louisehourcade Chapter 104 Intelligent Star Network
Chapter 104 Intelligent Star Network
The delivery review was conducted at Blue Bay Communications headquarters.
Zuo Cheng led Tang Xu and Ma Hao to the conference room at Blue Bay. The review committee consisted of seven people: Lu Mingyuan, the Vice President of Technology at Blue Bay Communications, the Chief Engineer of the Satellite Business Unit, and several external communications experts. The lineup was stronger than the previous two reviews, indicating Blue Bay's high level of attention to this review.
"President Zuo, please begin your report." Lu Mingyuan chaired the meeting, his expression very formal.
Zuo Cheng opened the PPT, and the first page only had six words: Intelligent Star Network Scheduling.
"Distinguished reviewers, we have upgraded the inter-satellite link scheduling algorithm for the third phase of the Tianqiong satellite network from a fixed strategy to an AI scheduling scheme based on deep reinforcement learning," said Zuo Cheng. "The spectrum utilization rate has increased from 62% to 82%, exceeding the design target by two percentage points. This means that the available capacity of the Tianqiong satellite network has increased by one-third."
The meeting room fell silent. The figure of 82% clearly exceeded the expectations of everyone on the review committee.
The VP of Technology spoke first: "Mr. Zuo, what is the principle behind your AI scheduling solution? As far as I know, there are currently no mature cases worldwide of deep reinforcement learning being applied in the field of satellite communications."
"Our solution is indeed a world first," Zuo Cheng said. "The principle is to allow the AI agent to learn through trial and error in a simulation environment, and guide it to find the optimal spectrum allocation strategy through a reward mechanism. Compared with traditional fixed strategies, the AI solution can perceive changes in link status in real time and make scheduling decisions at the millisecond level."
"Where does the training data come from?" asked the chief engineer of the satellite division.
"We built a high-fidelity simulation environment using data from the first two phases of the Sky Dome project," Tang Xu added. "The error between the simulation environment and the real environment is controlled within five percent."
How well does the model generalize? If the number of satellites expands to over a thousand, will the algorithm still be applicable?
Zuo Cheng was prepared: "We conducted scalability tests. In a simulation environment with 600 satellites, the spectrum utilization rate remained above 78%. Larger-scale tests are underway, but preliminary results demonstrate the algorithm's good scalability."
An elderly professor with gray hair, one of the external experts, spoke up: "Mr. Zuo, the interpretability of deep reinforcement learning has always been a concern in the industry. If AI makes a scheduling decision, can you explain why it does it? If problems arise, how do you trace the cause?"
This is a very pointed question and also the biggest obstacle to the application of AI in the industrial field.
"Good question," Zuo Cheng said. "We implemented a two-layer design to solve this problem. First, all decisions made by the model are recorded in a log, including input states, output actions, and reward values, allowing for retrospective analysis. Second, we designed a safety constraint module. The AI's scheduling decisions must be made within preset safety boundaries; decisions exceeding these boundaries are automatically intercepted, reverting to traditional strategies. In other words, the AI always operates within the fences set by humans."
The old professor nodded, seemingly satisfied with the answer. Another expert next to him asked several questions about training stability, which Zuo Cheng answered one by one.
The review lasted three hours. The review committee raised more than a dozen questions about the technical solution, which Zuo Cheng and Tang Xu answered one by one. Ma Hao supplemented the data section, showing the convergence curve of the training process and the comparison results of different architectures; the data was very solid.
Finally, Lu Mingyuan announced the review results: "The third phase of the Sky Dome delivery review has received a technical score of 93 points, passing. Congratulations to 402."
93 points. Slightly lower than the first and second rounds by 1 point, but considering the introduction of a completely new AI solution this time, the judging committee gave higher bonus points for the technical difficulty. Overall, it is an excellent result.
Zuo Cheng stood up and shook hands with each member of the review committee. The VP of Technology, shaking his hand, said, "Mr. Zuo, your AI scheduling solution has truly opened my eyes. To be honest, Blue Bay Communications is also researching the possibility of AI scheduling, but our progress is far behind yours. If this solution can operate stably, I suggest that it be promoted to Blue Bay's other satellite projects."
"Thank you," Zuo Cheng said. "We are willing to cooperate with Lanwan Communications to conduct more in-depth verification."
Stepping out of the conference room, Tang Xu let out a long sigh of relief: "93 points. I was nervous for three hours straight, my palms were sweaty."
Ma Hao patted him on the shoulder: "When are you going to get rid of this sweaty palm habit?"
"Try speaking to seven judges for three hours and see if your palms sweat."
Zuo Cheng smiled, not joining their banter. He was thinking about something else. The VP of Technology mentioned extending the solution to other satellite projects, meaning the commercial value of the AI scheduling solution extended far beyond the Tianqiong project. Lanwan Communications is one of the largest satellite operators in China; if the 402 AI solution could be deployed across Lanwan's entire satellite fleet, it would be a sustainable source of revenue.
He took out his phone and sent a message to Han Lu: "Sky Dome passed the delivery review with a score of 93. Also, the VP of Blue Bay Technology mentioned that we want to extend the AI solution to other satellite projects. Prepare a business plan; we'll discuss it at the next weekly meeting."
Han Lu replied: "Received! 93 points is fantastic! I'll prepare the plan right away!"
Zuo Cheng put away his phone and walked out of the Blue Bay Communications building. The sun was shining brightly, and the autumn air in Hangzhou was as fresh as if it had been washed. He stood at the building entrance and took a deep breath.
AI, 402, has finally crossed this threshold. From communications to IoT to AI, every step has been perfectly timed. The technology tree silently guides from behind, the team executes diligently from the front, and he stands in the middle, connecting the forces of both. This is his role: an unseen conductor.
My phone vibrated again; it was a message from Yu Ying: "How did the judging go?"
"93 points, passed."
"I knew you could do it! We have to celebrate tonight!"
Zuo Cheng smiled and replied with a "good" followed by a heart emoji.
He looked up at the sky. Although he couldn't see the satellites, he knew that hundreds of kilometers above him, 480 satellites were orbiting at high speed, their links flashing with invisible light under the control of AI. A brand new era of communication was quietly beginning.
The spark that ignited in 402 has begun to spread like wildfire. And this is just the beginning.