Neural network energy management method for plug-in hybrid electric harvester based on quasi-periodic process

Abstract

The present invention discloses a neural network energy management method for a plug-in hybrid electric harvester based on a quasi-periodic process. A quasi-periodic database is established based on the harvester's operating process and different feed rates. The overall machine power requirements for different quasi-periodic process samples in the quasi-periodic database are calculated and added to the quasi-periodic database, and an optimal engine power ratio table for each sample is obtained. A BP neural network is established, and the neural network is trained using the feed rate, quasi-periodic operating stage, battery state of charge, and grain quality in the granary as inputs and the optimal engine power ratio as output. The trained BP neural network serves as an optimal energy management strategy model. The feed rate, quasi-periodic operating stage, battery state of charge, and grain quality in the granary during the harvester's operation are obtained in real time and input into the optimal energy management strategy model to obtain the optimal engine power ratio, thereby calculating the optimal engine torque. The method of the present invention has the characteristics of high energy utilization and high real-time performance.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent control of agricultural machinery, and in particular relates to a neural network energy management method for a plug-in hybrid power harvester based on a quasi-periodic process. Background Art

[0002] In recent years, with increasing attention to energy efficiency and environmental protection, the concept of green agriculture has rapidly developed. Hybrid power technology, recognized internationally as an effective means of energy conservation and emission reduction, has been successfully applied in the automotive industry. Hybrid power harvesters have become a research hotspot in the rice combine harvester field. The key to achieving energy conservation and emission reduction in hybrid power harvesters lies in their energy management strategies. However, a global optimization method for energy management in plug-in hybrid power harvesters, which takes real-time performance into account across various operating scenarios, currently hinders its practical application. Summary of the Invention

[0003] In view of the shortcomings in the prior art, the present invention provides a neural network energy management method for a plug-in hybrid harvester based on a quasi-periodic process.

[0004] The present invention achieves the above technical objectives through the following technical means.

[0005] A neural network energy management method for a plug-in hybrid electric harvester based on a quasi-periodic process is characterized by:

[0006] According to the harvester operation process and different feeding amounts, a quasi-cycle database is established, and different battery initial charge states are set for the quasi-cycle database;

[0007] Based on the quasi-periodic power demand model and the conversion equivalent efficiency model, the whole-machine power demand of different quasi-periodic process samples in the quasi-periodic database is calculated, and the whole-machine power demand of all quasi-periodic process samples is added to the quasi-periodic database to form a quasi-periodic whole-machine power demand database;

[0008] The dynamic programming algorithm is used to solve the optimal engine power ratio table for different samples in the quasi-periodic engine power demand database;

[0009] A BP neural network was established, with the feed amount, quasi-periodic operation stage, battery state of charge, and grain quality in the granary as inputs and the optimal engine power ratio as output. The trained BP neural network was used as the optimal energy management strategy model.

[0010] The feed amount, quasi-periodic operation stage, battery charge status and grain quality in the granary during the harvester operation are obtained in real time, and the optimal energy management strategy model is input to obtain the optimal engine power ratio. The real-time engine power is calculated based on the optimal engine power ratio, and then the optimal engine torque is calculated to achieve engine control.

[0011] Furthermore, the quasi-periodic power demand model is:

[0012]

[0013] Among them, P h is the rated harvesting power requirement of the harvester, P d P is the power requirement of the harvester at standard speed without load. u is the harvester unloading power requirement, q is the harvester feeding amount, k ph is the power demand coefficient of feed quantity, v r is the driving speed, v s Harvester standard speed, m is the unloaded mass of the harvester, m grainmax is the fully loaded mass of the harvester, m grain is the grain quality of the granary during the harvester operation, P request The power required by the harvester components for quasi-periodic operation, s combine It is the harvester operation stage, and its values ​​1, 2, 3, and 4 correspond to harvesting operation, full-load transfer, unloading, and empty-load transfer, respectively.

[0014] Furthermore, the conversion equivalent efficiency model is:

[0015]

[0016]

[0017]

[0018]

[0019] Among them, n e For engine efficiency, is the fuel consumption rate, ω e is the engine speed, T e is the engine torque, H lhv Lower calorific value of fuel, n g is the generator efficiency, is the generator efficiency function, ω g is the generator speed, T g is the generator torque, n b is the battery efficiency, is the battery efficiency function, socb is the battery state of charge, P b is the battery power, n m is the motor efficiency, ω m is the motor speed, T m is the motor torque, is the battery efficiency function.

