WO2022021727A1

WO2022021727A1 - Urban complex electricity consumption prediction method and apparatus, electronic device, and storage medium

Info

Publication number: WO2022021727A1
Application number: PCT/CN2020/134505
Authority: WO
Inventors: 杨德州; 王赟中; 夏懿; 王飞; 李万伟; 李正辉; 魏勇; 胡安龙; 薛国斌; 李惠庸; 李敏; 彭婧; 梁魁; 平常; 万小花; 韩建锋; 陈庆胜; 李玉科; 李麟鹤; 杨昌海
Original assignee: 国网甘肃省电力公司; 国网甘肃省电力公司经济技术研究院; 华北电力大学
Priority date: 2020-07-29
Filing date: 2020-12-08
Publication date: 2022-02-03
Also published as: CN111985695A

Abstract

An urban complex electricity consumption prediction method and apparatus, an electronic device, and a storage medium. The method comprises: acquiring historical hourly electricity consumption sequences of each type of electricity consumption entities in an urban complex (101); performing splitting and dimensionality reduction on the historical hourly electricity consumption sequences according to preset calendar tabs to obtain first electricity consumption sequences for the calendar tabs (102); performing non-linear dimensionality reduction on the first electricity consumption sequences in a time dimension to obtain second electricity consumption sequences (103); inputting the second electricity consumption sequences into a preset predictive neural network for prediction to obtain a first electricity consumption prediction result based on the calendar tabs (104); and obtaining an electricity consumption prediction result of the complex by calculation on the basis of the first electricity consumption prediction result (105). The electricity consumption prediction result of the urban complex can thus be more accurate.

Description

Electricity consumption forecasting method, device, electronic equipment and storage medium for urban complex

technical field

The invention relates to the field of data processing, in particular to a method, device, electronic device and storage medium for predicting power consumption of an urban complex.

Background technique

The so-called "urban complex" is a kind of building complex that integrates the five core functions of business office, hotel and catering, commercial sales, apartment residence and cultural and entertainment complex in geographical space, and establishes a mutual relationship between each part. The dynamic relationship of interdependence finally forms a multi-functional and efficient "city within a city". The complex load structure and huge spatial scale make urban complexes become a large-scale electricity customer in the urban power grid. For the urban complex itself, accurate electricity forecast helps complex users to rationally and flexibly arrange energy consumption methods to achieve the purpose of energy conservation and emission reduction. For example, energy storage and distributed photovoltaics can be utilized in different periods according to the results of electricity forecasting; in addition, for power companies, accurate electricity forecasting helps them formulate flexible maintenance, scheduling and marketing plans, thereby ultimately reducing power supply costs.

At present, the monthly electricity consumption forecasting methods for urban complexes are generally divided into two categories according to different forecasting algorithms: forecasting methods based on mathematics or statistics (time series method, grey forecasting method) and artificial intelligence forecasting methods (neural network forecasting method). method, support vector machine regression method). Existing research shows that prediction methods based on mathematics or statistics are more interpretable, but lack flexibility and accuracy. On the contrary, compared with traditional forecasting methods, although the interpretability of artificial intelligence forecasting methods is poor, it is more suitable for high-dimensional, nonlinear, and high-complexity urban complex electricity consumption forecasting. Usually, when using the artificial intelligence prediction method, it is necessary to use the historical monthly electricity consumption data of the entire urban complex to train a single-step prediction model to predict the electricity consumption for a period of time in the future. This method has certain limitations: first, it directly predicts the overall monthly electricity consumption of the urban complex, and it is impossible to accurately grasp the characteristics of its internal load components; Low historical monthly data, the amount of this type of data is small, and the number of samples that can be constructed to train the model is limited, which increases the risk of overfitting the prediction model, which ultimately leads to low prediction accuracy; finally, the load in the urban complex The structure is complex, the electricity consumption behavior is highly volatile, and the predictability of the electricity consumption sequence is low, resulting in low prediction accuracy. Therefore, the existing electricity consumption forecasting methods for urban complexes have the problem of low forecasting accuracy.

SUMMARY OF THE INVENTION

The embodiment of the present invention provides a method for predicting the power consumption of an urban complex, which can improve the accuracy of the power consumption prediction of the urban complex.

In a first aspect, an embodiment of the present invention provides a method for predicting electricity consumption in an urban complex, including:

Obtain the historical hourly electricity consumption sequence of various types of electricity consumption entities in the urban complex;

Splitting the historical hourly power consumption sequence according to preset calendar tags to reduce the dimension to obtain a first power consumption sequence for each calendar tag;

Performing nonlinear dimensionality reduction on the first power consumption sequence in the time dimension to obtain a second power consumption sequence;

Inputting the second power consumption sequence into a preset prediction neural network for prediction, and obtaining a first power consumption prediction result based on the calendar tag;

Based on the first electricity consumption prediction result, the electricity consumption prediction result of the complex is obtained by calculation.

