Model and data driving combined user side load non-intrusive identification method

Abstract

The invention relates to the technical field of load management, in particular to a model and data driving combined user side load non-intrusive identification method. Comprising the following steps: extracting steady-state features and transient-state features of typical electric equipment; constructing a user side load non-intrusive identification architecture based on decision tree classification; the invention discloses a user side load non-intrusive identification algorithm based on multi-feature fusion. According to the design of the invention, through extraction of steady state features and transient state features of typical electric equipment, multi-dimensional feature quantity identification is carried out on load samples based on a decision tree classification criterion, and non-intrusive load identification is realized through feature quantity matching of similar loads; an identification criterion based on decision tree classification is established, and various types of electric equipment can be quickly decomposed from a user sample; the steady-state and transient-state multi-feature fusion of the electric equipment is applied to a load non-intrusive identification algorithm, so that the accuracy of the algorithm can be improved, and the calculation amount and time of the algorithm can be greatly reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of load management, and in particular to a non-intrusive identification method for user-side loads that combines a model with data driving. Background Art

[0002] As a key technology in non-intrusive load monitoring (NILM), load identification decomposes the total load signature based on load features extracted from load sampling data to identify various electrical devices. Currently, numerous methods for implementing non-intrusive load monitoring have been studied, most of which employ pattern recognition methods, such as clustering algorithms, evolutionary algorithms, and machine learning algorithms, to achieve load monitoring and identification. Existing cluster identification research assumes that the number and type of loads (i.e., the number of clusters) is known, and most use active and reactive power as load signatures. However, when the number and type of loads are unknown or when multiple appliances are operating simultaneously, they cannot accurately identify the appliance type and operating status. Evolutionary algorithms can achieve more accurate load identification results through iterative optimization, but they are computationally complex, and the accuracy of the algorithm's final iterative output is significantly affected by the optimization objective function. Existing studies have often used a single feature value as the optimization objective, resulting in certain limitations in the optimization results. Machine learning algorithms (such as neural networks) suffer from significant influence of weights and thresholds on the accuracy of neural network output, and random initialization results are unstable. Therefore, optimization of weights and thresholds is necessary to improve the accuracy of neural networks in load monitoring and identification.

[0003] In view of this, we propose a non-intrusive user-side load identification method that combines model and data driving. Summary of the Invention

[0004] The purpose of the present invention is to provide a non-intrusive identification method for user-side load that combines a model with data driving to solve the problems raised in the above background technology.

[0005] To solve the above technical problems, the present invention aims to provide a non-intrusive user-side load identification method that combines a model with data-driven approach, comprising the following steps:

[0006] S1. Extraction of steady-state and transient characteristics of typical electrical equipment: The steady-state characteristics of typical electrical equipment include power increment, steady-state power, third harmonic increment, and load operating time; the transient characteristics of typical electrical equipment include transient power surges;

[0007] S2. Construct a non-intrusive user-side load identification architecture based on decision tree classification;

[0008] S3. Applying a multi-feature fusion user-side load non-intrusive identification algorithm, specifically including the following steps:

[0009] S3.1. Collect and pre-process the voltage and current at the user's main incoming line to obtain the basic characteristic quantity of the load: active power sequence P m , reactive power sequence Q m , third harmonic current sequence I 3m ;

[0010] S3.2. Detect active power fluctuation and active power slow change of load samples to distinguish between fluctuating loads, slow-changing loads, and fixed-frequency loads;

[0011] S3.3. Extract load identification characteristics: active power increment sequence ΔP m , reactive power increment sequence ΔQ m , current third harmonic increment sequence ΔI 3m ;

[0012] S3.4. Complete load identification by detecting the transient process of the fixed-frequency load and matching the load characteristic quantity.

[0013] As a further improvement of the present technical solution, in S1, the method for extracting the power increment is:

[0014] The main steady-state power characteristics are: steady-state active power P and steady-state reactive power Q. The change of steady-state power during the operation of electrical equipment is characterized by the active power increment ΔP and the reactive power increment ΔQ. The power increment is obtained by calculating the discrete steady-state power sequence P. i , Q i Perform the difference operation and the calculation formula is as follows:

[0015]

[0016] Where i is the sampling point number of the power sequence, and j is a user-defined integer value, which is limited in selection.

