A non-intrusive evaluation method for load time-shifting capability based on Markov chain

Abstract

The present invention relates to a non-invasive evaluation method for load time-shifting capability based on a Markov chain. Based on a large amount of historical data, an improved Markov chain state transfer matrix and a load transfer power calculation method are constructed. After inputting the power curve of the day, the transferable power of the user at each moment can be calculated, providing a reference for the user's real-time power adjustment. Based on the present invention, the transferable and absorbable power of the day can be further calculated, the time-shifting capability of the user's load can be obtained, and the time-shifting capability of the user's daily electricity consumption can be evaluated. The method does not need to master the device data in the user in advance and has high accuracy, providing technical support for guiding the user to reasonably arrange the electricity consumption time and formulate energy-saving and economical electricity consumption plans.

Description

Technical Field

[0001] The invention relates to the field of energy consumption evaluation of power users, and in particular to a non-invasive evaluation method for load time-shifting capability based on a Markov chain. Background Art

[0002] Smart electricity consumption is an inevitable trend in the development of power grid technology and an inevitable requirement for building an economic, low-carbon, and environmentally friendly society. In recent years, in particular, the perception, communication, and computing capabilities of power equipment have developed rapidly, providing strong technical support for power companies to promote smart electricity consumption. One of the key links in implementing user smart electricity consumption is the need to analyze user energy consumption characteristics, understand the electricity consumption behaviors of various types of users, effectively evaluate the user's adjustable load potential, and guide users to reasonably arrange electricity consumption time, so as to formulate energy-saving and low-cost electricity consumption plans for users in a targeted manner. This technology complies with the clean and low-carbon social development needs of the power grid and the government. At present, based on the configuration of intelligent monitoring and collection systems, using big data analysis technology, constructing user energy analysis and evaluation methods has become an important technical direction.

[0003] At present, load detection and analysis are mainly divided into invasive and non-invasive types. Among them, non-invasive means that there is no need to enter the user's interior to detect each energy-consuming device, but to build an intelligent power consumption detection and analysis terminal at the home port to analyze the user's energy consumption data, which has the advantages of being more efficient, convenient and low-cost. At present, non-invasive load energy consumption analysis is mostly centered on the identification of user energy-consuming devices - by building a steady-state or transient waveform library of various user devices, such as active power, reactive power, voltage, current and other steady-state characteristics, establishing transient characteristics such as instantaneous active power, instantaneous current, harmonic starting waveforms; then collecting the current power waveform, based on the superposition principle, using dynamic time bending distance, similarity, clustering algorithms, etc., to identify the current energy-consuming device from the waveform library, so as to specifically grasp the user's energy consumption situation. However, in actual collection, the types, models, quantities, user behaviors and other aspects of energy-consuming devices involved by different users are different, and there will be large errors in the identification of the internal equipment situation of the user who is not clear in advance.

[0004] Regarding the evaluation of load time-shifting capability, existing research is based on massive historical data analysis, which divides loads into basic loads, time-shiftable loads, and adjustable loads. Wavelet method is used to extract data features such as the maximum adjustable rate, adjustment depth, and maximum adjustable time. Finally, fuzzy mean clustering is used to compare with the loads in the database to calculate the current proportion of time-shiftable loads. However, this method requires knowing the proportion of each type of load in the database at each moment in advance, has high requirements for data transparency, and will have errors in user identification of unknown load components.

[0005] In addition, there are also studies based on historical data, using hidden Markov models to simulate and analyze the status of household load switches, and obtaining a household switch status model, which can determine the start and stop status of household switches in real time. However, historical data also requires the number of household switches and the corresponding start and stop status to be known in advance.

[0006] Therefore, there is an urgent need for a load time-shifting capacity assessment method that does not require the user's internal equipment data to be known in advance and has high accuracy. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide a non-invasive evaluation method for load time-shifting capability based on a Markov chain, which does not require the user's internal equipment data to be known in advance and has high accuracy.

[0008] In order to solve the above problems, the present invention discloses a non-invasive evaluation method of load time-shifting capability based on a Markov chain, which includes:

[0009] Based on the historical load data of the user's smart meter, the load data at the same time every day is used to form a one-dimensional Markov chain, and then the load data at all times of the day are divided into M states according to the value size, and the corresponding state transfer matrix Hr of size M×M is constructed;

[0010] The diagonal transfer probability in the state transfer matrix is ​​set to zero, and the state points whose distance from the maximum probability state point exceeds M / 3 states and whose probability value is less than the threshold ε are also set to zero; after the zeroing is completed, the sum of the elements in each row of the matrix is ​​made 1, and the improved state transfer matrix Hr′ is obtained;

[0011] Obtain the load data for the day to be evaluated and determine the load data P at each moment. 0 (t) corresponds to the state, and combines the improved state transfer matrix Hr′ and the fluctuation power to calculate the corresponding theoretical transfer power P move (t), which is used to evaluate the time-shifting capability of loads.

