CN112734097B - Unmanned train energy consumption prediction method, system and storage medium - Google Patents
Unmanned train energy consumption prediction method, system and storage medium Download PDFInfo
- Publication number
- CN112734097B CN112734097B CN202011639921.4A CN202011639921A CN112734097B CN 112734097 B CN112734097 B CN 112734097B CN 202011639921 A CN202011639921 A CN 202011639921A CN 112734097 B CN112734097 B CN 112734097B
- Authority
- CN
- China
- Prior art keywords
- energy consumption
- train
- value
- neural network
- wolf
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000005265 energy consumption Methods 0.000 title claims abstract description 188
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000003860 storage Methods 0.000 title claims abstract description 7
- 230000008569 process Effects 0.000 claims abstract description 29
- 239000002245 particle Substances 0.000 claims description 128
- 238000013528 artificial neural network Methods 0.000 claims description 82
- 241000282461 Canis lupus Species 0.000 claims description 74
- 238000004422 calculation algorithm Methods 0.000 claims description 60
- 241000282421 Canidae Species 0.000 claims description 46
- 238000012549 training Methods 0.000 claims description 36
- 238000005457 optimization Methods 0.000 claims description 23
- 239000013598 vector Substances 0.000 claims description 22
- 238000004364 calculation method Methods 0.000 claims description 12
- 238000010845 search algorithm Methods 0.000 claims description 10
- 238000011156 evaluation Methods 0.000 claims description 8
- 230000001133 acceleration Effects 0.000 claims description 7
- 230000009194 climbing Effects 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 230000002028 premature Effects 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 4
- 238000009825 accumulation Methods 0.000 claims description 3
- 230000003094 perturbing effect Effects 0.000 claims description 2
- 230000006870 function Effects 0.000 description 12
- 230000002068 genetic effect Effects 0.000 description 8
- 230000007613 environmental effect Effects 0.000 description 7
- 238000012360 testing method Methods 0.000 description 7
- 238000006243 chemical reaction Methods 0.000 description 6
- 210000000349 chromosome Anatomy 0.000 description 5
- 238000011161 development Methods 0.000 description 5
- 230000004927 fusion Effects 0.000 description 5
- 238000012795 verification Methods 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 238000004140 cleaning Methods 0.000 description 3
- 238000013524 data verification Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000007781 pre-processing Methods 0.000 description 3
- 241000208422 Rhododendron Species 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000003071 parasitic effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000013486 operation strategy Methods 0.000 description 1
- 230000035515 penetration Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
- 238000009423 ventilation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Business, Economics & Management (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Human Resources & Organizations (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Physiology (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an unmanned train energy consumption prediction method, system and storage medium, which are used for realizing the minimum traction energy consumption prediction of an unmanned train with high reliability and high precision by fusing various parameters such as collected train operation data, in-train and station passenger data, in-train environment data and the like on the basis of ensuring the requirements of safety, comfort, timekeeping and the like in the operation process of the unmanned train.
Description
Technical Field
The invention relates to the field of artificial intelligence, in particular to an unmanned train energy consumption prediction method, an unmanned train energy consumption prediction system and a storage medium.
Background
Along with continuous breakthrough and innovation of the rail transit technology, the automation level of the rail vehicles is gradually improved. In the great background of increasingly serious energy consumption problems, development of efficient energy-saving technology and operation technology are urgently needed in application of novel unmanned trains. Meanwhile, the development of the efficient and environment-friendly rail transit system is not only required to actively cope with energy crisis, but also is required to meet the development requirement of building a green city smart city in the future. If the energy consumption of the rail transportation department is too high, huge pressure can be brought to national energy supply, and the economic benefit of enterprises is limited, so that sustainable development is influenced.
In the railway sector, the energy consumption of the transportation sector accounts for more than 80% of the railway energy consumption, and the train traction power consumption is a main mode of the energy consumption of a rail transit system. Therefore, the research on the effective energy-saving train control method has great significance for reducing railway energy consumption, and simultaneously promotes to improve the parking precision and the train punctuality, thereby realizing the continuous improvement of the train automatic control system. The method has important significance for reducing railway transportation cost, improving railway transportation industry efficiency and realizing sustainable development of railways.
In the running process of the unmanned train between stations, proper working conditions need to be selected, and factors of running energy consumption caused by passenger flow change and extra energy consumption caused by external environment change also need to be considered besides the energy consumption of the normal running of the train on a fixed track. Thus, during operation of the unmanned train, the number of operating conditions can be flexibly selected according to the operating experience of big data and the line conditions. CN102360401B patent proposes an energy-saving dispatching method for urban rail transit based on genetic algorithm, CN106740998a proposes a method for realizing energy-saving operation by finding a speed switching point through vehicular ATO, and although energy saving is realized by adjusting stop time or switching point, the internal and external influencing factors of the urban rail transit vehicle and the condition of operation on the whole line are ignored.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides the unmanned train energy consumption prediction method, the system and the storage medium, and the minimum traction energy consumption prediction of the unmanned train with high reliability and high precision is realized by fusing various parameters such as collected train operation data, in-train and station passenger data, out-of-train environment data and the like on the basis of ensuring the requirements of safety, comfort, timekeeping and the like in the operation process of the unmanned train.
In order to solve the technical problems, the invention adopts the following technical scheme: an unmanned train energy consumption prediction method comprises the following steps:
1) Collecting unmanned train operation data, train interior and station passenger data and train exterior environment data; the unmanned train operation data comprise stable running speed energy consumption values, running distances and road gradient loss power of the train in a specified time interval in the running process; the passenger data in the train and the station comprise passenger flow, riding comfort evaluation indexes, luggage weight consignment and in-train boarding rate; the train external environment data comprise rainfall resistance, road ponding resistance, wind resistance energy consumption, snow accumulation energy consumption and temperature energy consumption in a specified time interval;
2) Taking the operation data of the unmanned train as the input of a DBN deep belief neural network, training the DBN deep belief neural network, and obtaining an unmanned train vehicle running energy consumption control prediction model; taking the passenger data in the train and the station as the input of a BILSTM deep neural network, training the BILSTM deep neural network, and obtaining an unmanned train passenger flow energy consumption prediction model; taking the train external environment data as input of a GRU deep neural network, training the GRU deep neural network, and obtaining an unmanned train environment energy consumption prediction model;
3) The unmanned train vehicle running energy consumption control prediction model, the unmanned train passenger flow energy consumption prediction model and the unmanned train environment energy consumption prediction model are fused to obtain an energy consumption prediction model;
4) And inputting the unmanned train operation data, the train interior and station passenger data and the train exterior environment data acquired in real time into the energy consumption prediction model to predict the train energy consumption.
The method fully considers the influence factors of various man-machine loop combinations during the operation of the unmanned train, fuses the collected various parameters to realize the low traction energy consumption prediction and optimization process with high reliability and high precision, and can effectively solve the energy saving optimization problem of the unmanned train.