[0017] As a further improvement of the present technical solution, in S1, the method for extracting the steady-state power characteristics is:

[0018] Steady-state power characteristics include the slow change and fluctuation of steady-state active power;

[0019] First, the slow variation of active power:

[0020] First, the discrete active power sequence P i Preprocessing is performed to reduce the impact of fluctuations during the slow change process. The data preprocessing method is to perform preprocessing on the active power sequence P i Sum and average to obtain the active power sequence P Nk, and its calculation formula is as follows:

[0021]

[0022] Where n is the discrete active power sequence P used in each preprocessing calculation i the number of

[0023] Secondly, for the active power sequence P Nk Calculate the power increment, power increment ΔP Nk Will form the incremental sequence:

[0024] ΔP Nk =P Nk+1 -P Nk (3)

[0025] Finally, the active power slow change property is judged, that is, when there are continuous T N More than ΔP Nk+n The values ​​are all greater than the set threshold T P , then it is determined that there is a load with slowly varying power entering at this time, as shown in formula (4):

[0026] ΔP Nk+n >T P and n>T N (4)

[0027] Where n = 0, 1, 2, T P is the power increment threshold, T N It is also the threshold set after multiple trials and learning;

[0028] Second, active power fluctuation:

[0029] The main idea behind calculating active power fluctuation is to construct a sliding window to monitor the fluctuation within the window, then extend the sliding window's time scale, analyze the changes in power fluctuation values ​​over long time scales, and set a threshold to determine whether there is a load with large power fluctuations. The implementation steps are as follows:

[0030] A. Constructing the power sliding window W[P ik+1 :P i( k +1) ]Monitor the steady-state active power sequence P, k=0,1,2,..., sliding window W[P ik+1 :P i( k +1) ] contains i discrete active power points;

[0031] B. Calculate the difference between the maximum active power values ​​in the bilateral window within the sliding window D t , and record the value into the difference sequence D;

[0032] C. Reconstruct another difference sliding window W[D mt+1 :D m( t +1) ]Monitoring difference sequence D, t=0,1,2,.., sliding window W[D mt+1 :D m( t +1) ] contains m power difference values;

[0033] D. The power difference within the statistical difference sliding window is D>D T The number of this condition N, where D T This is the threshold set after many experiments. If N>m / 3, it is determined that a load with large active power fluctuation is running in this window.

[0034] As a further improvement of the present technical solution, in S1, the method for extracting the third harmonic increment is:

[0035] The harmonic current formula of discrete current is shown in formula (5):

[0036]

[0037] Among them, FFT represents the fast Fourier transform operation, N s is the number of samples in the current sequence participating in this FFT operation, i[:] is the current sequence participating in this FFT operation, F i [:] is the dimension of i[:] after FFT operation, which is N s The amplitude result, f i (j) is the actual frequency result sequence, i j is the jth steady-state current harmonic;

[0038] The amplitude of the third harmonic current is the largest compared to other frequency harmonics, and the third harmonic current increment ΔI 3m The calculation formula is as follows:

[0039] ΔI 3m =I 3(m+j) -I 3m (6)

[0040] Where, I 3m It is the steady-state third harmonic current, and the choice of j value has certain restrictions.

[0041] As a further improvement of this technical solution, in S1, the method for extracting the load working time is:

[0042] The load time characteristics of different electrical equipment are very different. The load working time can be divided into continuous working time and intermittent working time.

[0043] First, continuous working time: the non-electrical differences between different electrical appliances are reflected in the working time T w Therefore, the continuous working time can be used as a non-electrical load identification feature:

[0044] T w =T off -T on (7)

[0045] Among them, T off is the moment when the load stops running, T on is the moment when the load starts to operate;

[0046] Second, intermittent working time: The formula for the working time of the kth segment (the k value is also the number of intermittent load operations) of the electrical equipment during operation is as follows:

[0047] T k =T off(k) -T on(k) 8)

[0048] Where, T off(k) is the starting time of the kth stage of operation of the electrical equipment, T on(k) It is the stopping time of the electrical equipment during the kth segment of operation.

[0049] As a further improvement of the present technical solution, in S1, the transient power surge is the product of the instantaneous voltage u(t) and the instantaneous current i(t), and the calculation formula is as follows:

[0050] P t =u(t)×i(t) (9)

[0051] Among them, P t It is a transient power shock.

[0052] As a further improvement of the present technical solution, in S2, the process of constructing a non-intrusive user-side load identification architecture based on decision tree classification includes the following steps:

[0053] S2.1. Extract the characteristics of the load to be identified;

[0054] S2.2. Classify the load samples to be identified into two categories according to the slow variation of active power: fixed-frequency load and variable-frequency load;

[0055] S2.3. For fixed frequency load:

[0056] (1) Classify loads A and B according to the fluctuation of active power;

[0057] (2) According to the transient power impact classification, Class A loads are divided into Class C loads with constant frequency loads accompanied by power impacts during transient processes and Class D loads with constant frequency loads without transient power impacts;

[0058] (3) Based on the difference in power increments, including the characteristics of active power and reactive power changes during operation, Class C loads are determined as load e and load f;

[0059] (4) According to the load time characteristics, the Class D load is determined as load g for intermittent operation load and load h for continuous operation load;

[0060] S2.4. For variable frequency loads:

[0061] (1) According to the different power increments during operation, determine the variable frequency load B with stable power change and the class B load with large power change;

[0062] (2) According to the size of the third harmonic component, Class B loads are classified into loads C with no third harmonic changes during operation and loads D with large third harmonic component changes during operation.

