WO2021171303A1

WO2021171303A1 - A system and method for energy management of identical appliances using non-intrusive load monitoring technique

Info

Publication number: WO2021171303A1
Application number: PCT/IN2021/050145
Authority: WO
Inventors: Mukesh Kumar; Rajendiran GOPINATH; Prakash Chandra JOSHUA; Srinivas Kota; Guruswamy Sivanpackiam AYYAPPAN; Venganallur Padmanabhan ANAND
Original assignee: Council Of Scientific And Industrial Research
Priority date: 2020-02-26
Filing date: 2021-02-16
Publication date: 2021-09-02

Abstract

The present invention discloses status monitoring of two or more identical air conditioners of same make and same model connected to the same phase of the power supply using non-intrusive load monitoring (NILM) technique. In a home/office/laboratory environment they may exists two or more air conditioners of same make and same model. In such cases, developing an effective NILM system becomes a challenging problem. The general features such as real power and reactive power would not be sufficient to distinguish the identical loads using NILM. The present invention discloses the system and method of processing the power signals for extracting a new combination of features that makes suitable for monitoring and detecting the events, turn on and off of two or more identical air conditioners. The normalized feature set includes power, statistical and intrinsic features extracted during both transient and steady state electrical signals to learn the inherent characteristics of the identical air conditioners. Further, the new combination of features extracted from the disclosed system and method of the present invention is also scalable and robust in monitoring and identifying the events of the multiple identical deep freezers in NILM system.

Description

A SYSTEM AND METHOD FOR ENERGY MANAGEMENT OF IDENTICAL APPLIANCES USING NON-INTRUSIVE LOAD MONITORING TECHNIQUE

FIELD OF THE INVENTION

The present disclosure relates to a system and method for energy management and electrical power monitoring of multiple identical loads, air conditioners and deep freezers of same make and same model connected to the same phase of the power supply in buildings/industries.

BACKGROUND OF THE INVENTION

Energy conservation and management is gaining its attention among researchers and policy makers due to an increase in energy demand and depletion of fossil fuel. It also plays a vital role in industrial/commercial/domestic/power sectors for reducing carbon emission, the energy bill. To conserve energy, it is important to have a monitoring system with precise measurement and identification for the electrical utilities used in the industries and commercial/residential buildings.

Building sector is one of the key areas that contributes energy consumption significantly. Energy wastage can be avoided by monitoring energy consumption of buildings and report significant information back to consumers. It is reported that the total energy consumption information would not change the consumer’s energy usage behavior significantly. To enhance quality of services in existing monitoring technique, there is a need to monitor energy consumption of individual appliances in buildings/offices/industries. Hence, one meter for each appliance is necessary. However, this approach is not considered to be a cost effective due to installation of sensors for every individual appliance.

To overcome this problem, non-intrusive approach was introduced to monitor the loads using a single meter from the main power supply. However, intensive research works on NILM are being carried out only in recent years due to the gradual increase in usage of smart meters. Moreover, there is no NILM algorithm/device has been developed that can accurately disaggregate the total energy consumption of buildings/households. Effective monitoring of loads/appliances could also be used to detect the ageing of appliances and provide suggestions to consumers for replacing inefficient appliance and thereby saving energy and its cost. Further, NILM can also be used to identify malfunctioning appliances and perform maintenance of the appliance in order to save energy.

In most of the works, real power and reactive power are used as the features to detect appliances operating states. Appliance patterns are learnt using machine learning algorithms in a supervised or unsupervised manner, based on the customers requirement or applications. The US patent No. US4858141, discloses an apparatus to disaggregate the household energy usage. The appliance events are detected using the change in the admittance of the circuit. Cluster analysis was used to group similar on and off events to identify the appliances. However, appliance state identification becomes complex when two or more appliances are in similar power and identical. In this scenario, real and reactive power features may not help to identify the events effectively. Though many works on NILM have been carried out by numerous researchers, very few have addressed similar load monitoring. Moreover, monitoring and event detection of identical appliances connected to the same phase of the power supply is a challenging task and it is still unexplored.

