formula is updated. The calibration parameters are optimized through offline simulation to construct an optimal equivalent factor correction MAP table; the penalty function r(SOC) is introduced to correct the equivalent factor, and the equivalent fuel consumption is updated using formula (4): Among them, r(SOC) is used as a correction function to convert the current SOC(t) and the target SOC tar The numerical gap, SOC tar The SOC value of the equilibrium point is set. The equilibrium point range is set between 25% and 70%. The user sets it according to the needs. The penalty function is calculated using formula (5); Among them, SOC max is the upper limit of battery SOC, SOC min is the lower limit of SOC, ε(t) is the correction coefficient based on battery temperature at time t, and α is the correction index. Under the same battery SOC, different r(soc) values ​​have different penalty levels. When SOC is lower than the target SOC tar The penalty function value is less than 1, the battery cost is low, the vehicle state tends to discharge the battery, and when the SOC is higher than the target SOC tar When the value is greater than 1, the value of the penalty function is greater than 1, the battery usage cost is high, and the vehicle state tends to be charging; At the same SOC, when the battery temperature exceeds 45°C or is below 10°C, ε(t) is set to be greater than 1. When the battery temperature is greater than 10°C and less than 45°C, ε(t) is set to be equal to 1, and the equivalent factor correction coefficient is not corrected. ε(t) is obtained by linear table lookup based on the battery temperature. The correction coefficient MAP table is determined through offline simulation optimization and actual vehicle calibration. The battery temperature correction coefficient table is shown below. Step 6: Based on the equivalent factors determined in step 5, the equivalent fuel consumption minimization strategy is used to calculate the torque of the motor and engine that minimizes instantaneous fuel consumption under different operating modes; all feasible solutions are traversed to obtain the coupling matrix combination that minimizes the total fuel consumption cost, and finally the optimal target torque of the engine and motor is obtained.

2. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to claim 1 are implemented.

3. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to claim 1 are implemented.

Images