[0063] As a further improvement of the present technical solution, in S3.2, the active power fluctuation and active power gradual variation detection of the load sample specifically includes:

[0064] First, active power fluctuation detection:

[0065] Construct power sliding window W[P ik+1 :P i( k +1) ] detects active power sequence, k=0,1,2,..., a single sliding of the sliding window can detect i discrete active power points, that is, W[P ik+1 :P i( k +1) ] contains i power points; calculate the difference D of the maximum active power points in the sliding window T , and enter this value into the difference sequence D; then pass the difference sliding window W[D mt+1 :D m( t +1) ]Statistical sliding window satisfies D>D T If N>m / 3, it is determined that there is a load with large active power fluctuation; then the total working time T of the load with large fluctuation is calculated. w , if T w >T T Then determine that the washing machine is in T w Time period operation, where T T It is a threshold value adjusted through experiments, which can distinguish the length of load working time;

[0066] Second, active power slow change detection:

[0067] For active sequence P mThe sum of the i power points is averaged to obtain a new discrete active power sequence P Nk , and then P Nk The power ramp sequence ΔP is obtained by performing differential analysis on the sequence one by one. Nk ; If there are continuous T N More than ΔP Nk Satisfy ΔP Nk >T P When , it is determined that there is a load with slow power change starting at this time; calculate the total power increase P B , Q B , the total rise of the 3rd harmonic component I 3B and load starting time T B Detect reactive power increment Q B Does Q meet B >Q T If the condition is met, it is determined that the variable frequency air conditioner is started, otherwise it is determined that the IH rice cooker is running, where Q T It is a threshold value set through experiments and can distinguish the size of the load reactive power increment;

[0068] Third, fixed frequency load:

[0069] The load that does not meet the active power fluctuation detection and active power slow change detection is a fixed-frequency load. For this type of load, identification is achieved through steps S3.3-S3.4.

[0070] As a further improvement of the present technical solution, in S3.4, completing load identification by detecting the matching between the transient process of the fixed-frequency load and the load characteristic quantity specifically includes:

[0071] S3.4.1. Setting the transient power change threshold P AT , if the active power increment sequence satisfies ΔP m >P AT , it means that there is a load with a power step increase at this time, and the total increase in active power P is calculated. u , the total increase in reactive power Q u , the total rise of the third harmonic current I 3u and load starting time T u , and store these characteristic quantities in the entry list of the fixed frequency load; calculate the power impact coefficient k according to the formula p , if k p Greater than the threshold k set by test pt , it is judged that the normal air conditioner is started;

[0072] S3.4.2. If the active power increment sequence satisfies ΔP m <-P ATWhen , it means that there is a load with a power step decrease. Calculate the total decrease in active power Pd and the total decrease in reactive power Q d , the total drop of the third harmonic current I 3d and load stop time T d and store these values ​​in the load exit list.

[0073] As a further improvement of the technical solution, in S3.4.2, storing the calculated value in the load exit list specifically includes:

[0074] (A) Scan the characteristic quantities of the load entry list and the load exit list, and perform comparison and matching based on the characteristics that the power change value is equal when the load is put into and exited, that is, the same load P u ≈P d , Q u ≈Q d , I 3u ≈I 3d , thus completing the load identification, the load working time T W =T d -T u ;

[0075] (B) Differentiate and identify similar loads based on their temporal characteristics, classifying loads into intermittent workloads and continuous workloads;

[0076] a. Perform reactive power detection on continuous workload. If Q d >Q T , then the load is considered to be an induction cooker; if the load also meets Q d <-Q T and I 3d <-I 3T The load can be considered as air conditioning; if the power impact multiple k p Satisfy k p >k pt , then the load is judged to be a fixed frequency air conditioner, otherwise it is judged to be a variable frequency air conditioner. d <-Q T and I 3d <-I 3T , if the load working time T W Meet T W >T T , the load is considered to be an electric water heater, otherwise it is determined to be an electric kettle.

[0077] b. Intermittent loads can be divided into equal intermittent loads and non-equal intermittent loads according to the working interval length; the load operation mode conforms to equal intermittent operation, if Q is satisfied d <-Q T , I3d <-I 3T , then it is considered that there is a microwave oven, otherwise it is determined to be an electric oven; if the load operation mode meets the non-uniform intermittent operation, then it is determined that there is an electric rice cooker.

[0078] Compared with the prior art, the present invention has the following beneficial effects:

[0079] 1. This model-based, data-driven, non-intrusive user-side load identification method addresses the issue of achieving accurate and efficient non-intrusive user-side load identification. First, a method for extracting the steady-state and transient characteristics of typical electrical equipment is proposed. Next, a non-intrusive user-side load identification architecture based on decision tree classification is proposed. Finally, a multi-feature fusion non-intrusive user-side load identification algorithm is proposed. Finally, a case study is conducted on a typical residential user. By extracting the steady-state and transient characteristics of typical electrical equipment, the method identifies multi-dimensional features of load samples based on the decision tree classification criterion. Non-intrusive load identification is achieved by matching the features of similar loads.

