近年来，随着全球能源结构转型加速，页岩油作为非常规油气资源的重要组成部分，其高效开发成为保障国家能源安全的关键。长庆油田页岩油年产量占 全国页岩油总产量的50%以上，但受制于长水平井开发方式、原油物性、储层特征等综合因素，井下作业仍面临作业效率不高、效益管控不细、精准决策 不够等诸多挑战，直接影响着效益开发。 想要打开效益开发新局面，就必须发挥创新性思维助力产业结构优化，利用数智化技术推动业务模式重构、管理模式变革。长庆油田充分发挥产学研用一 体化生产主动性，联合专业数智公司，以昆仑大模型平台为底座，通过深度应用多模态数据与大模型算法，形成诊断专家知识库，打通井下作业"感知-诊 断-决策-执行-评价"的全链路，实现作业全过程化管控、全业务链规范、全标准化操作的功能模块自动化。 "该系统集成了井下压力、温度、流量等20多项参数动态感知，能'嗅'到油藏中流体的微妙变化。"长庆油田页岩油开发分公司技术与信息研究中心副主任 刘环宇说，该系统将"修井后分析"转变为"修井前预警"，井筒诊断响应速度从小时级缩短至秒级。 6月23日，在位于甘肃庆阳的中国石油长庆油田页岩油开发分公司生产指挥中心，技术人员张耀辉打开被称为"页岩 ...

Core Viewpoint - Ledong Robotics Co., Ltd. has submitted its prospectus for an IPO on the Hong Kong Stock Exchange, aiming to leverage its leading position in the intelligent robotics market, particularly in visual perception technology [2][12]. Business Overview - Founded in 2017, Ledong Robotics is a full-stack intelligent robotics company focused on developing and selling visual perception products and solutions [2]. - The company launched its first-generation laser radar and simultaneous localization and mapping algorithm in 2018, followed by the world's first consumer-grade Mini DTOF radar [2]. Market Position - By 2024, the number of intelligent robots equipped with Ledong's visual perception technology reached 6 million, with seven out of the top ten global home service robot brands being its clients, indicating a near-monopoly in the market [2]. - In 2024, Ledong Robotics achieved revenue of 467.3 million RMB, making it the largest company globally focused on visual perception technology in intelligent robotics [5]. Financial Performance - From 2022 to 2024, Ledong's revenue grew from 233.8 million RMB to 467.3 million RMB, with a compound annual growth rate (CAGR) exceeding 41% [5][6]. - Despite revenue growth, the company reported net losses of 73.1 million RMB, 68.5 million RMB, and 56.5 million RMB for the years 2022, 2023, and 2024, respectively [6][8]. - The gross profit margin has declined from 27.3% in 2022 to 19.5% in 2024, reflecting increasing cost pressures [8][9]. Product Development - In 2024, Ledong Robotics capitalized on the global smart lawn mower market, launching and mass-producing smart lawn mowers with sales exceeding 10,000 units [4]. - The second generation of smart lawn mowers, which integrates AI algorithms, was produced in 2025, with total sales surpassing 15,000 units from January to May 2024 [4]. Revenue Composition - The majority of Ledong's revenue comes from visual perception products, which accounted for 1.04 billion RMB, 1.67 billion RMB, and 3.41 billion RMB from 2022 to 2024 [6][7]. - Revenue from smart lawn mowers was 23.27 million RMB in 2024, representing approximately 5% of total revenue, indicating a potential second growth curve for the company [7]. Leadership and Investment - The company is led by founder Zhou Wei, a successful entrepreneur with a background in technology and innovation [10]. - Ledong Robotics has attracted significant investment from notable firms such as Huaye Tiancheng and Yuanjing Capital, with key shareholders holding 39.61% of the total issued share capital [11].

证券研究报告 计算机 DeepSeek GRM 带来新的推理 Scaling 路径 华泰研究 2025 年 5 月 06 日│中国内地 动态点评 近日，DeepSeek 团队的新论文《Inference-Time Scaling for Generalist Reward Modeling》发布，引入了一种自我原则点评调优（SPCT）的方法， 提出了通用奖励模型在推理阶段的新算法，验证了推理阶段的扩展策略在效 率与性能上的双重优势。同时基于此方法推出 DeepSeek GRM 模型，27B 的参数能跑出目前 R1 模型 671B 参数相当的性能。我们认为该方法带来了 推理阶段新的 Scaling 方法，或暗示了 DeepSeek R2 的技术路径，模型迭 代加速可期。 SPCT 通过生成的方法获得奖励信号，拓展了推理 Scaling 路径 目前，强化学习已广泛应用于 LLM 的大规模后训练阶段，现有的模型已经 证明采用合适的学习方法，可以实现推理的可拓展性。但强化学习过程中的 挑战在于在人工规则之外的一般领域难以获得 LLMs 的准确奖励信号，这决 定了强化学习的效果。DeepSeek 团队发现在奖励模型中采 ...