Early warning and recommendation system for the proactive management of wireless broadband networks

A non-transitory computer-readable medium comprising a computer program, that, when executed by a computer system, enables the computer system to perform the following method for providing early warning of future non-normal wireless network conditions and for providing recommendations for remedying predicted non-normal conditions, the method comprising:

• importing data from disparate sources into a data warehouse, wherein the imported data includes:
• (1) data related to network components;
• (2) KPI indicators for the network and associated time stamps;
• and (3) non-network data;
• building a semantic data model for the KPI indicators by enriching the KPI indicators with metadata;
• performing data cleansing operations on the imported data;
• using regression analysis and dimension reduction techniques on the cleansed, imported data to automatically identify significant network and non-network observational elements, network configuration parameters, and KPI indicators for use in defining a situation related to the network, wherein observational elements include network components IDs, location information for such network components, and environmental conditions in said locations;
• creating vectors that each define a situation related to the network, wherein each vector comprises values for the identified significant network and non-network observational elements, network configuration parameters, and KPI indicators;
• using unsupervised machine learning to build a statistical model that predicts the situations that result in non-normal network conditions, wherein building the statistical model comprises performing unsupervised clustering of the vectors based on network state;
• generating a situations table based on unsupervised clustering results, wherein the table lists distinct, labeled situations and the cluster to which each situation corresponds;
• creating a deep neural network model of situations (“the deep neural network model”) by using deep neural networks to reconstruct the situations table in memory, resulting in further microclustering of the vectors and identification of one or more new situations;
• importing new data from disparate sources into the data warehouse and creating a new situations table using the new data;
• comparing the new situations table to the table reconstructed using the deep neural network model (“the DNN table”) to determine how well the deep neural network model of situations predicted the new data;
• applying a Reinforcement Learning method to the new and previously-imported data to guide the deep neural network model to understand how changes to network configuration parameters affect situations;
• adjusting the hyperparameters of the deep neural network model based on the comparison of the new situations table to the DNN table accordingly;
• retraining the deep neural networks model with the adjusted hyperparameters;
• using the statistical model created from unsupervised clustering to predict a future non-normal network situation;
• using the deep neural network model and Reinforcement Learning results to identifying one or more changes to network configuration parameters that has at least a threshold probability of moving the predicted non-normal situation to a normal network state;
• and
• displaying in a user interface an alert for the predicted non-normal network situation and a corresponding recommendation for remedying the predicted non-normal network situation, wherein the recommendation is based on the one or more identified changes to network configuration parameters that has at least a threshold probability of moving the predicted non-normal situation to a normal network state.

A computer system for providing early warning of future non-normal wireless network conditions and for providing recommendations for remedying predicted non-normal conditions, the system comprising:

• one or more processors;
• one or more memory units coupled to the one or more processors, wherein the one or more memory units store instructions that, when executed by the one or more processors, cause the system to perform the operations of:
• importing data from disparate sources into a data warehouse, wherein the imported data includes:
• (1) data related to network components;
• (2) KPI indicators for the network and associated time stamps;
• and (3) non-network data;
• building a semantic data model for the KPI indicators by enriching the KPI indicators with metadata;
• performing data cleansing operations on the imported data;
• using regression analysis and dimension reduction techniques on the cleansed, imported data to automatically identify significant network and non-network observational elements, network configuration parameters, and KPI indicators for use in defining a situation related to the network, wherein observational elements include network components IDs, location information for such network components, and environmental conditions in said locations;
• creating vectors that each define a situation related to the network, wherein each vector comprises values for the identified significant network and non-network observational elements, network configuration parameters, and KPI indicators;
• using unsupervised machine learning to build a statistical model that predicts the situations that result in non-normal network conditions, wherein building the statistical model comprises performing unsupervised clustering of the vectors based on network state;
• generating a situations table based on unsupervised clustering results, wherein the table lists distinct, labeled situations and the cluster to which each situation corresponds;
• creating a deep neural network model of situations (“the deep neural network model”) by using deep neural networks to reconstruct the situations table in memory, resulting in further microclustering of the vectors and identification of one or more new situations;
• importing new data from disparate sources into the data warehouse and creating a new situations table using the new data;
• comparing the new situations table to the table reconstructed using the deep neural network model (“the DNN table”) to determine how well the deep neural network model of situations predicted the new data;
• applying a Reinforcement Learning method to the new and previously-imported data to guide the deep neural network model to understand how changes to network configuration parameters affect situations;
• adjusting the hyperparameters of the deep neural network model based on the comparison of the new situations table to the DNN table accordingly;
• retraining the deep neural networks model with the adjusted hyperparameters;
• using the statistical model created from unsupervised clustering to predict a future non-normal network situation;
• using the deep neural network model and Reinforcement Learning results to identifying one or more changes to network configuration parameters that has at least a threshold probability of moving the predicted non-normal situation to a normal network state;
• and
• displaying in a user interface an alert for the predicted non-normal network situation and a corresponding recommendation for remedying the predicted non-normal network situation, wherein the recommendation is based on the one or more identified changes to network configuration parameters that has at least a threshold probability of moving the predicted non-normal situation to a normal network state.