[0080] 2. In the non-intrusive user-side load identification method that combines this model with data-driven analysis, as living standards improve, more and more electrical devices appear in daily life. By establishing an identification criterion based on decision tree classification, various types of electrical devices can be quickly decomposed from user samples.

[0081] 3. In the user-side load non-intrusive identification method that combines this model with data-driven, the fusion of steady-state and transient multi-features of electrical equipment is applied to the load non-intrusive identification algorithm, which can improve the accuracy of the algorithm to a certain extent and significantly reduce the algorithm calculation amount and time. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 This is an exemplary architecture diagram of non-intrusive identification of user-side load based on decision tree classification in the present invention;

[0083] Figure 2 This is a flowchart of an exemplary multi-feature fusion user-side load non-intrusive identification algorithm in the present invention;

[0084] Figure 3 This is an active power variation diagram of a typical residential user load condition exemplified in the present invention;

[0085] Figure 4 This is a reactive power variation diagram of a typical residential user load condition exemplified in the present invention;

[0086] Figure 5 This is a diagram showing changes in the third harmonic component of a typical residential user load condition exemplified in the present invention. DETAILED DESCRIPTION

[0087] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0088] Example 1

[0089] like Figure 1-Figure 2 As shown, this embodiment provides a non-intrusive user-side load identification method that combines a model with data driving, including the following steps:

[0090] S1. Extraction of steady-state and transient characteristics of typical electrical equipment: The steady-state characteristics of typical electrical equipment include power increment, steady-state power, third harmonic increment, and load working time; the transient characteristics of typical electrical equipment include transient power impact;

[0091] In this step, the power increment is extracted as follows:

[0092] The main steady-state power characteristics are: steady-state active power P and steady-state reactive power Q. The change of steady-state power during the operation of electrical equipment is characterized by the active power increment ΔP and the reactive power increment ΔQ. The power increment is obtained by calculating the discrete steady-state power sequence P. i , Q i Perform the difference operation and the calculation formula is as follows:

[0093]

[0094] Where i is the sampling point number of the power sequence, and j is a user-defined integer value with limited selection. If j is too large, switching events in a very short time will not be captured, which will have a negative impact on the resolution of the algorithm. If j is too small, the load power change will be incompletely captured, and the algorithm's computational complexity will be greatly increased, reducing the program's computational speed.

[0095] Furthermore, the method for extracting the steady-state power characteristics is:

[0096] Steady-state power characteristics refer to the slow change and fluctuation of steady-state active power;

[0097] First, the slow variation of active power:

[0098] First, the discrete active power sequence P i Preprocessing is performed to reduce the impact of fluctuations during the slow change process. The data preprocessing method is to perform preprocessing on the active power sequence P i Sum and average to obtain the active power sequence P Nk, and its calculation formula is as follows:

[0099]

[0100] Where n is the discrete active power sequence P used in each preprocessing calculation i the number of

[0101] Secondly, for the active power sequence P Nk Calculate the power increment, power increment ΔP Nk Will form the incremental sequence:

[0102] ΔP Nk =P Nk+1 -P NK (3)

[0103] Finally, the active power slow change property is judged, that is, when there are continuous T N More than ΔP Nk+n The values ​​are all greater than the set threshold T P , then it is determined that there is a load with slowly varying power entering at this time, as shown in formula (4):

[0104] ΔP Nk+n >T P and n>T N (4)

[0105] Where n = 0, 1, 2, T P is the power increment threshold, T N It is also the threshold set after multiple trials and learning;

[0106] Second, active power fluctuation:

[0107] The main idea behind calculating active power fluctuation is to construct a sliding window to monitor the fluctuation within the window, then extend the sliding window's time scale, analyze the changes in power fluctuation values ​​over long time scales, and set a threshold to determine whether there is a load with large power fluctuations. The implementation steps are as follows:

[0108] A. Constructing the power sliding window W[P ik+1 :P i(k+1) ]Monitor the steady-state active power sequence P, k=0,1,2,..., sliding window W[P ik+1 :P i(k+1) ] contains i discrete active power points;

[0109] B. Calculate the difference between the maximum active power values ​​in the bilateral window within the sliding window D t , and record the value into the difference sequence D;

[0110] C. Reconstruct another difference sliding window W[D mt+1:D m( t +1) ]Monitoring difference sequence D, t=0,1,2,.., sliding window W[D mt+1 :D m( t +1) ] contains m power difference values;

[0111] D. The power difference within the statistical difference sliding window is D>D T The number of this condition N, where D T This is the threshold set after many experiments. If N>m / 3, it is determined that a load with large active power fluctuation is running in this window.

