User baseline load estimation method and device based on photovoltaic and electric vehicle identification

Abstract

The invention provides a user baseline load estimation method and device based on photovoltaic and electric vehicle identification, and belongs to the technical field of electric power. The method comprises the following steps: constructing a space-time two-dimensional feature vector according to user net load data, and identifying the existence condition of user distributed photovoltaic and electric vehicles; performing non-intrusive load decomposition on the user net load to obtain a user actual load, a distributed photovoltaic load and an electric vehicle load; the distributed photovoltaic loads and the electric vehicle loads obtained through decomposition are classified, and the load mode similarity in the classes is made to be the maximum; for users participating in demand response, estimating actual loads of the users, and for users containing distributed photovoltaic or electric vehicles, estimating distributed photovoltaic or electric vehicle loads of the users based on category division; and adding and subtracting the estimated actual load of the user, the distributed photovoltaic load and the electric vehicle load, and calculating the baseline load of the user. According to the method, the accuracy of estimating the base line load of the distributed photovoltaic and electric vehicle user can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric power, and particularly relates to baseline load estimation, specifically a method and device for estimating a user's baseline load based on the identification of photovoltaic power and electric vehicles. Background Art

[0002] The user baseline refers to the normal state or level of a user when not participating in a specific activity or project under specific conditions. In a power demand response project, the user's baseline load refers to the user's electricity consumption load calculated assuming that the user does not participate in the demand response.

[0003] Demand Response (DR for short) refers to a short-term behavior in which electricity users temporarily change their electricity consumption behaviors, reduce or increase electricity consumption according to price signals or incentive measures when the electricity market price rises significantly or there are risks to the system security and reliability, so as to promote the balance between electricity supply and demand, ensure the stable operation of the power grid, and suppress the rise of electricity prices.

[0004] In order to be able to give more reasonable economic compensation to users participating in incentive-based demand response, it is necessary to accurately estimate the demand response compensation amount of users. The demand response compensation amount is the load that the user should have consumed when not participating in the demand response minus the actual electricity consumption load of the user, and the load that the user should have consumed when not participating in the demand response is the user's baseline load.

[0005] In recent years, the proportion of photovoltaic power generation in the total power generation of the power grid has been increasing year by year, and new loads such as electric vehicles have also begun to emerge continuously. These phenomena have put forward higher requirements for the guarantee of the power grid. Specifically in the field of estimating the baseline load of residential users, after the increase in the distributed photovoltaic penetration rate and the electric vehicle ownership rate of residential users, it has brought a significant impact on the improvement of the accuracy of baseline load estimation.

[0006] The Chinese patent application with the application number 202110649562.9 proposed "A Method, Device and Terminal Equipment for Estimating the Baseline Load of Photovoltaic Users", and the scenario it focuses on only includes distributed photovoltaic, and the corresponding scenario is relatively primitive.

[0007] The Chinese patent application with the application number 202110692462.4 proposed "Baseline Load Estimation Method, Device and Terminal Equipment", in which the technical problem of "the photovoltaic output is 'invisible' to system operators and load aggregators" was proposed, but there is no corresponding solution.

[0008] The Chinese patent application with the application number 202211205382.2 proposed "Baseline Load Estimation Method, Device, Electronic Equipment and Storage Medium" for a small-scale distributed power system (distributed photovoltaic, distributed wind power).

[0009] The Chinese patent application with the application number 202410521119.7 proposes "A Method for Estimating the Baseline Load of Demand Response Considering the Output of Distributed Photovoltaic Power Generation" for photovoltaic power generation.

[0010] In the prior art, when estimating the baseline load, the impacts of distributed power sources and electric vehicles on the accuracy of baseline load estimation are not considered simultaneously. Summary of the Invention

[0011] The embodiments of the present invention propose a method and device for estimating the user's baseline load based on the identification of photovoltaic and electric vehicles to improve the accuracy of estimating the baseline load of users with distributed photovoltaic and electric vehicles.

[0012] In a first aspect, the embodiments of the present invention provide a method for estimating the user's baseline load based on the identification of photovoltaic and electric vehicles, including: Step S1: Construct a spatio-temporal two-dimensional feature vector according to the user's net load data to identify the presence of distributed photovoltaic and electric vehicles of the user.

[0013] Step S2: Based on the presence of distributed photovoltaic and electric vehicles of the user, perform non-intrusive load decomposition on the user's net load to obtain the user's actual load, distributed photovoltaic load, and electric vehicle load.

[0014] Step S3: Classify the decomposed distributed photovoltaic load and electric vehicle load respectively to maximize the similarity of load patterns within the class.

[0015] Step S4: Estimate the actual load of users participating in demand response. For users with distributed photovoltaic and / or electric vehicles, estimate the distributed photovoltaic load and / or electric vehicle load during their demand response periods based on the classification.

[0016] Step S5: Add and subtract the estimated user's actual load, distributed photovoltaic load, and electric vehicle load to calculate the user's baseline load.

[0017] In a possible implementation, Step S1 includes: Step S11: Analyze the influence mechanism of distributed photovoltaic and electric vehicles on the user's net load curve according to the distributed photovoltaic load and electric vehicle load of the user and the net load data of all users.