[0020] Furthermore, the overall power requirements of the different quasi-periodic process samples are:

[0021]

[0022] Among them, P required This is the overall power requirement for the quasi-periodic process sample.

[0023] Furthermore, when using the dynamic programming algorithm to solve the optimal engine power ratio table, SoC is used as the state variable, engine power ratio is used as the control variable, and the sum of fuel consumption and electricity consumption is used to establish the cost function of the dynamic programming algorithm; SoC is discretized with a step size of 0.01 to obtain [SoC min ,SoC final ], the engine torque is discrete into multiple states [Te min ,Te max ], constrain the engine power, engine speed, battery power, battery state of charge, and motor power, calculate different engine power proportions according to the overall power demand, then establish a corresponding initial cost table according to different engine power proportions, calculate the cost function values ​​under different engine power proportions, and fill the initial cost table, use the filled cost table to find the engine power proportion sequence corresponding to the minimum cost function value at the end point of the quasi-periodic process, repeat this process several times until the engine power proportion sequence between different iterative processes remains consistent, and use the engine power proportion sequence at this time as the optimal engine power proportion table; wherein, SoC min is the minimum battery state of charge, SoC final is the final battery state of charge, Te min is the minimum engine torque, Te max is the maximum engine torque.

[0024] Furthermore, the final battery state of charge is obtained by the relationship between the initial value of SoC, the duration of the quasi-periodic process and the endurance time:

[0025]

[0026] Among them, SoC initial is the initial battery state of charge, t endurance is the battery life of the plug-in hybrid harvester, t quasi_periodThe duration of the quasi-cyclic process.

[0027] Furthermore, the evaluation function used by the BP neural network is:

[0028]

[0029] Among them, MSE is the mean square error, N r is the number of output samples, P ereal is the engine power ratio calculated by BP neural network, P etar is the optimal engine power ratio in the sample.

[0030] Furthermore, the real-time acquisition of the feed amount during the harvester operation is estimated based on the harvester dynamics equation:

[0031]

[0032] Among them, q r is the feeding amount, M H is the header torque at the sampling moment, M bf is the reel idling friction torque, M gf is the friction torque of the cutter during idling, M jf is the idling friction torque of the cutting platform auger, i b 、i j are the transmission ratios from the driving shaft to the reel and the auger, J b is the moment of inertia of the reel, J j is the auger's moment of inertia, ω in is the cutting platform speed at the sampling time, k g 、k b 、k j are the feed load factors of the cutter, reel and auger respectively.

[0033] Furthermore, the optimal engine torque is:

[0034]

[0035] P engine_realtime =P realtime *R e

[0036] Among them, T engine_realtime is the optimal torque of the engine, P engine_realtime is the real-time power demand of the engine, ω engine_realtime is the real-time engine speed, P realtime is the real-time power demand of the whole machine, R e is the optimal engine power ratio.

[0037] Furthermore, the sampling range of the feeding amount is 3 kg / s-9 kg / s with an interval of 0.1 kg / s, and the sampling range of the initial state of charge of the battery is 20%-100% with an interval of 1%.

[0038] The beneficial effects of the present invention are:

[0039] (1) The present invention establishes a quasi-periodic database based on the harvester's operating process and different feeding amounts, and sets different battery initial charge states for the quasi-periodic database, thereby simplifying the power demand analysis of the harvester;

[0040] (2) The present invention solves the optimal engine power ratio table of different samples in the quasi-periodic whole machine power demand database based on a dynamic programming algorithm, takes the feed amount, quasi-periodic operation stage, battery charge state, and grain quality in the quasi-periodic process as input, and the optimal engine power ratio as output, and trains the neural network. The trained BP neural network is used as the optimal energy management strategy model. The optimal energy management strategy model is used to obtain the optimal engine power ratio when the harvester is operating in real time, and the real-time power of the engine is calculated by the optimal engine power ratio, and then the optimal engine torque is calculated, thereby controlling the engine; the fuel efficiency of the engine can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of the neural network energy management method for a plug-in hybrid electric harvester based on a quasi-periodic process according to the present invention;

[0042] Figure 2 This is a block diagram of the dynamic programming algorithm described in the present invention. DETAILED DESCRIPTION

[0043] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the protection scope of the present invention is not limited thereto.