[0112] Furthermore, the method for extracting the third harmonic increment is:

[0113] The harmonic current formula of discrete current is shown in formula (5):

[0114]

[0115] Among them, FFT represents the fast Fourier transform operation, N s is the number of samples in the current sequence participating in this FFT operation, i[:] is the current sequence participating in this FFT operation, F i [:] is the dimension of i[:] after FFT operation, which is N s The amplitude result, f i (j) is the actual frequency result sequence, i j is the jth steady-state current harmonic;

[0116] The amplitude of the third harmonic current is the largest compared to other frequency harmonics, and the third harmonic current increment ΔI 3m The calculation formula is as follows:

[0117] ΔI 3m =I 3(m+j) -I 3m (6)

[0118] Where, I 3m It is the steady-state third harmonic current, and the choice of j value has certain restrictions.

[0119] Furthermore, the method for extracting the load working time is:

[0120] The load time characteristics of different electrical equipment are very different. The load working time can be divided into continuous working time and intermittent working time.

[0121] First, continuous working time: the non-electrical differences between different electrical appliances are reflected in the working time T w Therefore, the continuous working time can be used as a non-electrical load identification feature:

[0122] T w =T off -T on (7

[0123] Among them, T off is the moment when the load stops running, T on is the moment when the load starts to operate;

[0124] Second, intermittent working time: The formula for the working time of the kth segment (the k value is also the number of intermittent load operations) of the electrical equipment during operation is as follows:

[0125] T k =T off(k) -T on(k) (8)

[0126] Where, T off(k) is the starting time of the kth stage of operation of the electrical equipment, T on(k) It is the stopping time of the electrical equipment during the kth segment of operation.

[0127] Furthermore, the transient power surge is the product of the instantaneous voltage u(t) and the instantaneous current i(t), and the calculation formula is as follows:

[0128] P t =u(t)×i(t) (9)

[0129] Among them, P t It is a transient power shock.

[0130] S2. Construct a non-intrusive user-side load identification architecture based on decision tree classification;

[0131] like Figure 2 As shown in FIG, the process of constructing a non-intrusive user-side load identification architecture based on decision tree classification includes the following steps:

[0132] S2.1. Extract the characteristics of the load to be identified;

[0133] S2.2. Classify the load samples to be identified into two categories according to the slow variation of active power: fixed-frequency load and variable-frequency load;

[0134] S2.3. For fixed frequency load:

[0135] (1) Classify loads A and B according to the fluctuation of active power;

[0136] (2) According to the transient power impact classification, Class A loads are divided into Class C loads (constant frequency loads with transient power impacts) and Class D loads (constant frequency loads without transient power impacts);

[0137] (3) Based on the difference in power increments, including the characteristics of active power and reactive power changes during operation, Class C loads are determined as load e and load f;

[0138] (4) According to the load time characteristics, the Class D load is divided into load g (intermittent operation load) and load h (continuous operation load);

[0139] S2.4. For variable frequency loads:

[0140] (1) According to the different power increments during operation, determine the variable frequency load B with stable power change and the class B load with large power change;

[0141] (2) According to the size of the third harmonic component, Class B loads are determined as load c (no third harmonic change during operation) and load d (large third harmonic component change during operation).

[0142] S3. Applying a multi-feature fusion user-side load non-intrusive identification algorithm, specifically including the following steps:

[0143] S3.1. Collect and pre-process the voltage and current at the user's main incoming line to obtain the basic characteristic quantity of the load: active power sequence P m , reactive power sequence Q m , third harmonic current sequence I 3m ;

[0144] S3.2. Detect active power fluctuation and active power slow change of load samples to distinguish between fluctuating loads, slow-changing loads, and fixed-frequency loads;

[0145] In this step, the active power fluctuation and active power gradual change detection of the load sample specifically includes:

[0146] First, active power fluctuation detection:

[0147] Construct power sliding window W[P ik+1 :P i(k+1) ] detects active power sequence, k=0,1,2,..., a single sliding of the sliding window can detect i discrete active power points, that is, W[P ik+1 :P i(k+1) ] contains i power points; calculate the difference D of the maximum active power points in the sliding window T , and enter this value into the difference sequence D; then pass the difference sliding window W[D mt+1 :D m(t+1) ]Statistical sliding window satisfies D>D T If N>m / 3, it is determined that there is a load with large active power fluctuation; then the total working time T of the load with large fluctuation is calculated. w , if Tw >T T Then determine that the washing machine is in T w Time period operation, where T T It is a threshold value adjusted through experiments, which can distinguish the length of load working time;

[0148] Second, active power slow change detection:

[0149] For active sequence P m The sum of the i power points is averaged to obtain a new discrete active power sequence P Nk , and then P Nk The power ramp sequence ΔP is obtained by performing differential analysis on the sequence one by one. Nk ; If there are continuous T N More than ΔP NK Satisfy ΔP NK >T P When , it is determined that there is a load with slow power change starting at this time; calculate the total power increase P B , Q B , the total rise of the 3rd harmonic component I 3B and load starting time T B Detect reactive power increment Q B Does Q meet B >Q T If the condition is met, it is determined that the variable frequency air conditioner is started, otherwise it is determined that the IH rice cooker is running, where Q T It is a threshold value set through experiments and can distinguish the size of the load reactive power increment;

[0150] Third, fixed frequency load:

[0151] The load that does not meet the active power fluctuation detection and active power slow change detection is a fixed-frequency load. For this type of load, identification is achieved through steps S3.3-S3.4.