[0018] Step S12: Analyze the output change law of distributed photovoltaic under the influence factors including at least different weather conditions and photovoltaic capacity; analyze the change of the charging behavior of electric vehicles corresponding to users under different travel modes.

[0019] Step S13: Considering the impacts of distributed photovoltaics and electric vehicles on the net load, based on the analysis results of steps S11 and S12, extract the features that can reflect the characteristics of user resources and the presence or absence of distributed photovoltaics and electric vehicles, and construct a feature vector.

[0020] Step S14: Use the feature vector as the model input and obtain the types of distributed resources contained in the user through training.

[0021] Further, in step S13, the basis for extracting the features that can reflect the characteristics of user resources and the presence or absence of distributed photovoltaics and electric vehicles includes: the ratio of the maximum value to the minimum value of the historical daily load, the average value of the load change ratio during the load climbing stage, the average value of the load change ratio during the load decreasing stage, the average value of the load change ratio within 24 hours, the ratio between the daily maximum load and the minimum load, the daily load variance, and the number of daily load mutation values.

[0022] In a possible implementation, step S2 includes: Step S21: Construct a non-intrusive load decomposition model based on CNN-BiLSTM.

[0023] Step S22: Use the user's net load data as the input, and obtain the user's distributed photovoltaic and electric vehicle load data through the non-intrusive load decomposition model. If the user does not have distributed photovoltaics or electric vehicles, the corresponding load data is 0.

[0024] In a possible implementation, step S3 includes: Step S31: For the distributed photovoltaic data of all users on all days obtained by decomposition, cluster the distributed photovoltaic data of users with distributed photovoltaics on each day to obtain the classification results of the distributed photovoltaic loads of users on different days, and assign the same class label to users in the same class.

[0025] Step S32: Regarding the differences in the classification results of the distributed photovoltaic loads on different days, perform a weighted average on the classification results of different days, and finally make the classification results meet the requirement of the maximum similarity of the load patterns within the class.

[0026] Step S33: For the electric vehicle data of all users on all days obtained by decomposition, divide the electric vehicle users into 3 categories according to the charging time, namely: daytime charging type, nighttime charging type, and mixed charging type. The charging times of the electric vehicle loads of users in the same class are relatively close.

[0027] Step S34: For the electric vehicle data of all users on all days obtained by decomposition, cluster the electric vehicle users according to the charging power. The charging powers of the electric vehicle users in the same class are relatively close.

[0028] In a possible implementation, step S4 includes: Step S41: For users participating in demand response, estimate the actual load during the demand response period on the demand response day using the historical actual load during the corresponding period on the non-demand response day of the user.

[0029] Step S42: Estimate the distributed photovoltaic load data of the user during the demand response period on that day using the distributed photovoltaic data of similar users during the demand response period on that day.

[0030] Step S43: Estimate the charging period of the electric vehicle of the user during the demand response period on that day using the electric vehicle data of similar users during the demand response period on that day, and estimate the charging power during the demand response period on that day using the historical charging power of the user's electric vehicle.

[0031] In a second aspect, an embodiment of the present invention provides a user baseline load estimation device based on photovoltaic and electric vehicle identification, including: An identification module, configured to identify the presence of distributed photovoltaic and electric vehicles of the user according to the constructed spatio-temporal two-dimensional feature vector.

[0032] A decomposition module, configured to perform non-intrusive load decomposition on the user's net load to obtain the actual load, distributed photovoltaic load, and electric vehicle load of the user.

[0033] A classification module, configured to classify the distributed photovoltaic and electric vehicle loads obtained by decomposition.

[0034] An estimation module, configured to estimate the actual load of the demand response user. For users with distributed photovoltaic and / or electric vehicles, estimate their distributed photovoltaic load and / or electric vehicle load based on the classification.

[0035] A calculation module, configured to add and subtract the estimated actual load, distributed photovoltaic load, and electric vehicle load of the user to calculate the user's baseline load.

[0036] Each module corresponds to each step of the above estimation method, and each module includes the specific content for implementing each step of the above method.

[0037] The beneficial effects of the present invention compared with the prior art are: By constructing a spatio-temporal two-dimensional feature vector, the present invention can more accurately identify the presence of distributed photovoltaic and electric vehicles of users, so as to achieve accurate decomposition and estimation of the actual load of users; adopting non-intrusive load decomposition technology, it avoids the complex steps of intervening in users or installing additional devices in traditional load decomposition methods, and simplifies the load decomposition process; by classifying users with distributed photovoltaic or electric vehicles, it can more accurately estimate the distributed photovoltaic or electric vehicle load during their demand response periods, which helps to formulate more accurate demand response strategies; by adding and subtracting the estimated actual load, distributed photovoltaic load and electric vehicle load of users, the baseline load of users can be calculated more precisely, improving the accuracy of baseline load calculation. In summary, compared with the prior art, this patent has advantages such as higher accuracy, lower complexity and more accurate demand response estimation, which helps to improve the operation efficiency and reliability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0039] Figure 1 is the implementation flowchart of the method for estimating the user baseline load based on the identification of photovoltaic and electric vehicles; Figure 2 is the structural schematic diagram of the device for estimating the user baseline load based on the identification of photovoltaic and electric vehicles. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] In the following description, specific details such as specific system structures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present invention. However, those skilled in the art should clearly understand that the present invention can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits and methods are omitted to avoid unnecessary details from interfering with the description of the present invention.