[0012] Preferably, the one-dimensional Markov chain is represented by X Dt ={x 1t ,x 2t ,…,x dt ,…,x Dt}, x dt Represents the data at time t on the dth day, where D is the total number of days; when dividing the states:

[0013]

[0014]

[0015] Where Y M (x dt ) is the element xdt The calculated state number; max{X Dt}、min{X Dt} are vectors X Dt The maximum and minimum values ​​in Δx t is the state distance; ceil{} is rounded up.

[0016] Preferably, the theoretical transfer power P is calculated by the following formula: move (t):

[0017]

[0018] Where λ(x) is the fluctuating power, and h′ is the element of the improved state transfer matrix Hr′.

[0019] Preferably, before officially using the historical load data, the method further includes:

[0020] The spline interpolation method is used to fill in the missing data of ultra-short term. If the missing data exceeds 10 consecutive collection moments, the data of the same period of the adjacent days are used to fill in the missing data.

[0021] Compared with the prior art, the present invention has the following advantages:

[0022] The present invention relies on massive historical electricity load data to improve the Markov state space. On the one hand, it eliminates the situation where the state is trapped and does not transfer itself. On the other hand, it screens and eliminates outlier state points far away from the concentrated area with high accuracy. Then, the improved Markov chain is used to analyze the user's electricity consumption state and state transition probability at each moment, and the user's time-shiftable power value is calculated.

[0023] During the formal evaluation, the load power curve of the day is input, and the reference value of the transferable power at each moment is calculated based on the state transfer matrix, state distance, and fluctuation component of the user at each moment; further, based on the present invention, the transferable and absorbable electricity of the day can be calculated, and the time-shiftability evaluation of the user's daily electricity consumption can be realized, providing a reference for user load regulation. This method does not require the user's equipment data to be mastered in advance and has high accuracy, providing a reference for exploring the user's load regulation potential and guiding the user to reasonably arrange electricity usage time. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings.

[0025] Figure 1 A flow chart of a method provided by an embodiment of the present invention.

[0026] Figure 2 A state transition space matrix provided by an embodiment of the present invention.

[0027] Figure 3 An improved state transition space matrix provided by an embodiment of the present invention.

[0028] Figure 4 The load curve and transferable power value provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0029] The core of the present invention is to propose a non-invasive method for evaluating the time-shifting capacity of user loads. This method obtains a large amount of historical data from the user's total table and conducts mining and analysis to evaluate the time-shifting capacity of the user's power load. It does not need to enter the user's interior to obtain the type, model, quantity, and power consumption data of the energy-consuming equipment, which can save a lot of manpower and physical costs.

[0030] refer to Figure 1 The embodiment of the present invention provides a non-intrusive evaluation method of load time-shifting capability based on a Markov chain, which specifically includes the following contents:

[0031] Step S101: Based on the historical load data of the user's smart meter, the load data at the same time every day is used to form a one-dimensional Markov chain, and then the load data at all times of the day are divided into M states according to the value size, and the corresponding state transfer matrix Hr is constructed.

[0032] (1) Establishing a user time-shiftable evaluation data matrix

[0033] Collect the total historical load data (i.e. active power data) of the user's smart meter, and then use the spline difference method to fill in the ultra-short-term missing data. If the missing data exceeds 10 consecutive collection moments, use the data of the same period of the adjacent day to fill it in. Arrange the filled load data in sequence, select the load data at the same time every day, and form a one-dimensional Markov chain X Dt ={x 1t ,x 2t ,…,x dt ,…,x Dt}, where element x dt Represents the data at time t on the dth day, where D is the total number of days.

[0034] (2) Classification of load states

[0035] Based on all the one-dimensional Markov chains constructed above, the Markov state vector is calculated as follows:

[0036]

[0037]

[0038] Where Y M (x dt) is the element x dt The calculated state number; max{X Dt}、min{X Dt} are vectors X Dt The maximum and minimum values ​​in the state space; M is the dimension of the defined state space; Δx t is the state distance; ceil{} is rounded up.

[0039] Through the above formula, all load data at time t in the history are divided into M states according to the value size.