[0152] S3.3. Extract load identification characteristics: active power increment sequence ΔP m , reactive power increment sequence ΔQ m , current third harmonic increment sequence ΔI 3m ;

[0153] S3.4. Complete load identification by detecting the transient process of fixed-frequency loads and matching the load characteristic quantities;

[0154] In this step, load identification is completed by detecting the transient process of the fixed-frequency load and matching the load characteristic quantity, specifically including:

[0155] S3.4.1. Setting the transient power change threshold PAT , if the active power increment sequence satisfies ΔP m >P AT , it means that there is a load with a power step increase at this time, and the total increase in active power P is calculated. u , the total increase in reactive power Q u , the total rise of the third harmonic current I 3u and load starting time T u , and store these characteristic quantities in the entry list of the fixed frequency load; calculate the power impact coefficient k according to the formula p , if k p Greater than the threshold k set by test pt , it is judged that the normal air conditioner is started;

[0156] S3.4.2. If the active power increment sequence satisfies ΔP m <-P AT When , it means that there is a load with a power step drop. Calculate the total drop in active power P d , the total drop in reactive power Q d , the total drop of the third harmonic current I 3d and load stop time T d and store these values ​​in the load exit list;

[0157] (A) Scan the characteristic quantities of the load entry list and the load exit list, and perform comparison and matching based on the characteristics that the power change value is equal when the load is put into and exited, that is, the same load P u ≈P d , Q u ≈Q d , I 3u ≈I 3d , thus completing the load identification, the load working time T W =T d -T u ;

[0158] (B) Differentiate and identify similar loads based on their temporal characteristics, classifying loads into intermittent workloads and continuous workloads;

[0159] a. Perform reactive power detection on continuous workload. If Q d >Q T , then the load is considered to be an induction cooker; if the load also meets Q d <-Q T and I 3d <-I 3T The load can be considered as air conditioning; if the power impact multiple k p Satisfy k p >k pt, then the load is judged to be a fixed frequency air conditioner, otherwise it is judged to be a variable frequency air conditioner. d <- QT and I 3d <-I 3I , if the load working time T W Meet T W >T T , the load is considered to be an electric water heater, otherwise it is determined to be an electric kettle.

[0160] b. Intermittent loads can be divided into equal intermittent loads and non-equal intermittent loads according to the working interval length; the load operation mode conforms to equal intermittent operation, if Q is satisfied d <-Q T , I 3d <-I 3T , then it is considered that there is a microwave oven, otherwise it is determined to be an electric oven; if the load operation mode meets the non-uniform intermittent operation, then it is determined that there is an electric rice cooker.

[0161] Case Study Examples

[0162] like Figure 3-Figure 5 As shown in the figure, a typical residential user is selected for case analysis. The specific steps are as follows:

[0163] Select typical residential user load conditions such as Figure 3-Figure 5 As shown, Figure 3 、 Figure 4 The power changes under this condition are shown. Figure 3 The A, B, C, D, E, and F marked in the figure refer to electric kettle, electric water heater, ordinary air conditioner, electric rice cooker, microwave oven, and induction cooker respectively. Figure 5 The identification results are shown in Table 1 below.

[0164] Table 1 Identification of load operating conditions

[0165]

[0166] As can be seen from Table 1, this technical solution accurately identifies the load switching moment with an error of 0s, and the load power identification accuracy is ≥96.54.

[0167] Those skilled in the art will appreciate that the process of implementing all or part of the steps of the above embodiments may be accomplished by hardware, or by instructing related hardware through a program.

[0168] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A non-intrusive user-side load identification method combining model and data drive, characterized in that: The steps include: S1. Extraction of steady-state and transient characteristics of typical electrical equipment: The steady-state characteristics of typical electrical equipment include power increment, steady-state power, third harmonic increment, and load operating time; the transient characteristics of typical electrical equipment include transient power surges; S2. Construct a non-intrusive user-side load identification architecture based on decision tree classification; S3. Applying a multi-feature fusion user-side load non-intrusive identification algorithm, specifically including the following steps: S3.

1. Collect and pre-process the voltage and current at the user's main incoming line to obtain the basic characteristic quantity of the load: active power sequence P m , reactive power sequence Q m , third harmonic current sequence I 3m ; S3.

2. Detect active power fluctuation and active power slow change of load samples to distinguish between fluctuating loads, slow-changing loads, and fixed-frequency loads; S3.

3. Extract load identification characteristics: active power increment sequence ΔP m , reactive power increment sequence ΔQ m , current third harmonic increment sequence ΔI 3m ; S3.

4. Complete load identification by detecting the transient process of the fixed-frequency load and matching the load characteristic quantity.