[0044] The present invention provides a neural network energy management method for a plug-in hybrid electric harvester based on a quasi-periodic process. The method divides the working process of the harvester into four stages, establishes a quasi-periodic database, calculates the whole-machine power requirements of the quasi-periodic process samples, and adds the whole-machine power requirements of all samples to the quasi-periodic database to form a quasi-periodic whole-machine power requirement database. The dynamic programming algorithm is then used to solve the optimal engine power ratio table for different samples in the quasi-periodic whole-machine power requirement database. The BP neural network is then trained with the battery state of charge, feed amount, grain quality in the granary, and operating stage as inputs and the optimal engine power ratio as output to obtain an optimal energy management strategy model. Finally, the method communicates with the main controller through the CAN network to transmit the real-time battery state of charge, feed amount, grain quality in the granary, and operating stage. The main controller outputs the real-time optimal engine power ratio of the harvester through the optimal energy management strategy model, and calculates the real-time power of the engine through the optimal engine power ratio, and then calculates the optimal torque of the engine, thereby controlling the engine, reducing the energy consumption of the harvester, and improving fuel efficiency. Figure 1 As shown, the specific steps include:

[0045] Step 1: Establish a quasi-periodic database based on the harvester's operating process and different feeding amounts, and set different battery initial charge states for the quasi-periodic database;

[0046] The operation process of the harvester is divided into four stages of quasi-cyclic process, namely harvesting operation, full-load transfer, grain unloading, and empty-load transfer; according to the feed amount range, the sampling range is set to 3kg / s-9kg / s and the interval is 0.1kg / s.

[0047] The established quasi-periodic database contains several quasi-periodic process samples. Each sample consists of four stages. The full-load transfer, grain unloading, and empty-load transfer times of each quasi-periodic process are preset values, and the duration of the harvesting operation is determined by the feed amount.

[0048] According to the working range of the battery, the sampling range of the battery initial state of charge is set to 20%-100% and the interval is 1%, which serves as different battery initial states of charge in the quasi-periodic database.

[0049] Step 2: Based on the quasi-periodic power demand model and the conversion equivalent efficiency model, calculate the overall power demand of different quasi-periodic process samples in the quasi-periodic database;

[0050] According to the four stages of the quasi-periodic process, a quasi-periodic power demand model is established, as follows:

[0051]

[0052] Among them, P his the rated harvesting power requirement of the harvester, P d P is the power requirement of the harvester at standard speed without load. u is the power requirement of the harvester for unloading grain, s combine is the harvester operation stage, and its values ​​1, 2, 3, and 4 correspond to harvesting operation, full-load transfer, unloading, and empty-load transfer, respectively. q is the harvester feed amount, and k ph is the power demand coefficient of feed quantity, v r is the driving speed, v s Harvester standard speed, m is the unloaded mass of the harvester, m grainmax is the fully loaded mass of the harvester, m grain is the grain quality of the granary during the harvester operation, P request The power requirements of harvester components for quasi-cyclic operation.

[0053] A conversion equivalent efficiency model is established, which mainly considers the engine operating efficiency, battery operating efficiency, engine operating efficiency and motor operating efficiency, as follows:

[0054]

[0055]

[0056]

[0057]

[0058] Among them, n e For engine efficiency, is the fuel consumption rate, ω e is the engine speed, T e is the engine torque, Hlhv is the fuel low calorific value, n g is the generator efficiency, is the generator efficiency function, ω g is the generator speed, T g is the generator torque, n b is the battery efficiency, is the battery efficiency function, soc b is the battery state of charge, P b is the battery power, n m is the motor efficiency, ω m is the motor speed, T m is the motor torque, is the battery efficiency function.

[0059] During the harvester operation, the power demand of the entire machine for the quasi-periodic process samples in the quasi-periodic database is calculated based on the quasi-periodic power demand model and the conversion equivalent efficiency model. The details are as follows:

[0060]

[0061] Among them, P required This is the overall power requirement for the quasi-periodic process sample.