[0040] (3) Construct the state transfer matrix Hr

[0041] Let the element in matrix St be s ij , the state transfer process is as follows:

[0042] s ij = St(Y M (x dt )=j|Y M (x dt-t )=i) (3)

[0043] In the formula, i and j represent two adjacent data x dt-t 、x dt The corresponding state; s ij is the total number of state transitions from state i to state j.

[0044] Then the element h in the state transfer matrix Hr ij The definition is as follows:

[0045]

[0046]

[0047] In the formula, the size of matrix Hr is M×M, h ij is the transition probability from state i to state j; the numerator s ij is the number of transitions from state i to state j; the denominator represents the total number of transitions from state i to other states.

[0048] Step S102, in the state transfer matrix, the diagonal transition probability is set to zero, and the state points whose distance from the maximum probability state point exceeds M / 3 states and the state points whose probability value is less than the threshold ε are also set to zero; after the zeroing is completed, the sum of the elements in each row of the matrix is ​​made 1, and the improved state transfer matrix Hr′ is obtained.

[0049] First, let’s take an example to illustrate the necessity of improvement. Take the data of a user at 0:00 for the whole year as an example. The state number M is 30. The load state space characteristics of the user at the same time are as follows: Figure 2As shown. Figure 2 It can be seen that the user's power load space state characteristics show "normalization", that is, no matter what the current load state i is, the value range of the state j after the next transfer is [1, 15], that is, the data has a tendency to return to stability. Figure 2 It can be seen that the state matrix has outlier state points far away from the concentrated area as shown by the “circle”, which are caused by abnormal mutation data and will interfere with the calculation of transfer power.

[0050] In order to solve the problem that the transfer probability is easy to fall into its own state, such as still being in state i after transferring from state i, which will affect the calculation of the load power transfer power value, the present invention improves the state transfer matrix and sets the diagonal transfer probability to zero. At the same time, in order to solve the problem of outliers in data distribution and improve the quality of state data, the present invention regards the state with a distance of more than M / 3 states from the maximum probability state point as an outlier state, and the probability value is less than the threshold ε as an outlier, and both are set to zero.

[0051] Furthermore, to ensure that the sum of each row of probabilities is 1 after the above operations, the state transition probability is recalculated based on formula (4).

[0052]

[0053] In the formula, h′ ij is the modified state transfer matrix; h i_max is the maximum state probability of the i-th row; loca(h i_max ) is the position function, indicating h i_max The column position in row i.

[0054] Step S103: Obtain the load data for the day to be evaluated and determine the load data P at each moment. 0 (t) corresponds to the state, and combined with the improved state transfer matrix Hr′ and the fluctuation power, the corresponding theoretical transfer power P is calculated move (t), which is used to evaluate the time-shifting capability of loads.

[0055] Because the power value represented by each state = state distance Δx t ×Number of states Y M (x dt ), so the same state corresponds to a fixed power value, which deviates from the actual load data and requires superposition of fluctuating power. The present invention uses t-locationscale distribution to simulate the fluctuating power in each state, and the formula is as follows:

[0056]

[0057] Where λ(x) is the random analog of active power, μ is the location parameter, δ is the scale parameter, and v is the shape parameter.

[0058] Input the load data of the day and calculate the load P at each moment according to formula (1) and formula (2): 0 (t) corresponds to the state Y M (P 0 (t)), then the improved state transfer matrix Hr′ and the fluctuation power at each moment are calculated to calculate P 0 (t) The formula for theoretical transfer power is as follows:

[0059]

[0060] Where P move (t) is the load P at time t 0 Theoretical transfer power of (t); m·Δx t +λ(x) represents the distance Δx between state m and state t The product of P and the superimposed power fluctuation λ(x) represents the active power value generated in the m state simulation; 0 The state corresponding to (t) is Y M (P 0 (t)), then In the improved state transfer matrix, it represents the state Y M (P 0 (t)) The probability of transitioning to state m; Then it means P 0 (t) is the theoretical power value transferred to state m. By summing them up, we can get the load P at time t. 0 Theoretical transfer power P of (t) move (t).

[0061] Furthermore, by calculating the transfer power at each moment, the transferable power P at each moment of the day can be calculated one by one according to formula (8): move (t), thus obtaining the transferable power and the absorbable transfer power for the whole day. The smallest absolute value of the two can be used to evaluate the load time-shifting capacity of the day. The relevant formula is as follows:

[0062] W move_up =∫P move (t)dt P move (t)>0 (9)

[0063] W move_down =∫P move (t)dt P move (t)<0 (10)

[0064] W move=min{W move_up ,|W move_down |} (11)

[0065] Where W move_up , W move_down and W move They respectively represent the load power that users can transfer to other times, the power that can be absorbed when transferred to other times, and the load power that can be shifted by end users.

