CN112734137A - Short-term building power load prediction method and system based on intelligent electric meter data - Google Patents
Short-term building power load prediction method and system based on intelligent electric meter data Download PDFInfo
- Publication number
- CN112734137A CN112734137A CN202110108591.4A CN202110108591A CN112734137A CN 112734137 A CN112734137 A CN 112734137A CN 202110108591 A CN202110108591 A CN 202110108591A CN 112734137 A CN112734137 A CN 112734137A
- Authority
- CN
- China
- Prior art keywords
- data
- load
- load data
- time
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 37
- 238000012545 processing Methods 0.000 claims abstract description 13
- 238000010219 correlation analysis Methods 0.000 claims abstract description 10
- 230000002159 abnormal effect Effects 0.000 claims abstract description 9
- 230000005611 electricity Effects 0.000 claims description 12
- 230000001364 causal effect Effects 0.000 claims description 7
- 210000002569 neuron Anatomy 0.000 claims description 7
- 238000013528 artificial neural network Methods 0.000 abstract description 6
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 5
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 5
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 230000003247 decreasing effect Effects 0.000 description 3
- 238000005265 energy consumption Methods 0.000 description 3
- 238000005406 washing Methods 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 239000003345 natural gas Substances 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 239000011800 void material Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- General Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Tourism & Hospitality (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a short-term building power load prediction method and a system based on intelligent electric meter data, wherein the method comprises the following steps: 1) data input: taking total table load data and internal load data as input data, wherein the internal load data comprises each sub table load data and unmeasured load data; 2) data processing: filling missing data, correcting abnormal values, and reserving internal load data with a correlation coefficient larger than a set value according to a correlation analysis result carried out in advance; 3) and (3) load prediction: inputting the time sequence formed by the internal load data and the total table load data obtained in the step 2) into a pre-trained stable time convolution network, and performing short-term prediction on the building power load at the target future time. According to the method, the building power load is predicted in a short term by adopting the time convolution neural network, and compared with the traditional method, the time for training to reach stability is obviously shortened, the error is obviously reduced, and the prediction performance is better.
Description
Technical Field
The invention relates to short-term building power load prediction, in particular to a short-term building power load prediction method and system based on intelligent electric meter data.
Background
With the improvement of living standard of ordinary individual residents, the individual building load level is also improved, and the load has great fluctuation due to the behavior of residents. During the peak period of power utilization, when an individual building uses a plurality of high-power electrical appliances at the same time, the power supply pressure of the power grid feeder line can be increased, and the power supply pressure of the whole power distribution network is further increased. Therefore, through customer-side demand response, the supplier can realize peak clipping of the power grid by adjusting the electricity price, and the residents can also reduce the electric bill. Short-term building load prediction is a key technology of customer-side demand response, and win-win between a power distribution department and a user can be realized.
The variability of the electricity utilization behaviors of the customers in the building has obvious correlation with the prediction effect, and the data size is huge, so that great challenges are brought to the learning of a time series model. The traditional prediction method adopts a long-short-term memory recurrent neural network, and learns the habits of residents in using household appliances based on household appliance use data under the condition of considering household appliance consumption, so that the prediction performance superior to that of multiple references is obtained. However, the recurrent neural networks such as long-short-term memory cannot process the input sequence in parallel, so the overhead of model training increases with the increase of the length of the input sequence, and long-term historical information is forgotten, so that the prediction result is not correct.
Disclosure of Invention
The invention aims to provide a short-term building power load prediction method and system based on intelligent electric meter data, which have better prediction performance.
In order to achieve the purpose, the short-term building power load prediction method based on the intelligent electric meter data comprises the following steps:
1) data input: taking total table load data and internal load data as input data, wherein the internal load data comprises each sub table load data and unmeasured load data;
2) data processing: searching missing data and abnormal data for the total table load data and the internal load data, filling the missing data, correcting an abnormal value, keeping the internal load data with the correlation coefficient larger than a set value according to a correlation analysis result of the internal load data and the total table load data which are performed based on historical load data in advance, and deleting other internal load data;
3) and (3) load prediction: inputting the time sequence formed by the internal load data and the total table load data obtained in the step 2) into a pre-trained stable time convolution network (TCN for short), and performing short-term prediction on the building power load at the target future time (the time when the next group of power consumption data is generated).