In step 2), the specific training process of the unmanned train vehicle running energy consumption control prediction model comprises the following steps: and the train stable running energy consumption value, the climbing energy consumption value and the downhill energy consumption value in the unmanned train running data, the road gradient loss power and the running distance are input by a DBN deep confidence neural network, the vehicle running energy consumption after the interval time T is output by the DBN deep confidence neural network, the optimal weight and the threshold value of the DBN deep confidence neural network are searched by using a gray wolf optimization algorithm, and the DBN deep confidence neural network corresponding to the optimal weight and the threshold value is the unmanned train running energy consumption control prediction model.
The model is mainly used for predicting different historical data in the running energy consumption of the unmanned train, the DBN deep confidence neural network can enable the whole neural network to generate training data according to the maximum probability by training weights among neurons of the DBN deep confidence neural network, and meanwhile, the data characteristics can be effectively extracted and the prediction accuracy can be improved by combining the application of an optimization algorithm.
The specific implementation process for searching the optimal weight and threshold of the DBN deep confidence neural network by using the gray wolf optimization algorithm comprises the following steps:
A1, randomly placing the position of each wolf in the wolf group in a solution space, setting the population quantity of the wolf as m, the characteristic quantity of a vehicle running energy consumption original data set as d, setting the position matrix of the wolf group as a binary matrix in m multiplied by d dimension, and setting the position of the ith wolf as CG i ＝(CG i1 ,CG i2 ,...CG id )，CG i1 ,CG i2 ,...CG id Coding each dimension of each gray wolf in the wolf group; processing the input data set of the DBN deep confidence neural network according to the position information in the individual wolf to obtain a new data set, calculating and evaluating the fitness value fitness of each individual wolf, and using the individual wolf to pass through the maximum iteration number item max The position obtained by the iteration is used as a local optimal solution, namely the weight and the threshold of the DBN neural network;
a2, sequentially finding and recording the positions X of alpha wolves, beta wolves and delta wolves from large to small according to the fitness value α 、X β 、X δ ；
A3, perturbing the positions of the alpha wolves, the beta wolves and the delta wolves, comparing the perturbed positions of the alpha wolves, the beta wolves and the delta wolves with the original positions of the alpha wolves, the beta wolves and the delta wolves, reserving the positions which are closer to the optimal solution, and gradually approaching the optimal solution by utilizing the subsequent iteration process;
a4, calculating the positions of alpha wolves, beta wolves and delta wolves of each wolf individual in the wolf group by using a wolf algorithm, and obtaining the updated positions of the wolf individuals which are closer to the optimal solution by using the following formula:
wherein ,representing the position value of the d dimension of the ith gray wolf individual in the process of the t+1st iteration, X α 、X β 、X δ The positions of alpha wolves, beta wolves and delta wolves are respectively, and random is 0,1]Random numbers in between;
a5, randomly updating the positions of the wolf population according to a preset probability value Pa, calculating the updated probability r of the found wolf prey, and if r > Pa, finding the wolf prey and updating the positions of the wolf individuals; c, according to the quality of the updated position and the original position, reserving a position which is closer to the optimal solution, and entering a step A6; otherwise, directly entering the step A6;
a6, judging whether the iteration times reach the maximum iteration times, if so, stopping iteration, and outputting a global optimal value obtained according to the alpha wolf position and the fitness value fitness of the alpha wolf position; otherwise, returning to the step A2 until a global optimal value is found; the global optimal value is the optimal weight and threshold of the DBN deep belief neural network.
The inherent defect that the artificial neural network has low learning speed and local minimum points is effectively overcome. And the position of the population is updated in the iterative process, so that the algorithm is prevented from entering local optimum, and the optimizing precision and the convergence speed of the algorithm are improved.
In the step 2), the acquisition process of the passenger flow energy consumption prediction model of the unmanned train comprises the following steps: and searching the optimal weight and threshold value of the BILSTM depth neural network by utilizing a quantum particle swarm algorithm, wherein the BILSTM depth neural network corresponding to the optimal weight and threshold value is the unmanned train passenger flow energy consumption prediction model.
The model is mainly used for predicting different historical data in the passenger flow energy consumption of the unmanned train, and the BILSTM deep neural network can better capture information transmitted in two directions relative to the unidirectional LSTM neural network, so that the effectiveness of extracting the passenger flow data characteristics is enhanced, and meanwhile, the prediction accuracy is further improved by combining the application of an optimization algorithm.
The specific implementation process for searching the optimal weight and threshold of the BILSTM deep neural network by the Li Yongliang sub-particle swarm algorithm comprises the following steps:
b1, taking a position vector of each quantum particle individual in the quantum particle group as a weight and a threshold value of the BILSTM depth neural network, and initializing a position vector parameter of the quantum particle group individual as a random number of [ -1,1 ]; initializing input parameters of a quantum particle swarm algorithm;
B2, setting an fitness function, and determining an initial optimal quantum particle individual position vector and iteration times; substituting the weight value and threshold value corresponding to the quantum particle individual position vector into an unmanned train passenger flow energy consumption prediction model based on a BILSTM depth network, determining the type of identification vector labels by utilizing the unmanned train passenger flow energy consumption prediction model based on the BILSTM depth network determined by the quantum particle individual position vector, and determining the weight calculation results of passenger flow, riding comfort evaluation index, baggage weight and in-train boarding rate in train and station passenger data by utilizing the quantum particle individual position;
if the population fitness variance of the quantum particle swarm is smaller than the premature convergence judging threshold, the particles with the worst fitness and the population extremum particles in the quantum particle swarm are mutated, and the particles with the best current fitness are used as global optimal quantum particle individuals, namely the extremum of the population;
b4, when the iteration times are greater than the elite population iteration times, extracting extremum values of the populations to construct the elite population through information sharing among the populations, and turning to a step B6, otherwise turning to a step B6;
b5, updating the position parameters of each particle swarm particle by comparing the individual extremum with the global extremum, namely recalculating and comparing the fitness value of each particle, and updating the individual extremum if the fitness value is better than the current individual extremum; comparing the global extremum particles, if the particle fitness value is better than the current population extremum, updating the global extremum particles, adding 1 to the iteration times, and turning to the step B3; the global extremum particles are extremum obtained by evolution in all particle individuals including elite populations;
And B6, continuing to iterate and evolve elite population according to the steps B3-B5, judging whether the maximum iteration times are met, if yes, ending, otherwise, adding 1 to the iteration times, and turning to the step B3 until a global optimal value is found, wherein the global optimal value is the optimal weight and the threshold of the BILSTM depth neural network.
The quantum particle swarm algorithm is an improved algorithm of the particle swarm algorithm, effectively solves the problem of insufficient global convergence performance of the particle swarm algorithm, and enables an example to search in the space of the whole feasible solution, thereby searching the global optimal solution and improving the global convergence and searching capacity of the algorithm.