Optionally, the calendar label is a week label, and the step of dividing the historical hourly electricity consumption sequence according to preset calendar labels to reduce the dimension to obtain the first electricity consumption sequence for each calendar label Specifically include:

Splitting the historical hourly electricity consumption sequence according to the preset week label for dimension reduction to obtain the electricity consumption sequence for each week label;

Align the power consumption sequences for each week label to obtain a first electricity consumption sequence for each week label.

Optionally, the step of performing nonlinear dimensionality reduction on the first power consumption sequence in the time dimension to obtain the second power consumption sequence specifically includes:

Perform nonlinear dimensionality reduction on the time dimension of the first electricity consumption sequence by using a preset self-encoding network;

Wherein, the preset self-encoding network is obtained by training with preset dimension reduction parameters.

Optionally, the self-encoding network includes an encoding network, and the encoding network includes a first input layer, a first hidden layer, and a first output layer, wherein the output of the first input layer is the same as the first hidden layer. The input of the containing layer is connected, the output of the first hidden layer is connected to the input of the first output layer, the first input layer includes M×k neurons, and the first output layer includes M neurons, The dimension reduction parameter is k, and M and k are both positive integers. According to the dimension reduction parameter, nonlinear dimension reduction is performed on the first power consumption sequence in the time dimension through a preset auto-encoding network. The steps specifically include:

The first power consumption sequence is sequentially input to the first input layer, the first hidden layer and the first output layer to obtain an M-dimensional second power consumption sequence.

Optionally, the self-encoding network further includes a decoding network, and the second power consumption sequence is input into a preset prediction neural network for prediction, and a first power consumption prediction result based on the calendar tag is obtained. The steps include:

Inputting the M-dimensional second power consumption sequence into a prediction neural network for prediction to obtain an M-dimensional prediction result;

The M-dimensional prediction result is input into a preset decoding network for decoding, and the M×k-dimensional first electricity consumption prediction result is obtained.

Optionally, the decoding network includes a second input layer, a second hidden layer, and a second output layer, wherein the output of the second input layer is connected to the input of the second hidden layer, and the first The output of the two hidden layers is connected to the input of the second output layer, the second input layer includes M neurons, the second output layer includes M×k neurons, and the M-dimensional prediction The result is input into a preset decoding network for decoding, and the steps of obtaining the M×k-dimensional first electricity consumption prediction result specifically include:

The M-dimensional second power consumption sequence is sequentially input to the second input layer, the second hidden layer, and the second output layer to obtain an M×k-dimensional first power consumption prediction result.

Optionally, the step of calculating the electricity consumption forecast result of the complex based on the first electricity consumption forecast result specifically includes:

According to the first electricity consumption prediction result, calculating and obtaining the second electricity consumption prediction result of various electricity consumption entities based on the calendar;

The second electricity consumption prediction results corresponding to the various types of electricity consumption entities are added to obtain the electricity consumption prediction result of the complex.

In a second aspect, an embodiment of the present invention further provides a device for predicting electricity consumption in an urban complex, wherein the device includes:

The acquisition module is used to acquire the historical hourly electricity consumption sequence of various types of electricity consumption entities in the urban complex;

a first dimension reduction module, configured to split the historical hourly electricity consumption sequence according to preset calendar tags for dimension reduction, to obtain a first electricity consumption sequence for each calendar tag;

A second dimension reduction module, configured to perform nonlinear dimension reduction of the first electricity consumption sequence in the time dimension to obtain a second electricity consumption sequence;

a prediction module, configured to input the second power consumption sequence into a preset prediction neural network for prediction, and obtain a first power consumption prediction result based on the calendar tag;

A calculation module, configured to calculate the electricity consumption prediction result of the complex based on the first electricity consumption prediction result.

In a third aspect, an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, when the processor executes the computer program The steps in the method for predicting the electricity consumption of an urban complex provided by the embodiment of the present invention are implemented.

In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the urban complex electricity consumption provided by the embodiment of the present invention is realized. Steps in the Quantitative Forecasting Method.

In the embodiment of the present invention, the historical hourly electricity consumption sequence of various types of electricity-consuming entities in the urban complex is obtained; the historical hourly electricity consumption sequence is divided according to preset calendar labels to reduce the dimension, to obtain a specific calendar for each calendar. the first electricity consumption sequence of the tag; perform nonlinear dimension reduction on the first electricity consumption sequence in the time dimension to obtain a second electricity consumption sequence; input the second electricity consumption sequence into a preset prediction The neural network performs prediction to obtain a first electricity consumption prediction result based on the calendar tag; and based on the first electricity consumption prediction result, calculates and obtains the electricity consumption prediction result of the complex. By using the historical hourly electricity consumption sequence to predict the electricity consumption of each hour in the future, the resolution of the training samples can be effectively increased, thereby preventing the over-fitting of the prediction model and improving the accuracy of electricity consumption forecasting. Predicting various types of electricity-consuming entities, and finally obtaining the overall electricity consumption forecasting results of urban complexes, compared to directly using the electricity consumption of urban complexes for forecasting, it can make the electricity consumption forecasting results more accurate. more precise.