[0041] To make the objectives, technical solutions and advantages of the present invention clearer, the following will be illustrated through specific embodiments with reference to the drawings.

[0042] In this embodiment, taking the transformer substation area as a unit, a spatio-temporal two-dimensional feature vector is constructed according to the user's net load data to identify the presence of distributed photovoltaic (PV) and electric vehicles (EVs) among the users in the transformer substation area; non-intrusive load decomposition is performed on the user's net load to obtain the user's actual load, distributed PV load, and EV load; the distributed PV load and EV load obtained by decomposition are respectively classified to maximize the similarity of the load patterns within the class; for users participating in demand response, their actual load is estimated, and for users with distributed PV or EVs, their distributed PV load or EV load is estimated based on the classification; the estimated user's actual load, distributed PV load, and EV load are added and subtracted to calculate the user's baseline load. The present invention can improve the accuracy of estimating the baseline load of users with distributed PV and EVs.

[0043] The dataset used in this embodiment is the electricity load data of 300 residential users in Sydney, Australia in 2012. Each user has a rooftop PV with different capacities. This dataset contains the net load data and PV output power data of 300 households throughout the year, with a sampling interval of 1 hour. At the same time, some load data for EVs charging at night are obtained through Monte Carlo simulation for the generation of EV users.

[0044] See Figure 1 , the method for estimating the user's baseline load based on the identification of PV and EVs includes the following steps: Step S1: Construct a spatio-temporal two-dimensional feature vector according to the user's net load data to identify the presence of distributed PV and EVs of the user.

[0045] Step S2: Based on the presence of distributed PV and EVs of the user, perform non-intrusive load decomposition on the user's net load to obtain the user's actual load, distributed PV load, and EV load.

[0046] Step S3: Classify the decomposed distributed PV load and EV load respectively to maximize the similarity of the load patterns within the class.

[0047] Step S4: Estimate the actual load of users participating in demand response. For users with distributed PV and / or EVs, estimate their distributed PV load and / or EV load during the demand response period based on the classification.

[0048] Step S5: Add and subtract the estimated user's actual load, distributed PV load, and EV load to calculate the user's baseline load.

[0049] In step S1, the distributed photovoltaic and electric vehicle loads of some users within the area (i.e., within the substation area) and the net load data of all users are obtained, and a spatio-temporal two-dimensional feature vector is constructed based on the user net load data to identify the presence of distributed photovoltaic and electric vehicles of users.

[0050] The above-mentioned "some users" refer to users who have distributed photovoltaic or electric vehicles.

[0051] In this embodiment, step S11: Analyze the influence mechanism of distributed photovoltaic and electric vehicles on the user net load curve according to the distributed photovoltaic load and electric vehicle load of users and the net load data of all users; step S12: Analyze the output change law of distributed photovoltaic under influencing factors including at least different weather conditions and photovoltaic capacity; analyze the change of charging behavior of electric vehicles corresponding to users under different travel modes.

[0052] Specifically, the output of distributed photovoltaic is greatly affected by natural conditions such as weather factors. Under different weather conditions, the output of photovoltaic will be significantly different. The output is higher on sunny days and lower on cloudy and rainy days. In addition, there are certain differences in the distributed photovoltaic capacity installed by different users. When other conditions are certain, the larger the installed distributed photovoltaic capacity, the greater the photovoltaic output. The distributed photovoltaic installed behind the meter cannot be sensed by the power grid, but will be reflected in the power load measurement results in the form of reducing the user's electricity consumption. With the change of daily weather types and the differences in the installed photovoltaic capacity of specific users, the reduction value of user electricity consumption caused by photovoltaic output will also have significant fluctuations, and this kind of fluctuation will have a significant impact on the net load curve.

[0053] There are significant differences between new loads such as electric vehicles and traditional power loads. As distributed resources on the power consumption side, electric vehicles are mainly used for activities such as commuting and leisure and entertainment of residential users. The charging time is generally from when the user arrives at the workplace in the morning to before getting off work in the afternoon and from getting off work and going home to going to work the next morning, and the charging time at home mainly focuses on the latter. Therefore, the charging behavior of users for electric vehicles will increase the load at night and early morning, and the electricity peak value of users during this period will increase significantly. In addition, due to the possible different travel demands of users every day, the charging period of electric vehicles will also have certain fluctuations, resulting in uncertain fluctuations in the net load curve.

[0054] In a possible implementation, considering that the above-mentioned distributed photovoltaic power output is affected by factors such as weather conditions and photovoltaic capacity, and electric vehicles are affected by factors such as user travel patterns, it may have varying degrees of impact on the user's net load curve. A series of power consumption characteristics, power ratio characteristics, real-time characteristics, statistical characteristics, etc. are respectively extracted from the user's net load curves under different weathers, different photovoltaic capacities, and different travel patterns, and features with higher discrimination for the presence of distributed photovoltaics and electric vehicles of users are selected from them for subsequent construction of feature vectors.