The US patent No. 2014/0336831A1 discloses an apparatus and method to monitor the similar power appliances. The features, apparent power and power factor were used to monitor the status of appliances. It was reported that event detection becomes complex when two or more appliances having similar power factors were operated simultaneously. It is also noted that power factors have little variations when it is in similar range. To overcome this problem, apparent power information was used along with the power factor. However, apparent power and power factor may not be sufficient when two or more appliances are identical.

To overcome the aforementioned limitations, there is a need to identify the robust features to detect the events of identical appliances. The present invention describes a new set of features that are robust to identical appliances irrespective of brand/make. Multiple identical air conditioners are considered for event detection using new feature set. The developed NILM system is also validated across different air conditioners brand as well as with the deep freezers to check robustness of the features.

OBJECTIVES OF THE INVENTION

The present disclosure aims to monitor identical loads in a non-intrusive manner for energy management of the appliances. The present invention considers typical office/home environment where two or more air conditioners or deep freezers loaded in the same phase of the power supply which are identical (same make and same model). In this scenario, non-intrusive load monitoring (NILM) system may not recognize the events of identical appliances correctly. Features such as real power, and reactive power may not be sufficient to distinguish the loads effectively when two or more identical loads exists. The present disclosure aims to develop NILM system which has robust features that are extracted from electrical signals sampled at high sampling rate. The robust features need to be capable of capturing unique signature/pattern from the identical loads for identifying the load conditions across different make/brand.

SUMMARY OF THE INVENTION

The present disclosure reveals the system for monitoring identical loads using non-intrusive approach. Multiple identical air conditioners are considered for this study. Three phase voltage and current signals are measured from the incoming main power supply using the voltage and current sensors. The NILM system for identical appliance monitoring comprises of various functional modules that are processed sequentially. The functional modules include, voltage and current sensor module, data acquisition, signal conditioning module, feature calculation, feature fusion and feature normalization using one or more processors, interfaces, identical appliance expert knowledge storage module and appliance status display modules. The feature characteristics and status of the identical appliances can be transferred to the server using one or more communication protocols. Initially, electrical signals are sampled at high sampling rate (in kHz) to capture the behavior of identical air conditioners. The acquired signals are preprocessed by signal conditioning unit to filter out the noises presented in the electrical signals. The noise filtered signals are then processed through moving window unit to segment the voltage and current signals for every cycle. The moving window unit can be configured to fixed and overlapping intervals of the signals. The segmented data is given as an input to feature extractor unit, where different set of features are extracted from the electrical signals.

The feature set consists of power, statistical and intrinsic features derived from power features, to capture turn on and turn off conditions of the identical air conditioners. The features are also extracted during transient and steady states to obtain patterns effectively for load identification. The features power factor (FI), apparent power (F2), current form factor (F3) and instantaneous current coefficient of kurtosis (F4) are extracted for every cycle of input power signal. Form factor is derived from the root mean square of the current signal (Irms) whereas the coefficient of kurtosis is derived from raw instantaneous current signal (I). Further, the features are then expanded by extracting intrinsic relationship between power features using geometric mean. The intrinsic relationship between FI and F2, FI and F3, F2 and F3, and FI, F2 and F3 are extracted. The features are extracted for the transient and steady state separately and then feature fusion has been carried out to make the input feature matrix. Further, the extracted features are normalized using z-score normalization to make the features in a similar scale for distinguishing the identical appliances effectively. The normalized features are fed to supervised machine learning classifier, support vector machine (SVM) for training the model. The trained model is an expert knowledge base which will be stored in the storage module of the NILM system. The expert knowledge will be accessed to identify unknown signals from the main power supply in the testing field.