Preferably, the step 1) further includes a data collection step, namely, collecting time series electricity utilization data of the building general table and each sub-table (such as a washing machine, a plug and the like) in a period of time before the predicted time, and obtaining unmeasured load data of each group of electricity utilization data according to a difference value of the sum of the load data measured by the general table and the load data measured by each sub-table, so as to obtain the input data.
Preferably, in step 1), the length of the acquired time series is equal to the length of the time series input to the time convolution network each time during training.
Preferably, in the step 3), the convolution layer includes a causal convolution and a hole convolution.
Preferably, the basic structure of the time convolution network is a residual error network formed by 2-4 residual error layers, and each residual error layer comprises 1-2 basic layers, 1-2 convolution layers and 1 direct connection layer.
Preferably, in the step 3), the convolution layer includes a causal convolution and a hole convolution.
Preferably, in the step 3), the number of neurons of the time convolution network is 250-350.
Preferably, in the step 3), the pre-training of the time convolution network includes the following steps: inputting total table load data and internal load data of a past period of time according to training requirements; filling and correcting the load data by adopting the method in the step 2), then carrying out correlation analysis on each internal load data and the total table load data, selecting the internal load data with the correlation coefficient larger than a set value (preferably 0.1) as auxiliary input data, combining the auxiliary input data with the total table load data, and training the time convolution network.
Preferably, the training parameters of the time convolutional network are set as: the batch is set as 60, the learning rate is 0.00001, 10 rounds of training are carried out on the model, the training frequency of each round is set as 20 epochs, each epoch training is carried out once on the whole data set, 10 rounds of training are carried out to obtain 10 scores in total, and the score average value is obtained to serve as the prediction accuracy index.
The invention also provides a short-term building power load prediction system designed for realizing the method, which comprises the following steps:
a data acquisition module configured to implement the data acquisition step;
a data processing module configured to implement the data processing steps;
a load prediction module configured to implement the load prediction step.
Preferably, the load prediction module adopts a time convolution network, the time convolution network comprises 2 residual error layers, and each residual error layer comprises 1-2 base layers, 1-2 convolution layers and 1 direct connection layer.
Compared with the prior art, the invention has the beneficial effects that:
according to the method, the time convolution neural network is adopted, the sub-meter and the total meter of the intelligent electric meter are combined, the total meter load data, each sub-meter load data and the unmeasured load data are used as input data of a prediction task, correlation analysis is conducted on the input data, and internal load data related to the total load serving as a target task are extracted and used as auxiliary input data.
Drawings
Fig. 1 is a block diagram illustrating a short-term building electrical load prediction system according to embodiment 1 of the present invention.
Fig. 2 to 4 are schematic structural diagrams of causal convolution, hole convolution and residual connection of the time convolution network provided in embodiment 1 of the present invention.
Fig. 5 is a bar chart of correlation coefficients of each internal load and total table load data obtained by performing correlation analysis in embodiment 2 of the present invention.
Fig. 6 is an iterative comparison graph of the performance of TCN networks with different structural parameters and the conventional LSTM network in embodiment 2 of the present invention.
Fig. 7 is a comparison graph of the predicted results of TCN networks with different structural parameters and the conventional algorithm in embodiment 2 of the present invention.
Fig. 8 is an enlarged detail view of the box in fig. 7.
Wherein: data acquisition module 1, data processing module 2 and load prediction module 3
Detailed Description
The invention is described in further detail below with reference to the figures and the specific embodiments.
Example 1
The present embodiment describes the construction, training and application of the neural network separately.
1. Constructing a time convolution network and setting parameters
Selecting the length (n +1) of a history input sequence, setting the size k of a convolution kernel, setting the step length stride to be 1, and setting and operating the following parameters in order to ensure that the length of an output sequence of a network is the same as the length of an input sequence and cause-effect convolution is required: the convolution parameter zero padding (padding) is determined to be k-1, the output sequence length is (n + k-1), and a tail division operation is performed to remove the last future output sequence, i.e., the tail division number is equal to padding, as shown in fig. 2.