[0062] The whole-machine power requirements of all quasi-periodic process samples are added to the quasi-periodic database to form a quasi-periodic whole-machine power requirement database.

[0063] Step 3: Use the dynamic programming algorithm to solve the optimal engine power ratio table for different samples in the quasi-periodic machine power demand database, and record the corresponding feed amount, quasi-periodic operation stage, battery state of charge, and grain quality in the granary as sample variables;

[0064] Taking SoC as the state variable and engine power ratio as the control variable, the specific method of using dynamic programming algorithm to solve the optimal engine power ratio in different quasi-periodic processes is as follows (see Figure 2 ):

[0065] The cost function of the dynamic programming algorithm is established based on the sum of fuel consumption and electricity consumption as follows:

[0066]

[0067] Among them, J is the cost function, C fuel is the fuel consumption, C ele is the power consumption, N1 is the number of stages, and △t is the sampling time interval.

[0068] Considering the overall power demand, engine speed constraint, engine power constraint, motor power constraint, battery state of charge constraint, and battery power constraint, define the following constraints:

[0069]

[0070] Among them, P required is the power requirement of the whole machine, P e is the engine power, P M is the motor power, ω e is the engine speed, ω emin 、ω emax are the minimum and maximum engine speeds, P emax is the maximum engine power, P Mmin 、P Mmax is the minimum and maximum motor power, SoC is the battery state of charge, SoC min , SoC max is the minimum and maximum battery state of charge, P bat is the battery power, P batCmax 、PbatDmax The maximum charging and discharging power for the battery.

[0071] The relationship between the initial value of SoC, the duration of the quasi-periodic process and the endurance time is used to obtain the termination value of SoC in the quasi-periodic scenario, which is expressed as follows:

[0072]

[0073] Among them, SoC final is the final battery state of charge, SoC initial is the initial battery state of charge, t endurance is the battery life of the plug-in hybrid harvester, t quasi_period The duration of the quasi-cyclic process.

[0074] Discretize SoC with a step size of 0.01 to get [SoC min ,SoC final ], the engine torque is discrete into multiple states [Te min ,Te max ], constrain the engine power, engine speed, battery power, battery state of charge, and motor power, calculate different engine power proportions according to the whole machine power demand of the quasi-periodic process sample obtained in step 2, and then establish the corresponding initial cost table according to the different engine power proportions (the process of establishing the cost table is the existing technology), calculate the cost function values ​​under different engine power proportions, and fill the initial cost table, use the filled cost table to find the engine power proportion sequence corresponding to the minimum cost function value at the end point of the quasi-periodic process, repeat this process several times until the engine power proportion sequence between different iterative processes is consistent, and use the engine power proportion sequence at this time as the optimal engine power proportion table; wherein, Te min is the minimum engine torque, Te max is the maximum engine torque.

[0075] Step 4: Establish a BP neural network, using the feed amount, quasi-periodic operation stage, battery state of charge, and grain quality in the granary as inputs and the optimal engine power ratio as output. The trained BP neural network is used as the optimal energy management strategy model.

[0076] First, normalize the optimal data from step 3. The specific method is:

[0077]

[0078] Among them, Y is the normalized data, Y max 、Y minis the maximum normalized data and the minimum normalized data, X is the initial data, including feed amount, battery charge state, grain quality in granary, quasi-periodic operation state and optimal engine power ratio, X max 、X min are the maximum and minimum values ​​of the initial data.

[0079] Normalized data can improve the training efficiency and performance of the model and reduce the occurrence of gradient disappearance or explosion.

[0080] After that, define the evaluation function used by the BP neural network:

[0081]

[0082] Among them, MSE is the mean square error, N r is the number of output samples, P ereal is the engine power ratio calculated by the neural network, P etar is the optimal engine power ratio in the sample obtained in step 3.

[0083] The activation function of the BP neural network is selected as follows:

[0084]

[0085] The adaptive learning function is defined as follows:

[0086] d w =M c d wprev +(1-M c )l r g w

[0087] Among them, d w is the weight, d wprev is the final weight value, M c is the momentum constant, l r is the learning rate, g w is the gradient size.