Further, the identical appliance behavior and its characteristics would also change gradually over the period of time. Therefore, the identical appliance signatures are also updated at regular intervals to make the decision support system accurate. The experiments with the present system reveal that the process of obtaining the new feature set is found to be robust in identifying the event detection (on/off conditions) of the multiple identical air conditioners connected in the same phase of the power supply. Further, similar experiments are also carried out for different brands of identical air conditioners as well as two identical deep freezers to check the effectiveness. The experimental results suggest that the new feature set is capable of finding the intrinsic characteristics of the identical air conditioners across different brands as well as the identical deep freezers which helps to distinguish the event detection.

BRIEF DESCRIPTION OF THE DRAWING

The following description describes the drawings/block diagrams that are used for illustrating the present disclosure. The description can be better understood along with the appended drawings. The detailed description has been described with the reference numbers and figure numbers.

FIG 1 illustrates three identical air conditioners of same brand and same model have been loaded to the same phase of three-phase power supply. It also includes three-phase measuring system for voltage and current signals from the main power supply.

FIG 2 is a diagram to describe load configurations where three different brand air conditioners are connected. For each brand, two identical air conditioners are connected to the same phase of power line.

FIG 3 illustrates the overall architecture of hardware system for identical appliance monitoring system that comprises of microcontrollers, ADC, storage and display units.

FIG 4 describes the complete system that includes different brands of identical loads connected to the power supply and process of feature extraction, feature fusion and normalization method involved during signal processing and appliance status detection.

FIG 5 illustrates two identical deep freezers of same brand and same model have been loaded to the same phase of three-phase power supply. It also includes three-phase measuring system for voltage and current signals from the main power supply.

FIG 6 illustrates apparent power of the two identical air conditioners (Brand A: A1 and A2) during turn on and off conditions. FIG 7 illustrates power factor of the two identical air conditioners (Brand A: A1 and A2) during turn on and off conditions.

FIG 8 illustrates current form factor of Irms of the two identical air conditioners (Brand A: A1 and A2) during turn on and off conditions.

FIG 9 illustrates instantaneous current coefficient of kurtosis for the two identical air conditioners (Brand A: A1 and A2) during turn on and off conditions.

Table 1 lists identical appliance detection performance (confusion matrix) for the air conditioners (Brand A: A1 and A2) using the traditional feature set (power factor and apparent power).

Table 2 lists identical appliance detection performance (confusion matrix) for the air conditioners (Brand A: A1 and A2) using the new disclosed feature set.

Table 3 lists identical appliance detection performance (confusion matrix) for the three different brand air conditioners (Brand A: AC1, AC2, AC3; Brand B: AC4, AC5; Brand C: AC6, AC7) using the traditional feature set (power factor and apparent power).

Table 4 lists identical appliance detection performance (confusion matrix) for the air conditioners (Brand A: AC1, AC2, AC3; Brand B: AC4, AC5; Brand C: AC6, AC7) using the new disclosed feature set.

Table 5 lists identical appliance detection performance (confusion matrix) for the deep freezers (Brand A: A1 and A2) using the traditional feature set (power factor and apparent power).

Table 6 lists identical appliance detection performance (confusion matrix) for deep freezers (Brand A: A1 and A2) using the new disclosed feature set. DETAILED DESCRIPTION OF THE INVENTION

The disclosure reveals the system and method for monitoring turn on/off events of multiple identical air conditioners connected to the same phase of power supply in a non-intrusive manner. The system comprises of two or more identical air conditioners of different brands to check the effectiveness of the developed system. The developed system is also tested for different configuration of loads connected to the same phase of the three-phase power supply. Further, the developed method is also tested for the monitoring the events of two or more identical deep freezers connected in the same phase of the power supply.

FIG. 4 illustrates one type of configurations used for the analysis. In this configuration, three air conditioners brand (2, 3 and 4) are connected in the three-phase power supply. Each brand has two or more identical air conditioners connected to the same phase of the power supply. Single NILM meter 1 is used for measuring the three-phase current and voltage signals. The acquired electrical signals are fed to signal conditioning unit 8 to minimize the noises presented in the signals. The denoised signals are given as an input to analog to digital converter (ADC) 10. The acquired signals are sampled at the high sampling rate (in kHz) to capture more information from the signal. The converted signal is then segmented using moving window method for every cycle of electrical signal in the data segmentation unit 22. The segmented data is fed to feature extraction unit 18.