Description of drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.

1 is a flowchart of a method for predicting electricity consumption in an urban complex provided by an embodiment of the present invention;

2 is a structural diagram of an apparatus for predicting electricity consumption in an urban complex provided by an embodiment of the present invention;

3 is a structural diagram of a first dimension reduction module provided by an embodiment of the present invention;

4 is a structural diagram of a prediction module provided by an embodiment of the present invention;

5 is a structural diagram of a computing module provided by an embodiment of the present invention;

FIG. 6 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention.

detailed description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

Please refer to FIG. 1. FIG. 1 is a flowchart of a method for predicting electricity consumption in an urban complex provided by an embodiment of the present invention. As shown in FIG. 1, the method includes the following steps:

101. Obtain the historical hourly electricity consumption sequence of various types of electricity consumption entities in the urban complex.

In the embodiment of the present invention, the electricity consuming entities in the urban complex can be classified to obtain different types of electricity consuming entities, and the above electricity consuming entities can be understood as buildings or areas that need electricity. The above-mentioned electricity-consuming entities can be classified according to the functions of buildings or areas, for example, they can be divided into buildings or areas such as business, catering, commercial, residential, and cultural and entertainment complexes. In the embodiment of the present invention, a building is used as an example for description. Different building functions lead to large differences in electricity consumption characteristics of different types of buildings. Classifying the buildings in the urban complex according to their building functions and then predicting them separately can effectively improve the prediction accuracy.

For each type of building, collect its historical hourly electricity consumption sequence, which is expressed as:

in,

Represents the historical hourly electricity consumption sequence, and N represents the number of historical hourly points that can be collected.

At this time, the electricity consumption sequence of a certain type of building for a period of time in the future can be expressed as:

in,

Indicates the electricity consumption in the future period, and T indicates the total number of hours in the future period. For example, if the future period is the next month, then

Represents the electricity consumption in the next month, and T represents the total hourly points in the next month.

The above historical hourly electricity consumption sequence can be used as input data in forward reasoning, and can be used as a sample data set when training the prediction model to train the prediction model.

For example, select a city complex as the collection object of the electricity consumption sequence, and collect the historical hourly electricity consumption data of various types of buildings in its interior for a total of 36 months from January 1, 2017 to December 31, 2019. The data sampling interval is 1 hour. The electricity consumption data from the 1st to 22nd months are used as the training set, the data from the 23rd to 29th months are used as the validation set, and the electricity consumption data from the 30th to 36th months are used as the test set.

Using the prediction model constructed by the historical hourly electricity consumption sequence to predict the electricity consumption of T hours in the next month (the prediction step size is T) can effectively increase the training samples of the prediction model, that is, to increase the resolution of electricity consumption. rate, thereby preventing overfitting of the predictive model.

102. Split the historical hourly power consumption sequence according to preset calendar tags to reduce the dimension, and obtain a first power consumption sequence for each calendar tag.

In this embodiment of the present invention, the preset calendar label may be a weekday label or a ten-day label, and the above-mentioned weekday label is, for example, a Monday label, a Tuesday label, a Wednesday label, a Thursday label, a Friday label, a Saturday label, and a Sunday label. Etc., the above ten-day labels are for example: early-day labels, mid-day labels, late-day labels, etc.

The above-mentioned splitting and dimensionality reduction is, for example, splitting the historical hourly electricity consumption sequence into seven first electricity consumption sequences corresponding to the weekday label according to the week label. For another example, according to the ten-day label, the historical hourly electricity consumption sequence is divided into three first electricity consumption sequences corresponding to the ten-day label.

It should be understood that the prediction accuracy of a prediction model tends to decrease as the prediction step size (ie, the dimension of the sequence) increases.

When using the hourly electricity consumption sequence of the calendar label to predict the hourly electricity consumption corresponding to the calendar label of the next month, the historical hourly electricity consumption sequence will be decomposed into multiple parts, such as when decomposing by week label, the historical hour The electricity consumption sequence will be decomposed into seven parts, each part is T _i , i=1,...,7. At this time, the monthly electricity consumption forecast results of various types of buildings can be obtained by accumulating 7 parts of the forecast results.

Further, in terms of the week label, the above-mentioned first electricity consumption sequence can be expressed as:

Among them, Wi represents the week " _i ",

Represents the first hour-based electricity consumption sequence of the historical week "i", and represents the hourly data points of the historical week "i" that can be collected. When i=1,

The first power consumption sequence for Monday can be represented.

103. Perform nonlinear dimension reduction on the first electricity consumption sequence in the time dimension to obtain a second electricity consumption sequence.

In the embodiment of the present invention, in order to ensure the prediction accuracy of the multi-step prediction, it is necessary to further reduce the prediction step size for each type of electricity consumption sequence, that is, to further reduce the dimension of the first electricity consumption sequence.