[0088] During the training process, 70% of the input and output data are set as training data, 15% of the input and output data are set as test data, and 15% of the input and output data are set as confidence test data.

[0089] Step 5: Real-time acquisition of the harvester's feed volume, quasi-cycle operation phase, battery charge status, and grain quality in the granary;

[0090] According to the header dynamic equation:

[0091]

[0092] M qr =(k g +k b +k j )*q r

[0093] Among them, M H is the header torque at the sampling moment; M bf M is the friction torque of the reel wheel during idling; gf M is the friction torque of the cutter during idling operation; jf is the idling friction torque of the cutting platform auger; i b 、i j are the transmission ratios from the driving shaft to the reel and the auger respectively; J b is the moment of inertia of the reel; ω in J is the cutting head speed at the sampling time; j M is the moment of inertia of the auger; qr Feed load torque; q r is the feeding amount; k g 、k b 、k j are the feed load factors of the cutter, reel and auger respectively.

[0094] The method for estimating the feeding amount is:

[0095]

[0096] The grain quality in the granary is estimated using a granary grain storage height sensor. In this embodiment, the sensor uses an ultrasonic sensor to obtain the grain storage height of the grain bin, and then estimates the grain quality in the granary.

[0097] The quasi-periodic operation stage is obtained according to the operation status of the harvester motor, which can be seen in Table 1.

[0098] Table 1 Harvester motor operating status during quasi-periodic operation

[0099]

[0100]

[0101] The battery charge status is obtained through the CAN network of the whole machine.

[0102] Step 6: Use the real-time information as the input of the optimal energy management strategy model to obtain the optimal engine power ratio, and calculate the real-time engine power through the optimal engine power ratio, and then calculate the optimal engine torque to control the engine.

[0103] The optimal engine power ratio is used to control the engine. The specific method is as follows: obtain the real-time whole-machine power demand through the CAN bus, calculate the real-time engine power demand based on the optimal engine power ratio, and then obtain the optimal engine torque based on the real-time engine power demand and the real-time engine speed:

[0104] P engine_realtime =P realtime *R e

[0105]

[0106] Among them, P realtime is the real-time power demand of the whole machine, R e is the optimal engine power ratio, P engine_realtime is the real-time power demand of the engine, ω engine_realtime is the real-time engine speed, T engine_realtime Optimal torque for the engine.

[0107] The embodiments described are preferred implementations of the present invention, but the present invention is not limited to the above implementations. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essence of the present invention are within the scope of protection of the present invention.

Claims

1. A neural network energy management method for a plug-in hybrid electric harvester based on a quasi-periodic process, characterized by: According to the harvester operation process and different feeding amounts, a quasi-cycle database is established, and different battery initial charge states are set for the quasi-cycle database; Based on the quasi-periodic power demand model and the conversion equivalent efficiency model, the whole-machine power demand of different quasi-periodic process samples in the quasi-periodic database is calculated, and the whole-machine power demand of all quasi-periodic process samples is added to the quasi-periodic database to form a quasi-periodic whole-machine power demand database; The dynamic programming algorithm is used to solve the optimal engine power ratio table for different samples in the quasi-periodic engine power demand database; A BP neural network was established, with the feed amount, quasi-periodic operation stage, battery state of charge, and grain quality in the granary as inputs and the optimal engine power ratio as output. The trained BP neural network was used as the optimal energy management strategy model. The feed amount, quasi-periodic operation stage, battery charge status and grain quality in the granary during the harvester operation are obtained in real time, and the optimal energy management strategy model is input to obtain the optimal engine power ratio. The real-time engine power is calculated based on the optimal engine power ratio, and then the optimal engine torque is calculated to achieve engine control.

2. The plug-in hybrid electric harvester neural network energy management method according to claim 1, characterized in that: The quasi-periodic power demand model is: Among them, P h is the rated harvesting power requirement of the harvester, P d P is the power requirement of the harvester at standard speed without load. u is the harvester unloading power requirement, q is the harvester feeding amount, k ph is the power demand coefficient of feed quantity, v r is the driving speed, v s Harvester standard speed, m is the unloaded mass of the harvester, m grainmax is the fully loaded mass of the harvester, m grain is the grain quality of the granary during the harvester operation, P request The power required by the harvester components for quasi-periodic operation, s combine It is the harvester operation stage, and its values ​​1, 2, 3, and 4 correspond to harvesting operation, full-load transfer, unloading, and empty-load transfer, respectively.