[0066] Experimental verification

[0067] First, we obtain the historical electricity consumption data of users. We take residential users as the verification object and collect 5 years of electricity consumption data from the user's total meter. The data interval is 15 minutes. The data volume is 96×1825. Then, we construct load vectors for the data at the same time throughout the year, forming a total of 96 load sequences with a length of 1825. We construct a one-dimensional Markov chain X for each load sequence. Dt ={x 1t ,x 2t ,…,x dt ,…,x Dt}, where D is 1825 days.

[0068] The present invention selects the power load time as 0:00, sets the total state number M to 30, and sets the outlier probability threshold ε to 0.01. Then, the Markov state vector Y is calculated according to the formula M (x dt ) and state distance Δx t , where Δx t The value is 0.1753. Furthermore, the improved load state transfer space matrix is ​​calculated according to the improved Markov chain mathematical model. The state space distribution is as follows: Figure 3 shown.

[0069] contrast Figure 2 It can be seen that the improved state transition space matrix removes the discrete transition state, improves the quality of the state data, and sets the diagonal probability to zero, which is convenient for the next step of calculating the state transition power. At the same time, repeating the above steps at different times can obtain 96 state transition matrices based on the improved Markov chain.

[0070] The next step is to input the user load sequence to be evaluated and evaluate the time-shifting capability of the load on that day. 0 (t) is 96×1, and 96 independent transfer power calculations are performed, and the load P is calculated each time. 0 (t) corresponds to the state Y M (P 0 (t)) and the transferable power P moveAfter calculating the transfer power at each moment, we can obtain the load transferable power, the absorbable power, and the load power that can be shifted by the end user on that day. The load curve and the transferable power value are as follows: Figure 4 shown.

[0071] Figure 4 The bar graph in the upper part of the middle axis indicates the load power that can be transferred by users at each moment, which basically matches the load curve value of the day. The lower part of the axis indicates the power value that can be absorbed at each moment and transferred from other moments, from which the power value that can be shifted at each moment of the day can be evaluated. The calculation result is that the transferable power for the whole day is 39.6170kWh, and the absolute value of the transferable power is 15.5614kWh. Therefore, the user's ability to shift the power on the day takes the minimum value of the two, which is 15.5614kWh.

[0072] The technical solution provided by the present invention is described in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, the present invention can also be improved and modified in a number of ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A non-invasive evaluation method for load time-shifting capability based on Markov chains, characterized in that: The method includes: Based on the historical load data of the user's smart meter, the load data at the same time every day is constructed into a one-dimensional Markov chain, and then the load data at all times of the day are divided into M states according to the value size, and the corresponding state transfer matrix Hr of size M×M is constructed; The diagonal transfer probability in the state transfer matrix is ​​set to zero, and the state points whose distance from the maximum probability state point exceeds M / 3 states and the state points whose probability value is less than the threshold ε are also set to zero; after the zeroing is completed, the sum of the elements in each row of the matrix is ​​made 1, and the improved state transfer matrix Hr' is obtained; Obtain the load data for the day to be evaluated and determine the load data at each moment The corresponding state is calculated by combining the improved state transfer matrix Hr' and the fluctuation power to obtain the corresponding theoretical transfer power , used to evaluate the time-shifting capacity of loads.

2. The method according to claim 1, characterized in that: The one-dimensional Markov chain is expressed as , Indicates d Heaven t Time data, D is the total number of days; when dividing the state: In the formula, For elements The calculated state number sequence number; , Respectively, vector The maximum and minimum values ​​in ; is the state distance; To round up.

3. The method according to claim 2, characterized in that The theoretical transfer power is calculated by the following formula : In the formula, is the fluctuating power, h ' is the element of the improved state transfer matrix Hr'; is the load at time t Theoretical transfer power: Represents the distance between state m and state The product of superimposed power fluctuations , represents the active power value generated in the m state simulation; The corresponding status is ,but In the improved state transfer matrix, the state The probability of transitioning to state m; It means Theoretical power value for transition to state m.

4. The method according to claim 1, characterized in that Before the historical load data is officially used, the following steps are also required: The spline interpolation method is used to fill in the missing data of ultra-short term. If the missing data exceeds 10 consecutive collection moments, the data of the same period of the adjacent days are used to fill in the missing data.

Images