應用機器學習演算法設計具非等向熱傳性質電池儲能系統之熱管理策略

Abstract

本研究探討電池模組的非等向熱傳導性質對充放電過程中電池溫度分布之影響。設定體發熱模擬方法，將電池模組視為均勻的非等向性材料，計算貨櫃型電池儲能系統的電池溫度分布。提出平行吹拂式冷卻系統之創新設計，強化電池容易散熱表面的散熱強度，使散熱效果評斷參數大幅增加76.8 %。利用機器學習演算法，建立冷卻氣體控制策略及電池溫度預測模型，降低電池最高溫度及最大溫差；且電池溫度預測誤差僅0.16 %，顯著提升儲能貨櫃熱管理系統的安全性能及商業價值。 商用電池儲能貨櫃所使用的電池模組多為大型層壓式鋰離子電池，內部熱傳性質各方向相異，需合併分析各層材料的熱傳現象來模擬溫度變化過程。若不考慮電池內部熱傳，熱能無法由內部傳遞至外界，表面溫度受冷卻氣體流場主導，產生顯著溫差。本文提出體發熱模擬方法，將電池視為均勻的非等向熱傳材料。結果顯示，電池溫度受內部熱傳性質影響顯著，高溫區主要出現在熱傳導係數高方向的表面。若增加流經此方向表面的冷卻氣體，可有效提升散熱效果。 第二部份考量具非等向熱傳性質的電池模組，設計平行吹拂式冷卻系統，調整內裝配置，使冷卻氣體可均勻流經電池熱傳導係數高方向的表面，並讓迴流氣體加強主要熱點的散熱。結果顯示，平行吹拂式冷卻系統在冷卻氣體流量為6 m3/s、溫度為25 ℃的情況下，與FS-CR冷卻系統相比，最高溫度和最大溫差分別下降11.3 %和28.7 %，評斷參數Ind提升76.8 %，儲能貨櫃的散熱效果大幅提升。 第三部份運用機器學習演算法，最佳化電池儲能系統熱管理策略。使用決策樹演算法，分析冷卻氣體條件對電池溫度分布之影響。結果得知，降低冷卻氣體溫度和提升流量，可分別有效改善電池模組的最高溫度和最大溫差。也建立人工神經網絡模型，預測電池在不同冷卻氣體條件下的溫度。經驗證誤差僅0.16 %，預測時間僅14秒，節省99.9 %分析時間，大幅提升電池儲能系統熱管理策略的商業價值。

In this study, we investigate the influence of the anisotropic thermal properties on the temperature distribution within container-type battery energy storage systems (BESS). The volumetric heating method is employed for simulating the heating process, under the assumption that the battery modules are consisting of single materials with anisotropic thermal parameters. Based on the findings, an innovative design of the parallel-cooling system is introduced for the battery thermal management system (BTMS). The system enhances the cooling airflow around the surfaces in the directions with higher thermal conductivity. With the implement of the parallel-cooling system, the performance index of the thermal management system can be significantly increased by 76.8 %. In addition, with the utilization of two machine learning algorithms, decision tree and artificial neural network, we proposed the thermal management strategies and the temperature predictive model for the BESS. The predictive model achieves an impressive error of only 0.16 %, significantly improving the safety and the commercial profitability of the BTMS. Commercial BESS predominantly use large-format laminated lithium-ion batteries with anisotropic thermal conductivities. The internal thermal properties of these batteries are various with the directions. In this study, the volumetric heating method is applied, simulating the heat transfer process within the BESS by treating the battery as a homogenous material with anisotropic thermal conductivities. The results reveal that the temperature distribution is mainly influenced by the internal anisotropic thermal properties, while the hot spots primarily occurs on the surfaces with high thermal conductivity. By enhancing the cooling airflow around these specific surfaces, the cooling performance of the BTMS can be effectively improved. Considering the temperature distribution of the batteries with anisotropic thermal properties, the parallel-cooling system is proposed. In the system, batteries are arranged in series and cooling air flows parallelly to them. This configuration intensifies cooling air flow through the surfaces with high thermal conductivities. In addition, the recirculating cooling air enhances the heat dissipation of the hotspots in the BESS. With the parallel cooling system, the highest temperature and the temperature difference across the entire BESS are reduced by 11.3 % and 28.7 %, respectively. The performance index of the BTMS is effectively increased by 76.8 %, resulting in a significant improvement in the safety of the BESS. Two machine learning algorithms are implemented to establish the thermal management of BESS. The decision tree algorithm clarifies the relationship between the cooling air conditions and the battery temperature distributions. Based on the results, we introduced the cooling strategy that the highest temperature and the maximum temperature difference of the batteries can be effectively decreased by reducing the temperature and increasing the flow rate of the cooling air, respectively. Furthermore, the artificial neural network is utilized to forecast battery temperatures under various cooling air conditions. The model demonstrates an impressive error of only 0.16 %, accurately predicting the battery temperatures. Moreover, regardless of the amount of data, the prediction time takes just 14 seconds, significantly reducing the analysis time by 99.9 %. These advancements remarkably enhance the commercial value of the BTMS.