ENERES Co., Ltd.
KDDI AGILE DEVELOPMENT CENTER CORPORATION

Eneres Co., Ltd. and KDDI AGILE DEVELOPMENT CENTER CORPORATION(hereinafter KAG) began a joint demonstration in mid-May 2025 to incorporate HCAI (Human-Centered AI) into anomaly detection systems for solar power generation forecasting.
The project aims to realize AI systems that can be used with confidence by continuously improving AI accuracy through a cycle where AI-derived results are analyzed and evaluated by another AI, which is then evaluated by humans who provide improvement instructions.
In this demonstration, KAG provides knowledge and technology for generative AI and HCAI, Eneres implements generative AI and HCAI development, and both companies jointly conduct system construction.

As “generative AI” rapidly penetrates business, Eneres has actively incorporated AI into its supply-demand management operations and power generation forecasting. One example is the prediction AI “anomaly detection system” that detects “anomalies” in solar power generation forecast values and “deviations from actual values,” then uses generative AI to report, analyze, and propose solutions.
However, generative AI has challenges such as “hallucination” where it generates plausible misinformation, and “unclear reasoning for conclusions making responsibility for judgments ambiguous.” To utilize generative AI in the energy field that supports social infrastructure, these challenges must be resolved.

To address these challenges, Eneres and KAG are jointly working to build a system that incorporates another generative AI into the process of objectively evaluating “how accurate and reliable” AI output results are, with humans ultimately evaluating and judging the evaluation results returned and providing feedback to each^[1] AI.
KAG handles software development and provision within the KDDI Group and is a DX-specialized organization that has continued to thoroughly commit to “agile development”[1]. It possesses leading knowledge in Japan in the generative AI and HCAI fields. KAG excels at providing value using the latest trending technologies such as LLMOps^[2] and AI agent product applications.

Eneres’ founding business is power supply-demand management, and the company has worked on advancing power generation and demand forecasting using prediction AI to support power generation operators from an early stage.
While this demonstration focuses on solar power generation forecasting, we aim to expand HCAI to other fields in the future.

Through this demonstration, Eneres and KAG will contribute to realizing a rich future society where humans and AI coexist.

■Demonstration Overview

Purpose	Verification of HCAI system effectiveness for anomaly detection in solar power generation forecasting
Duration	Mid-May 2025 – End of the year
Roles	– Eneres – Provide knowledge on energy management, provide solar power generation anomaly detection logic, develop and implement analysis generative AI system, provide and evaluate demonstration field experiments – KAG – Provide knowledge and technology on HCAI, develop and implement evaluation generative AI system, develop AI evaluation and visualization technology, examine AI agent technology applications, system-wide coordination and optimization
Expected Effects	・Improve prediction AI forecast accuracy ・Speed up and improve efficiency of anomaly cause investigation ・Promote utilization by improving generative AI reliability ・Economic benefits such as avoiding imbalance charges in power generation forecasting operations

Roles of Each AI and Humans in the Demonstration

Action	Entity	Roles in This Demonstration and Overview
Prediction	AI	Prediction AI predicts power generation based on data such as solar power plant generation records, weather data, and equipment information. Systems that Eneres has already developed and is operating in practice will be used.
Analysis	Generative AI	Analysis generative AI comprehensively analyzes the values forecast by the prediction AI and actual values by comparing them with past cases and related data, investigates possible causes, and generates and proposes specific countermeasures. Mainly developed by Eneres.
Evaluation	Generative AI	Evaluation generative AI evaluates the content proposed by the analysis generative AI (basis for anomaly detection, validity of cause analysis, consistency of proposals, etc.) from multiple perspectives including past data, consistency, and possibility of hallucination. Mainly developed by KAG.
Judgment	Humans/Experts	Confirm the evaluation results from the evaluation generative AI and the thought processes of the analysis generative AI through easily understandable visualization based on scoring. From this, make situational judgments, identify causes, and provide countermeasure instructions. Human feedback is accumulated as data and used for continuous improvement of analysis generative AI accuracy.

---

[1] Agile development: A development method that completes software by repeating design, development, and testing for each function over short periods of 1-2 weeks. Even when changes occur in scope or user needs during development, priorities can be flexibly changed to handle the changes.

[2] LLMOps (Large Language Model Operations): A collective term for practices and workflows to streamline the development, operation, and management of applications using large language models (LLMs). Refers to systems, infrastructure, and methods for smoothly advancing the entire LLM lifecycle, including data preparation, model fine-tuning, deployment, and monitoring.