Specifically, for an hourly electricity consumption sequence [D] _1,N×24 containing N days, a sequence compression mapping function f can be defined such that:

f:[D] _1,N×24 →[D] _1,N×M

Among them, M=24/k, k∈N+, M∈N+, k represents the dimensionality reduction parameter (also called the compression scale), and k and M are both positive integers, denoted k∈K.

At this time, the first power consumption sequence

The dimension of will be reduced from w _i to w _i /k, and the second electric quantity sequence can be expressed as:

in,

Represents the second power consumption sequence of the week "i"".

It can be understood that when the value of k is different, the power consumption sequence under different compression scales can be obtained. When k=1, it means that the sequence dimension reduction is not performed; when k=24, it means that the power consumption data per 24 hours is reduced to one power consumption data.

104. Input the second power consumption sequence into a preset prediction neural network for prediction, and obtain a first power consumption prediction result based on the calendar tag.

In this embodiment of the present invention, the prediction neural network may select a BP neural network. Taking the dimension reduction parameter as k=6 as an example for illustration, the dimension of the second electricity consumption sequence is T _i /6, i=1, . . . , 7, and the value of T _i may be different in different months. Therefore, a multi-step prediction model of BP neural network based on rolling prediction can be constructed. First, a multi-input, single-output BP neural network prediction model is trained on low-dimensional encoded sequences. Then, according to the requirements of the to-be-predicted step size T _i in different months, the model is used to roll forecast T _i /6 times to obtain multi-step prediction results.

Specifically, the input and output of the above-mentioned BP neural network are as follows:

Among them, Input is the input and Output is the output. The output of the BP neural network can be used as the first electricity consumption prediction result.

105. Based on the first electricity consumption prediction result, calculate and obtain the electricity consumption prediction result of the complex.

In the embodiment of the present invention, after obtaining the first electricity consumption prediction result corresponding to the second electricity consumption sequence, the first electricity consumption prediction result based on the calendar tag for various types of electricity consumption entities within a period of time can be obtained. For example, the seven first electricity consumption forecast results of various types of electricity consumption entities from Monday to Sunday in the next month are obtained, that is, the electricity forecast results of each day from Monday to Sunday in the next month. Accumulate the seven first electricity consumption forecast results of each type of electricity consumption entity from Monday to Sunday in the next month to obtain the electricity consumption forecast result corresponding to one type of electricity consumption entity, and then calculate the electricity consumption of all electricity consumption entities. The electricity forecast results are added together, and finally the total electricity consumption forecast result of the urban complex is obtained.

Optionally, the forecast in this embodiment of the present invention may be a monthly forecast.

The above calendar label is a week label, and the above step 102 may include:

Divide the historical hourly electricity consumption sequence according to the preset week label to reduce the dimension, and obtain the electricity consumption sequence for each week label. Align the power consumption sequences for each week label to obtain the first electricity consumption sequence for each week label. By aligning the electricity consumption sequence of each week label, the electricity consumption sequence is fixed structure data, and then the prediction model can be batched.

In this embodiment of the present invention, the historical hourly electricity consumption sequence can be

Decomposition, and then reduce the to-be-predicted step size (also called sequence dimension) of the split sequence of each class of electricity-consuming entities. have to. Specifically, considering the calendar effect of electricity consumption behavior, that is, electricity consumption behaviors on the same working day or weekend off days in different weeks are often similar, the historical hourly electricity consumption sequence is

It is divided into seven series according to its week label: the historical hourly electricity consumption sequence of all Mondays to the historical hourly electricity consumption series of all Sundays. The seven historical power consumption sequences can also be referred to as the first power consumption sequence, and the first power consumption sequence can be expressed as:

Among them, Wi represents the week " _i ",

The first power consumption sequence for Monday can be represented.

At this point, the electricity consumption in the next month can be expressed as:

Among them, T _i represents the hour number of the week "i" in the next month, and the following conditions are met:

In this embodiment of the present invention, when using the hourly electricity consumption sequence of the historical week "i" (ie, the first electricity consumption sequence) to predict the hourly electricity consumption of the week "i" in the next month, the prediction step size ( sequence dimension) will be decomposed into 7 parts, each part is T _i , i=1,...,7. At this time, the monthly electricity consumption forecast results of various types of electricity consumption entities can be obtained by accumulating 7 parts of the forecast results. Record the hourly electricity consumption sequence to be predicted for week "i" in the next month as:

Optionally, the foregoing step 103 may include:

Non-linear dimensionality reduction is performed on the first power consumption sequence in the time dimension through a preset self-encoding network; wherein, the above-mentioned self-encoding network is obtained by training with preset dimensionality reduction parameters.

f:[D] _1,N×24 →[D] _1,N×M

At this time, the first power consumption sequence

in,

Represents the second power consumption sequence of the week "i"".

It can be understood that when the value of k is different, the power consumption sequence under different compression scales can be obtained. When k=1, it means that the sequence dimension reduction is not performed; when k=24, it means that the electricity consumption data per 24 hours is reduced to one electricity consumption data, that is, a one-dimensional electricity consumption sequence.