In step 2), the obtaining process of the unmanned train environment energy consumption prediction model comprises the following steps: and the energy consumption of rainfall resistance, the energy consumption of road ponding resistance and the energy consumption of wind resistance are used for controlling the energy consumption of the road ponding in the environmental data outside the train, the ponding energy consumption data is input into the GRU deep neural network, the energy consumption value of the train environment after the interval time T is output from the GRU deep neural network, the optimal weight and the threshold value of the GRU deep neural network are searched by utilizing a mixed particle swarm algorithm and a universal gravitation search algorithm, and the GRU deep neural network corresponding to the optimal weight and the threshold value is the unmanned train passenger flow energy consumption prediction model.
Compared with LSTM, the GRU algorithm is easier to train, the operation speed is improved, the training requirement can be rapidly met in the process of a large amount of unmanned train environment data, various important data features are reserved, and the output of the prediction result is convenient for optimization training.
In step 3), the energy consumption prediction modelThe expression of (2) is: /> wherein ,/>The prediction results of the unmanned train running energy consumption control prediction model, the prediction results of the unmanned train passenger flow energy consumption prediction model and the prediction results of the unmanned train environment energy consumption prediction model are respectively obtained; w (w) 1 、w 2 、w 3 Is a weight coefficient. The fused model can further improve the prediction precision.
The specific implementation process for searching the optimal weight and threshold of the GRU deep neural network by using the mixed particle swarm algorithm and the universal gravitation search algorithm comprises the following steps:
c1, initializing input parameters of a PSOGSA gravitation search algorithm and a particle swarm algorithm;
c2, taking the position vector of each particle as the weight and the threshold of the GRU depth neural network, setting the individual extremum of each particle as the current position, taking the reciprocal of the mean square error MSE of the predicted value and the actual value as the fitness function, calculating the fitness value of each particle by using the fitness function, and taking the individual extremum corresponding to the particle with the highest fitness value as the initial global extremum; substituting parameter values corresponding to individual particle position vectors into an unmanned train environment energy consumption prediction model based on a GRU depth network in sequence, and calculating the sum of environment energy consumption by utilizing the weight calculation results of train rainfall resistance energy consumption, road ponding resistance energy consumption, wind resistance energy consumption and ponding energy consumption determined by the individual quantum particle positions;
C3, calculating the fitness value of each particle after each iteration according to the fitness function set by C2, namely the reciprocal of the mean square error MSE of the predicted value and the actual value;
c4, comparing the fitness value of the single particle with the fitness value of the individual extremum, if the fitness value of the single particle is better, updating the individual extremum, otherwise, keeping the original value;
c5, updating the gravitation coefficient and the inertial mass of the particles, and calculating the speed and the acceleration based on the fitness value of each particle to update the position of each particle;
c6, calculating a global optimal value of the particle position and the fitness by using a GSA optimization algorithm;
and C7, judging whether a termination condition is met, if yes, exiting, otherwise, turning to a step C3 until a global optimal value is found, wherein the global optimal value is the optimal weight and the threshold of the GRU deep neural network.
The hybrid particle swarm algorithm and the universal gravitation search algorithm are improved algorithms of the particle swarm algorithm, and the global optimizing capability and the particle swarm algorithm of the universal gravitation search algorithm are utilized to improve the optimizing capability of the hybrid algorithm for the characteristics of increasing the memory and the social information exchange capability of particles, and meanwhile the defect of algorithm stagnation is effectively overcome.
The invention also provides an unmanned train energy consumption prediction system, which comprises computer equipment; the computer device is configured or programmed to perform the steps of the method of the invention.
The present invention also provides a computer storage medium storing a program; the program is configured or programmed to perform the steps of the method of the present invention.
Compared with the prior art, the invention has the following beneficial effects:
1. according to the invention, the energy consumption conditions of the unmanned vehicle under various complex environments inside and outside the train are considered in addition to the common driving energy consumption, the comprehensive calculation and prediction of the influence parameters of the unmanned train are realized, the real-time detection data of the train, the passenger data and the external environment influence are organically combined, and a more complete consideration factor is provided for the energy-saving research and development field of the unmanned vehicle.
2. According to the invention, man-machine loop fusion construction is performed outside traditional train mechanics analysis and train running state research, subsystem elements are fully considered on the basis of actual train running experience, and man-machine-loop matching conditions and standards are provided, so that a multi-target unmanned train energy-saving optimization model is established. Meanwhile, different energy-saving regulation strategies are selected through collected data, and comprehensive detection measures and control schemes are provided. The unmanned train is operated in an energy-saving optimization mode based on personnel, vehicles and environmental parameters, so that the operation cost of the train can be saved, and the convenience and energy-saving benefit of the unmanned train are improved to the greatest extent.
3. The invention comprehensively considers the speed energy consumption value, the driving distance, the road gradient loss power, the load and comfort energy consumption in the unmanned train, the accumulated snow, the accumulated water wind power and other factors at the stable driving speed of the unmanned train, and utilizes various prediction models and optimization algorithms to carry out weighted fusion, establishes a traction energy consumption model based on parameters to carry out energy consumption analysis and prediction, thereby ensuring the effectiveness of energy consumption strategy selection and real-time energy consumption adjustment instructions. The configuration of unmanned train energy consumption multi-objective optimization and the minimum energy consumption of the unmanned train driving route are realized, and the energy consumption of the whole urban rail transit network can be reduced.
Drawings
Fig. 1 is a schematic block diagram of an embodiment of the present invention.
Detailed Description
As shown in fig. 1, the embodiment of the present invention mainly includes the following steps:
step 1: unmanned train man-machine ring parameter signal acquisition
The method comprises the steps of train operation data acquisition, in-car and station passenger data acquisition and inside and outside environment data acquisition, and comprises the following specific contents:
1) Unmanned train operation data acquisition
The training data are collected through the vehicle-mounted equipment, the trackside equipment and the like of the unmanned train, and the training data comprise stable driving speed energy consumption values, driving distances and road gradient loss power of the unmanned train in a designated time interval in the driving process.
The real-time travel speed and the travel distance within the interval time T of the unmanned train can be obtained by the in-vehicle apparatus. The road gradient loss power calculation is calculated by combining vehicle body offset angle information collected by a level meter arranged at the bottom of a vehicle and the running speed of the vehicle. The time is obtained by integrating the climbing factor of the unmanned train in the interval time T.
2) Data acquisition of passengers in unmanned train and station
The training data including passenger flow, riding comfort evaluation index, luggage weight, in-car boarding rate, etc. are collected by the unmanned train on-board equipment, station fixing equipment, etc. The comfort evaluation is calculated by fusing temperature data measured by a temperature sensor and train ventilation quantity, and the rest is calculated by a load sensor to obtain train load data.
