Research on the Prediction of Energy-saving Potential of Traffic Tunnel Based on BP Neural Network

. Accurate grasp of underground building environmental temperature and humidity and cooling and heating loads is the basis for energy-saving analysis, ventilation and air conditioning system design and operation control. In order to come up with a method that can quickly and accurately predict the environmental temperature and humidity and energy-saving potential of underground power station traffic tunnel, this paper takes Xiangyou pumped storage power station traffic tunnel as the research object and designs a BP neural network model for traffic tunnel temperature and relative humidity prediction with respect to the main environmental influencing factors. Comparing with the test data, it is verified that the model has high prediction accuracy. At the same time, the prediction model was used to calculate the energy saving potential of traffic tunnel ventilation. The results showed that the relative error of predicting the energy saving potential of traffic tunnel in summer was less than 2% and in winter was less than 0.3%. This prediction method can obtain the traffic tunnel air temperature and humidity and ventilation energy saving potential simply and accurately, so that the design can be developed in the direction favorable to the energy saving of underground buildings.


Introduction
China's building energy consumption accounted for about a quarter of the total energy consumption, ranking first in energy consumption, among which the energy consumption of ventilation and air conditioning systems inside buildings accounted for about 40-50% of building energy consumption [1] . The plant of largescale hydropower projects were often built underground, and the internal heat and humidity environment was complex and changeable. It was necessary to maintain suitable environmental parameters with the help of air-conditioning equipment to improve the working efficiency of the mechanical and electrical equipment in the plant and ensured the health of the staff. In order to ensured that the thermal and humid environment in the plant meets the requirements, the air conditioning system often consumed a large amount of energy. Therefore, it was increasingly important to reduce the energy consumption of ventilation and air conditioning in underground power stations.
When the outside air passed through the underground traffic tunnel, it exchanged heat with the rock mass wall and was sent into the plant. The underground traffic tunnel rock mass taken away the heat in the air in summer and cools the air, and in winter, the air was heated. Most of the traditional studies do not consider the dynamic changes of outdoor air temperature, and the temperature of ventilation is usually set to a constant value during the study. In fact, the inlet air temperature of the traffic tunnel varied continuously with the fluctuation of the outdoor air temperature [2] . The treated air was sent directly into the plants to regulate the ambient temperature of each plant and reduce the load on the air conditioning units in the plants, thus achieving the purpose of energy saving. Therefore, the underground hydropower station had better energy-saving advantages of using corridor ventilation, and should be given full use in the design of the ventilation and air conditioning system of the underground plant of the power station [3] .
At present, there were two calculation methods for heat transfer in traffic tunnel ventilation, which were numerical simulation method and theoretical analysis method. The prediction of the heat and humidity exchange of air exiting a traffic tunnel using conventional methods was not ideal due to the extremely correlated nonlinear relationship between the temperature and relative humidity of the air and other parameters [4][5] . The BP neural network had a strong approximation function, learning and associative memory ability, which could better handle multiparameter nonlinear problems. The unconstrained nature of the input parameters of the BP neural network was used to avoid tedious mathematical analysis and to achieve multi-parameterization of the nonlinear regression [6] . The neural network prediction Fund Project The research is supported by Natural Science Basic Research Program of Shaanxi (Program No. 2022JQ-465) had been widely used in various fields, not only as an important tool in the study of building energy consumption, but also as a basis for energy-saving optimization of air conditioning systems. This paper combined with the actual situation of the traffic tunnel, using a large amount of actual measurement data, and fully studied the heat and humidity exchange capacity of air during ventilation in the traffic tunnel of pumped storage power station under the condition of dynamic change of outdoor temperature by BP neural network model.

BP Neural Network Model and Data
Pre-processing

Basic Principle of BP Neural Network
BP neural network is a forward multilayer network structure. Its learning process consists of forward propagation of the signal and backward propagation of the error [7] . In fig.1, L 1 , L 2 , and L 3 are the input, hidden, and output layers, respectively, X and Y are the network input and output vectors, W 1 is the weight matrix of the input and hidden layers, W 2 is the weight matrix of the hidden and output layers, f (z) is the activation function, which is generally a sigmoid function, a and A are the output vectors of the hidden and output layers, respectively, and the hidden and output layers are connected to each other by the purelin transfer function and according to the principle that the error moves in the decreasing direction [8][9] .