Based on the causal convolution, to adaptively adjust the depth of the time convolution network, a hole convolution (scaled convolution) is used. Specifically, an input sequence X is givenn+1＝{x0,x1,...,xnWhere f (i) is a convolution function, the hole convolution can be defined as:
d is the hole factor, and m-i.d is the input sequence number of the element selected for convolution. When d is 1, the hole convolution is degenerated to a general one-dimensional convolution. As the number of network layers increases, d grows exponentially, and the higher the level, the more historical information is covered by the network sequence. As shown in fig. 3.
Constructing residual connections based on causal convolution and hole convolution, specifically a first branch, wherein if the length of an output sequence is not equal to that of an input sequence, an input layer is connected to the output layer through 1x1 convolution, otherwise, the input layer is connected to the output layer through identity mapping; and the second branch, namely, the input sequence is connected to a network formed by sequentially forming a plurality of convolution layers, an activation layer and a dropout layer and superposed by a same structure. As shown in fig. 4.
On the basis, the model depth design is determined according to the sequence length, the cavity convolution and the residual error connection mechanism. Specifically, initially, assume that the depth of the network is L, the depth of the input layer is 0, and each layer is a residual layer. The longest history memory information that can be contained in the last output sequence in the n-th layer of neurons can be derived by the following formula:
ILis the longest input sequence information on which layer L residual neuron last sequence values can depend. When load prediction is performed, the output sequence of the neuron in the previous layer is used as the input sequence of the current layer, and only the final value of the output sequence is used as the output of the residual error network after finally passing through all the residual error layers upwards, as shown in fig. 4.
The residual structure shown in fig. 4 superimposes the base layer, and may superimpose the convolutional layer as needed, or superimpose the base layer and the convolutional layer at the same time.
And finally, taking the output of each parallel residual error network as the input of a feedforward neural network, and outputting the input as the prediction result of the future time step.
2. Training of a time convolutional network
1) Data acquisition and input: acquiring time sequence electricity consumption data (the length of a single sequence is n +1) of a building total table and each branch table (such as a washing machine, a plug and the like) in a past period of time before a prediction time according to training requirements, obtaining unmeasured load data of each group of electricity consumption data according to the difference value of the sum of load data measured by the total table and load data measured by each branch table, and taking the total table load data and internal load data as input data; wherein, the sub-meter load data and the unmeasured load data are collectively called as internal load data;
2) data processing: searching missing data and abnormal data for the load data, filling the missing data, correcting abnormal values, performing correlation analysis on each sub-table and unmeasured load data and total table load data on the basis (after filling and correction), sorting according to the correlation coefficients from large to small, and selecting internal load data and total table load data with the correlation coefficients larger than a set value (selected to be 0.1 in the embodiment) as auxiliary input data of the time convolution network.
3) Model training
And training the time convolution network by combining the summary load data and the internal load data.
The specific training parameters are set as follows: the batch is set to 60, the learning rate is set to 0.00001, 10 rounds of training are performed on the model, the number of times of each round of training is set to 20 epochs, each epoch training is performed once on the whole data set, 10 rounds of training are performed to obtain 10 scores in total, and the score average value is obtained to serve as the prediction accuracy index (namely, each error index in table 2).
Setting model parameters: the hidden layer neuron is set to 300, and since the number of residual layers is set to 2, we select an input sequence length of 7. The original residual layer base layer (B) is 1, the causal void convolution layer (C) in the base layer is 1, and several improved versions are designed simultaneously in this embodiment, which is detailed in table 1.
3. Short-term building electrical load prediction
The steps of adopting the trained time convolution network model to carry out load prediction are as follows:
1) data acquisition and input: collecting time sequence electricity utilization data (the sequence length is n +1 and is the same as the training stage) of a building general table and each sub-table (such as a washing machine, a plug and the like) in a period of time before the prediction time, obtaining unmeasured load data of each group of electricity utilization data according to the difference value of the sum of the load data measured by the general table and the load data measured by each sub-table, and taking the load data of the general table and the internal load data as input data; wherein, the sub-meter load data and the unmeasured load data are collectively called as internal load data;
2) data processing: searching missing data and abnormal data for the total table load data and the internal load data, filling the missing data, correcting an abnormal value, reserving the internal load data with the correlation coefficient larger than a set value of 0.1 and deleting other internal load data according to a correlation analysis result (see a training part of a time convolution network) of the internal load data and the total table load data which is performed based on historical load data in advance;
3) and (3) load prediction: inputting the time sequence formed by the internal load data and the total table load data obtained in the step 2) into a pre-trained stable time convolution network (TCN for short), and performing short-term prediction on the building power load at the target future time (the time when the next group of power consumption data is generated).