3) Unmanned train external environment data acquisition
The training data are collected through the vehicle-mounted equipment of the unmanned train, the railway monitoring station and the like, and comprise rainfall resistance, road ponding resistance, wind resistance energy consumption, snow accumulation energy consumption, temperature energy consumption and the like in a specified time interval.
The rainfall resistance calculation module of the vehicle body obtains rainfall resistance in the vehicle body surface interval time T by utilizing data fusion collected by a force sensor arranged on the surface of the unmanned vehicle body.
The vehicle-mounted camera is utilized to collect road ponding images, and the infrared detector beside the rail is matched with the ponding information to save the water penetration resistance of the vehicle and calculate the ponding energy consumption.
The wind resistance energy consumption is obtained by combining a trackside wind measuring station and an auxiliary wind measuring station thereof to collect wind speed data in real time, obtaining a wind speed sample set along a railway and accumulating the relative wind speeds in an interval time T.
The energy consumption generated by the accumulated snow on the road surface is obtained by acquiring the road surface image by using an image acquisition device, performing triangulation by using an infrared image and a reference image to obtain a depth image, and performing recognition and feature extraction on the reconstructed three-dimensional information by using a three-dimensional reconstruction method.
The temperature energy consumption value is obtained by accumulating the difference value between the temperature inside the unmanned vehicle and the temperature outside the unmanned vehicle in the interval time T;
step 2: unmanned train man-machine loop parameter signal transmission and preprocessing
Each carriage of the unmanned train is provided with a wireless transmission device which is used for connecting a vehicle-mounted data acquisition module, an off-board data acquisition module and a platform data processing center, so that data transmission of acquisition data storage and different modules is realized, and wireless network transmission can be adopted.
And a central computer is arranged on each train of unmanned trains to form a data processing model to receive key data collected from a monitoring range, respectively perform data preprocessing and model training, and output model training results in real time. Meanwhile, the man-machine interaction end is used for receiving the instruction sent by the platform data center and displaying the instruction on the interaction port to guide the next operation of the train.
And preprocessing the original data by removing error information, supplementing missing values, designing labels and the like. The validity and diversity of the original data are ensured. Meanwhile, the original data is divided into a training set, a verification set and a test set. The neural network is fully trained, the performance of the model is accurately tested, and the optimization algorithm is utilized to further integrate and process the multi-class data, so that the prediction performance is improved.
Step 3: unmanned aerial vehicle train man-machine ring parameter information processing model
Step 3.1: train unmanned train vehicle running energy consumption control prediction model
Training the energy consumption of the unmanned train adopts the DBN deep belief neural network to train the historical data. The input of the model is the steady running energy consumption value, the climbing and descending energy consumption value, the loss power and the running distance history data of the train. 2000 time sequences of the running energy of the unmanned train are collected, and the obtained data are divided into a training set, a verification set and a test set. Based on DBN deep confidence neural network prediction, the weight process of the unmanned train operation energy consumption value obtained by integrating and optimizing the input parameters by adopting a sirius optimization (CS-GWO) algorithm of fusion azalea search is as follows:
A1: acquiring historical data of running energy consumption of unmanned train
The energy consumption historical data comprises a train stable running energy consumption value, climbing and descending energy consumption values, loss power and running distance. And loading the driving data set of the unmanned train and pre-cleaning the training data.
A2: and taking the stable running energy consumption value, the climbing and descending energy consumption value, the loss power and the running distance of the train in the historical running data verification set as input data and the energy consumption after the interval time T as output data, wherein the weight and the threshold of the DBN neural network in the unmanned vehicle energy consumption prediction model based on the DBN neural network are trained and optimized by adopting a gray wolf optimization (CS-GWO) algorithm of fusion azalea search.
Setting the maximum iteration number m of an algorithm, setting the size of the wolf population, and setting the found probability Pa of the wolf population; the dimension n of the dataset is set.
A3: randomly placing the position of each wolf in the wolf group in the solution space, setting the population number of the wolf as m, the characteristic number of the original data set of the vehicle driving energy consumption as d, setting the position matrix of the wolf group as a binary matrix in m multiplied by d dimension, and setting the position of the ith wolf as CG i ＝(CG i1 ,CG i2 ,...CG id )，CG i1 ,CG i2 ,...CG id Coding each dimension of each gray wolf in the wolf group; processing the input data set of the DBN deep confidence neural network according to the position information in the individual wolf to obtain a new data set, calculating and evaluating the fitness value fitness of each individual wolf, and using the individual wolf to pass through the maximum iteration number item max The position obtained by the iteration is used as a local optimal solution, namely the weight and the threshold of the DBN neural network;
a4: sequentially finding and recording the positions X of alpha wolves, beta wolves and delta wolves according to the fitness value from large to small α 、X β 、X δ 。
A5: the positions of alpha wolves, beta wolves and delta wolves in the step 4 are disturbed by the following method, and the value of the gray wolves is decimal number
wherein ,xi (t+1) is the position of the ith nest after updating, alpha is the variable of the control step length, L (lambda) is the random number of the representative Lewy distribution, and the formula is as follows
L*u＝t -λ ,1＜i≤n
The updated wolf position is then mapped to binary using the following disclosure. And comparing the disturbed position with the original position, and keeping the position which is more approximate to the prey.
in the formula ,representing the value of the d-th dimension component of the ith parasitic nest from the algorithm to the ith generation, and random is [0,1]Random numbers in between; conversion of decimal numbers into binary values is achieved by mapping the next generation value of the parasitic nest into 0 or 1
A6: for each individual wolf in the wolf group, the position X of alpha wolf, beta wolf and delta wolf is calculated by using GWO algorithm update formula α 、X β 、X δ At this time, their values are decimal numbers, and the updated position of the wolf is obtained using the following formula.
Representing the value of the d dimension of the ith gray wolf individual in the process of the t+1th iteration, and random is [0,1]Random number in between, and the individual value of the wolf is mapped to 0,1 through conversion]To represent the probability that they take 0 or 1, the updated binary value for each dimension of the wolf individual is obtained.
A7: and calculating the probability Pa of finding the updated wolf prey, and if r > Pa, finding the wolf prey and updating the position of the wolf individual. And (5) according to the quality of the updated position and the original position, reserving a position which is closer to the optimal solution of the prey. Otherwise, directly enter the next step.
A8: the positions of alpha wolves, beta wolves and delta wolves are updated and recorded. Judging whether the iteration times of the algorithm reach the maximum iteration times max, if so, stopping the algorithm from iterating, and outputting a global optimal value obtained according to the alpha wolf position and the fitness value fitness of the alpha wolf position; otherwise, executing the step A5 again until the global optimal value is found, and outputting the optimal weight and the threshold value of the DBN neural network.
A9: inputting the data of the unmanned driving history driving data training set into the DBN neural network for training, outputting a prediction result and evaluating the model performance traveling by using a test set.