[0055] Based on the above analysis, in this embodiment, step S13: Considering the impact of distributed photovoltaics and electric vehicles on the net load, according to the analysis results of steps S11 and S12, extract features that can reflect the user's resource characteristics and the presence or absence of distributed photovoltaics and electric vehicles, and construct a feature vector.

[0056] In a possible implementation, the following features are used to construct a feature vector for identifying photovoltaic users.

[0057] The basis for extracting features that can reflect the user's resource characteristics and the presence or absence of distributed photovoltaics and electric vehicles includes: the ratio of the maximum and minimum values of the historical daily load, the average value of the load change ratio during the load climbing stage, the average value of the load change ratio during the load decreasing stage, the average value of the load change ratio within 24 hours, the ratio between the daily maximum load and the minimum load, the daily load variance, the number of load mutation values per day, etc. The specific definitions and construction methods of the features are as follows: In this embodiment, calculating features requires a dataset for the past month.

[0058] Post-metering photovoltaic power generation will cause a significant reduction in the power consumption in the user's load curve, but the corresponding photovoltaic power output size will fluctuate greatly with changes in weather conditions. Therefore, the daily total load of residential users with post-metering photovoltaics will vary greatly with the change of weather types. In this paper, the ratio between the maximum and minimum values of the historical daily load is used as the first feature, and the calculation formula is as follows: In the formula, P represents the load of a time period. According to the numerical precision, if the statistics are taken once an hour (sampling interval of one hour), there are 24 time periods per day. If the statistics are taken once every 15 minutes, there are 96 time periods per day. ΣP represents the daily load of a certain day, ( ΣP ) represents the set of daily loads corresponding to each day within a time period. In this embodiment, the time period is one month. ( Σ P ) max And ( ΣP ) minThe results corresponding to the maximum and minimum values ​​of daily load are shown respectively. Due to the reduction effect of photovoltaic output on the load curve, the minimum value of the total daily load corresponding to photovoltaic users will be significantly smaller than that of non-photovoltaic users, resulting in a larger result of feature 1. Therefore, photovoltaic users can be identified through this feature.

[0059] The time period from the beginning of the load climb to the end of the load climb is often relatively fixed. However, since the time period when the photovoltaic power generation starts to output and the time period when the residential users' load starts to climb have a certain overlap, the load climbing rate of the residential users after the installation of the photovoltaic meter will offset the photovoltaic power generation to a certain extent during the load climbing period. Therefore, the corresponding load change ratio average value is small during the load climbing period of each day of the month. Therefore, the sum of the corresponding changes can be used as the second feature, and the calculation formula is as follows: In the formula, t is the corresponding time when the load increases. For residential users, the load increases usually occur between 6 and 9 o'clock every day. t It is the load curve index corresponding to the 6 o'clock moment. P i for i The active power value of the load curve at the moment, Δ t is the corresponding time interval. In this embodiment, the time interval is set to 3 hours, that is, the calculation time is from 6 o'clock to 9 o'clock when the load climbs. The corresponding average value of photovoltaic users will be smaller than that of users without photovoltaic installation. This feature can be used to identify photovoltaic users. The interval between i and i-1 determines the resolution based on specific data, such as 15 minutes, one hour, etc.

[0060] Perform the above calculations on the daily data for a month and then sum them up.

[0061] The time period from the beginning of load reduction to the end of load reduction is often relatively fixed. However, since the period when photovoltaic power ends and the period when residential users' load begins to decrease have a certain overlap, the rate of load reduction of residential users with photovoltaic power installed after the meter will offset the photovoltaic power generation to a certain extent during the load reduction period. Therefore, the average value of the corresponding load change ratio during the load reduction period of each day of the month is relatively small. Therefore, the sum of the corresponding changes can be used as the third feature, and the calculation formula is as follows: In the formula, t is the corresponding load reduction period start time. For residential users, the load reduction period is usually from 20:00 to 23:00 every day. t It is the load curve index corresponding to the 20 o'clock moment. P i is the active power value of the load curve at time i, Δt For the corresponding time interval, the time interval is set to 3 hours in this embodiment, that is, the calculation time is from 20:00 to 23:00 when the load is climbing. The average value of photovoltaic users will be less than that of users without installed photovoltaic. The identification of photovoltaic users can be achieved through this feature. The data determination resolution is the same as above.

[0062] Similarly, the above calculations are performed on the data of each day within a month, and then summed up.

[0063] The calculation formulas of the above two features are the same. The time period corresponding to T2 is the load climbing period, and the time period corresponding to T3 is the load decreasing period.

[0064] In addition to affecting the load climbing period and decreasing period of residential users, the load climbing rate of residential users in other periods will also be affected by the behind-the-meter photovoltaic. The corresponding impact can be represented by Feature Four, and the calculation is as follows: In the formula, R represents the resolution of the data, in hours. For example, if the resolution is 15 minutes, then R = 1 / 4, and 24 represents the time period of a day. After accumulating the daily load period changes, the load period identification of non-photovoltaic users and photovoltaic users can be achieved.

[0065] The above calculations are performed on the data of each day within a month, and then summed up.