FIG. 3 describes the detailed block diagram of hardware modules integrated in the NILM system. Signal conditioning unit 8 comprises of three phase voltage conditioning 8a and current conditioning 8b sub-units where voltage and current signals are preprocessed separately. The power electronic devices, switched mode power supply (SMPS) 6 and DC- DC converter 7 are the inputs to supply the power to ADC 10. The ADC comprises of analog and digital circuits for processing the signals. Analog circuit 10a is powered by the SMPS 6, whereas digital and buffer circuits 10b are powered by the DC-DC converter 7.

The converted signals are then fed to one or more microcontroller units. Multiple microcontrollers are used to carry out different process in parallel and thereby computation time would be minimized. Different modules such as identical air conditioner status monitoring display 14, expert knowledge storage/data logger 15 and MODBUS 12 are interfaced with microcontrollers for further process (FIG. 3). The expert knowledge storage module is used to store the identical air conditioners signature database/features characteristics and model parameters of the machine learning algorithm. The trained model may have many parameters that are learnt from the input features using machine learning algorithms and thereby significant to distinguish the identical air conditioners.

FIG. 4 illustrates the feature extraction and feature normalization in detail. The feature extraction unit 18 extracts a set of features during transient and steady states from the preprocessed electrical signal, voltage and current for every cycle. The present disclosure has the capability to extract transient and steady state features for the data sampled at higher sampling rate. Therefore, the extracted feature has the pattern for both transient and steady level which would be effective in learning inherent information from the raw signals of the appliances.

The feature extraction method 18 has two types of the features, namely power features 18a, statistical features 18b, and intrinsic features 18c. In this approach, first the features are extracted based on the power and statistical features and then the intrinsic features are derived from the power features to capture the turn on and turn off events of the identical air conditioners at transient and steady state level. The three types of features are then parallelly fused 23 to form feature matrix which will be fed to the next process. Further, the variations in the power, statistical and intrinsic features would affect the accurate event detection of identical air conditioners. Therefore, features are normalized to make the power and statistical features in a similar range for effective identification of the events of the identical air conditioners. Moreover, feature normalization would lead to less computation time for training the model. The present disclosure has feature normalization unit 24 to scale the features in similar magnitudes using z-score normalization, which preserves the feature values with an average/mean of zero and standard deviation of one. The normalized features are fed to intelligent decision support system 25 for appliance detection. In 25, machine learning algorithm, support vector machine (SVM) is used for decision-making process. However, the present disclosure is not limited to single algorithm alone and it may also have one or more models developed from the multiple machine learning algorithms to make the decision by majority voting approach. Prior to the testing of unknown signals, expert knowledge/historical information 15 about the identical air conditioners are obtained by training the model. The normalized features from the identical air conditioners for the on and off state events during transient and steady states are used for developing the model using supervised learning approach.

The present disclosure reveals a new set of features that comprises of power related, statistical and intrinsic features extracted during transient and steady states of the electrical signals for every cycle. In 18a, power related features, apparent power, power factor and current form factor are extracted. Subsequently, statistical feature 18b, coefficient of kurtosis is extracted from the instantaneous current signal. Further, the features are expanded by deriving intrinsic features from the power related features. The intrinsic features help to identify intrinsic relationship between the power related features for effective event detection of identical air conditioners. From state of the art, it is observed that the feature combination, power factor and apparent power are effective in detecting the events even when the appliances have similar power factors appliance. However, these features are not sufficient when one or more physical parameters of the appliances are identical. Moreover, event detection becomes complex when two or more identical appliances are connected in the same phase of the power line. In this case, the current signatures will look same for the identical appliances. To overcome this, the present disclosure illustrates the process of feature extraction and feature normalization for making the events of the identical air conditioners distinguishable.