Similarly, the dimension of the hourly electricity consumption sequence to be predicted is also reduced from w _i to w _i /k, which can be written as:

It should be noted that, first of all, the choice of compression scale directly affects the final prediction accuracy. The sequence dimension reduction parameter is too large, which will cause

The number of data points included is reduced, which in turn leads to a large reduction in the number of training samples, which ultimately reduces the accuracy of the prediction model; if the sequence compression scale is too small, it cannot be effectively reduced

The dimension of , that is, the step size to be predicted is still large, and the accuracy of the multi-step prediction model cannot be guaranteed at this time. Secondly, the different sequence compression mapping methods obtained

The predictability is also different. Since the focus of the present invention is the latter, rather than finding the optimal compression scale, the present invention selects the value of the compression scale k to be 6.

In a possible embodiment, the dimensionality reduction of the first electricity consumption sequence can be realized by additive aggregation, for example, it can be performed by the following formula:

In this way, multiple data points can be accumulated to obtain one data point, thereby realizing dimensionality reduction.

In addition, the embodiment of the present invention also provides a nonlinear dimension reduction based on an auto-encoding network, which realizes the dimension reduction of the time series of electricity consumption by extracting nonlinear features in the electricity consumption sequence. The self-encoding network consists of an encoding network (the first half of the self-encoding network) and a decoding network (the second half of the self-encoding network). In this embodiment of the present invention, an encoding network can be used to non-linearly map high-dimensional input data (that is, the first power consumption sequence) to a low-dimensional encoding sequence (that is, the second power consumption sequence), and the encoding sequence can be decoded by decoding The network reconstructs the input data. Therefore, the encoded sequence after dimension reduction can accurately represent the input data, that is, the loss of information is smaller, and the predictability of the sequence after dimension reduction is improved.

Optionally, the encoding network includes a first input layer, a first hidden layer, and a first output layer, wherein the output of the first input layer is connected to the input of the first hidden layer, and the first hidden layer is The output of the first output layer is connected to the input of the first output layer, the first input layer includes M×k neurons, the first output layer includes M neurons, the dimension reduction parameter is k, and M and k are both positive integers, The above-mentioned steps of performing nonlinear dimensionality reduction on the first power consumption sequence in the time dimension through a preset self-encoding network according to the dimensionality reduction parameters specifically include:

The first power consumption sequence is input into the first input layer, the first hidden layer and the first output layer in sequence to obtain an M-dimensional second power consumption sequence.

In this embodiment of the present invention, first, an auto-encoding network may be trained for the first power consumption sequence obtained in step 102 . Wherein, in the encoding network of the auto-encoding network, the ratio of the number of neurons in the first input layer to the number of neurons in the first hidden layer is the dimensionality reduction parameter k (in the embodiment of the present invention, k=6 is used as an example) , the number of neurons in the first input layer can be selected as M×k=24, and the number of neurons in the first output layer is M=4.

Optionally, the above step 103 may include:

Input the M-dimensional second electricity consumption sequence into the prediction neural network for prediction, and obtain the M-dimensional prediction result; input the M-dimensional prediction result into the preset decoding network for decoding, and obtain the M×k-dimensional first electricity consumption volume forecast results.

Optionally, the above-mentioned decoding network may include a second input layer, a second hidden layer, and a second output layer, wherein the output of the second input layer is connected to the input of the second hidden layer, and the second hidden layer The output of the second output layer is connected to the input of the second output layer, the second input layer includes M neurons, the second output layer includes M × k neurons, and the M-dimensional prediction result is input into the preset decoding network for decoding, The steps of obtaining the M×k-dimensional first electricity consumption prediction result may include:

In a possible embodiment, the above-mentioned encoding network, prediction network, and decoding network can jointly form a prediction model. Taking M×k=24, M=4, and k=6 as an example for illustration, there are: the encoding network and the The decoding network can jointly form a self-encoding network symmetrical about the middle layer. The first output layer of the encoding network corresponds to the middle layer 1, and the second input layer of the decoding network corresponds to the middle layer 2, as shown in Table 1:

第一输入层神经元个数The number of neurons in the first input layer	24twenty four
第一隐含层神经元个数The number of neurons in the first hidden layer	1212
第一输出层(中间层1)神经元个数Number of neurons in the first output layer (middle layer 1)	44
第二输入层(中间层2)神经元个数Number of neurons in the second input layer (middle layer 2)	44
第二隐含层神经元个数Number of neurons in the second hidden layer	1212
第二输出层神经元个数Number of neurons in the second output layer	24twenty four
各层神经元激活函数The activation function of each layer of neurons	ReLUReLU
优化器optimizer	AdamAdam
学习率learning rate	0.010.01

Table 1

The input and output (equal to the input) training samples of the neural network are as follows:

After obtaining the trained neural network, use its first half (self-encoding network) to reduce the dimension of the first power consumption sequence, that is, the high-dimensional first power consumption sequence

Each 24 points in the coder are input into the encoder in turn, and then the output of the first output layer is extracted, which is the low-dimensional coding sequence, that is, the second power consumption sequence. The second power consumption sequence is predicted by the prediction network to obtain the prediction result, and then the prediction result is input into the decoding network for decoding, which can effectively improve the prediction accuracy.

