stella.sensor
Class PredictiveSensor

java.lang.Object
  util.PropertyContainer
      util.PropertyResources
          util.PropertyBundles
              util.rmi.RmiPropertyContainer
                  util.rmi.AbstractRmiCaster
                      stella.sensor.AbstractSensor
                          stella.sensor.TimerTaskSensor
                              stella.sensor.DataSensor
                                  stella.sensor.CumulativeSensor
                                      stella.sensor.PredictiveSensor

All Implemented Interfaces:

Cloneable, Remote, Caster, DataCaster, RmiDataCaster, RmiErrorCaster, Cumulating, Predicting, Telemetering, ExitCleaning, Initializable, LocalizedSupplying, PropertySupplying, ResourceSupplying, RmiCaster, RmiPropertyBearing

Direct Known Subclasses:

SmoothingSensor

public abstract class PredictiveSensor
extends CumulativeSensor
implements Predicting

A prediciting sensor adds the capability of predicting future sensor readings. This is in particular interesting for deducing bad weather situations ahead from their occurance. Of cause this is not a straight-forward task, so we do not aim at predicting days ahaed, but merely for times comparable to the closing time of the STELLA roof. This class handles this task by utilizing the extrapolating capabilities of the Extrapolation class and its daugther classes. Three keys are needed to correctly adapt this class.

• The KEY_EXTRAPOLATE key must point to the class of the extrapolator to use (resolvable class name). Currently, LinearExtrapolation and ParabolicExtrapolation are supported.
• The KEY_BASE key points to the number of past data points to use for the prediction base. This number must not exceed the cumulative power of the sensor. Smaller number gives you faster reaction, but overreaction may be a problem. This base number should be normally be higher for parabolic extrapolation. For linear extrapolation, a number comparable to the number of data points to predict is probably sufficient.
• The KEY_FUTURE key points to the time in the future (in ms) this sensor should predict its value. The number of data points to predict is hidden from the user, but it is this number the extrapolator needs. Therefore, the predictive sensor calculates the number of data points to predict by dividing this time span by its respawn intervall. The default value for this long is DEFFUTURE which is shortly longer than the closing time of the roof.
• The KEY_CHI key should point to a value that allows the prediciting sensor to judge the reliability of the prediction. If the residuals to the values used in the extrapolator is above this value, the predicition is treated unreliable. This is useful if the predictor should only be used to find certain, rather well-predictable events. If this key points to zero, the extrapolation is considered always reliable.
• The KEY_PREDICTADVISE key may be used to allow only advisory function of the predictive behaviour. In this case, the weather sensor will never produce bad weather events from its predictive edge, bad weather will only show up as a remark in the syslog.
• In version 1.2, the possibility to specify more than a single base was introduced. Multiple extrapolators give you a better handle on judging the reliability. For the first attempt, the predicted value is simply the average of the extrapolators.
• From version 1.2 on, the data set is now only copies once to the extrapolator, when the required quantity is reached for the first time. Additional sensor data is now added to the extrapolators Extrapolation.shiftForward(double) method, greatly reducing CPU-overhead.

See Also:

Shelf, Extrapolation

Nested Class Summary

Nested classes/interfaces inherited from class stella.sensor.TimerTaskSensor

TimerTaskSensor.SensorRead

Nested classes/interfaces inherited from class util.PropertyResources

PropertyResources.URLResource

Field Summary

private List<Integer>	bases The list of extrapolator bases.
private static String	DEFBASE The default number of past measurements for predicition.
private static double	DEFCHI The default reliability measure, meaning always reliable.
private static long	DEFFUTURE The default prediction time span.
private static boolean	DEFPREDICTADVISE The default advisory behaviour.
private Extrapolation[]	extrapolate The extrapolating algorithms to use.
private int	fut The number of steps to extrapolate.
static String	KEY_BASE The key to the list of number of past datas used for prediction.
static String	KEY_CHI The key to juding the reliability of the prediction.
static String	KEY_EXTRAPOLATE The key mapping to the extrapolating class.
static String	KEY_FUTURE The key to the time span to extrapolate.
static String	KEY_PREDICTADVISE The key defining only advisory functionality if set to true.
private int	maxbase The maximum base size.
private int	minbase The minimum base size.
protected Shelf<Double>	predict Stores the predicted values, i.e.