[0066] The charging behavior of electric vehicle users will cause a significant increase in the load during the corresponding charging period. There will be differences in the load charging periods on different historical days. Therefore, there will be differences in the change magnitude of the load between electric vehicle users at the same time on different historical days and the load change magnitude of users without installed electric vehicles. In addition, there will also be significant differences between the load during the charging period of electric vehicle users and the load during the non-charging period in the adjacent period. Therefore, the identification of electric vehicle users can be carried out according to the corresponding changes.

[0067] There is a ratio between the maximum load and the minimum load of the daily load. Between the maximum load and the minimum load of electric vehicles, due to the existence of the charging power of electric vehicles, the maximum load will be significantly higher than the maximum load of users without purchased electric vehicles. The corresponding difference is represented as Feature Five: In the formula, P max is the maximum value of the daily load, P min is the minimum value of the daily load. In this embodiment, the historical load one month away from the corresponding day of the user to be identified is selected for analysis, and the one with the largest ratio is taken as Feature Five.

[0068] The corresponding ratio of electric vehicle users will be greater than that of users without installed electric vehicles, so the identification of corresponding electric vehicle users can be achieved.

[0069] There is a certain range of variation in the daily load curve. The power change caused by the charging behavior of electric vehicle users will lead to a sudden change in the user's electricity load. Therefore, in the historical load curve of electric vehicle users, the load will change significantly in a short period of time. The corresponding change can be judged by the load variance, which is expressed as Feature Six: In the formula, P t represents the load at each time period of the historical day, t is the corresponding user historical time period range. In this embodiment, the historical load one month before the corresponding day of the user to be identified is selected for analysis. Calculate the variance of the historical load at the corresponding time period. The corresponding variance of electric vehicle users will increase significantly. Therefore, electric vehicle users can be identified through the corresponding feature.

[0070] There is a certain range of variation in the load curve of each month. The power change caused by the charging behavior of electric vehicle users will lead to a sudden change in the user's electricity load. Due to the fixed charging demand of electric vehicle users' load every month for corresponding electric vehicle users, the number of mutation values will be different compared with the historical load of users who have not purchased electric vehicles. The corresponding number can be used as the seventh feature, which is expressed as follows: In the formula, P i represents the load at i time period of the historical day, and calculate the standard deviation of the corresponding time period P var , according to the 3σ principle, the load exceeding the load average value P avg three times the standard deviation P var within the range can be identified as outliers. Among electric vehicle users, due to the existence of load outliers caused by charging, the corresponding number of mutation values of electric vehicle users will increase significantly. Therefore, electric vehicle users can be identified through this feature.

[0071] In this embodiment, step 14: Use the feature vector composed of the above seven features as the model input, and use the classification and recognition model of the improved LightGBM to identify various types of users in the sub-clusters of users. Through training, the types of distributed resources contained in the users are obtained. Each user corresponds to four possible existence states of distributed resources: no distributed resources, only distributed photovoltaic, only electric vehicle, and both distributed photovoltaic and electric vehicle, which is convenient for further decoupling the user load based on the identification results of the loads of photovoltaic and electric vehicle users in the future.

[0072] The distributed resources here refer to distributed photovoltaic and electric vehicle.

[0073] LightGBM fully implements the Gradient Boosting Decision Tree (GBDT) and integrates the Exclusive Feature Bundling (EFB). In addition, it also adopts the Gradient-based One-Side Sampling (GOSS) to narrow the search range of split points, adopts the histogram-based algorithm to find the best split point, and adopts the leaf-by-leaf growth strategy with depth limit, thus having the advantages of fast prediction, low memory consumption, and high accuracy.

[0074] In practice, the household load will be interfered by various factors, resulting in abnormal fluctuations in the net load curve, and there are abnormal data in the extracted features, which leads to overfitting of the model. To better solve the above problems, a regularization term is introduced into the loss function of the LightGBM algorithm in this embodiment to reduce overfitting.

[0075] Using the identification model proposed in this embodiment for identification, and using accuracy, precision, recall rate, and balanced F-score for method performance evaluation, the following results are obtained: Table 1 Identification accuracy indicators for various cluster users It can be seen that the identification model proposed in this embodiment has good performance: the evaluation indicators identified in each estimation scenario are all above 98%.

[0076] In step S2, based on the existence of various distributed resources of users, non-intrusive load decomposition is performed on the user's net load to obtain the user's actual load, distributed photovoltaic load, and electric vehicle load.

[0077] The load here refers to the load after demand response.

[0078] In this embodiment, there are specifically two steps. Step S21: Construct a non-intrusive load decomposition model based on CNN-BiLSTM. Step S22: Use the user's net load data as the input, and obtain the user's distributed photovoltaic and electric vehicle load data through the non-intrusive load decomposition model. If the user does not have distributed photovoltaic or electric vehicle, the corresponding load data is 0.

[0079] Step S21: Establish a non-intrusive load decomposition model based on CNN-BiLSTM.

