Energy optimization method for connected vehicles on a cloud database Flah aymen Photovoltaic, wind and geothermal unit National school of engineering of Gabès, Tunisia University of Gabès, Tunisia firstname.lastname@example.org Chokri MAHMOUDI Photovoltaic, wind and geothermal unit National School of Engineering of Gabès, University Of Gabès, Tunisia email@example.com Abstract— In this paper, we expose an energy optimization method for an ensemble of electrical vehicles. The idea is based on vehicles information sharing in a cloud database. Installed sensors in a vehicle will allow us to know the needed vehicle information. The regrouped data will be used then for building an optimal energy experience which will be shared with the new vehicles. The neural network technique is used here for the learning phases. The proposed application for this work was validated for five connected cars. Matlab Simulink was used for simulating the results and given the present statistics. Keywords—Energy, Hybrid electrical vehicles, power management, Matlab, neural network, Cloud database, optimization. I. INTRODUCTION Energy is the needed factor for any physical system for continuing life in this word. Human or any physical system can't proceed without this issue. We can't move or do anything without energy. Our building physical systems can't also move or work without energy. It's essential to find the best solution for managing the existing energy in the specific situation. Try moving from a point to another destination, without knowing your energy resources and without estimating your energy needs. It will be hardness and difficult. Also using a vehicle based on any resource of energy can't be flexible if we can't calculate or estimate our needs. For a long time, engineers and researchers search to build a multi-energy source system for helping users face the energy sources dissipation. Photovoltaic and other renewable energy were used in numerous applications, according to this objective. We can see in  and in  that researchers search to build a multi-source of energy for this application. Using a multi-source of energy will be benefited if we can manage this quantity of available power. In this context, this paper exposes a problem and a solution of the power management system inside an important electrical field, which contains at least two sources of energy, the combustion, and the electrical energy. Which are inside the hybrid electric vehicle. The internal combustion engine will use the combustion energy for doing its touch and the installed Lassaad SBITA Photovoltaic, wind and geothermal unit Director National School of Engineering of Gabès, University of Gabès, TUNISIA firstname.lastname@example.org electrical motor will also use the electricity for running. Here, we can find several version of the hybrid vehicle as the serial and the parallel electrical vehicle. This is depending on the placement of the engines. In our previous work cited in , we have exposed those versions. Different forms of power management algorithms and principle were exposed to various cited work as ,  and our last work related to this field is exposed as . Basing on the last statistics appeared in 2018 we can conclude that it is possible that we have more than 22000 electrical vehicles on the road in the same city. We can see the example of Canada in this link . If each vehicle has its conductor philosophy, its positions, its own internal parameters and its specific condition. This can be a very important database about this vehicle and we can benefit from this stored information. The idea is based on the vehicle's information share. Each vehicle can allocate its own experience for a specific condition with another vehicle or with an ensemble of the vehicle. Using a Cloud database, we can store all those information, we can build a learning algorithm basing on an optimization technique for helping other vehicles defining rapidly the needed decision and the best performance for minimizing the energy losses due to an unknowing information. As we know that approximately, in all the latest versions of cars, intelligent sensors were used and installed inside. Some systems are related to the facing detection, where we can be used for driver fatigue system. This system can detect the driver mode if it is happy, angry or normal etc.… Same sensors used in Lane Keeping Assist System (LKAS)  to oblige driver to keep hands on the steering wheel, can provide heartbeats rate and determine if the driver is nervous. Traditional vehicle sensors as speed sensors, fuel sensors, battery sensors, etc.… will be useful for knowing the energy state of the car. GPS is also installed in all the new cars; this will be a benefit for detecting the vehicle position. More details, about this part, were exposed in our previous work cited in . So in this paper, we will try to regroup all those sensors information for building a vehicle database, which will be utilized, in a second stage, in a neural-cloud database for building an optimal energy experience for the corresponding specific condition. vehicle connected to the cloud database. Each vehicle will p g expose their information as it is described in figure (2). So, the objective of this paper will be attempted after a general introduction and after showing the essential components of a hybrid vehicle model. Where we explained before, the needed database architecture and its components, then we will expose the communication protocol for the cars and the existed database. The principle of neural network learning algorithm inside the cloud database will be then explained and the obtained simulated results, for a tangible prototype example, will be discussed. We conclude this paper with a principal conclusion. II. HYBRID AND TOTAL ELECTRICAL VEHICLE COMPONENTS Electrical vehicles were developed on two models; Totally electrical vehicle (TEV) and hybrid electric vehicle (HEV). At least we will find an electrical motor, which will be used as the main