In the scenario shown in FIG. 2 and 4, complexity of the NILM system increases further when two or more brands having multiple identical air conditioners are connected to the same phase of the power line. It is noted that the event detection of identical air conditioners becomes quite simple when they are connected in different phases of the power line due to the variations in the phases. The present disclosure addresses these complex scenarios for event detection of the identical air conditioners. It is well known that every appliance has its own signature, and it also applies to the identical appliances. The present disclosure intended to capture the unique signature of the identical appliances (air conditioners) using the new feature combination derived from the process of functional modules of the present system.

First, the power related features, apparent power (FI), power factor (F2) and current form factor (F3) are extracted for every cycle of electrical signals. Current form factor is derived from the instantaneous Irms current signals. The statistical feature, coefficient of kurtosis (F4) is derived from the instantaneous current signals. The shape of the current signal for the identical air conditioners is captured in both power and statistical perspective at the transient as well as the steady state. Current form factor helps to capture significant information about the shape of current signal derived from root mean square of the current signal. Coefficient of kurtosis feature measures the shape information in terms of peakness in a statistical perspective. Subsequently, intrinsic features are extracted from the power features. The intrinsic relationship between apparent power (FI) and power factor (F2), apparent power (FI) and current form factor (F3), power factor (F2) and current form factor (F3), and apparent power (FI), power factor (F2) and current form factor (F3) are extracted using geometric mean for obtaining intrinsic features. The intrinsic features help to reveal unique information about the identical air conditioners. The three types of feature set are fused 23 and then feature set is normalized using z-score normalization for maintaining the features in the same scale and thereby the features can be distinguishable between identical air conditioners. The intrinsic features (IF) are extracted using the following mathematical expressions.

7F12 = Antilog (log FI + logF2)j (1)

7F13 = Antilog (log FI + logF3)j (2)

7F123 = Antilog (logFl + logF2 + logF3)j (4)

Therefore, the new normalized feature set or normalized feature matrix can be expresses as Fnorm = [FI, F2, F3, F4, 7F12, 7F13, 7F23, 7F123] The illustration of the power and statistical features for the two identical AC’s (Brand A: AC1 and AC2) during on and off states are shown in FIGS. 6-9. In this load configuration, identical AC’s are connected to the same phase of the power supply. The figures also describe the features at both transient and steady state levels of the identical AC’s during different operating conditions.

The process of obtaining expert knowledge or historical information 15 of identical appliances is the crucial/important step in the NILM algorithm development. The present disclosure has the capability to store the trained model parameters/expert knowledge about the identical air conditioners which are obtained in offline process. The present disclosure uses non-linear kernel, radial basis function (RBF) for mapping the features from lower dimensional space (input feature space) to higher dimensional space where the appliance states are classified linearly. The RBF kernel parameters are fine-tuned empirically to improve the appliance identification accuracy. The pattern of every individual identical appliances is learnt separately during the training phase and then the optimized trained model parameters are stored in the storage module of the NILM system. The model parameters/expert knowledge are also updated at regular interval whenever the new identical appliances are commissioned in the monitoring/home/office environment. The air conditioners performance characteristics may also change over the period of time. Therefore, the existing identical air conditioners signatures are also need to be updated at fixed time intervals.

The developed system has been tested for different load configurations. Experiments have been carried in similar environmental and operating conditions across different brands for checking the robustness of the disclosed feature set. In one of the load configurations (FIG. 1), three identical air conditioners of brand A has been loaded to the same phase of three- phase power line. The air conditioners are connected to the same phase for checking the robustness of the features in distinguishing identical air conditioners. Experiments and results for this configuration has been found be significant improvement for the disclosed new feature set when compared to the features, power factor and apparent power used in the state of the art. It is also observed that the length of the wire connected to every individual appliance from the source (single meter in the three-phase power line) plays a significant role in distinguishing the identical air conditioners due to the effect of additional resistance of the wire added in the acquired signal.