[0080] Taking the active power as an example of the data type, the total active power at a certain moment is the sum of the active powers consumed by each electrical appliance in the residence at that moment. The calculation formula is as follows: In the formula, y ( t ) is the total active power collected by the household total electricity meter at the sampling time point t moment, N is the type of load to be decomposed, y i ( t ) is the active power of the i th load at the t moment, e ( t ) is the noise generated in the measurement environment. Here, N = 3, representing three categories of distributed photovoltaic, electric vehicle, and other loads respectively.

[0081] Non-intrusive load monitoring is to solve y ( t ) given y i ( t ). Therefore, by training a neural network, the training result ŷ i ( t ) can be used to approximately represent y ( t ), that is: In the formula, f ( y ( t )) is the non-linear relationship established during the neural network learning process.

[0082] Considering the correlation between each load and the time period, based on the above formula, a load decomposition mathematical model based on CNN-BiLSTM is established: In the formula, tanh () is the hyperbolic tangent function, is the weight connecting the input layer y ( t ) and the j th hidden neuron, and M is the number of hidden neurons. is the time series feature extracted by the CNN layer. v i,j is the output after normalization by the tanh () function, j indicating the j th hidden neuron. V i ( t ) is the output set of each hidden neuron, which is actually a state matrix. z i is the connection between the hidden neuron and the output V i ( t ) The weight between. Through V i ( t ) and z i ( t ), a linear combination can represent the estimated value of the true value .

[0083] In the CNN-BiLSTM model, CNN is mainly used to extract time series features during the load operation process, and LSTM is mainly used to predict the load power. The main steps are as follows: 1) Divide the input data into a training set and a test set according to a certain proportion, and reconstruct it into the dimensions required by the CNN-BiLSTM model; 2) Extract the features in the input data through the convolutional layer; 3) Input the result after convolution into BiLSTM, and learn the time correlation of the load under different operating states through the forget gate, input gate, and output gate, and predict it; 4) Finally, output the predicted result through the fully connected layer. Through the above steps, the CNN-BiLSTM model can identify the deep features in the input data and complete the tasks of load decomposition and prediction.

[0084] Through the non-intrusive load decomposition based on CNN-BiLSTM, the user's actual load, distributed photovoltaic load, and electric vehicle load processes are obtained, and accuracy metrics (MAE), deviation metrics (Bias), and robustness metrics (RER) are used for evaluation. The results are as follows: Table 2 Decoupling Accuracy Metrics of Distributed Resources for Various Cluster Users It can be seen that the identification model proposed in this embodiment has good performance: the evaluation metrics identified in each estimation scenario are all above 98%, and the accuracy is relatively high.

[0085] In step S3, the decomposed distributed photovoltaic load and electric vehicle load are classified so that the similarity of the load patterns within the class is maximized.

[0086] There are fewer types of distributed photovoltaics and the load patterns are relatively consistent. However, there are more types of electric vehicles, and the similarity of the load patterns within the class is maximized. Different parameters can be set in subsequent calculations, resulting in higher accuracy.

[0087] In this embodiment, step S31: For all the distributed photovoltaic data of all users obtained by decomposition, the distributed photovoltaic data of each day containing distributed photovoltaic users is clustered to obtain the classification results of the distributed photovoltaic loads of different days for users, and the same class label is assigned to users in the same class.

[0088] In a possible implementation, the DBSCAN algorithm is used to cluster the distributed photovoltaic data of users. The basic principle and steps are as follows: (1) Normalize the initial data, and set the neighborhood radius r and the minimum density M min ; (2) Randomly select a point O that has not been clustered 1 , and calculate the number M of points whose Euclidean distance from O 1 is less than r according to the following formula, and cluster these points and O l into one class; 1 In the formula, D E ( x 1, x 2) is the Euclidean distance between vectors x 1 and x 2, x 1,i and x 2,i are the i-th dimensions of vectors x 1 and x 2, and I is the total number of dimensions.

[0089] (3) If M l ≥ M min , go to (4), otherwise go to (6); (4) Traverse the other points in the current class, calculate the number of unclustered points in the neighborhood r of each point respectively, and include these unclustered points in the current class; (5) If the number of unclustered points in the neighborhood of all points in the current class is less than M min , go to (6), otherwise go back to (4); (6) If there are still unclustered points, go back to (2), otherwise the clustering is completed.

[0090] In this embodiment, step S32: For the differences in the classification results of distributed PV loads on different days, the classification results of different days are weighted and averaged, and finally the classification results meet the requirement of the maximum similarity of the load patterns within the class.

[0091] Table 3 Clustering effect indicators of PV users In this embodiment, step S33: For all-day electric vehicle data of all users obtained by decomposition, electric vehicle users are divided into three categories according to the charging time, namely: daytime charging type, nighttime charging type, and mixed charging type. The electric vehicle users are clustered according to the charging power, and the charging powers of electric vehicles of users in the same category are relatively close.

[0092] Step S34: For all-day electric vehicle data of all users obtained by decomposition, the electric vehicle users are clustered according to the charging power, and the charging powers of electric vehicles of users in the same category are relatively close. In a possible implementation, the K-means clustering algorithm is used to cluster the electric vehicle data of users. In the first clustering, the clustering criterion is that the charging times are close, and in the second clustering, the clustering criterion is that the charging powers are close. The clustering effects are as follows: Table 4 Clustering effect indicators of electric vehicle users It can be seen from the data in Table 3 and Table 4 that through the first clustering and the second clustering, users are divided into clusters with very similar PV output powers / electric vehicle output powers, thus providing a technical basis for subsequent baseline load estimation.

[0093] In step S4, estimate the actual load of users participating in demand response. For users with distributed PV and / or electric vehicles, estimate their distributed PV loads and / or electric vehicle loads during the demand response period based on the category division.

[0094] In this embodiment, it specifically includes: step S41: For users participating in demand response, use the historical actual load corresponding to the non-demand response period of the users to estimate the actual load during the demand response period on the demand response day.

[0095] In a possible implementation, the actual load during the demand response period on the demand response day should be equal to the average load of the non-demand response days corresponding to the same period in the past month.

[0096] Step S42: Use the distributed PV data of users in the same category during the demand response period on this day to estimate the distributed PV load data of this user during the demand response period on this day.

[0097] In a possible implementation, the distributed photovoltaic load of a user during the demand response period on the demand response day should be equal to the average value of the distributed photovoltaic load data of the same type of distributed photovoltaic users at the corresponding time period on the same day in the past.

[0098] Step S43: Estimate the charging period of the electric vehicle of the user during the demand response period on the day using the electric vehicle data of the same type of users during the demand response period on the day, and estimate the charging power during the demand response period on the day using the historical charging power of the electric vehicle of the user.

[0099] In a possible implementation, the start time and end time of the electric vehicle charging of the user during the demand response period on the demand response day should be equal to the average values of the charging start time and end time of the same type of electric vehicle users on the same day, and the charging power of the electric vehicle of the user during the demand response period on the demand response day should be equal to the average value of the charging power of the same type of electric vehicle users on the same day. Through numerical simulation using actual data, the accuracies of the charging start time, charging end time, and total charging duration of the electric vehicle reach 90.16%, 88.24%, and 92.51% respectively.

[0100] In step S5, add and subtract the estimated actual load, distributed photovoltaic load, and electric vehicle load of the user to calculate the user's baseline load.

[0101] In a possible implementation, the user's baseline load can be calculated as follows: where respectively represent the baseline load, actual load, distributed photovoltaic load, and electric vehicle load of the user i at the k th demand response day t moment, and the electric vehicle load is the charging power.

[0102] It is known that are obtained from steps S42 and S43 respectively.

[0103] When using the decomposed users for baseline load estimation, the MAE of the estimation result is 0.1253, the Bias is 0.0258, and the RER is 0.0221. By comparing with other methods, it is found that the method proposed in this embodiment can significantly improve the accuracy of baseline load estimation, and the estimation effect is better.

[0104] The method proposed by the present invention is applicable to user sub-clusters composed of multiple resources, with high identification and decoupling accuracy. The MAE of baseline load estimation can be improved by 33% - 47%, and the mean absolute percentage error (MAPE) index of different data sets does not exceed 15%.

[0105] It should be understood that the sequence numbers of the steps in the above embodiments do not indicate the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0106] The following is the device embodiment of the present application. For the details not described in detail, reference may be made to the corresponding method embodiments above.

[0107] Figure 2 The structural schematic diagram of the user baseline load estimation device provided by the embodiment of the present application based on photovoltaic and electric vehicle identification is shown. For the convenience of description, only the parts related to the embodiments of the present application are shown and are described in detail as follows: As Figure 2 shown, the user baseline load estimation device based on photovoltaic and electric vehicle identification includes: An identification module 21, configured to identify the presence of distributed photovoltaic and electric vehicles of a user according to the constructed spatio-temporal two-dimensional feature vector.

[0108] A decomposition module 22, configured to perform non-intrusive load decomposition on the user's net load to obtain the actual load, distributed photovoltaic load, and electric vehicle load of the user.

[0109] A classification module 23, configured to classify the distributed photovoltaic and electric vehicle loads obtained by decomposition.

[0110] An estimation module 24, configured to estimate the actual load of a demand response user, and for a user with distributed photovoltaic and / or electric vehicles, estimate its distributed photovoltaic load and / or electric vehicle load based on the classification.

[0111] A calculation module 25, configured to add and subtract the estimated actual load, distributed photovoltaic load, and electric vehicle load of the user to calculate the user's baseline load.

[0112] Each module corresponds to each step of the above-mentioned user baseline load estimation method based on photovoltaic and electric vehicle identification, and each module may include the specific content of each step of the above method.

[0113] In the above embodiments, the descriptions of the respective embodiments have their own emphases. For the parts not described in detail or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0114] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0115] The above-described embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application and should all be included within the protection scope of this application.

Claims

1. A user baseline load estimation method based on photovoltaic and electric vehicle identification, characterized in that: include: Step S1: constructing a spatiotemporal two-dimensional feature vector based on the user's net load data to identify the presence of the user's distributed photovoltaic and electric vehicles; Step S2: Based on the existence of the user's distributed photovoltaic and electric vehicles, the user's net load is non-intrusively decomposed to obtain the user's actual load, distributed photovoltaic load and electric vehicle load; Step S3: Classify the decomposed distributed photovoltaic loads and electric vehicle loads into categories respectively, so that the similarity of load patterns within the category is maximized; Step S4: estimating the actual load of users participating in demand response. For users with distributed photovoltaics and / or electric vehicles, estimating the distributed photovoltaic load and / or electric vehicle load during the demand response period based on the classification; Step S5: The estimated actual user load, distributed photovoltaic load and electric vehicle load are added and subtracted to calculate the user baseline load.