active motor or a reserve active motor. This is possible in the hybrid version . The system of battery and the electronic power system is present in two versions of cars. The fuel tank and the ICE are present only on the hybrid version. The position of an electrical motor with the hybrid one gives the nomination of series or parallel hybrid electrical vehicle . Figure (1), indicates the relation between those blocs for a hybrid and a pure electric vehicle model. Fig. 2. Information on vehicle shared in the cloud database Knowing the GPS position of the vehicle will allow us to know the vehicle slope and then the possible needed. Using the face detection system will help to see the conductor mode of the vehicle. Also knowing the batteries and the fuel status will permit us knowing if the hybrid mode must be launched or not. The vehicle weight is equally important that knowing the applied torque on the machine in the TEV architecture. So each information will be useful for optimizing the power consumption. IV. COMMUNICATION PROTOCOLS FOR VEHICLES AND THE CLOUD DATABASE Actually, we can't have a database if we don't have vehicles. The database doesn't exist without a vehicle for shared information. But, if the database exists, this is indicating that we can find at least one vehicle connected to this base. However, it is possible for some other vehicles to join this community. This is must be checked by the driver from the beginning. Then each new vehicle must give responses to numerous requests. More description of this point is explained as follows: Fig. 1. TEV and HEV architecture The hybrid nomination will be valid if, at least, two energy sources are used. It is possible to use more than two sources, we can enlarge also with the renewable energy, as the solar energy. More than work was presented on this subject as presented in  and . This is will help the system autonomy. Basing on the conventional electric architecture we will develop in the coming parts, the proposed energy management architecture. III. ESSENTIAL COMPONENTS FOR BUILDING THE NEEDED DATABASE As we have exposed in the introduction section, each vehicle, needs an optimal energy experience, must share in the corresponding cloud database some needed information. Those data will be regrouped and classified from each connected vehicle for building the energy experience database. Actually, it's hard to build a perfect archive. Because we can't find in the same zone, the similar vehicles models with the same installed sensors and the same architecture. Therefore, for simplifying the application we will suppose in this first version of this work, that all the connected vehicles have the exact same characteristics. So, for starting creating a database, we must have more than If the original vehicle is fresh for this database, the system will ask firstly the car model, if it is HEV or TEV, the vehicle position, and the target trajectory. It will classify the car as the corresponding category. Inside, the database, the neural network program will compare the vehicle's information faces the other existed info. Two scenarios can be founded, if the vehicle condition has a similarity in the database, the algorithm will choose, from the existed, the optimal one. Else if the sent information is novel, the system will adjust the database, according to this vehicle category, and a new learning phase will be started. Then, a novel optimal energy condition will have existed. This vehicle will preserve its energy experience until the system gives a different optimal condition. For the HEV model, the car will be asked to give its energy experience as fuel and electricity consumption form and the level of its stored energy. Also, the vehicle speed will be asked. All of that information must be accessible to 300Km. Else if the car is classified in the TEV model, the system will ask, basically, the electrical power consumed, the vehicle speed and acceleration during the last 100km. If one of that essential information is absent the car does not have an order to join the database. The presented figure (3), explains this protocol. algorithm will start again. This is the step C. Else; step B will be executed. Vehicles who built the database HEV TEV Electrical Power Batterie Statuse EX1 Electrical Power Mechanical Power Fuel consumed Batterie Statuse EX2 EX3 EX4 New vehicle EX5 Step A Database with neural network Algorithm Optimal experience Fig. 3. TEV and HEV communication with the Neural network database Step B Learning from vehicles 1 TO 4 Sent the existed optimal energie experience for the vehicle 5 Optimal energy exepreince O.EX V. THE LEARNING ALGORITHM DESCRIPTION As we have multi-input information in the database, from multi-connected vehicles, the system will organize those parameters into several vectors input. The number of input will configure the neural network. For the neural network output, this is linked to the car model, if it is TEV or HEV. In this work, we will expose, only, the neural architecture of a TEV. TABLE I. NEURAL NETWORK ARCHITECTURE FOR TEV Vehicle Model Input Layer Number of layers Output Layer Learning function TEV 5 2 1 Sigmoid The number of neurons in the input layer depends on the number of attached vectors. As we have 5 parameters from the TEV and 7 from the HEV, the number of neurons was selected as it is exposed in the previous table. In the output layer, we have to fix one neuron for the TEV, because we have an alone energy source. However, two sources of energy exist in the HEV model. Then the references of needed power, which is equal to the optimal energy experience, will be injected into the vehicle. Algorithm Adding the Vehicle 5 data to the database and adjust the energy experinece Step C Fig. 4. Description for the algorithm function In the rest of this part, we will expose some simulated data from four total electrical vehicles. The results for the hybrid model will be the object of a future work. The stored data will be utilized in the learning