Fields inherited from class stella.sensor.CumulativeSensor

KEY_AVERAGE, KEY_MEMORY, KEY_PACK

Fields inherited from class stella.sensor.DataSensor

DEFRAWFORMAT, DEFREPORT, DEFSIGMAFORMAT, KEY_RAWFORMAT, KEY_REPORT, KEY_SIGMAFORMAT

Fields inherited from class stella.sensor.TimerTaskSensor

KEY_RESPAWN

Fields inherited from class stella.sensor.AbstractSensor

KEY_BIND, KEY_BINDNAME, KEY_CONVERT, KEY_DESCRIPTION, KEY_SENSORNAME

Fields inherited from class util.rmi.AbstractRmiCaster

ALTERNATEHOST, ALTERNATEPORT, KEY_EXPORTPORT, KEY_REGISTRYHOST, KEY_REGISTRYPORT

Fields inherited from class util.PropertyBundles

KEY_LOCALECOUNTRY, KEY_LOCALELANGUAGE, KEY_RESOURCEBUNDLES

Fields inherited from class util.PropertyResources

KEY_NOINITONCREATE, localurl, locate, POSTFIX_DIR, POSTFIX_EXT, POSTFIX_FILE, POSTFIX_LIST, POSTFIX_URL, urlset

Fields inherited from class util.PropertyContainer

KEY_LISTSEPARATOR, KEY_MAPKEYVALUECHAR, KEY_MAPSEPARATOR

Fields inherited from interface stella.rmi.RmiDataCaster

NAMING_EXTENSION

Fields inherited from interface util.PropertySupplying

CONFIG, KEY_CLASS

Fields inherited from interface util.ResourceSupplying

KEY_URLRESOURCES, KEY_URLUSECONFIG, KEY_URLUSECURRENT, KEY_URLUSEHOME

Constructor Summary

protected

PredictiveSensor(Map<String,String> qualify) Constructs a fully-qualified predictive sensor.

Method Summary

protected DataEvent	generateDataEvent() Takes the last sensor predictions and add it to the CumulativeDataEvent produced by the parental method.
Extrapolation[]	getExtrapolation() Returns the extrapolating instance used.
List<Double>	getLastPredict(int nr) The interface method for retrieving the last nr predicted raw values.
String	getPredictToggleInfo() Returns extra info on a smooth toggle event.
boolean	isPredictAdvise() Returns true if the predictive functionality of this sensor is only meant for advisory behaviour.
boolean	isPredicting() Returns true if the extrapolate object is non-null and enough measurements are already made.
boolean	isReliable() Returns a measure if this prediction is trustable.
protected void	newData(double raw, Date time) Conducts a single measure by calling this class's super method.
void	predictFrom(double lastraw) Predicts from the actual data point to the near future.
double	regetPredict() Returns the most-recently added predicted raw value as a primitive double.

Methods inherited from class stella.sensor.CumulativeSensor

cumulateOne, getLastRaw, getLastTime, getMaxSize, getMemory, getSize

Methods inherited from class stella.sensor.DataSensor

addDataListener, addRmiDataListener, conductOneMeasure, exit, getRawFormat, getSigmaFormat, removeDataListener, removeRmiDataListener, report

Methods inherited from class stella.sensor.TimerTaskSensor

cancel, equals, getRawToggleInfo, getReading, getRespawn, getTimeOfRead, hashCode, initSensorTimerTask, isReady, readSensor, regetRaw, setTimeOfRead, validReading

Methods inherited from class stella.sensor.AbstractSensor

addRmiErrorListener, createSensor, getBindName, getConvert, getSensorDescription, getSensorInitInfo, getSensorName, init, notifyErrorListener, removeRmiErrorListener, setConvert, status, toString

Methods inherited from class util.rmi.AbstractRmiCaster

bindAlternate, bindToRegistry, bindToRegistry, bindToRegistry, casterExit, casterExit, getBindName, isValid, unbindFromRegistry, unbindFromRegistry

Methods inherited from class util.rmi.RmiPropertyContainer

containsRemoteProperty, getRemoteProperties, getRemoteProperty, setRemoteProperty

Methods inherited from class util.PropertyBundles

clone, getLocalized, getLocalized, getLocalizedString, getLocalizedString, loadResource

Methods inherited from class util.PropertyResources

createFrom, createFrom, createFrom, getApplet, getAsResources, getLocalClassLoader, getPropertiesToKey, getPropertiesToKey, getResource, getResourceAsStream, getResourceFromKey, getResources, keyCreate, keyCreate, reload, setApplet

Methods inherited from class util.PropertyContainer

augment, augment, augment, defaultBoolean, defaultChar, defaultDouble, defaultFloat, defaultInt, defaultLong, defaultObject, defaultObject, defaultProperties, defaultProperty, getAsBoolean, getAsChar, getAsDouble, getAsEnums, getAsFloat, getAsInt, getAsList, getAsLong, getAsMap, getAsMap, getAsObject, getAsObject, getProperties, getProperty, has, isNew, parseObject, reload, removeProperty, rescanned, setObject, setProperties, setProperty, stringProperties

Methods inherited from class java.lang.Object

finalize, getClass, notify, notifyAll, wait, wait, wait

Methods inherited from interface stella.sensor.Cumulating

cumulateOne, getLastRaw, getLastTime, getMaxSize, getSize, getTimeOfRead, regetRaw

Methods inherited from interface util.rmi.RmiCaster

bindToRegistry, isValid, unbindFromRegistry

Methods inherited from interface stella.sensor.Telemetering

validReading

Methods inherited from interface util.PropertySupplying

defaultBoolean, defaultChar, defaultDouble, defaultFloat, defaultInt, defaultLong, defaultObject, defaultObject, defaultProperties, defaultProperty, getAsBoolean, getAsChar, getAsDouble, getAsFloat, getAsInt, getAsList, getAsLong, getAsMap, getAsObject, getAsObject, getProperties, getProperty, has, parseObject, removeProperty, setObject, setProperty, stringProperties

Methods inherited from interface util.ResourceSupplying

getResource, getResourceAsStream, getResources

Field Detail

KEY_EXTRAPOLATE

public static final String KEY_EXTRAPOLATE

The key mapping to the extrapolating class.