The embodiment also provides a short-term building power load prediction system based on the method, which comprises a data acquisition module 1, a data storage module and a data processing module, wherein the data acquisition module is configured to realize the data acquisition step; a data processing module 2 configured to implement the data processing steps; a load prediction module 3 configured to implement the load prediction step.
The load prediction module 3 adopts a time convolution network, the time convolution network comprises 1 direct connection layer and 2 residual error layers, each residual error layer comprises 1-2 basic layers and 1-2 convolution layers, and specific parameters are shown in table 1.
Example 2
This example trains the time-time convolutional network given in example 1 based on the AMPds2 dataset and evaluates and compares its predicted performance after the training has stabilized. The method comprises the following specific steps:
1. experimental data set
This example uses The AMPds2 dataset (Stephen makonin. ampds2: The Almanac of minutely Power dataset (Version 2),2016) to train and evaluate The time-convolution network disclosed in example 1, as well as several common control algorithms.
From month 4 2012 to month 3 2014, the AMPds2 data set collected various energy consumption in their authors' homes, including measurements of electricity, water, and natural gas per minute, for a total of 1051200 readings, relating to meters including 21 smartmeters, 2 water meters, and 2 natural gas meters (with additional equipment instructions).
The embodiment selects the smart meter data set with granularity per minute as the original data set. The full House meter reading consists of three parts, namely a Master-House meter reading (MHE), a Rental Suite (RSE) electric sub-meter reading and a Garage (GRE) electric sub-meter reading. The MHE consists of the remaining 18 electrical (AE) meter readings and unmeasured energy consumption (UNE). Thus, the relationship between data set readings can be described as:
WHE＝MHE+RSE+GRE
MHE＝AE+UNE
2. experimental setup
In the experiment, 1/4/2012, 0: 00-9/4/2012, 24:00 was used as the training set, and 10/4/1440 minutes/day was selected as the test set. After the correlation analysis, as shown in fig. 5, Master-House readings (MHE), electric Heat Pump (HPE), dryer (CDE), oven blower and thermostat (FRE), unmeasured energy consumption (UNE), basement plug and light (BME), kitchen fireplace (WOE), home office electricity (OFE), TV/PVR/amp (tve), Dishwasher (DWE) were selected as input data for the model.
Each model was set as follows:
(1) k is adjacent to each other: the kd-Tree is 30, the weights are uniform and consistent, and the neighbor is 15.
(2) Gradient Boost Decision Tree (GBDT): estimators is 100, the maximum depth of the decision tree is 6, the minimum sample size required for the internal node repartitioning is 60, and the rest are default values.
(3) Support Vector Regression (SVR): kernel is RBF, gamma is 0.03, C is 5.66, and others are defaults.
(4) LSTM: and adopting the set optimal parameters, and referring to: weicong Kong, Zhao Yang Dong, David J Hill, Fengji Luo, and Yan xu. short-term residual load for evaluating based on residual latent residual learning. IEEE Transactions on Power Systems,33(1):1087, 1088,2017.
In addition to LSTM, another benchmark test is to grid search for the optimal parameters in the sklern module of python.
In the training of the time-time convolution network, the batch is set to be 60, the learning rate is set to be 0.00001, the epoch times are set to be 20, Adam is selected in the optimization method, all models are trained for 10 times, and the average value is taken. At the same time, we have also tested several improved TCN architectures, whose structural parameters are detailed in table 1.
TABLE 1 TCN structural parameters
| Method | Residual layer | Base layer | Convolutional layer | Hidden neuron |
| TCN(B1C1) | 2 | 1 | 1 | 300 |
| TCN(B1C2) | 2 | 1 | 2 | 300 |
| TCN(B2C1) | 2 | 2 | 1 | 300 |
| TCN(B2C2) | 2 | 2 | 2 | 300 |
3. Results and analysis of the experiments
The following error indicators were used to evaluate the predictive performance of the model, where ytIs the sum of observed values
Is an expected value. The smaller their value, the better the model performance.