2. The non-intrusive user-side load identification method combining a model and data-driven approach according to claim 1 is characterized by: In S1, the method for extracting the power increment is: The main steady-state power characteristics are: steady-state active power P and steady-state reactive power Q. The change of steady-state power during the operation of electrical equipment is characterized by the active power increment ΔP and the reactive power increment ΔQ. The power increment is obtained by calculating the discrete steady-state power sequence P. i , Q i Perform the difference operation and the calculation formula is as follows: Where i is the sampling point number of the power sequence, and j is a custom integer value.

3. The non-intrusive user-side load identification method combining a model and data-driven approach according to claim 2 is characterized by: In S1, the method for extracting the steady-state power characteristics is: Steady-state power characteristics include the slow change and fluctuation of steady-state active power; First, the slow variation of active power: First, the discrete active power sequence P i Preprocessing is performed to reduce the impact of fluctuations during the slow change process. The data preprocessing method is to perform preprocessing on the active power sequence P i Sum and average to obtain the active power sequence P Nk , and its calculation formula is as follows: Where n is the discrete active power sequence P used in each preprocessing calculation i the number of Secondly, for the active power sequence P Nk Calculate the power increment, power increment ΔP Nk Will form the incremental sequence: ΔP Nk =P Nk+1 -P Nk (3) Finally, the active power slow change property is judged, that is, when there are continuous T N More than ΔP Nk+n The values ​​are all greater than the set threshold T P , then it is determined that there is a load with slowly varying power entering at this time, as shown in formula (4): ΔP Nk+n >T P and n > T N (4) Where n = 0, 1, 2, T P is the power increment threshold, T N It is also the threshold set after multiple trials and learning; Second, active power fluctuation: The main idea behind calculating active power fluctuation is to construct a sliding window to monitor the fluctuation within the window, then extend the sliding window's time scale, analyze the changes in power fluctuation values ​​over long time scales, and set a threshold to determine whether there is a load with large power fluctuations. The implementation steps are as follows: A. Constructing the power sliding window W[P ik+1 :P i(k+1) ]Monitor the steady-state active power sequence P, k=0,1,2,..., sliding window W[P ik+1 :P i(k+1) ] contains i discrete active power points; B. Calculate the difference between the maximum active power values ​​in the bilateral window within the sliding window D t , and record the value into the difference sequence D; C. Reconstruct another difference sliding window W[D mt+1 :D m(t+1) ]Monitoring difference sequence D, t=0,1,2,.., sliding window W[D mt+1 :D m(t+1) ] contains m power difference values; D. The power difference within the statistical difference sliding window is D>D T The number of this condition N, where D T This is the threshold set after many experiments. If N>m / 3, it is determined that a load with large active power fluctuation is running in this window.

4. The non-intrusive user-side load identification method combining a model and data-driven approach according to claim 3 is characterized by: In S1, the method for extracting the third harmonic increment is: The harmonic current formula of discrete current is shown in formula (5): Among them, FFT represents the fast Fourier transform operation, N s is the number of samples in the current sequence participating in this FFT operation, i[:] is the current sequence participating in this FFT operation, F i [:] is the dimension of i[:] after FFT operation, which is N s The amplitude result, f i (j) is the actual frequency result sequence, i j is the jth steady-state current harmonic; The amplitude of the third harmonic current is the largest compared to other frequency harmonics, and the third harmonic current increment ΔI 3m The calculation formula is as follows: ΔI 3m =I 3(m+j) -I 3m (6) Where, I 3m It is the steady-state third harmonic current, and the choice of j value has certain restrictions.

5. The non-intrusive user-side load identification method combining a model and data-driven approach according to claim 4 is characterized by: In S1, the method for extracting the load working time is: The load time characteristics of different electrical equipment are very different. The load working time can be divided into continuous working time and intermittent working time. First, continuous working time: the non-electrical differences between different electrical appliances are reflected in the working time T w Therefore, the continuous working time can be used as a non-electrical load identification feature: T w =T off -T on (7) Among them, T off is the moment when the load stops running, T on is the moment when the load starts to operate; Second, intermittent working time: The formula for the working time of the kth segment of the electrical equipment during operation is as follows: T k =T off(k) -T on(k) (8) Where, T off(k) is the starting time of the kth stage of operation of the electrical equipment, T on(k) It is the stopping time of the electrical equipment during the kth segment of operation.

6. The non-intrusive user-side load identification method combining a model and data-driven approach according to claim 5 is characterized by: In S1, the transient power surge is the product of the instantaneous voltage u(t) and the instantaneous current i(t), and the calculation formula is as follows: P t =u(t)×i(t) (9) Among them, P t It is a transient power shock.

7. The non-intrusive user-side load identification method combining a model and data-driven approach according to claim 6 is characterized by: In S2, the process of constructing a non-intrusive user-side load identification architecture based on decision tree classification includes the following steps: S2.

1. Extract the characteristics of the load to be identified; S2.

2. Classify the load samples to be identified into two categories according to the slow variation of active power: fixed-frequency load and variable-frequency load; S2.