The multi-step prediction results of the 7 second electricity consumption series are obtained through the prediction network

Afterwards, the second half of the trained neural network (ie, the decoding network) is used to decode the prediction result into the electricity consumption prediction result to obtain the first electricity consumption prediction result. That is, every 4 points in the prediction result of the low-dimensional coding sequence are input into the decoding network in turn, and then the result of the second output layer is extracted to obtain the second power consumption prediction result, and the second power consumption The forecast result is taken as the forecast result of hourly electricity consumption of week "i" in the next month.

Optionally, the above step 105 may include:

According to the above-mentioned first electricity consumption forecast results, the second electricity consumption forecast results of various types of electricity consumption entities based on the above calendar are calculated and obtained; the second electricity consumption forecast results corresponding to the above various types of electricity consumption entities are added to obtain Electricity consumption forecast results for urban complexes.

Specifically, after obtaining the first electricity consumption prediction result corresponding to the second electricity consumption sequence, the first electricity consumption prediction result of each type of electricity consumption entity based on the calendar tag within a period of time can be obtained. For example, the seven first electricity consumption forecast results of various types of electricity consumption entities from Monday to Sunday in the next month are obtained, that is, the electricity forecast results of each day from Monday to Sunday in the next month. Accumulate the seven first electricity consumption forecast results of each type of electricity consumption entity from Monday to Sunday in the next month to obtain the electricity consumption forecast result corresponding to one type of electricity consumption entity, and then calculate the electricity consumption of all electricity consumption entities. The electricity forecast results are added together, and finally the total electricity consumption forecast result of the urban complex is obtained.

In the embodiment of the present invention, the historical hourly power consumption sequence of various types of power-consuming entities in the urban complex is obtained; the historical hourly power consumption sequence is divided according to preset calendar tags to reduce the dimension, and obtain the specific calendar for each calendar. the first electricity consumption sequence of the tag; perform nonlinear dimension reduction on the first electricity consumption sequence in the time dimension to obtain a second electricity consumption sequence; input the second electricity consumption sequence into a preset prediction The neural network performs prediction to obtain a first electricity consumption prediction result based on the calendar tag; and based on the first electricity consumption prediction result, calculates and obtains the electricity consumption prediction result of the complex. By using the historical hourly electricity consumption sequence to predict the electricity consumption of each hour in the future, the resolution of the training samples can be effectively increased, thereby preventing the over-fitting of the prediction model, thereby improving the prediction accuracy; Compared with the direct use of urban complex electricity consumption to forecast, it can make the electricity forecast result more accurate

It should be noted that the method for predicting the electricity consumption of an urban complex provided by the embodiment of the present invention can be applied to devices such as mobile phones, monitors, computers, and servers that can predict the electricity consumption of an urban complex.

Please refer to FIG. 2. FIG. 2 is a schematic structural diagram of an apparatus for predicting electricity consumption in an urban complex provided by an embodiment of the present invention. As shown in FIG. 2, the apparatus includes:

The obtaining module 201 is used to obtain the historical hourly electricity consumption sequence of various types of electricity consuming entities in the urban complex;

A first dimension reduction module 202, configured to perform dimension reduction by dividing the historical hourly electricity consumption sequence according to preset calendar labels to obtain a first electricity consumption sequence for each calendar label;

The second dimension reduction module 203 is configured to perform nonlinear dimension reduction of the first electricity consumption sequence in the time dimension to obtain a second electricity consumption sequence;

A prediction module 204, configured to input the second power consumption sequence into a preset prediction neural network for prediction, and obtain a first power consumption prediction result based on the calendar tag;

The calculation module 205 is configured to calculate the electricity consumption prediction result of the urban complex based on the first electricity consumption prediction result.

Optionally, as shown in FIG. 3 , the calendar label is a week label, and the first dimension reduction module 202 includes:

A splitting unit 2021, configured to split the historical hourly electricity consumption sequence according to preset week labels to reduce dimensions, and obtain electricity consumption sequences for each week label;

The aligning unit 2022 is configured to align the power consumption sequence for each week tag to obtain a first power consumption sequence for each week tag.

Optionally, the second dimensionality reduction module 203 is specifically further configured to perform nonlinear dimensionality reduction on the time dimension of the first electricity consumption sequence through a preset self-encoding network;

Optionally, the self-encoding network includes an encoding network, and the encoding network includes a first input layer, a first hidden layer, and a first output layer, wherein the output of the first input layer is the same as the first hidden layer. The input of the containing layer is connected, the output of the first hidden layer is connected to the input of the first output layer, the first input layer includes M×k neurons, and the first output layer includes M neurons, The dimensionality reduction parameter is k, and M and k are both positive integers. The 203 is also used to input the first power consumption sequence into the first input layer, the first hidden layer and the first output in sequence. layer to obtain the M-dimensional second electricity consumption sequence.