In another load configuration (FIG.2 and 4), multiple identical air conditioners of different brand (A, B and C) are used for the testing. In this configuration, brand A has three identical air conditioners and, brand B and C have two identical air conditioners respectively. It is to be noted that all the identical air conditioners of different brands have been connected to the same phase of three-phase power line. The experiments and results reveal that the turn on and off states of the seven air conditioners have been identified effectively. The performance measure of the NILM system has been evaluated using the confusion matrix. The appliance identification performance of identical air conditioners for the traditional feature set (power factor and apparent power) and disclosed new feature set is listed in Tables 1-4. The numbers in these tables represents the number of data samples that were classified correctly, or misclassified as some other appliances in the context of confusion matrix. It is to be noted that the data samples of the turn off events will be less compared to the turn on events. Table 1 and 2, lists the classification performance (confusion matrix) of the identical air conditioners (brand A: A1 and A2) connected to the same phase of the power line. The experiments and results show that the disclosed new feature set outperforms the tradition feature sets for the identical appliances detection. Subsequently, experiments with the identical air conditioners of three different brands (Brand A: Al, A2, A3; Brand B: Bl, B2; Brand C: Cl, C2) are carried out to check the robustness of the disclosed feature set. The classification performance of the identical air conditioners of three brands for the traditional and new feature set is listed in Table 3 and 4 respectively. It was observed that the disclosed new feature set performs better than the traditional feature set.

In another configuration (FIG. 5), multiple deep freezers of same brand and same model (identical) has been connected to the same phase of the power supply. In this configuration two identical deep freezers are monitoring using the same method and system which are used in monitoring the identical air conditioners. The experiments and results (Table 5 and 6) reveal that the new disclosed feature set is also scalable and robust in identifying the events of the multiple deep freezers effectively. Therefore, the new disclosed feature set can be further adapted for various identical electrical utilities in the non-intrusive load monitoring system.

EXAMPLES

A system for monitoring the appliances in home/office buildings is much needed for effective energy management based on the feedback provided to the consumers. Non- intrusive load monitoring (NILM) has been introduced to monitor the appliance effectively using a single power meter at the entry point of the building. Though NILM works better for different appliances, it fails to recognize the events of the identical appliances accurately due to less variations in the power features. The present disclosure intends to monitor multiple identical appliances, specifically multiple identical air conditioners or multiple deep freezers of same make and same model connected to the same phase of the power line which will be a challenging task. A new set of features are derived from the power features to effectively distinguish the turn on and turn off events of the identical air conditioners. The present disclosure can also be used for various applications including multiple identical refrigerator monitoring at specific use cases where maintaining the temperature of refrigeration system and its expected operational performance is crucial.

ADVANTAGES

The present disclosure able to distinguish the turn on/off events of multiple identical appliances of same make and same model connected to the same phase of the power supply. The developed system has been tested successfully for the multiple identical air conditioners of different brands and it can be further extended to different electrical utilities/appliance. The disclosed process of deriving the new feature set (combination) for the NILM system has been found to be robust across different brands of air conditioners as well as for the deep freezers. Moreover, the feature set has the ability to distinguish the appliances even when all the identical appliances are connected to the same phase of the power line which is considered to be a challenging problem, where the variations of the electrical parameters would be minimum. The major advantage of the disclosed system is that the appliances states can be identified even at the transient level due to the effect of the feature learning from the highly sampled electrical signals.

Claims

We Claim:

1. A system for monitoring two or more identical air conditioners connected to the same phase of the power supply in a non-intrusive method, the system comprising: an intelligent decision support system to detect on and off status of two or more identical air conditioners for every cycle of electrical signals acquired by the three- phase measurement system in a non-intrusive manner using a method of processing the signals for extracting group of power, statistical and intrinsic features during transient and steady state of power signals. a sequence of multiple functional modules are configured to process the voltage and current signals, analog to digital conversion of input signals, voltage and current signal (data) segmentation per cycle, extraction of two or more types of features, fusion of multiple features types, and normalization of the input features for monitoring two or more identical air conditioners of same brand and same model connected to the same phase of the power supply; a three phase voltage and current measurement system at the entry point of the building, configured to acquire aggregated power signals from the two or more identical air conditioners connected to the same phase of the power supply; wherein: the three-phase single meter voltage and current measurement system further comprises of voltage and current conditioning units to denoise the aggregated signals acquired from the two or more identical air conditioners connected to the same of the power supply separately; wherein: one of the functional modules for non-intrusive load monitoring system of identical air conditioners comprises of at least one switched mode power supply (SMPS) and DC-DC converter to supply the power to the analog and digital circuits respectively in the analog to digital converter module.