3. For fixed frequency load: (1) Classify loads A and B according to the fluctuation of active power; (2) According to the transient power impact classification, Class A loads are divided into Class C loads with constant frequency loads accompanied by power impacts during transient processes and Class D loads with constant frequency loads without transient power impacts; (3) Based on the difference in power increments, including the characteristics of active power and reactive power changes during operation, Class C loads are determined as load e and load f; (4) According to the load time characteristics, the Class D load is determined as load g for intermittent operation load and load h for continuous operation load; S2.

4. For variable frequency loads: (1) According to the different power increments during operation, determine the variable frequency load B with stable power change and the class B load with large power change; (2) According to the size of the third harmonic component, Class B loads are classified into loads C with no third harmonic changes during operation and loads D with large third harmonic component changes during operation.

8. The non-intrusive user-side load identification method combining model and data drive according to claim 7 is characterized by: In S3.2, the active power fluctuation and active power gradual change detection of the load sample specifically includes: First, active power fluctuation detection: Construct power sliding window W[P ik+1 :P i(k+1) ] detects active power sequence, k=0,1,2,..., a single sliding of the sliding window can detect i discrete active power points, that is, W[P ik+1 :P i(k+1) ] contains i power points; calculate the difference D of the maximum active power points in the sliding window T , and enter this value into the difference sequence D; then pass the difference sliding window W[D mt+1 :D m(t+1) ]Statistical sliding window satisfies D>D T If N>m / 3, it is determined that there is a load with large active power fluctuation; then the total working time T of the load with large fluctuation is calculated. w , if T w >T T Then determine that the washing machine is in T w Time period operation, where T T It is a threshold value adjusted through experiments, which can distinguish the length of load working time; Second, active power slow change detection: For active sequence P m The sum of the i power points is averaged to obtain a new discrete active power sequence P Nk , and then P Nk The power ramp sequence ΔP is obtained by performing differential analysis on the sequence one by one. Nk ; If there are continuous T N More than ΔP Nk Satisfy ΔP Nk >T P When , it is determined that there is a load with slow power change starting at this time; calculate the total power increase P B , Q B , the total rise of the 3rd harmonic component I 3B and load starting time T B Detect reactive power increment Q B Does Q meet B >Q T If the condition is met, it is determined that the variable frequency air conditioner is started, otherwise it is determined that the IH rice cooker is running, where Q T It is a threshold value set through experiments and can distinguish the size of the load reactive power increment; Third, fixed frequency load: The load that does not meet the active power fluctuation detection and active power slow change detection is a fixed-frequency load. For this type of load, identification is achieved through steps S3.3-S3.

4.

9. The non-intrusive user-side load identification method combining model and data drive according to claim 8 is characterized by: In S3.4, completing load identification by detecting the matching between the transient process of the fixed-frequency load and the load characteristic quantity specifically includes: S3.4.

1. Setting the transient power change threshold P AT , if the active power increment sequence satisfies ΔP m >P AT , it means that there is a load with a power step increase at this time, and the total increase in active power P is calculated. u , the total increase in reactive power Q u , the total rise of the third harmonic current I 3u and load starting time T u , and store these characteristic quantities in the entry list of the fixed frequency load; calculate the power impact coefficient k according to the formula p , if k p Greater than the threshold k set by test pt , it is judged that the normal air conditioner is started; S3.4.

2. If the active power increment sequence satisfies ΔP m <-P AT When , it means that there is a load with a power step drop. Calculate the total drop in active power P d , the total drop in reactive power Q d , the total drop of the third harmonic current I 3d and load stop time T d and store these values ​​in the load exit list.

10. The non-intrusive user-side load identification method combining model and data drive according to claim 9 is characterized in that: In S3.4.2, storing the calculated value in the load exit list specifically includes: (A) Scan the characteristic quantities of the load entry list and the load exit list, and perform comparison and matching based on the characteristics that the power change value is equal when the load is put into and exited, that is, the same load P u ≈P d , Q u ≈Q d , I 3u ≈I 3d , thus completing the load identification, the load working time T W =T d -T u ; (B) Differentiate and identify similar loads based on their temporal characteristics, classifying loads into intermittent workloads and continuous workloads; a. Perform reactive power detection on continuous workload. If Q d >Q T , then the load is considered to be an induction cooker; if the load also meets Q d <-Q T and I 3d <-I 3T The load can be considered as air conditioning; if the power impact multiple k p Satisfy k p >k pt , then the load is judged to be a fixed frequency air conditioner, otherwise it is judged to be a variable frequency air conditioner. d <-Q T and I 3d <-I 3T , if the load working time T W Meet T W >T T , the load is considered to be an electric water heater, otherwise it is determined to be an electric kettle. b. Intermittent loads can be divided into equal intermittent loads and non-equal intermittent loads according to the working interval length; the load operation mode conforms to equal intermittent operation, if Q is satisfied d <-Q T , I 3d <-I 3T , then it is considered that there is a microwave oven, otherwise it is determined to be an electric oven; if the load operation mode meets the non-uniform intermittent operation, then it is determined that there is an electric rice cooker.

Images