Optionally, as shown in FIG. 4 , the self-encoding network further includes a decoding network, and the prediction module 204 includes:

The first prediction unit 2041 is configured to input the M-dimensional second power consumption sequence into a prediction neural network for prediction, and obtain an M-dimensional prediction result;

The second prediction unit 2042 is configured to input the M-dimensional prediction result into a preset decoding network for decoding to obtain an M×k-dimensional first electricity consumption prediction result.

Optionally, the decoding network includes a second input layer, a second hidden layer, and a second output layer, wherein the output of the second input layer is connected to the input of the second hidden layer, and the first The output of the two hidden layers is connected to the input of the second output layer, the second input layer includes M neurons, the second output layer includes M×k neurons, and the second prediction unit 2042 specifically further It is used to input the M-dimensional second power consumption sequence into the second input layer, the second hidden layer and the second output layer in sequence, so as to obtain an M×k-dimensional first power consumption prediction result.

Optionally, as shown in FIG. 5 , the computing module 205 includes:

a first calculation unit 2051, configured to calculate and obtain second power consumption prediction results of various power consumption entities based on the calendar according to the first power consumption prediction result;

The second calculation unit 2052 is configured to add the second electricity consumption prediction results corresponding to the various types of electricity consumption entities to obtain the electricity consumption prediction result of the complex.

It should be noted that the apparatus for predicting the power consumption of an urban complex provided by the embodiment of the present invention can be applied to devices such as mobile phones, monitors, computers, and servers that can predict the power consumption of urban complexes.

The apparatus for predicting the power consumption of an urban complex provided by the embodiment of the present invention can realize the various processes implemented by the method for predicting the power consumption of an urban complex in the above method embodiments, and can achieve the same beneficial effects. In order to avoid repetition, details are not repeated here.

Referring to FIG. 6, FIG. 6 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. As shown in FIG. 6, it includes: a memory 602, a processor 601, and a memory 602 and a processor A computer program running on 601, which:

The processor 601 is used for calling the computer program stored in the memory 602, and performs the following steps:

Based on the first electricity consumption prediction result, the electricity consumption prediction result of the urban complex is obtained by calculation.

Optionally, the calendar label is a week label, and the processor 601 performs the division and dimension reduction of the historical hourly electricity consumption sequence according to the preset calendar label to obtain the first usage for each calendar label. The steps of the power sequence include:

Optionally, the step performed by the processor 601 to perform nonlinear dimensionality reduction of the first power consumption sequence in the time dimension to obtain a second power consumption sequence specifically includes:

Optionally, the self-encoding network includes an encoding network, and the encoding network includes a first input layer, a first hidden layer, and a first output layer, wherein the output of the first input layer is the same as the first hidden layer. The input of the containing layer is connected, the output of the first hidden layer is connected to the input of the first output layer, the first input layer includes M×k neurons, and the first output layer includes M neurons, The dimensionality reduction parameter is k, and M and k are both positive integers. According to the dimensionality reduction parameter executed by the processor 601, the first power consumption sequence is analyzed in the time dimension through a preset self-encoding network. The steps for nonlinear dimensionality reduction include:

Optionally, the self-encoding network further includes a decoding network, and the processor 601 performs the inputting the second power consumption sequence into a preset prediction neural network for prediction, and obtains the first power consumption based on the calendar tag. The steps of the electricity prediction result include:

Optionally, the decoding network includes a second input layer, a second hidden layer, and a second output layer, wherein the output of the second input layer is connected to the input of the second hidden layer, and the first The outputs of the two hidden layers are connected to the inputs of the second output layer, the second input layer includes M neurons, the second output layer includes M×k neurons, and the processing performed by the processor 601 to perform the The M-dimensional prediction result is input into a preset decoding network for decoding, and the steps of obtaining the M×k-dimensional first electricity consumption prediction result specifically include:

Optionally, the step performed by the processor 601 to obtain the power consumption forecast result of the complex based on the first power consumption forecast result specifically includes:

It should be noted that the above-mentioned electronic devices may be devices such as mobile phones, monitors, computers, servers, etc., which can be applied to predict the electricity consumption of urban complexes.

The electronic device provided in the embodiment of the present invention can realize the various processes realized by the method for predicting the electricity consumption of the urban complex in the above method embodiment, and can achieve the same beneficial effect. In order to avoid repetition, details are not repeated here.

Embodiments of the present invention further provide a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, each of the methods for predicting electricity consumption in urban complexes provided by the embodiments of the present invention is implemented. process, and can achieve the same technical effect, in order to avoid repetition, it is not repeated here.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. During execution, the processes of the embodiments of the above-mentioned methods may be included. The storage medium may be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM) or a random access memory (Random Access Memory, RAM for short), and the like.