Step 3.2: training unmanned train passenger flow energy consumption prediction model
The passenger flow historical data of the unmanned train is trained by adopting a BILSTM deep neural network, and the BILSTM is an abbreviation of a bidirectional long-short-term memory and structurally consists of a forward LSTM and a backward LSTM. The BILSTM may connect two hidden layers with different directions for output, meaning that it contains both forward and reverse data. The output layer may obtain past and future information in the input data through the structure. BILSTM deep neural networks are capable of efficiently analyzing time series data with long-term and short-term dependencies.
The training data including passenger flow, riding comfort evaluation index, luggage weight consignment and in-car seating rate are collected through unmanned train vehicle-mounted equipment, station fixing equipment and the like. The method comprises the steps of collecting 2000 passenger flows of the unmanned train in time sequence, and dividing the obtained data into a training set, a verification set and a test set. The process of integrated optimization of the predictions of the input parameters by adopting quantum particle swarm algorithm based on BILSTM deep neural network is as follows:
b1: acquiring passenger flow energy consumption historical data of unmanned train
And loading an unmanned train passenger flow energy consumption data set, and pre-cleaning training data. And taking the passenger flow in the historical passenger flow energy consumption data verification set, the riding comfort evaluation index, the weight of the checked baggage and the in-vehicle seating rate as input data, taking the energy consumption after the interval time T as output data, and training and optimizing the weight and the threshold of the BILSTM neural network in the passenger flow energy consumption prediction model of the unmanned vehicle based on the BILSTM neural network by adopting a quantum particle swarm algorithm.
B2: taking the position vector of each quantum particle individual in the quantum particle group as the weight and the threshold value of the BILSTM neural network, and initializing the position vector parameters of the quantum particle group individual as the random number of [ -1,1 ]; the value range of the number of the particle swarm group is [10,100], the value range of the number of the particles of the particle swarm group is [3,60], the value range of the maximum iteration number is [200,1200], the value range of the iteration number of the elite group is [20,200], the value range of the premature convergence judgment threshold is [0.02,0.5], and the value range of the worst particle variation ratio of the group is [1%,5% ];
b3: setting a fitness function, determining an initial optimal quantum particle individual position vector and iteration times t, wherein t=1, sequentially bringing parameter values corresponding to the quantum particle individual position vector into the fitness function, utilizing weight calculation results of each train energy consumption parameter determined by the quantum particle individual position, and taking the reciprocal of a mean square error MSE of the calculation results and an actual value as a second fitness function f 2 (x)，f 2 (x)＝1/MSE；
B4: calculating the group fitness variance of each quantum particle swarm, and judging the premature convergence;
if the population fitness variance of the quantum particle swarm is smaller than the premature convergence judging threshold value gamma, the worst fitness particle and population extremum particle of delta% in the quantum particle swarm are mutated, and the particle with the best current fitness is used as a global optimal quantum particle individual;
B5: judging whether to build elite population; when the iteration times are greater than the elite population iteration times, extracting extremum of each population to construct elite population through information sharing among the populations, and turning to step B9, otherwise turning to step B6;
b6: updating various population particle parameters;
b7: recalculating and comparing fitness values of each particle, and updating the individual extremum if the fitness value is superior to the current individual extremum; comparing the global extremum particles, if the particle fitness value is better than the current population extremum, updating the global extremum particles, and turning to the step B4, wherein t=t+1;
b8: the elite population continues to evolve;
b9: and (3) judging whether the maximum iteration times are met, if yes, exiting, otherwise, enabling t=t+1 to enter a step B4 until a global optimal value is found, and outputting the optimal weight and the threshold value of the passenger flow energy consumption prediction model of the corresponding BILSTM depth network.
B10: inputting the data of the unmanned passenger flow energy consumption data training set into the BILSTM neural network for training, outputting a prediction result and evaluating the model performance by using a test set.
Step 3.3: training unmanned train environment energy consumption prediction model
The energy consumption prediction caused by the natural environment based on the GRU deep neural network is adopted, and the model is input into rainfall resistance energy consumption, road ponding resistance energy consumption, wind resistance energy consumption and ponding energy consumption data during the operation of the unmanned train. The unmanned train environment energy time sequence is 2000, and the acquired data is divided into a training set, a verification set and a test set. The weight of each parameter prediction result of the model is obtained by adopting a mixed particle swarm algorithm and a universal gravitation search algorithm (PSOGSA) to carry out optimization selection, and the process is as follows:
C1: acquiring environmental energy consumption historical data of unmanned train
And loading an unmanned train passenger flow energy consumption data set, and pre-cleaning training data. And taking rainfall resistance energy consumption, road ponding resistance energy consumption, wind resistance energy consumption and ponding energy consumption which are concentrated in the historical passenger flow energy consumption data verification as input data and taking the energy consumption after an interval time T as output data, wherein the weight and the threshold of the GRU neural network in the unmanned vehicle environment energy consumption prediction model based on the GRU neural network are trained and optimized by adopting a mixed particle swarm algorithm and a universal gravitation search algorithm.
C2: initializing input parameters of a PSOGSA particle swarm algorithm, setting a population scale as N=50, setting a maximum iteration number as T=1000, setting a function dimension as d=30, setting an inertia weight as w=0.9, and setting an acceleration factor as c1=0.5 and c2=1.5; the calculation formula of PSOGSA is as follows:
V i (t+1)＝w×V i (t)+d 1 ×rand×ac i (t)+d 2 ×rand×(gbest-X i (t))
X i (t+1)＝X i (t)+V i (t+1)
in the above, V i Indicating speed, X i (t) is the current position of the ith particle, t is the number of iterations, the inertial weight is the weight, d 1 and d2 Is the acceleration coefficient, rand is [0,1 ]]Random variable, ac i (t) is the acceleration of the ith particle in t iterations, gbest is the current best solution, V i Is the speed of the ith particle.
And C3: calculating fitness value, taking the position vector of the individual particles as the weight and the threshold of the GRU neural network, setting the individual extremum of each particle as the current position, calculating the fitness value of each particle by using a fitness function, and taking the individual extremum corresponding to the particle with good fitness as the initial global extremum. Sequentially bringing parameter values corresponding to individual position vectors of particles, utilizing weight calculation results of energy consumption parameters of each train determined by individual positions of quantum particles, and taking mean square error MSE of the calculation results and actual values as fitness function
And C4: calculating the fitness value of each particle after each iteration according to the fitness function of the particle;
c5: comparing the fitness value of each particle with the fitness value of each individual extremum, if the fitness value is more optimal, updating the individual extremum, otherwise, keeping the original value;
c6: updating gravity coefficient and inertial mass of particle, calculating velocity and acceleration to update particle position
C7: using global optimum values for calculating particle position and fitness
And C8: and judging whether the termination condition is met, if yes, exiting, otherwise, turning to the step C4 until a global optimal value is found, and outputting the optimal weight and the threshold of the environment energy consumption prediction model of the corresponding GRU depth network.
C9: inputting the data of the unmanned environment energy consumption data training set into the GRU neural network for training, outputting a prediction result and evaluating the model performance by using a test set.