Neural Network Building
For the Xianyou pumped storage power plant with natural ventilation combined with mechanical exhaust, its traffic tunnel air heat and humidity exchange is affected by outdoor meteorological conditions and fluctuates with changes in external air parameters. The amount of heat and humidity exchange between outdoor air entering the traffic tunnel and the wall is influenced by a variety of factors and is closely related to many factors such as the temperature and humidity of the inlet air, the length of the traffic tunnel and the air volume inside the tunnel [10] . Therefore, the dry bulb temperature of the air at the entrance of the traffic https://doi.org/10.1051/e3sconf/202235602027 E3S Web of Conferences 356, 02027 (2022) ROOMVENT 2022 tunnel, the relative humidity of the air at the entrance, the air volume inside the tunnel, the temperature of the wall, the length of the traffic tunnel and the heat and humidity transfer coefficient of the rock at the wall are taken as the input quantities, and the air temperature at the exit of the traffic tunnel and the relative humidity of the air are taken as the output quantities.
For BP neural networks, the input and output quantities need to be data normalized in order to make the samples better learned by the network model and to prevent the phenomenon of data overflow [16]. The measured data are converted to values in the range [-1,1] by normalizing the mapminmax function. The number of hidden layer nodes l is determined according to the formula l=2N+1, where N is the number of input layer nodes. In this paper, the number of hidden layer nodes is selected as 15. Set the network initialization parameters, the learning rate is 0.01, and the maximum error of the designed network is 1e-5.      Fig.6 shows the comparison between the predicted and actual values of air temperature and humidity at the exit of the traffic tunnel in summer based on the BP neural network model after the above training. As shown in the figure, the predicted value of air temperature at the exit of the traffic tunnel fluctuates gently at the beginning and end time periods, and the measured air relative humidity curve suddenly shows two spikes near the time points of 9:30 and 14:30. This is because the air relative humidity inside the traffic tunnel is subject to the transient influence of many factors, such as wall humidity, fogging inside the https://doi.org/10.1051/e3sconf/202235602027 E3S Web of Conferences 356, 02027 (2022) ROOMVENT 2022 tunnel and the monitoring equipment itself. The predicted values are not affected by such occasional spikes and depressions, and the overall fluctuation of the exit air relative humidity is relatively flat. The predicted values show the correct trend of the exit air relative humidity of the traffic tunnel, and the predicted results are more stable. The comparison between the predicted and actual summer temperature and humidity values shows that the errors are within the allowed range, the relative humidity prediction results have small errors, the predicted values have smooth transitions, and the overall fluctuations are smooth. Fig.7 shows the comparison between the predicted and actual values of air temperature and humidity at the exit of the traffic tunnel in winter based on the BP neural network model after the above training. The predicted air temperature at the exit of the traffic tunnel fluctuates greatly, and the prediction accuracy has decreased. The overall curve of the measured air relative humidity at the exit of the traffic tunnel is flat, and its accuracy is not affected by sporadic deviations. By comparing the predicted and actual values of winter temperature and humidity, it can be seen that the predicted results fluctuate in local time periods, but the overall development trend is consistent with the actual measured values, and the errors arising from the prediction are within the allowable range. In general, the predicted results show the correct trend of air temperature and humidity at the exit of the traffic tunnel, the accuracy of the prediction is satisfactory, and the prediction model of temperature and humidity at the exit of the traffic tunnel in summer and winter can be considered to have good prediction ability.