The above disclosures are only preferred embodiments of the present invention, and of course, the scope of the rights of the present invention cannot be limited by this. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

A method for predicting electricity consumption in an urban complex, comprising the following steps:

Obtain the historical hourly electricity consumption sequence of various types of electricity consumption entities in the urban complex;

Splitting the historical hourly power consumption sequence according to preset calendar tags to reduce the dimension to obtain a first power consumption sequence for each calendar tag;

Performing nonlinear dimensionality reduction on the first power consumption sequence in the time dimension to obtain a second power consumption sequence;

Inputting the second power consumption sequence into a preset prediction neural network for prediction, and obtaining a first power consumption prediction result based on the calendar tag;

Based on the first electricity consumption prediction result, the electricity consumption prediction result of the urban complex is obtained by calculation.
The method according to claim 1, wherein the calendar label is a week label, and the historical hourly electricity consumption sequence is divided and dimension-reduced according to a preset calendar label to obtain the calendar label for each calendar label. The steps of the first electricity consumption sequence specifically include:

Splitting the historical hourly electricity consumption sequence according to the preset week label for dimension reduction to obtain the electricity consumption sequence for each week label;

Align the power consumption sequences for each week label to obtain a first electricity consumption sequence for each week label.
The method according to claim 1 or 2, wherein the step of performing nonlinear dimensionality reduction of the first power consumption sequence in the time dimension to obtain the second power consumption sequence specifically includes:

Perform nonlinear dimensionality reduction on the time dimension of the first electricity consumption sequence by using a preset self-encoding network;

Wherein, the preset self-encoding network is obtained by training with preset dimension reduction parameters.
The method of claim 3, wherein the self-encoding network comprises an encoding network, the encoding network comprises a first input layer, a first hidden layer and a first output layer, wherein the first input The output of the layer is connected with the input of the first hidden layer, the output of the first hidden layer is connected with the input of the first output layer, the first input layer includes M × k neurons, the first An output layer includes M neurons, the dimension reduction parameter is k, and M and k are both positive integers. The steps of non-linear dimensionality reduction in the time dimension specifically include:

The first power consumption sequence is sequentially input to the first input layer, the first hidden layer and the first output layer to obtain an M-dimensional second power consumption sequence.
The method according to claim 4, wherein the self-encoding network further comprises a decoding network, and the second power consumption sequence is input into a preset prediction neural network for prediction, and the calendar label is obtained based on the calendar label. The steps of the first electricity consumption prediction result specifically include:

Inputting the M-dimensional second power consumption sequence into a prediction neural network for prediction to obtain an M-dimensional prediction result;

The M-dimensional prediction result is input into a preset decoding network for decoding, and the M×k-dimensional first electricity consumption prediction result is obtained.
The method of claim 4, wherein the decoding network comprises a second input layer, a second hidden layer and a second output layer, wherein the output of the second input layer is the same as the second hidden layer The input of the containing layer is connected, the output of the second hidden layer is connected to the input of the second output layer, the second input layer includes M neurons, and the second output layer includes M×k neurons, The step of inputting the M-dimensional prediction result into a preset decoding network for decoding to obtain the M×k-dimensional first electricity consumption prediction result specifically includes:

The M-dimensional second power consumption sequence is sequentially input to the second input layer, the second hidden layer, and the second output layer to obtain an M×k-dimensional first power consumption prediction result.
The method according to claim 1, wherein the step of calculating the electricity consumption forecast result of the complex based on the first electricity consumption forecast result specifically comprises:

According to the first electricity consumption prediction result, calculating and obtaining the second electricity consumption prediction result of various electricity consumption entities based on the calendar;

The second electricity consumption prediction results corresponding to the various types of electricity consumption entities are added to obtain the electricity consumption prediction result of the complex.
A device for predicting electricity consumption in an urban complex, characterized in that the device comprises:

The acquisition module is used to acquire the historical hourly electricity consumption sequence of various types of electricity consumption entities in the urban complex;

a first dimension reduction module, configured to split the historical hourly electricity consumption sequence according to preset calendar tags for dimension reduction, to obtain a first electricity consumption sequence for each calendar tag;

A second dimension reduction module, configured to perform nonlinear dimension reduction of the first electricity consumption sequence in the time dimension to obtain a second electricity consumption sequence;

a prediction module, configured to input the second power consumption sequence into a preset prediction neural network for prediction, and obtain a first power consumption prediction result based on the calendar tag;

A calculation module, configured to calculate the electricity consumption prediction result of the urban complex based on the first electricity consumption prediction result.
An electronic device, characterized by comprising: a memory, a processor, and a computer program stored on the memory and running on the processor, the processor implementing the computer program as claimed in claim 1 when the processor executes the computer program Steps in the urban complex electricity consumption forecasting method described in any one of to 7.
A computer-readable storage medium, characterized in that, a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the urban integration according to any one of claims 1 to 7 is realized Steps in the body electricity consumption prediction method.