Schedule Logo

Poster Session I

4:30 pm – 6:30 pm, Tuesday October 14 Session DT4 COEX, Lobby E
Topics:
PrevNext

Application of Value Iterative Processes in Reinforcement Learning: Feature Analysis of Reaction Networks in Silane Thermal Decomposition and Silane Plasma

Poster 60
Presenter: Issei Yonemoto (The University of Shiga Prefecture)
Authors: Kota Hamano (The University of Shiga Prefecture), Taku Iguchi (The University of Shiga Prefecture), Shigeyuki Miyagi (The University of Shiga Prefecture), Osamu Sakai (The University of Shiga Prefecture)

A new method to quantitatively evaluate the impact of target species on the overall reaction network is proposed, focusing on the fact that the value iterative process of reinforcement learning yields a feature of the network structure. For thermal decomposition of silane, which consists of 109 species and 767 reaction pathways identified from the literature, the particle densities and reaction rate quantities were calculated by numerical simulation, and a weighted network model was constructed in which these reaction rate quantities were used as edge weights. By applying the value iterative process to this model, the return distribution of each species was derived, and principal component analysis and clustering using the k-means method were performed to classify the species into five groups. The results showed that the principal component score strongly correlates with the weighted degree. The clustering analysis revealed that features such as the proximity of the reaction pathway to the target species and its association with the presence or absence of dangling bonds were evaluated. Furthermore, a reaction network for silane plasma was also created based on the work of Hamano et al. [1], and compared with this thermal decomposition reaction network of silane. Comparing the degree distributions of both networks, we found that the thermal decomposition of silane has the characteristics of a random network, while the degree distribution of silane plasma is similar to that of a scale-free network.

Funding acknowledgement

This study is partially supported by Grant-in-Aid for Scientific Research from the Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT/JSPS KAKENHI) with Grant Nos. JP24K21198 and JP24H00036.

POSTERS (97)