, wherein: the preprocessed aggregated signal of two or more identical air conditioners from the signal processing unit are determined to segment at the fixed intervals using moving window function for each cycle during both transient and steady state level at high sampling rates and consequently configured to fed into feature extractor unit, said moving window function configured to segment aggregated power signals in an overlapping manner and different time intervals based on the user preferences; and a feature extractor unit (module) is configured to extract three types of features for every cycle of current and voltage signals acquired at transient and steady state operating conditions from the two or more identical air conditioners connected to the same phase of the power supply, said feature extraction module for monitoring and event detection of two or more identical air conditioners comprising of power related features, statistical feature and intrinsic features derived from power features using geometric mean.

2. The system of claim 1, wherein: the power related features comprises of apparent power, power factor and current form factor for monitoring and event detection of two or more identical air conditioners, wherein the current form factor is derived from the root mean square of the current signals at the higher sampling rate and includes the information of shape of the signal along with the apparent power and power factor for identical air conditioners on and off states;

3. The system of claim 1, wherein: the statistical feature, includes coefficient of kurtosis of the current signal to get the peakness (shape) of the instantaneous current signal during transient and steady states of the two or more identical air conditioners in a statistical perspective.

4. The system of claim 1 , wherein: the apparent power feature includes the information about the differences between power consumption due to the power loss in the wire used in the particular wiring path while two or more identical loads connected to same phase.

5. The system of claim 1, wherein: the intrinsic features, (IF12, IF13, IF23, andIF123) are derived from the power features, apparent power (FI), power factor (F2), and current form factor (F3) using geometric mean and used to obtain the intrinsic relationship between the power features that help to distinguish the identical air conditioners; wherein the intrinsic features (IF), IF 12 are derived from apparent power (FI) and power factor (F2); and IF 13 is derived from apparent power (FI) and current form factor (F3); IF23 is derived from power factor (F2) and current form factor (F3); IF 123 is derived from apparent power (FI), power factor (F2) and current form factor (F3).

6. The system of claim 1, wherein: the total new feature set comprises of apparent power (FI), power factor (F2), current form factor (F3), current coefficient of kurtosis (F4), intrinsic feature (IF12), intrinsic feature (IF13), intrinsic feature (IF23), and intrinsic feature (IF 123) for event detection, on and off states of two or more identical air conditioners connected to the same phase of the power supply.

7. The system of claim 1, wherein: a feature fusion module for monitoring and event detection of two or more identical air conditioners is configured to combine the power features, statistical feature and intrinsic features extracted in the feature extraction module during the transient and steady state level for forming input feature matrix before input to the intelligent decision support system.

8. The system of claim 7, wherein: a feature normalization module for monitoring and event detection of two or more identical air conditioners is configured to normalize the input feature matrix using z-score normalization method to make all the features in similar range of values in order to distinguish events of the identical air conditioners.

9. The system of claim 1, wherein: the display module is configured to display the graphical visualization of the decisions made by the intelligent decision support system by illustrating the on and off status along with the time of the events of identical air conditioners separately., and wherein the display module is also configured to display monitored measured electrical parameters and two or more feature types and its values.

10. The system of claim 1 , wherein: the monitoring and event detection of two or more identical air conditioners connected to the same phase of the power supply have at least two or more provisions to send the monitored electrical parameters and features and as well as the event status with respect to time, through two or more communication protocols.