3. The neural network energy management method for a plug-in hybrid electric harvester according to claim 2, characterized in that: The conversion equivalent efficiency model is: Among them, n e For engine efficiency, is the fuel consumption rate, ω e is the engine speed, T e is the engine torque, H lhv Lower calorific value of fuel, n g is the generator efficiency, is the generator efficiency function, ω g is the generator speed, T g is the generator torque, n b is the battery efficiency, is the battery efficiency function, soc b is the battery state of charge, P b is the battery power, n m is the motor efficiency, ω m is the motor speed, T m is the motor torque, is the battery efficiency function.

4. The neural network energy management method for a plug-in hybrid electric harvester according to claim 3, characterized in that: The overall power requirements of the different quasi-periodic process samples are: Among them, P required This is the overall power requirement for the quasi-periodic process sample.

5. The neural network energy management method for a plug-in hybrid electric harvester according to claim 1, characterized in that: When using the dynamic programming algorithm to solve the optimal engine power ratio table, SoC is used as the state variable, engine power ratio is used as the control variable, and the sum of fuel consumption and electricity consumption is used to establish the cost function of the dynamic programming algorithm; SoC is discretized with a step size of 0.01 to obtain [SoC min ,SoC final ], the engine torque is discrete into multiple states [Te min ,Te max ], constrain the engine power, engine speed, battery power, battery state of charge, and motor power, calculate different engine power proportions according to the overall machine power demand, then establish a corresponding initial cost table according to the different engine power proportions, calculate the cost function values ​​under different engine power proportions, and fill the initial cost table, use the filled cost table to find the engine power proportion sequence corresponding to the minimum cost function value at the end point of the quasi-periodic process, repeat this process several times until the engine power proportion sequences between different iterative processes are consistent, and use the engine power proportion sequence at this time as the optimal engine power proportion table; Among them, SoC min is the minimum battery state of charge, SoC final is the final battery state of charge, Te min is the minimum engine torque, Te max is the maximum engine torque.

6. The plug-in hybrid electric harvester neural network energy management method according to claim 5, characterized in that: The final battery state of charge is obtained by the relationship between the initial value of SoC, the duration of the quasi-periodic process and the endurance time: Among them, SoC initial is the initial battery state of charge, t endurance is the battery life of the plug-in hybrid harvester, t quasi_period The duration of the quasi-cyclic process.

7. The neural network energy management method for a plug-in hybrid electric harvester according to claim 1, characterized in that: The evaluation function used by the BP neural network is: Among them, MSE is the mean square error, N r is the number of output samples, P ereal is the engine power ratio calculated by BP neural network, P etar is the optimal engine power ratio in the sample.

8. The neural network energy management method for a plug-in hybrid electric harvester according to claim 1, characterized in that: The real-time acquisition of the feed amount during the harvester operation is estimated based on the harvester dynamics equation: Among them, q r is the feeding amount, M H is the header torque at the sampling moment, M bf is the reel idling friction torque, M gf is the friction torque of the cutter during idling, M jf is the idling friction torque of the cutting platform auger, i b 、i j are the transmission ratios from the driving shaft to the reel and the auger, J b is the moment of inertia of the reel, J j is the auger's moment of inertia, ω in is the cutting platform speed at the sampling time, k g 、k b 、k j are the feed load factors of the cutter, reel and auger respectively.

9. The neural network energy management method for a plug-in hybrid electric harvester according to claim 1, characterized in that: The optimal engine torque is: P engine_realtime =P realtime *R e Among them, T engine_realtime is the optimal torque of the engine, P engine_realtime is the real-time power demand of the engine, ω engine_realtime is the real-time engine speed, P realtime is the real-time power demand of the whole machine, R e is the optimal engine power ratio.

10. The plug-in hybrid electric harvester neural network energy management method according to claim 1, characterized in that: The sampling range of the feeding amount is 3kg / s-9kg / s with an interval of 0.1kg / s, and the sampling range of the initial state of charge of the battery is 20%-100% with an interval of 1%.

Images