Percent mean absolute error (MAPE):
mean Absolute Error (MAE):
root Mean Square Error (RMSE):
the predicted performance (MAPE, MAE, RMSE) of each model is shown in table 2 below:
TABLE 2 prediction of building load
As can be seen from the above table, the MAPE of TCN (B1C2) was the lowest and the MAPE decreased by 11.90% compared to the best baseline LSTM; TCN (B2C2) possessed the lowest RMSE, which decreased by 0.8%; TCN (B1C2), TCN (B2C1) and TCN (B2C2) all have the lowest MAE, and the MAE is reduced by 7.5%; the worst TCN (B1C1) MAPE decreased by 9.13%, MAE by 7.5%, and RMSE leveled off.
The performance variation of the test set during LSTM and TCN training is shown in FIG. 6. Both achieved a relatively steady state during training, but TCN reached a steady state almost 50% less time than LSTM. Thus, the TCN architecture not only enables optimal performance of the network with greater efficiency, but ultimately better network performance than LSTM.
The results of the four network predictions are shown in fig. 7. To better highlight the advantages of TCN over each benchmark test, some details in fig. 7 are shown in fig. 8. In the case of large residential load fluctuation, the four TCN architectures have stronger capability of tracking the residential load than the LSTM, which means that the TCN architecture has great potential in predicting the residential load fluctuation in real time.
Claims (10)
1. A short-term building power load prediction method based on intelligent electric meter data is characterized by comprising the following steps: the method comprises the following steps:
1) data input: taking total table load data and internal load data as input data, wherein the internal load data comprises each sub table load data and unmeasured load data;
2) data processing: searching missing data and abnormal data for the total table load data and the internal load data, filling the missing data, correcting an abnormal value, keeping the internal load data with the correlation coefficient larger than a set value according to a correlation analysis result of the internal load data and the total table load data which are performed based on historical load data in advance, and deleting other internal load data;
3) and (3) load prediction: inputting the time sequence formed by the internal load data and the total table load data obtained in the step 2) into a pre-trained stable time convolution network, and performing short-term prediction on the building power load at the target future time.
2. The short term building electrical load prediction method of claim 1, wherein: the step 1) further comprises a data acquisition step, namely acquiring time series electricity utilization data of the building general table and each sub-table in a period of time before the prediction time, and obtaining unmeasured load data of each group of electricity utilization data according to the difference value of the sum of the load data measured by the general table and the load data measured by each sub-table, so as to obtain the input data.
3. The short term building electrical load prediction method of claim 2, wherein: in the step 1), the length of the acquired time sequence is equal to the length of the time sequence input into the time convolution network each time during training.
4. The short term building electrical load prediction method of claim 1, wherein: in the step 3), the basic structure of the time convolution network is a residual error network formed by 2-4 residual error layers, and each residual error layer comprises 1-2 basic layers, 1-2 convolution layers and 1 direct connection layer.
5. The short term building electrical load prediction method of claim 4, wherein: in the step 3), the time convolution network comprises a causal convolution and a cavity convolution, and the number of neurons in each layer is 250-350.
6. The short-term building electrical load prediction method as claimed in any one of claims 1 to 5, wherein: in the step 3), the pre-training of the time convolution network includes the following steps: inputting total table load data and internal load data of a past period of time according to training requirements; filling and correcting the load data by adopting the method in the step 2), then carrying out correlation analysis on each internal load data and the total table load data, selecting the internal load data with the correlation coefficient larger than a set value as auxiliary input data, combining the auxiliary input data with the total table load data, and training the time convolution network.
7. The short term building electrical load prediction method of claim 6, wherein: the set value of the correlation coefficient is 0.1.
8. The short term building electrical load prediction method of claim 6, wherein: the training parameters of the time convolution network are set as follows: the batch is set as 60, the learning rate is set as 0.00001, 10 rounds of training are carried out on the model, the training frequency of each round is set as 20 epochs, each epoch training is carried out once on the whole data set, 10 rounds of training are carried out to obtain 10 scores in total, and the score average value is obtained to serve as the prediction accuracy index.