See Also:

Constant Field Values

KEY_BASE

public static final String KEY_BASE

The key to the list of number of past datas used for prediction.

See Also:

Constant Field Values

KEY_FUTURE

public static final String KEY_FUTURE

The key to the time span to extrapolate.

See Also:

Constant Field Values

KEY_CHI

public static final String KEY_CHI

The key to juding the reliability of the prediction.

See Also:

Constant Field Values

KEY_PREDICTADVISE

public static final String KEY_PREDICTADVISE

The key defining only advisory functionality if set to true.

See Also:

Constant Field Values

DEFFUTURE

private static final long DEFFUTURE

The default prediction time span.

See Also:

Constant Field Values

DEFBASE

private static final String DEFBASE

The default number of past measurements for predicition.

See Also:

Constant Field Values

DEFCHI

private static final double DEFCHI

The default reliability measure, meaning always reliable.

See Also:

Constant Field Values

DEFPREDICTADVISE

private static final boolean DEFPREDICTADVISE

The default advisory behaviour.

See Also:

Constant Field Values

predict

protected Shelf<Double> predict

Stores the predicted values, i.e. the near future of the data.

bases

private List<Integer> bases

The list of extrapolator bases. Consits of Integer objects.

minbase

private int minbase

The minimum base size.

maxbase

private int maxbase

The maximum base size.

extrapolate

private Extrapolation[] extrapolate

The extrapolating algorithms to use.

fut

private int fut

The number of steps to extrapolate. Initialized only if extrapolting.

Constructor Detail

PredictiveSensor

protected PredictiveSensor(Map<String,String> qualify)

Constructs a fully-qualified predictive sensor. Use of this constructor guarantees that the KEY_FUTURE and the KEY_BASE are set. Additionally, the extrapolating instance is constructed, if the KEY_EXTRAPOLATE is present and names a valid Extrapolation class. In the latter case, fut is assigned.

Method Detail

newData

protected void newData(double raw,
                       Date time)

Conducts a single measure by calling this class's super method. Additionally a value is predicted and stored in the predictors shelf as soon as its number of base point is reached.

Overrides:

newData in class CumulativeSensor

predictFrom

public void predictFrom(double lastraw)

Predicts from the actual data point to the near future. In fact, the value returned is the average of all extrapolators used. Note that this method can only predict after the required number of measurements KEY_BASE has been made.

Specified by:

predictFrom in interface Predicting

Parameters:

lastraw - The current raw measurement

regetPredict

public double regetPredict()

Returns the most-recently added predicted raw value as a primitive double. If no prediction is possible, Double.NaN is returned.

Specified by:

regetPredict in interface Predicting

Returns:

The predicted raw value or NaN.

getLastPredict

public List<Double> getLastPredict(int nr)

The interface method for retrieving the last nr predicted raw values.

Specified by:

getLastPredict in interface Predicting

Parameters:

nr - The number of predicted values to be retrieved.

Returns:

A list of Doubles.

See Also:

Predicting

getExtrapolation

public Extrapolation[] getExtrapolation()

Returns the extrapolating instance used.

isPredicting

public boolean isPredicting()

Returns true if the extrapolate object is non-null and enough measurements are already made.

Specified by:

isPredicting in interface Predicting

isPredictAdvise

public boolean isPredictAdvise()

Returns true if the predictive functionality of this sensor is only meant for advisory behaviour. In this case, bad weather events will only show up in the syslog, but will never produce bad weather events.

isReliable

public boolean isReliable()

Returns a measure if this prediction is trustable. In this version, the residuals of all extrapolators of the linear fit are checked if they lie beneath a certain threshold, defined with the KEY_CHI key. If this key points to zero, the sensor is trusted always, otherwise the χ² per data point must lie below this value for trustworthyness.
Note that the data should be set prior to calling this method.

Specified by:

isReliable in interface Predicting

getPredictToggleInfo

public String getPredictToggleInfo()

Returns extra info on a smooth toggle event. If no additional info is required, return null.

generateDataEvent

protected DataEvent generateDataEvent()

Takes the last sensor predictions and add it to the CumulativeDataEvent produced by the parental method. If no extrapolators are specified, the undisturbed data event, a CumulativeDataEvent is returned. If the CumulativeSensor.KEY_PACK property points to a value greater than one, the data event produced carries the average of the last predictions.

Overrides:

generateDataEvent in class CumulativeSensor

Returns:

A PredictiveDataEvent