Icon

02_​LSTM_​Network

This workflow predicts time series (energy consumption) by an LSTM network with lagged values as input. The trained model is then used for out-of-sample forecasting. The forecasted values are compared to the actual values, and the performance of the forecast is reported via scoring metrics and a line plot.



Data Loading Data Cleaning Time Series AnalysisLSTM NetworkSummary:In this workflow we'll define an LSTM Network Architecture to train and deploy on theTime Series. Instructions:1) Run the workflow up through the Partitioning node, we'll start from here2) Before we can train our network we need to define its architecture. First place theKeras Input Layer node, this defines the shape of the input data. Use shape [ ? , 200 ]for our 200 lagged values.3) The next layer will be the LSTM layer, place a Keras LSTM Layer node to representthis. Use 100 units, activation function ReLU, and recurrent activation function Sigmoid.4) Place the Keras Dense Layer as the last layer in the network architecture. Use 1 unitand activation function ReLU.5) To implement the LSTM model, use the Keras Network Learner node and train thenetwork on the training set, i.e. the bottom output of the Partitioning node. Use theList[...] column as input column, cluster_26 as target column, and MSE as lossfunction.6) Forecast values in the test set with the Deployment Loop component. Check theperformance with the Numeric Scorer and Line Plot. Network Architecture Create Input Vector convertdate/timeinto Date&Time objectsIntroducemissinghourstraining withMSE loss function[200] tensor for200 lagged inputs80/20 split[1] tensor foroutputLSTM with 100units and ReLUcombine intolistlag 200 valuesEnergyusagedata Missing Value String to Date&Time Column Filter Timestamp Alignment Line Plot Keras NetworkLearner Keras Input Layer Partitioning Keras Dense Layer Numeric Scorer Keras LSTM Layer Column Aggregator Lag Column Deployment Loop CSV Reader Data Loading Data Cleaning Time Series AnalysisLSTM NetworkSummary:In this workflow we'll define an LSTM Network Architecture to train and deploy on theTime Series. Instructions:1) Run the workflow up through the Partitioning node, we'll start from here2) Before we can train our network we need to define its architecture. First place theKeras Input Layer node, this defines the shape of the input data. Use shape [ ? , 200 ]for our 200 lagged values.3) The next layer will be the LSTM layer, place a Keras LSTM Layer node to representthis. Use 100 units, activation function ReLU, and recurrent activation function Sigmoid.4) Place the Keras Dense Layer as the last layer in the network architecture. Use 1 unitand activation function ReLU.5) To implement the LSTM model, use the Keras Network Learner node and train thenetwork on the training set, i.e. the bottom output of the Partitioning node. Use theList[...] column as input column, cluster_26 as target column, and MSE as lossfunction.6) Forecast values in the test set with the Deployment Loop component. Check theperformance with the Numeric Scorer and Line Plot. Network Architecture Create Input Vector convertdate/timeinto Date&Time objectsIntroducemissinghourstraining withMSE loss function[200] tensor for200 lagged inputs80/20 split[1] tensor foroutputLSTM with 100units and ReLUcombine intolistlag 200 valuesEnergyusagedata Missing Value String to Date&Time Column Filter Timestamp Alignment Line Plot Keras NetworkLearner Keras Input Layer Partitioning Keras Dense Layer Numeric Scorer Keras LSTM Layer Column Aggregator Lag Column Deployment Loop CSV Reader

Nodes

Extensions

Links

Download

To use this workflow in KNIME, download it from the below URL and open it in KNIME:

Download Workflow
On NodePit since: 2024-06-15
Last update: 2024-11-17
Created with KNIME version: v4.6.2
Tags: Time SeriesPredictionEnergy UsageLSTMDeep LearningNeural NetworkKeras

Deploy, schedule, execute, and monitor your KNIME workflows locally, in the cloud or on-premises – with our brand new NodePit Runner.

Try NodePit Runner!