algorithm for having the best energy experience. In figure (6), we expose the battery outputted power from four different cars, according to the acceleration form given in figure (5) and related to a distinctive exterior climate. We suppose that the weight of the cars is the same and it is equivalent to 1200Kg, the vehicle trajectory is also the same and it is fixed at 500 meters. VI. PROTOTYPE DESCRIPTION AND STATISTICS We will attempt to explain the running algorithm using five connected cars, four of them who construct the database and the fifth one will be a guest, which will ask an optimal energy experience. We suppose that all the cars are placed on the same trajectory. According to the figure (4), if the car number five is not available, the learning algorithm will use the data of the four existed cars. This is the step A. The algorithm will learn from those cars and it will try to generate the optimal energy experience (EX). When the fifth vehicle is present and requests an optimal energy experience, better than the existing, two possible happens. If existed energy experience is better, the vehicle data will be added to the database and the learning Fig. 5. Acceleration form for the four existing Cars The neural network algorithm will take those data and it will try to learn about this vehicle type comportment for a given acceleration as it is shown in figure (5). The learning phase was elaborated for 1000 iterations, which is corresponding to the minimum possible learning error and a reasonable time execution. The neural architecture is built on a back-propagation learning algorithm. circuit (SOCF). Also, we attach the positive acceleration (P.acce) time for each car according to the maximum acceleration slop (Acc.Slop) during this period. By comparing the case of the first and the fifth car, we can see that the results related to the fifth card are better. We have saved 0.2% of power antonym, in this case, face the case where the vehicle five construct its own energy form. TABLE II. STATISTICS RELATED TO THE CONSUMED POWER FOR THE FIVE CARS Fig. 6. Battery power for the four existing Cars, (a) is related to the EX1, (b) is related to EX2, (c) is related to EX3, (d) is related to EX4 After having the corresponding neural bloc, the overall algorithm will be ready. It is easy now for any vehicle who asks an optimal energy experience that has its need. In figure (7) and (8), we expose the results related to the guest vehicle. Figure (7) is linked to the fifth vehicle acceleration form. Figure (8) is related to the power battery for this car. Car 1 2 3 4 5 SOCI 100% 100% 100% 100% 100% SOCF 99.3% 98% 99.5% 99% 99,2% P.acce 70 sec 50 sec 100 sec 80 sec 68 sec Acc.slop 14% 50% 12% 31% 30% VII. CONCLUSION As it is outlined in this paper, a smart learning database basing on neural network algorithm was used in a power management problem related to the electric vehicle. The vehicles here are related to a cloud database and classified into two categories. The proposed method will use some essential information, associated with the car and driver. For detecting the best and the optimal energy experience, which will be used then by the guest cars. This optimal energy experience will be a benefit for reducing losses power related to electricity inside the electric vehicle. The given results and figures, account for this principle of this application and prove its efficiency. VIII. REFERENCES  Fig. 7. Acceleration form for the guest Cars     Fig. 8. Battery outputted power for the guest Car In the table (2) we expose the related statistics for the five cars depending on the consumed energy. The statistics were simulated for only 500 meters. We expose for each vehicle the battery SOC from the beginning (SOC I ) and at the end of the  A. R. Bhatti, Z. Salam, M. J. B. A. Aziz, K. P. Yee, and R. H. Ashique, “Electric vehicles charging using photovoltaic: Status and technological review,” Renew. Sustain. Energy Rev., vol. 54, pp. 34–47, 2016. A. Romero, M. Carvalho, and D. L. Millar, “Optimal Design and Control of Wind-Diesel Hybrid Energy Systems for Remote Arctic Mines,” J. Energy Resour. Technol., vol. 138, no. 6, pp. 62004–62010, Jun. 2016. C. Mahmoudi, A. Flah, and S. Lassaad, “An overview of electric Vehicle concept and power management strategies,” in International Conference on Electrical Sciences and Technologies in Maghreb, 2014, p. 1—8. J. P. Trovão, P. G. Pereirinha, H. M. Jorge, and C. H. Antunes, “A multi-level energy management system for multi-source electric vehicles - An integrated rulebased meta-heuristic approach,” Appl. Energy, vol. 105, pp. 304–318, 2013. J. Moreno, M. E. Ortuzar, and J. W. Dixon, “Energymanagement system for a hybrid electric vehicle, using ultracapacitors and neural networks,” IEEE Trans. Ind. Electron., vol. 53, no. 2, pp. 614–623, 2006. A. Flah, “Internal Fuzzy Hybrid Charger System for a Hybrid Electrical Vehicle,” J. Energy Resour. Technol., vol. 140, no. 1, pp. 12003–12008, Aug. 2017.      Si.-P. Rioux, “Statistiques on electical vehicle in Canada,” 2018. [Online]. Available: https://www.aveq.ca/actualiteacutes/category/statistiqu es. Mahmoudi, Chokri, A. Flah, and L. Sbita, “Novel concept of Power Management Architecture based on Smart EV Learning DataBase,” in Recent Advances on Electroscience and Computers, 2015, pp. 191–196. C. Mahmoudi, A. Flah, and L. Sbita, “Prototype design of a compact plug-in solar electric vehicle application for smart power management architecture,” in International Conference on Green Energy and Conversion Systems, GECS 2017, 2017. A. Rassõlkin, “An Overview of Electrical Vehicle and Hybrid Electrical Vehicle Drives,” in 13th International Symposium “Topical problems in the field of electrical and power engineering. Doctoral school of energy and geotechnology II,” 2013, pp. 76– 80. D. B. Richardson, “Electric vehicles and the electric grid: A review of modeling approaches, Impacts, and renewable energy integration,” Renew. Sustain. Energy Rev., vol. 19, pp. 247–254, 2013.