9. A short term building electrical load prediction system designed to implement the method of any one of claims 1 to 8, comprising:
a data acquisition module (1) configured to implement said data acquisition step;
a data processing module (2) configured to implement the data processing steps;
a load prediction module (3) configured to implement the load prediction step.
10. The short term building electrical load prediction system of claim 9, wherein: the load prediction module (3) adopts a time convolution network, the time convolution network comprises 2 residual error layers, and each residual error layer comprises 1-2 basic layers, 1-2 convolution layers and 1 direct connection layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110108591.4A CN112734137B (en) | 2021-01-27 | 2021-01-27 | Short-term building power load prediction method and system based on intelligent electric meter data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110108591.4A CN112734137B (en) | 2021-01-27 | 2021-01-27 | Short-term building power load prediction method and system based on intelligent electric meter data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112734137A true CN112734137A (en) | 2021-04-30 |
| CN112734137B CN112734137B (en) | 2022-12-16 |
Family
ID=75594108
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110108591.4A Active CN112734137B (en) | 2021-01-27 | 2021-01-27 | Short-term building power load prediction method and system based on intelligent electric meter data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112734137B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113408815A (en) * | 2021-07-02 | 2021-09-17 | 湘潭大学 | Deep learning-based traction load ultra-short-term prediction method |
| CN114118568A (en) * | 2021-11-19 | 2022-03-01 | 国网河南省电力公司经济技术研究院 | Multi-attention-based power consumption prediction method for convolutional neural network |
| CN114399032A (en) * | 2022-03-23 | 2022-04-26 | 武汉格蓝若智能技术有限公司 | Method and system for predicting metering error of electric energy meter |
| CN116108320A (en) * | 2023-04-13 | 2023-05-12 | 佳源科技股份有限公司 | Intelligent electric energy meter operation control method |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014042147A1 (en) * | 2012-09-12 | 2014-03-20 | 日本電気株式会社 | Data concentration prediction device, data concentration prediction method, and program thereof |
| CN110298509A (en) * | 2019-06-28 | 2019-10-01 | 佰聆数据股份有限公司 | A kind of big industrial trade Electric optimization of combination short-term load forecasting |
| CN110472779A (en) * | 2019-07-30 | 2019-11-19 | 东莞理工学院 | A kind of power-system short-term load forecasting method based on time convolutional network |
| CN110852655A (en) * | 2019-11-29 | 2020-02-28 | 湖南科技大学 | Household power load decomposition system with solar power supply system and decomposition method |
| CN111289942A (en) * | 2020-01-21 | 2020-06-16 | 北京市腾河电子技术有限公司 | Method and system for analyzing error of measurement domain based on single load jump and storage medium |
| CN111461436A (en) * | 2020-01-13 | 2020-07-28 | 长沙理工大学 | Method for predicting space-time dynamic distribution of charging load of electric automobile |
| CN111461463A (en) * | 2020-04-30 | 2020-07-28 | 南京工程学院 | Short-term load prediction method, system and equipment based on TCN-BP |
| CN111967679A (en) * | 2020-08-20 | 2020-11-20 | 江苏师范大学 | Ionized layer total electron content forecasting method based on TCN model |
-
2021
- 2021-01-27 CN CN202110108591.4A patent/CN112734137B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014042147A1 (en) * | 2012-09-12 | 2014-03-20 | 日本電気株式会社 | Data concentration prediction device, data concentration prediction method, and program thereof |
| CN110298509A (en) * | 2019-06-28 | 2019-10-01 | 佰聆数据股份有限公司 | A kind of big industrial trade Electric optimization of combination short-term load forecasting |
| CN110472779A (en) * | 2019-07-30 | 2019-11-19 | 东莞理工学院 | A kind of power-system short-term load forecasting method based on time convolutional network |
| CN110852655A (en) * | 2019-11-29 | 2020-02-28 | 湖南科技大学 | Household power load decomposition system with solar power supply system and decomposition method |
| CN111461436A (en) * | 2020-01-13 | 2020-07-28 | 长沙理工大学 | Method for predicting space-time dynamic distribution of charging load of electric automobile |
