Energy savings in cellular networks
A method may be disclosed for energy savings in cellular networks, comprising: collecting performance data from the cellular network; sharing the performance data with a batch process to train a prediction model; loading the prediction model into a closed-loop process; using the performance data in combination with the trained prediction model to predict cellular access network load for a particular cell; using the predicted cellular access network load to determine a recommended action; and executing the recommended action. The method may be implemented using a public cloud. The trained prediction model may be used to classify different cells and determine different recommended actions for the different cells.
1 . A method for energy savings in cellular networks, comprising:
collecting performance data from a cellular network;
sharing the performance data with a batch process to train a prediction model;
loading the prediction model into a closed-loop process;
using the performance data in combination with the trained prediction model to predict cellular access network load for a particular cell;
training a decision model to generate a power control recommended action for a specific predicted load condition;
determining, using the decision model and using the predicted cellular access network load, a recommended action; and
executing the recommended action using an application program interface (API) to send instructions to the particular cell.
2 . The method of claim 1 , wherein model training is performed using a public cloud, and wherein determining using the decision model is performed either using a public cloud or in a cellular operator network.
3 . The method of claim 1 , wherein the trained prediction model is used to classify different cells and determine different recommended actions for the different cells.
4 . The method of claim 1 , further comprising benchmarking by quantifying a value of an energy saving system into a reduction in energy per cell or per region.
5 . The method of claim 1 , wherein the benchmarking is based on comparing before and after energy usage.
6 . The method of claim 1 , using a load prediction model to predict load parameters for a time t+n.
7 . The method of claim 1 , wherein the prediction model uses a recurrent neural network (RNN).
8 . The method of claim 1 , wherein the prediction model uses one or more of DRNN (deep recurrent neural network), LSTM (long short-term memory), GRU (gated recurrent unit) models.
9 . The method of claim 1 , further comprising, by the decision model, a decision to grow or shrink cell energy saving time slots for capacity layer cells in a layered cell deployment architecture.
10 . The method of claim 1 , further comprising, by the decision model, a decision to increase or decrease signal power for capacity layer cells in a layered cell deployment architecture.
11 . A non-transitory computer-readable medium comprising instructions which, when executed in a network architecture at a processor, cause the network architecture to perform steps further comprising:
collecting performance data from a cellular network;
sharing the performance data with a batch process to train a prediction model;
loading the prediction model into a closed-loop process;
using the performance data in combination with the trained prediction model to predict cellular access network load for a particular cell;
training a decision model to generate a power control recommended action for a specific predicted load condition;
determining, using the decision model and using the predicted cellular access network load, a recommended action; and
executing the recommended action using an application program interface (API) to send instructions to the particular cell.
12 . The non-transitory computer-readable medium of claim 11 , wherein model training is performed using a public cloud, and wherein determining using the decision model is performed either using a public cloud or in a cellular operator network.
13 . The non-transitory computer-readable medium of claim 11 , wherein the trained prediction model is used to classify different cells and determine different recommended actions for the different cells.
14 . The non-transitory computer-readable medium of claim 11 , further comprising benchmarking by quantifying a value of an energy saving system into a reduction in energy per cell or per region.
15 . The non-transitory computer-readable medium of claim 11 , wherein the benchmarking is based on comparing before and after energy usage.
16 . The non-transitory computer-readable medium of claim 11 , using a load prediction model to predict load parameters for a time t+n.
17 . The non-transitory computer-readable medium of claim 11 , wherein the prediction model uses a recurrent neural network (RNN).
18 . The non-transitory computer-readable medium of claim 11 , wherein the prediction model uses one or more of DRNN (deep recurrent neural network), LSTM (long short-term memory), GRU (gated recurrent unit) models.
19 . The non-transitory computer-readable medium of claim 11 , further comprising, by the decision model, a decision to grow or shrink cell energy saving time slots for capacity layer cells in a layered cell deployment architecture.
20 . The non-transitory computer-readable medium of claim 11 , further comprising, by the decision model, a decision to increase or decrease signal power for capacity layer cells in a layered cell deployment architecture.