Analysis of Energy-saving Potential of Traffic Tunnel
In the process of external air being sent to the underground power plant through the traffic tunnel, the heat exchange between the air and the wall of the traffic tunnel is the change in the internal energy of the air, and the formula for the heat exchange per unit time of air flow through the traffic tunnel is [11] , , Where, q represents the heat exchange between the wall of the traffic tunnel and the air per unit time, kw. c represents the specific heat of air, kJ/(kg/౯). A represents the cross-sectional area of the cavity, m 2 ; v represents the air flow rate inside the tunnel, m/s; ȡ p represents the average value of air density in the import and export, kg/m 3 ; t w,Ĳ represents the air temperature at the entrance of the traffic tunnel, ౯; t o,Ĳ represents the air temperature at the exit of the traffic tunnel, ౯.  Table 1 shows the predicted results of the energy saving potential. In the summer working condition of the traffic tunnel, the power station does not use the mechanical exhaust air method, but uses the top arch air conditioning air supply to regulate the air in the plant, so the wind speed in the traffic tunnel is low and its energy saving potential is limited. The energy saving potential was calculated based on the average wind speed (0.06m/s average wind speed throughout the day in summer). The actual heat exchange of the traffic tunnel in summer is 5.916×10 6 KJ based on the above formula for calculating the heat exchange throughout the day, and the heat exchange based on the predicted value of the BP neural network model is 5.815×10 6 KJ. In the winter conditions of the traffic tunnel, the power station uses natural air inlet combined with mechanical exhaust, and the wind speed in the traffic tunnel is high, and its energy-saving potential is large. The energy saving potential was calculated based on the average wind speed (1.04m/s average wind speed throughout the day in winter). The actual heat exchange of the traffic tunnel in winter is 5.258×10 7 KJ based on the above formula for calculating the heat exchange throughout the day, and the heat exchange based on the predicted value of the BP neural network model is 5.265×10 7 KJ. The cooling energy efficiency ratio of the air conditioning system of the plant of Xianyou Pumped Storage Power Station is 3.2 on average for chillers with Grade 2 energy efficiency ratio. It uses a traffic tunnel to pre-treat the incoming outdoor air, and if the total heat exchange of its incoming air throughout the day is provided by a secondary energy efficiency ratio chiller. The actual savings of about 513.52 kW·h in 24h in summer are calculated by the predicted value of heat exchange, and the expected savings of about 504.81 kW·h in 24h in summer. In winter, the actual saving of 24h is about 4564.18 kW·h. By the predicted value, the saving of 24h in winter is about 4570.22 kW·h. Comparing the actual and predicted values of heat exchange between traffic tunnel and outdoor air in winter and summer, the relative error of prediction in summer does not exceed 2%, and the relative error of prediction in winter does not exceed 0.3%, which is relatively minor. In summary, the heat exchange per unit time between the traffic tunnel and outdoor air intake in summer and winter can be calculated as well as predicted based on the above-mentioned BP neural network prediction model, which provides reference for the design and operation management of air conditioning systems in actual projects and brings great economic benefits to the initial investment and later operation management of engineering construction.

Conclusion
The following conclusions can be drawn from the field measurement of the ventilation effect of the traffic tunnel of Xianyou Pumped Storage Power Station and the analysis and study of its energy-saving potential using the method of BP neural network prediction (1)Based on the BP neural network traffic tunnel temperature and humidity prediction model, the average relative error of predicted temperature in summer is 1.06% and humidity is 1.74%. The average relative error of temperature prediction in winter is 2.47% and humidity is 3.71%. The model can predict the time by time changes of ambient temperature and humidity in the traffic tunnel of underground power station more accurately.
(2)Based on the results of BP neural network prediction, the energy saving potential of the traffic tunnel ventilation of Xiangyou pumped storage power station is calculated, and the total heat exchange in the traffic tunnel in summer is about 5.815×10 6 KJ/day, which can save about 504.81 kW·h /day of air conditioning. The total heat exchange in winter is about 5.265×10 7 KJ/day, and the power consumption of air conditioning can be saved about 4570.22 kW·h /day. The relative error of the prediction in summer does not exceed 2%, and the relative error of the prediction in winter does not exceed 0.3%.
(3)The traffic tunnel air temperature and humidity predicted by using BP neural network model can be used to predict the energy-saving potential of underground power station corridors in the future, providing a new idea for the energy-saving optimization design of their ventilation and air conditioning systems, which has a wide application prospect.