| CN111289942A (en) * | 2020-01-21 | 2020-06-16 | 北京市腾河电子技术有限公司 | Method and system for analyzing error of measurement domain based on single load jump and storage medium |
| CN111461463A (en) * | 2020-04-30 | 2020-07-28 | 南京工程学院 | Short-term load prediction method, system and equipment based on TCN-BP |
| CN111967679A (en) * | 2020-08-20 | 2020-11-20 | 江苏师范大学 | Ionized layer total electron content forecasting method based on TCN model |
Non-Patent Citations (2)
| Title |
|---|
| 曹渝昆等: "基于深度可分离卷积的轻量级时间卷积网络设计", 《计算机工程》 * |
| 苗宏佳等: "基于负荷分解与聚类融合的短期负荷预测研究", 《电子测量技术》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113408815A (en) * | 2021-07-02 | 2021-09-17 | 湘潭大学 | Deep learning-based traction load ultra-short-term prediction method |
| CN114118568A (en) * | 2021-11-19 | 2022-03-01 | 国网河南省电力公司经济技术研究院 | Multi-attention-based power consumption prediction method for convolutional neural network |
| CN114399032A (en) * | 2022-03-23 | 2022-04-26 | 武汉格蓝若智能技术有限公司 | Method and system for predicting metering error of electric energy meter |
| CN116108320A (en) * | 2023-04-13 | 2023-05-12 | 佳源科技股份有限公司 | Intelligent electric energy meter operation control method |
| CN116108320B (en) * | 2023-04-13 | 2023-06-09 | 佳源科技股份有限公司 | Intelligent electric energy meter operation control method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112734137B (en) | 2022-12-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112734137B (en) | Short-term building power load prediction method and system based on intelligent electric meter data | |
| Aprillia et al. | Statistical load forecasting using optimal quantile regression random forest and risk assessment index | |
| CN107591844B (en) | Active power distribution network robust reconstruction method considering node injection power uncertainty | |
| CN111754026B (en) | Photovoltaic power station group power prediction method and device, computer equipment and storage medium | |
| Kong et al. | Effect of automatic hyperparameter tuning for residential load forecasting via deep learning | |
| Chen et al. | Short-term load forecasting based on deep learning for end-user transformer subject to volatile electric heating loads | |
| CN110380444B (en) | Capacity planning method for distributed wind power orderly access to power grid under multiple scenes based on variable structure Copula | |
| Razghandi et al. | Short-term load forecasting for smart home appliances with sequence to sequence learning | |
| Peng et al. | Short-term residential load forecasting based on smart meter data using temporal convolutional networks | |
| Hinterstocker et al. | Evaluation of the effects of time-of-use pricing for private households based on measured load data | |
| Ahmadi et al. | Optimal allocation of multi-type distributed generators for minimization of power losses in distribution systems | |
| CN116578551A (en) | GRU-GAN-based power grid data restoration method | |
| Cai et al. | Forecasting customers' response to incentives during peak periods: A transfer learning approach | |
| Visser et al. | The importance of predictor variables and feature selection in day-ahead electricity price forecasting | |
| Mai et al. | Exploring regression models to enable monitoring capability of local energy communities for self‐management in low‐voltage distribution networks | |
| US20240144004A1 (en) | Trained Models for Discovering Target Device Presence | |
| Ferraz et al. | Analysis of distribution system considering load uncertainties: a k-means clustering approach | |
| Dandea et al. | K-means clustering-based data mining methodology to discover the prosumers’ energy features | |
| CN115622034A (en) | Internet of things meter power load identification method based on improved genetic neural network | |
| Parrado-Duque et al. | A comparative analysis of machine learning methods for short-term load forecasting systems | |
| CN114881176A (en) | Non-invasive load identification method based on self-adaptive optimization random forest | |
| Firoozi et al. | Flexibility Forecast at Local Energy Community Level | |
| Ming et al. | Electrical load prediction in energy internet via linear correlation coefficient approach | |
| Foster et al. | Short-term load forecasting with high levels of distributed renewable generation | |
| Wang et al. | Operational Scenario Generation and Forecasting for Integrated Energy Systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |