Carbon emission is a top concern of people around the world. Countries and corporations are all motivated to reduce Carbon emissions. Power saving has long been a tough challenge since 2G era and becoming even tougher with upcoming 5G, as 5G brings up massive MIMO and large output power requirements, power consumption of 5G based station can be up to three times of 4G base stations, while over 70% of the total power consumption comes from RAN. Take China’s mobile network as an example, over four million radio sites are in service, with 80 million tons of carbon emission per year, accounting for nearly 1% of China’s annual emission.
The traditional energy-saving scheme lacks effective coordination within 2G/3G/4G/5G and between adjacent sites, and lacks timely adaptive energy-saving strategy according to network traffic load, which sometimes results in negative effect on network performance. On average, only 5% power would be saved with traditional energy saving solution, which is far from enough for operators to reduce OPEX. Therefore a smarter and more efficient solution is in desperate need to address the fore-mentioned and ever-increasing challenges of climate change. That’s why ZTE’s AI-based power saving solution comes into play, by leveraging the latest development of AI, big data and radio access technologies.
Intelligent network Q&M became reality through ZTE AI Engine Platform
For more sustainable development, operators and equipment manufacturers have launched energy saving and carbon emission reduction plans. However, as Shandong consumes the most power in China and the number of thermal power plants outrun all the other provinces, China Unicom Shandong branch feels the urge to go further on reducing carbon emissions with the advantages of AI.
With the help of three AI capabilities (data perception capability, AI analysis capability, and intent insight capability), AI accelerators are introduced to provide rapid AI training for intelligent network equipment operation and maintenance. Network load prediction, strategy adjustment and optimization, real-time KPI & performance monitoring would form a closed loop in AI based power saving solution, which is helpful for operators to find a balance between power saving and network performance with much less CAPEX.
The solution can intelligently self-identifies the real customer usage to reduce the power emission, with the precise insight of network behaviors as well as customer behaviors.
Successful Commercialization with Excellent Results
ZTE and China Unicom have deployed AI-based intelligent power saving in Shandong Province, China since June, 2019. After three phases’ commercial trial, it was verified that effective activation time of power saving has been increased by 150% to 300%, around 2.5 times power has been saved compared to that of non-AI power saving solutions.
- With rapid AI learning capability, it only takes one week to achieve the balance of power saving and network performance with AI-based power saving solution, which usually costs half a year through conventional power saving solutions, significantly reduces labor cost.
- More than 21,000Kwh per week have been saved for the whole network (10,000+ cells) with AI-based power saving solution, which translates into 8% to 15% power saving in average and great labor-cost saving, without significant degradation of network performance. In rural area, up to 4 times power saving has been achieved.
If 60% of all the 4 million sites in China enable AI-based power saving, $1.1 billion USD and nearly 50 billion tons of carbon emissions would be saved each year.
China Unicom Shandong Branch is delighted with both the live network performance and the power saving results, suggesting further deployment in more locations.
Maximize power saving with less impact on network performance
The innovation is based on big data analysis of historic and real time traffic and AI self- learning, with the accumulated knowledge of power saving solution experiences since 2G era, including valuable and comprehensive schemes and algorithms verified in live networks in the last ten years. The power saving management granularity is refined to cell level, matching differentiated traffic models based on AI self-learning and accurate traffic prediction. Power saving optimization is conducted in a timely and self-adaptable manner to respond to changing network performances.
Several AI algorithms are comprehensively involved with telecommunication architecture adaptation: ARIMA (Auto-regressive Integrated Moving Average Model), LSTM (Long Short-Term Memory), GBDT (Gradient Boosting Decision Tree) and FNNM (Feed-forward Neural Network Model), i.e. advanced deep learning theory in AI.
- Scenario recognized, cell differentiated strategy: different threshold trigger points are trained for corresponding scenarios through traffic prediction (RRC connection, PRB utilization and data throughput). The accuracy of predicted traffic matching with real-time traffic exceeds 90%. The precise cell-level scenario-based triggering and parameters setting replaced traditional inflexible parameter sets for the whole network, which significantly prolongs the applied time of power saving.
- KPI ensured optimization strategy: the network KPIs are monitored in real-time. Power saving strategy is performed in every 15 minutes and real-time roll-back is allowed to ensure power saving gain without impacting network performance.
The solution is applicable in 2G/3G/4G as well as 5G with only software upgrade required.
Since ZTE and China Unicom had conducted the whole network verification in Shandong province for 3 months, abundant data was accumulated and the results had become a reference for other provinces. In the meantime, more applications have been undergoing with China Unicom Chongqing branch, China Telecom Chengdu branch and China Mobile Henan branch since 2019Q4.
Ultimate target: closed-loop of fully self-adaptive capability
Whole network intelligence is difficult to be achieved overnight, a long-term development is required. However, with the continuous accumulation of data from commercial networks, AI-based power saving solution will improve gradually with the help of machine-learning algorithms, while the AI algorithms themselves will evolve iteratively to realize higher efficiency and accurate strategy adaption with changing network topology and traffic model.
Along with the speed-up of 5G commercialization in all regions, AI-based power saving solution for 5G will be imperative for operators. 4G/5G collaboration will also become a challenge for AI-based power saving strategy. ZTE sees the need of more power saving strategies in terms of scenarios recognition, service differentiation and collaboration among technologies and grabs the opportunity. A sustainable and booming deployments of the AI-powered solution can be expected in the near future.