Step 4: unmanned train energy consumption adjustment
After information such as the running energy consumption, the passenger flow energy consumption, the environmental energy consumption and the like of the unmanned train is obtained, a vehicle-mounted central computer and a platform data center of the unmanned train output real-time energy consumption information, and factors such as a central energy-saving adjustment instruction, a central departure interval, a horizontal line condition, a vertical line condition, a vehicle braking mode, a train running position and the like which influence expert strategies are defined in unmanned train comprehensive comparison, so that the next running of the train is guided in time according to the existing prediction result and the line real-time state, and a framework based on genetic evolution of a plurality of groups is adopted, wherein the MPGA exceeds the framework only by a single group, and a plurality of groups are introduced to perform optimized searching simultaneously.
And 3, three deep neural networks are used for completing the prediction of different types of energy consumption sequences, and have stronger learning and modeling capabilities, unlike the traditional shallow neural network. The multi-population genetic algorithm is used for integrating the prediction results of three deep networks. The final prediction results are obtained by integrating the prediction results of three depth networks, and the model set is realized by setting the weight coefficients of the prediction results of the depth networks. In addition, the integration of multiple deep learning methods can effectively improve the adaptability and robustness of the model.
In the operation process of the unmanned train, the main operation phases are an acceleration phase, a cruising phase, an idle running phase and a braking phase. The energy-saving control of the unmanned train is realized by completing a conversion instruction based on the running energy consumption, the passenger flow energy consumption and the environmental energy consumption of the unmanned train caused by the influence of real-time man-machine loop parameters during the running of the train and adjusting the running state of the train. The MPGA is used for calculating the working condition conversion points, integrating the energy consumption predicted values to feed back the center to determine the optimal energy consumption position, namely the conversion position of each stage.
The algorithm for solving the energy-saving operation strategy model of the unmanned train comprises the following steps:
d1: and reading the basic simulation data and calculating corresponding parameters. And reading corresponding predicted values of the running energy consumption, the passenger flow energy consumption and the environmental energy consumption of the unmanned train. The predicted value weights of the three optimized neural networks are trained and optimized by adopting MPGA (Lv Hui, zhou Cong, juan, zheng Jinhua) based on various genetic algorithms [ J ]. Computer engineering and application, 2010,46 (28): 57-60.) of group evolution.
D2: and randomly generating a plurality of initial populations, setting the population number, the initial population individual number and the individual length, generating an initial population P (t), and taking the position represented by the chromosome in the essence population as an optimal weight coefficient. According to the existing information, the method is divided into various groups:
P(t)＝{P 1 (t),P 2 (t),P 3 (t)} (1)
D3: and (5) determining control parameters. And dispersing individuals in the initial population according to different running conditions of the unmanned train, and taking different control parameters to ensure the differential evolution of various populations. The main control parameter is the crossover probability P c Probability of variation P b The value maintains the balance of the algorithm global search and the local search, and the calculation formula is as follows
P co ，P bo The initial crossover probability and the variation probability are respectively; g is population number; c, b is the interval length of the crossover and mutation operation; f (f) rand As a function of the generation of random numbers. P (P) c Generally in [0.7,0.9 ]]Randomly generate P in interval b Generally in [0.001,0.05 ]]Randomly generating in intervals, and taking the optimal chromosome in a plurality of genetic algorithms as the human-computer loop energy consumption integration weight.
D4: optimizing each population by adopting a genetic algorithm, transferring chromosomes among the populations, and selecting optimal chromosomes from the optimized populations respectively and preferentially, and adding the optimal chromosomes into elite populations, so that the elite populations contain not only local optimal solutions but also global optimal solutions.
D5: elite individuals are selected according to elite retention strategies and new populations are generated. The MPGA decides the algorithm termination and extracts the globally optimal solution according to the essence population. Comparing the optimized result with the result before optimization, and selecting the optimal weight coefficient as the global optimal solution of the optimization problem.
D6: integrating the running energy consumption, the passenger flow energy consumption and the environment energy consumption information of the unmanned train according to the results obtained by MPGA of a plurality of swarm genetic algorithms to generate an energy-saving solution, wherein w is as follows i Is a weight coefficient of three kinds of deep networks,is a prediction result for each depth network.
And feeding back the operation result to the vehicle-mounted central computer of the unmanned train and the platform data center to output a real-time driving state conversion instruction, thereby effectively controlling the traction energy consumption value of the unmanned train. Compared with the traditional genetic algorithm, the MPGA is not easy to sink into local optimum, and the running energy consumption, the passenger flow energy consumption and the environmental energy consumption information of the unmanned train are integrated comprehensively aiming at different actual running environments of the unmanned train, so that the self-adaptive adjustment of the optimal solution of each data characteristic is realized, the overall optimization of the correlation characteristic is realized, and the reliability of the optimal solution is enhanced.