2. The estimation method according to claim 1, characterized in that Step S1 includes: Step S11: Analyze the impact mechanism of distributed photovoltaic and electric vehicles on the user's net load curve based on the user's distributed photovoltaic load and electric vehicle load and the net load data of all users; Step S12: analyzing the output variation of distributed photovoltaic under at least different weather conditions and photovoltaic capacity factors; analyzing the charging behavior changes of users corresponding to electric vehicles under different travel modes; Step S13: Considering the impact of distributed photovoltaics and electric vehicles on net load, based on the analysis results of steps S11 and S12, extract features that can reflect user resource characteristics and the presence or absence of distributed photovoltaics and electric vehicles, and construct feature vectors; Step S14: Using the feature vector as a model input, and obtaining the distributed resource types of the user through training.

3. The estimation method according to claim 2, characterized in that: In step S13, the basis for extracting the features that can reflect the user resource characteristics and the existence of distributed photovoltaic and electric vehicles includes: the ratio of the historical daily maximum and minimum loads, the average value of the load change ratio in the load climbing stage, the average value of the load change ratio in the load reduction stage, the average value of the load change ratio within 24 hours, the ratio between the daily maximum load and the minimum load, the daily load variance, and the number of daily load mutation values.

4. The estimation method according to claim 2, characterized in that: In step S14, the classification and recognition model based on the improved LightGBM is used to train and test the types of distributed resources contained in the user. Each user corresponds to four possible distributed resource existence states: no distributed resources, only distributed photovoltaics, only electric vehicles, and both distributed photovoltaics and electric vehicles.

5. The estimation method according to claim 1, characterized in that: Step S2 includes: Step S21: constructing a non-intrusive load decomposition model based on CNN-BiLSTM; Step S22: using the user's net load data as input, obtaining the user's distributed photovoltaic and electric vehicle load data through a non-intrusive load decomposition model; if the user does not have distributed photovoltaic or electric vehicles, the corresponding load data is 0.

6. The estimation method according to claim 1, characterized in that: Step S3 includes: Step S31: clustering the distributed photovoltaic data of all users on all days obtained by decomposition, and obtaining the distributed photovoltaic load classification results of users on different days, and assigning the same category label to users of the same type; Step S32: for the differences in the classification results of distributed photovoltaic loads on different days, weighted average is performed on the load classification results on different days, so that the classification results finally meet the requirement of maximum similarity of load patterns within the class; Step S33: for all the electric vehicle data of all users on all days, the electric vehicle users are divided into three categories according to the charging time, namely: daytime charging type, nighttime charging type and mixed charging type. The electric vehicle load charging time of the same type of users is relatively close; Step S34: for the decomposed electric vehicle data of all users on all days, the electric vehicle users are clustered according to the charging power, and the charging power of electric vehicles of the same type of users is relatively close.

7. The estimation method according to claim 1, characterized in that: Step S4 includes: Step S41: for users participating in demand response, the actual load of the demand response period on the demand response day is estimated by using the historical actual load of the user in the corresponding period on the non-demand response day; Step S42: using the distributed photovoltaic data of similar users during the demand response period of the day to estimate the distributed photovoltaic load data of the user during the demand response period of the day; Step S43: using the electric vehicle data of similar users during the demand response period on that day to estimate the charging period of the electric vehicle of the user during the demand response period on that day, and using the historical charging power of the electric vehicle of the user to estimate the charging power during the demand response period on that day.

8. The estimation method according to claim 7, characterized in that: The start and end times of electric vehicle charging during the demand response period on the demand response day should be equal to the average of the start and end times of charging for similar electric vehicle users on the same day. The charging power of electric vehicles during the demand response period on the demand response day should be equal to the average of the charging power for similar electric vehicle users on the same day.

9. The estimation method according to claim 1, characterized in that: Step S5 includes: The estimated actual user load, distributed photovoltaic load and electric vehicle load are added and subtracted to calculate the user baseline load: In the formula, Representing users i In the k Demand response days t The baseline load, actual load, distributed photovoltaic load and electric vehicle load at the moment.

10. A user baseline load estimation device based on photovoltaic and electric vehicle identification, characterized in that: include: An identification module is used to identify the presence of user distributed photovoltaic and electric vehicles based on the constructed spatiotemporal dual-dimensional feature vector; A decomposition module is used to perform non-intrusive load decomposition on the user's net load to obtain the user's actual load, distributed photovoltaic load and electric vehicle load; A classification module is used to classify the decomposed distributed photovoltaic and electric vehicle loads; An estimation module, used for estimating the actual load of demand response users, and for users with distributed photovoltaics and / or electric vehicles, estimating their distributed photovoltaic loads and / or electric vehicle loads based on category classification; The calculation module is used to add and subtract the estimated actual user load, distributed photovoltaic load and electric vehicle load to calculate the user baseline load.

Images