Claims (6)
1. The unmanned train energy consumption prediction method is characterized by comprising the following steps of:
1) Collecting unmanned train operation data, train interior and station passenger data and train exterior environment data; the unmanned train operation data comprise stable running speed energy consumption values, running distances and road gradient loss power of the train in a specified time interval in the running process; the passenger data in the train and the station comprise passenger flow, riding comfort evaluation indexes, luggage weight consignment and in-train boarding rate; the train external environment data comprise rainfall resistance, road ponding resistance, wind resistance energy consumption, snow accumulation energy consumption and temperature energy consumption in a specified time interval;
2) Taking the operation data of the unmanned train as the input of a DBN deep belief neural network, training the DBN deep belief neural network, and obtaining an unmanned train vehicle running energy consumption control prediction model; taking the passenger data in the train and the station as the input of a BILSTM deep neural network, training the BILSTM deep neural network, and obtaining an unmanned train passenger flow energy consumption prediction model; taking the train external environment data as input of a GRU deep neural network, training the GRU deep neural network, and obtaining an unmanned train environment energy consumption prediction model;
3) The unmanned train vehicle running energy consumption control prediction model, the unmanned train passenger flow energy consumption prediction model and the unmanned train environment energy consumption prediction model are fused to obtain an energy consumption prediction model;
preferably, the method further comprises:
4) Inputting the operation data of the unmanned train, the passenger data in the train and the station and the external environment data of the train, which are acquired in real time, into the energy consumption prediction model to predict the energy consumption of the train;
in step 2), the specific training process of the unmanned train vehicle running energy consumption control prediction model comprises the following steps: the train stable running energy consumption value, the climbing energy consumption value and the downhill energy consumption value in the unmanned train running data, the road gradient loss power and the running distance are input by a DBN deep confidence neural network, the vehicle running energy consumption after the interval time T is output by the DBN deep confidence neural network, an optimal weight and a threshold value of the DBN deep confidence neural network are searched by using a gray wolf optimization algorithm, and the DBN deep confidence neural network corresponding to the optimal weight and the threshold value is the unmanned train running energy consumption control prediction model;
The specific implementation process for searching the optimal weight and threshold of the DBN deep confidence neural network by using the gray wolf optimization algorithm comprises the following steps:
a1, randomly placing the position of each wolf in the wolf group in a solution space, setting the population quantity of the wolf as m, the characteristic quantity of a vehicle running energy consumption original data set as d, setting the position matrix of the wolf group as a binary matrix with m multiplied by d dimension, and setting the position of the ith wolf as CG i ＝(CG i1 ,CG i2 ,...CG id )，CG i1 ,CG i2 ,...CG id Coding each dimension of each gray wolf in the wolf group; processing the input data set of the DBN deep confidence neural network according to the position information in the individual wolf to obtain a new data set, calculating and evaluating the fitness value fitness of each individual wolf, and using the individual wolf to pass through the maximum iteration number item max The position obtained by the iteration is used as a local optimal solution, namely the weight and the threshold of the DBN neural network;
a2, sequentially finding and recording the positions X of alpha wolves, beta wolves and delta wolves from large to small according to the fitness value α 、X β 、X δ ；
A3, perturbing the positions of the alpha wolves, the beta wolves and the delta wolves, comparing the perturbed positions of the alpha wolves, the beta wolves and the delta wolves with the original positions of the alpha wolves, the beta wolves and the delta wolves, and reserving the positions which are closer to the optimal solution;
A4, calculating the positions of alpha wolves, beta wolves and delta wolves of each wolf individual in the wolf group by using a wolf algorithm, and obtaining the updated positions of the wolf individuals which are closer to the optimal solution by using the following formula:
wherein ,representing the position value of the d dimension of the ith gray wolf individual in the process of the t+1st iteration, X α 、X β 、X δ The positions of alpha wolves, beta wolves and delta wolves are respectively, and random is 0,1]Random numbers in between;
a5, randomly updating the positions of the wolf population according to a preset probability value Pa, calculating the updated probability r of the found wolf prey, and if r > Pa, finding the wolf prey and updating the positions of the wolf individuals; according to the quality of the updated position and the original position, reserving a position which is closer to an optimal solution; otherwise, directly entering the step A6;
a6, judging whether the iteration times reach the maximum iteration times, if so, stopping iteration, and outputting a global optimal value obtained according to the alpha wolf position and the fitness value fitness of the alpha wolf position; otherwise, returning to the step A2 until a global optimal value is found; the global optimal value is the optimal weight and threshold of the DBN deep belief neural network;
in step 2), the obtaining process of the unmanned train environment energy consumption prediction model comprises the following steps: the method comprises the steps that rainfall resistance energy consumption, road ponding resistance energy consumption and wind resistance energy consumption are used in the operation period of an unmanned train in the train external environment data, ponding energy consumption data are input into a GRU deep neural network, train environment energy consumption values after interval time T are output from the GRU deep neural network, a mixed particle swarm algorithm and a universal gravitation search algorithm are used for searching an optimal weight and a threshold value of the GRU deep neural network, and the GRU deep neural network corresponding to the optimal weight and the threshold value is an unmanned train passenger flow energy consumption prediction model;
The specific implementation process for searching the optimal weight and threshold of the GRU deep neural network by using the mixed particle swarm algorithm and the universal gravitation search algorithm comprises the following steps:
c1, initializing input parameters of a PSOGSA gravitation search algorithm and a particle swarm algorithm;
c2, taking the position vector of each particle as the weight and the threshold of the GRU depth neural network, setting the individual extremum of each particle as the current position, taking the reciprocal of the mean square error MSE of the predicted value and the actual value as the fitness function, calculating the fitness value of each particle by using the fitness function, and taking the individual extremum corresponding to the particle with the highest fitness value as the initial global extremum; substituting parameter values corresponding to the individual position vectors of the particles into an unmanned train environment energy consumption prediction model based on the GRU depth network in sequence for iteration;
c3, calculating the fitness value of each particle after each iteration according to the fitness function set by the C2;
c4, comparing the fitness value of the single particle with the fitness value of the individual extremum, if the fitness value of the single particle is better, updating the individual extremum, otherwise, keeping the original value;
c5, updating the gravitation coefficient and the inertial mass of the particles, and calculating the speed and the acceleration based on the fitness value of each particle to update the position of each particle;
C6, calculating a global optimal value of the particle position and the fitness by using a GSA optimization algorithm;
and C7, judging whether a termination condition is met, if yes, exiting, otherwise, turning to a step C3 until a global optimal value is found, wherein the global optimal value is the optimal weight and the threshold of the GRU deep neural network.
2. The method for predicting the energy consumption of an unmanned train according to claim 1, wherein in the step 2), the process of obtaining the unmanned train passenger flow energy consumption prediction model includes: and searching the optimal weight and threshold value of the BILSTM depth neural network by utilizing a quantum particle swarm algorithm, wherein the BILSTM depth neural network corresponding to the optimal weight and threshold value is the unmanned train passenger flow energy consumption prediction model.
3. The unmanned train energy consumption prediction method according to claim 2, wherein the specific implementation process of searching the optimal weight and threshold of the BILSTM depth neural network by using a quantum particle swarm algorithm comprises the following steps:
B1, taking a position vector of each quantum particle individual in the quantum particle group as a weight and a threshold value of the BILSTM depth neural network, and initializing a position vector parameter of the quantum particle group individual as a random number of [ -1,1 ]; initializing input parameters of a quantum particle swarm algorithm;
b2, setting an fitness function, and determining an initial optimal quantum particle individual position vector and iteration times by taking the reciprocal of a mean square error MSE of a predicted value and an actual value as the fitness function; substituting a weight value and a threshold value corresponding to the quantum particle individual position vector into an unmanned train passenger flow energy consumption prediction model based on a BILSTM depth network, determining the type of identification vector labels by utilizing the unmanned train passenger flow energy consumption prediction model based on the BILSTM depth network determined by the quantum particle individual position vector, determining the passenger flow, riding comfort evaluation index, consignment luggage weight and weight calculation result of the in-train boarding rate in train and station passenger data by utilizing the quantum particle individual position, and calculating the sum of passenger flow energy consumption;
if the population fitness variance of the quantum particle swarm is smaller than the premature convergence judging threshold, the worst fitness particles and the population extremum particles in the quantum particle swarm are mutated, and the particles with the best current fitness are used as global optimal quantum particle individuals, namely the extremum of the population;
B4, when the iteration times are greater than the elite population iteration times, extracting extremum values of the populations to construct the elite population through information sharing among the populations, and turning to a step B6, otherwise turning to a step B6;
b5, updating the position parameters of each particle swarm particle by comparing the individual extremum with the global extremum, namely recalculating and comparing the fitness value of each particle, and updating the individual extremum if the fitness value is better than the current individual extremum; comparing the global extremum particles, if the particle fitness value is better than the current population extremum, updating the global extremum particles, adding 1 to the iteration times, and turning to the step B3; the global extremum particles are extremum obtained by evolution in all particle individuals including elite populations;
and B6, repeating the steps B3 to B5, judging whether the maximum iteration times are met, if yes, ending, otherwise, adding 1 to the iteration times, and turning to the step B3 until a global optimal value is found, wherein the global optimal value is the optimal weight and the threshold of the BILSTM depth neural network.
4. A method for predicting energy consumption of an unmanned train according to any one of claims 1 to 3, wherein in step 3), the energy consumption prediction model isThe expression of (2) is:
wherein ,/>The prediction results of the unmanned train running energy consumption control prediction model, the prediction results of the unmanned train passenger flow energy consumption prediction model and the prediction results of the unmanned train environment energy consumption prediction model are respectively obtained; w (w) 1 、w 2 、w 3 Is a weight coefficient.
5. An unmanned train energy consumption prediction system is characterized by comprising computer equipment; the computer device being configured or programmed for performing the steps of the method of one of claims 1 to 3.
6. A computer storage medium, characterized in that it stores a program; the program being configured or programmed to perform the steps of the method of one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011639921.4A CN112734097B (en) | 2020-12-31 | 2020-12-31 | Unmanned train energy consumption prediction method, system and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011639921.4A CN112734097B (en) | 2020-12-31 | 2020-12-31 | Unmanned train energy consumption prediction method, system and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112734097A CN112734097A (en) | 2021-04-30 |
| CN112734097B true CN112734097B (en) | 2023-09-05 |
Family
ID=75608917
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011639921.4A Active CN112734097B (en) | 2020-12-31 | 2020-12-31 | Unmanned train energy consumption prediction method, system and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112734097B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113361792B (en) * | 2021-06-21 | 2022-10-21 | 吉林大学 | Urban electric bus travel energy consumption estimation method based on multivariate nonlinear regression |
| CN113822473B (en) * | 2021-09-03 | 2023-12-26 | 浙江中控信息产业股份有限公司 | Traction energy consumption reasonable interval prediction method based on multidimensional data |
| CN113687242A (en) * | 2021-09-29 | 2021-11-23 | 温州大学 | Lithium ion battery SOH estimation method for optimizing and improving GRU neural network based on GA algorithm |
| CN113997915B (en) * | 2021-11-26 | 2022-06-24 | 北京大象科技有限公司 | Big data-based automatic train operation ATO (automatic train operation) accurate parking control method |
| CN114545280B (en) * | 2022-02-24 | 2022-11-15 | 苏州市职业大学 | New energy automobile lithium battery life prediction method based on optimization algorithm |
| CN117195484B (en) * | 2023-08-08 | 2024-05-03 | 广东贝能达交通设备有限公司 | Rail transit management method and system |
| CN117408720B (en) * | 2023-11-14 | 2024-05-07 | 武汉亿量科技有限公司 | Method, device, equipment and medium for preventing complaint marks of outer calls of electric pins |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108572648A (en) * | 2018-04-24 | 2018-09-25 | 中南大学 | A kind of automatic driving vehicle power supply multi-source fusion prediction technique and system |
| CN108846571A (en) * | 2018-06-08 | 2018-11-20 | 福州大学 | A kind of net connectionization electric car macroscopic view energy consumption estimation method |
| CN110852498A (en) * | 2019-10-31 | 2020-02-28 | 西安交通大学 | Method for predicting data center energy consumption efficiency PUE based on GRU neural network |
| CN111356620A (en) * | 2017-08-24 | 2020-06-30 | 图森有限公司 | System and method for autonomous vehicle control to minimize energy costs |
-
2020
- 2020-12-31 CN CN202011639921.4A patent/CN112734097B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111356620A (en) * | 2017-08-24 | 2020-06-30 | 图森有限公司 | System and method for autonomous vehicle control to minimize energy costs |
| CN108572648A (en) * | 2018-04-24 | 2018-09-25 | 中南大学 | A kind of automatic driving vehicle power supply multi-source fusion prediction technique and system |
| CN108846571A (en) * | 2018-06-08 | 2018-11-20 | 福州大学 | A kind of net connectionization electric car macroscopic view energy consumption estimation method |
| CN110852498A (en) * | 2019-10-31 | 2020-02-28 | 西安交通大学 | Method for predicting data center energy consumption efficiency PUE based on GRU neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112734097A (en) | 2021-04-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112734097B (en) | Unmanned train energy consumption prediction method, system and storage medium | |
| CN109508751B (en) | Deep neural network model modeling method for high-speed railway train late time prediction | |
| CN112686453B (en) | Intelligent prediction method and system for locomotive energy consumption | |
| CN110458456B (en) | Demand response type public transportation system scheduling method and system based on artificial intelligence | |
| CN108510737B (en) | Unmanned vehicle power supply real-time monitoring method and device integrating wind environment | |
| CN108647813B (en) | High-speed train dynamic interval energy-saving optimization method based on elastic force adjustment | |
| CN112230675B (en) | Unmanned aerial vehicle task allocation method considering operation environment and performance in collaborative search and rescue | |
| Guo et al. | A novel energy consumption prediction model with combination of road information and driving style of BEVs | |
| CN110299011A (en) | A kind of traffic flow forecasting method of the highway arbitrary cross-section based on charge data | |
| CN103280100A (en) | Bus IC card passenger transfer station judgment method combining intelligent dispatching data | |
| CN110288212A (en) | Electric taxi based on improved MOPSO creates charging station site selecting method | |
| CN110220725A (en) | A kind of wheel for metro vehicle health status prediction technique integrated based on deep learning and BP | |
| CN115063978B (en) | Bus arrival time prediction method based on digital twins | |
| CN111063204B (en) | Expressway vehicle speed prediction model training method based on toll station flow | |
| CN114492919A (en) | Electric automobile charging load prediction system and method based on traffic flow | |
| CN112734094B (en) | Intelligent city intelligent rail vehicle fault gene prediction method and system | |
| CN114510778A (en) | Track irregularity prediction method based on hybrid intelligent optimization LSTM | |
| CN116700296A (en) | Intelligent planning method and system for autonomous charging unmanned electric vehicle based on deep learning | |
| CN115907076A (en) | Mountain city charging load probability prediction system and method | |
| CN108510130B (en) | Intelligent unmanned vehicle limit mileage intelligent multi-source evaluation method and device | |
| CN109493449A (en) | A kind of lorry loading method for estimating state based on lorry GPS track data and high speed transaction data | |
| CN116523093A (en) | Grid demand sensing system and method of energy system based on random source load prediction | |
| Grubwinkler et al. | A modular and dynamic approach to predict the energy consumption of electric vehicles | |
| Kim et al. | A machine learning method for ev range prediction with updates on route information and traffic conditions | |
| CN114498632A (en) | Power distribution station load prediction method and device based on new energy and charging facility |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |