A method, system, device and medium for predicting residential power load based on space-time conduction network

Abstract

The present invention relates to the technical field of power load prediction, and in particular to a method, system, device and medium for predicting residential power load based on a space-time conduction network. The method firstly integrates collected residential power load data and external multi-source data of residential power load to obtain integrated residential power load data. Then, the integrated residential power load data is analyzed to obtain residential power load data components, and the influence of external disturbance is analyzed according to the residential power load data components. Finally, an LSTM network is called to predict the residential power load data components, and the weights of the predicted residential power load data components are calculated, and the residential power load is obtained according to the weight prediction. External reference factors are added to the prediction of residential power load, and the method has the characteristics of comprehensive data analysis, and the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power load prediction, and in particular to a method, system, device and medium for predicting residential power load based on a time-space conduction network. Background Art

[0002] Residential power load (RPL) refers to the sum of the electrical energy used by electrical equipment in residential areas and household users at a certain moment. It is an important part of the power grid load and has the characteristics of large scale, large volatility, and large peak-to-valley difference. RPL prediction is the basis for the economic operation of the power system and is related to the balance between power supply and demand. By accurately predicting RPL, power supply companies can better arrange power generation scheduling plans and optimize power grid resource allocation, which has positive significance for reducing the operating costs of power supply companies and improving the reliability of power supply in the power grid.

[0003] Many scholars have done a lot of research on RPL prediction. The document "A RPL prediction model based on threshold autoregressive moving average algorithm" calculates the temperature mutation point through the Markov model, and predicts RPL by combining the long short-term memory network with the threshold autoregressive moving average algorithm. The document "A RPL prediction method based on graph convolutional neural network" divides the RPL types through K-means clustering, and uses graph convolutional neural grid to construct the correlation of each PRL sequence, thereby improving the RPL prediction accuracy. The document "A RPL prediction method based on deep conditional probability density function" calculates the relationship between RPL prediction and observation value through the quartic kernel function, and improves the RPL prediction accuracy through cross-validation. The document "A RPL prediction method based on long short-term neural network" filters out the noise of RPL historical data through the hard threshold function, and uses the long short-term memory network to predict various types of user load data.

[0004] It can be seen that there are various RPL prediction methods. However, with the continuous advancement of power system reform and the construction of new power systems, multi-source loads such as rooftop photovoltaics, residential user energy storage, and electric vehicles have developed rapidly, and a large number of RPLs with electricity consumption characteristics have emerged, and load characteristics have undergone profound changes. The above methods are insufficient in the fusion of multi-source information, and the RPL prediction accuracy is low in the source-grid-load-storage interaction scenario. Summary of the invention

[0005] In view of the problem that the existing residential electricity load forecasting methods are insufficient in the fusion of multi-source information and the prediction accuracy is low in the source-grid-load-storage interaction scenario, the present invention proposes a residential electricity load forecasting method, system, equipment and medium based on a space-time conduction network; firstly, the fusion of multi-source heterogeneous data such as social economy, meteorological environment and the whole chain process tracing are carried out to provide supporting data for the residential electricity load forecast. Secondly, a multimodal method is used to decompose the electricity consumption components of the residential electricity load, and a space-time conduction network is used to analyze the propagation mechanism of the change of residential electricity load demand under the disturbance of external factors. Finally, a long short-term memory network is used to predict the residential electricity load of each component, and the weighted results are formed to form the residential electricity load forecasting results; thus, the prediction accuracy of the residential electricity load is improved.

[0006] The specific implementation contents of the present invention are as follows:

[0007] A method for predicting residential electricity load based on a spatiotemporal conduction network specifically comprises the following steps:

[0008] Step S1: integrating the collected residential electricity load data and external multi-source data of residential electricity load to obtain residential electricity load integrated data;

[0009] Step S2: Analyze the residential power load fusion data to obtain the residential power load data components, and analyze the impact of external disturbances based on the residential power load data components;

[0010] Step S3: Call the LSTM network to predict the residential electricity load data components, calculate the weights of the predicted residential electricity load data components, and obtain the residential electricity load based on the weight prediction.

[0011] In order to better implement the present invention, further, the step S1 specifically includes the following steps:

[0012] Step S11: Acquire historical data of residential electricity load from the power consumption information collection system of the power supply company to obtain residential electricity load data;

[0013] Step S12: calling the OCR algorithm, calculating the connection weights between the two convolutional layers of the external multi-source data image of the residential power load, outputting the recognized characters, and obtaining the external multi-source data of the residential power load;

[0014] Step S13: Calling linear regression to repair the external multi-source data of residential power load to obtain the repaired external multi-source data of residential power load;

[0015] Step S14: extracting multi-source data features from the external multi-source data of the repaired residential power load, and calling a decision function to fuse the external multi-source data of the repaired residential power load and the residential power load data to obtain residential power load fusion data.

[0016] In order to better implement the present invention, further, the step S2 specifically includes the following steps:

[0017] Step S21: calling the FCM clustering method to analyze the residential power load fusion data components and obtain the membership of different types of electrical appliances in the residential power load fusion data;

[0018] Step S22: calling the network analysis method to construct a dynamic network for external disturbance transmission of residential power loads, and obtaining the network density of residential power loads and the spatial-temporal transmission network correlation of residential power loads;

[0019] Step S23: Obtain external disturbance impact data based on the membership of different types of electrical appliances in the residential power load fusion data, the network density of the residential power load, and the spatial-temporal conduction network correlation of the residential power load.

[0020] In order to better implement the present invention, further, the step S3 specifically includes the following steps:

[0021] Step S31: calling the LSTM network to predict the components of the residential electricity load data;

[0022] Step S32: using the entropy weight method to calculate the impact weight of the residential electricity load data component to obtain the optimal impact weight value;

[0023] Step S33: Predict the residential electricity load based on the impact weight.

[0024] In order to better realize the present invention, further, the specific operation of step S31 is: calling the LSTM network, and predicting the lth residential electricity load component data according to the residential electricity load fusion data, external disturbance impact data, the set activation function, the photovoltaic power generation power of the residential households, and the energy storage discharge power of the residential households.

[0025] In order to better implement the present invention, further, the specific operation of step S32 is: calling the entropy weight method, calculating the conditional entropy of the residential electricity load component prediction and the information entropy of the residential electricity load component prediction, obtaining the joint entropy of the residential electricity load component prediction, obtaining the influence weight of the residential electricity load data component according to the joint entropy, and calculating the optimal value of the influence weight.

[0026] Based on the above-mentioned residential electricity load forecasting method based on time-space transmission network, in order to better realize the present invention, further, a residential electricity load forecasting system based on time-space transmission network is proposed, which is used to execute the above-mentioned residential electricity load forecasting method based on time-space transmission network, including a fusion unit, an analysis unit, and a forecasting unit;

[0027] The fusion unit is used to fuse the collected residential electricity load data and external multi-source data of residential electricity load to obtain fused residential electricity load data;

[0028] The analysis unit is used to analyze the residential power load fusion data to obtain the residential power load data components, and analyze the impact of external disturbances according to the residential power load data components;

[0029] The prediction unit is used to call the LSTM network to predict the residential electricity load data components, calculate the weights of the predicted residential electricity load data components, and obtain the residential electricity load according to the weight prediction.

[0030] Based on the above-mentioned residential electricity load forecasting method based on time-space conduction network, in order to better realize the present invention, further, an electronic device is proposed, including a memory and a processor; a computer program is stored on the memory; when the computer program is executed on the processor, the above-mentioned residential electricity load forecasting method based on time-space conduction network is implemented.

[0031] Based on the above-mentioned residential electricity load forecasting method based on time-space conduction network, in order to better realize the present invention, further, a computer-readable storage medium is proposed, characterized in that computer instructions are stored on the computer-readable storage medium; when the computer instructions are executed on the above-mentioned electronic device, the above-mentioned residential electricity load forecasting method based on time-space conduction network is implemented.

[0032] The present invention has the following beneficial effects:

[0033] (1) The present invention integrates multi-source heterogeneous data such as social economy and meteorological environment, adds external reference factors to the prediction of residential electricity load, and has the characteristics of comprehensive data analysis.

[0034] (2) The present invention uses a spatiotemporal conduction network to analyze the propagation mechanism of changes in residential electricity load demand under external disturbances, and identifies the impact of temperature and per capita GDP on changes in residential electricity load.

[0035] (3) The present invention uses LSTM to perform multi-time scale residential electricity load forecasting, effectively correlating the impact of data at different time scales on residential electricity load forecasting and improving the accuracy of residential electricity load forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic diagram of the residential electricity load prediction process provided by the present invention. DETAILED DESCRIPTION

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. It should be understood that the described embodiments are only part of the embodiments of the present invention, not all of the embodiments, and therefore should not be regarded as limiting the scope of protection. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technical personnel in this field without making creative work are within the scope of protection of the present invention.

[0038] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "disposed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0039] Embodiment 1:

[0040] This embodiment proposes a residential electricity load prediction method based on a spatiotemporal conduction network, which specifically includes the following steps:

[0041] Step S1: Fuse the collected residential electricity load data and external multi-source residential electricity load data to obtain residential electricity load fusion data.

[0042] The step S1 specifically includes the following steps:

[0043] Step S11: Acquire historical data of residential electricity load from the power consumption information collection system of the power supply company to obtain residential electricity load data;

[0044] Step S12: calling the OCR algorithm, calculating the connection weights between the two convolutional layers of the external multi-source data image of the residential power load, outputting the recognized characters, and obtaining the external multi-source data of the residential power load;

[0045] Step S13: Calling linear regression to repair the external multi-source data of residential power load to obtain the repaired external multi-source data of residential power load;

[0046] Step S14: extracting multi-source data features from the external multi-source data of the repaired residential power load, and calling a decision function to fuse the external multi-source data of the repaired residential power load and the residential power load data to obtain residential power load fusion data.

[0047] Step S2: Analyze the residential power load fusion data to obtain the residential power load data components, and analyze the impact of external disturbances based on the residential power load data components.

[0048] Furthermore, the step S2 specifically includes the following steps:

[0049] Step S21: calling the FCM clustering method to analyze the residential power load fusion data components and obtain the membership of different types of electrical appliances in the residential power load fusion data;

[0050] Step S22: calling the network analysis method to construct a dynamic network for external disturbance transmission of residential power loads, and obtaining the network density of residential power loads and the spatial-temporal transmission network correlation of residential power loads;

[0051] Step S23: Obtain external disturbance impact data based on the membership of different types of electrical appliances in the residential power load fusion data, the network density of the residential power load, and the spatial-temporal conduction network correlation of the residential power load.

[0052] Step S3: Call the LSTM network to predict the residential electricity load data components, calculate the weights of the predicted residential electricity load data components, and obtain the residential electricity load based on the weight prediction.

[0053] Furthermore, the step S3 specifically includes the following steps:

[0054] Step S31: calling the LSTM network to predict the components of the residential electricity load data.

[0055] Furthermore, the specific operation of step S31 is: calling the LSTM network, and predicting the lth residential electricity load component data according to the residential electricity load fusion data, external disturbance impact data, the set activation function, the photovoltaic power generation power of the residential households, and the energy storage discharge power of the residential households.

[0056] Step S32: using the entropy weight method to calculate the impact weight of the residential electricity load data component to obtain the optimal impact weight value.

[0057] Step S33: Predict the residential electricity load based on the impact weight.

[0058] Furthermore, the specific operation of step S32 is: calling the entropy weight method, calculating the conditional entropy of the residential electricity load component prediction and the information entropy of the residential electricity load component prediction, obtaining the joint entropy of the residential electricity load component prediction, obtaining the influence weight of the residential electricity load data component according to the joint entropy, and calculating the optimal value of the influence weight.

[0059] Working principle: This embodiment first integrates multi-source heterogeneous data such as social economy, meteorological environment, and chain-based full-process tracing to provide supporting data for residential electricity load forecasting. Secondly, a multimodal method is used to decompose the electricity consumption components of residential electricity load, and a spatiotemporal conduction network is used to analyze the propagation mechanism of changes in residential electricity load demand under external factors. Finally, a long short-term memory network is used to predict the residential electricity load of each component, and weighted to form the residential electricity load forecast result; thus, the residential electricity load forecast accuracy is improved.

[0060] Embodiment 2:

[0061] This embodiment is based on the above embodiment 1. Figure 1 As shown, a specific embodiment is described in detail.

[0062] The residential power load prediction method based on the spatiotemporal conduction network in this embodiment is divided into three steps: multi-source data fusion of residential power load, analysis of the impact of external disturbances on residential power load, and prediction of residential power load. Figure 1 shown.

[0063] Step S1: Multi-source data fusion: The collected residential power load data and external multi-source data of residential power load are integrated to obtain residential power load fusion data.

[0064] First, the power supply company collects residential power load data through the power collection system, and collects multi-source data such as social economy and meteorological environment through optical character recognition (OCR); secondly, it conducts data quality assessment on residential power load and external multi-source data, and repairs data with unqualified quality. Finally, it integrates multi-source heterogeneous data, realizes data chain traceability, and realizes the association of various types of data.

[0065] Step S11: Collecting multi-source data of residential electricity load.

[0066] The prediction and deduction of residential electricity load involves multi-source data such as historical residential electricity load data, social economy, meteorological environment, etc. Among them, residential electricity load data can be directly obtained from the power supply company's electricity information collection system, while the social economic data and meteorological data published on the Internet by relevant institutions are not integrated with the power supply company, so they are obtained through OCR.

[0067] The historical load of residential electricity consumption obtained by the power supply company through the electricity consumption information collection system for:

[0068] (1)

[0069] in: The total number of time points of historical data of residential electricity load; , ,......, It is the residential electricity load at different time points of historical data.

[0070] Guided training of connectionist temporal classification (GTC) is an OCR text recognition algorithm that uses a temporal attention mechanism to train a connectionist temporal classification (CTC) text recognition model and uses a graph convolution network (GCN) to characterize the correlation between elements of text recognition feature sequences, thereby improving text recognition accuracy.

[0071] GTC consists of four parts: correction module, backbone convolutional neural network, CTC classifier and attention classifier. The correction module converts the curved text image into a straight text image. The backbone convolutional neural network extracts the convolution features of the text image; the CTC classifier integrates the context information; and the attention classifier reduces the text recognition error.

[0072] GTC calculates the connection weights between the two convolutional layers of the multi-source data image of residential electricity load. Since pixels close to each other correspond to the same multi-source data character of residential electricity load, we first calculate and The distance between two pixels :

[0073] (2)

[0074] Where: exp is the function of natural exponential in multi-source data text recognition of residential electricity load; The tilt angle of characters in the text recognition of multi-source data of residential electricity load; Identify the straight-line distance between two pixels for multi-source data of residential electricity load.

[0075] In GTC text recognition, pixels with the same convolution feature also correspond to the same multi-source data text characters of residential electricity load. The cosine similarity of the convolution feature of two pixels in GTC text recognition is calculated. .

[0076] (3)

[0077] Where: , They are and Convolutional features of two pixels; for and The formula for calculating the cosine similarity of two pixels.

[0078] After obtaining the connection weights between the two convolutional layers of the multi-source data image of residential electricity load, the recognized characters are output .

[0079] (4)

[0080] Where: V Provides convolution input features for GTC text recognition; The selection weight in GTC text recognition.

[0081] Step S12: Repairing multi-source data of residential electricity load.

[0082] Affected by user meter failures, transmission channel abnormalities and other situations, there are missing and abnormal situations in the multi-source data collection of residential electricity load, which will cause deviations in the residential electricity load forecast. Therefore, the collected multi-source data of residential electricity load is repaired to eliminate the impact of missing data on load forecasting.

[0083] Linear regression is a data statistics and repair method that uses regression analysis to determine the dependency between multiple variables and repair missing or abnormal data.

[0084] Multi-source data of residential electricity load restored using linear regression for:

[0085] (5)

[0086] Where: The original residential electricity load multi-source data contains missing and abnormal data; is the intercept of the original residential electricity load multi-source data on the y-axis; is the slope of the original multi-source data of residential electricity load.

[0087] Step S13: Fusion of multi-source heterogeneous data of residential electricity load.

[0088] The fusion of internal and external multi-source heterogeneous data required for residential electricity load forecasting is the key to residential electricity load forecasting. Multi-source data of residential electricity load is usually unstructured, semi-structured and other data. Therefore, it is necessary to fuse multi-source data of residential electricity load to improve the prediction performance of residential electricity load.

[0089] Multi-source heterogeneous data fusion refers to the process of integrating and analyzing residential electricity load data from different sources, in different formats, and of different types. Multi-source heterogeneous data fusion can provide more comprehensive and accurate information for residential electricity load forecasting, thereby improving forecast accuracy.

[0090] Standardize the multi-source heterogeneous data to extract the characteristics of multi-source data. The characteristic value of the multi-source data k of residential electricity load in z attribute values for:

[0091] (6)

[0092] Where: The number of attributes of multi-source data of residential electricity load; It is multi-source data of residential electricity load with different attributes.

[0093] Then, the multi-source heterogeneous data are fused through the decision function, and the fused data for:

[0094] (7)

[0095] Where: is the number of multi-source heterogeneous data; is the kernel function; is the mean of different multi-source heterogeneous data; The optimal value for different multi-source heterogeneous data; is the characteristic value of multi-source heterogeneous data; It is the bias for multi-source heterogeneous data fusion.

[0096] Through the fusion of multi-source heterogeneous data, a complete data chain traceability is formed, realizing the association of various types of data on residential electricity load.

[0097] Step S2: Analysis of the impact of external disturbances on residential power loads: Analyze the residential power load fusion data to obtain residential power load data components, and analyze the impact of external disturbances based on the residential power load data components.

[0098] Firstly, the model decomposes the multimodal residential electricity load data to obtain the components of the residential electricity load data; secondly, the propagation mechanism of external disturbance data such as social economy and meteorological environment is analyzed; finally, the impact of external disturbance transmission and multi-temporal and multi-component residential electricity load demand is analyzed.

[0099] Step S21: Analyze the components of residential electricity load.

[0100] The purpose of analyzing the composition of residential electricity load is to understand the composition of residents' loads and the changes in various types of loads under different meteorological and economic conditions. Since the power of household appliances such as air conditioners and televisions is not equipped with intelligent switches, the load types of households cannot be directly counted. Therefore, the capacitive, resistive and inductive characteristics of various electrical equipment can be used to identify the type of electrical appliances and decompose the load components.

[0101] State Grid Corporation of China launched the construction of a new generation of smart meters in 2019, collecting data such as active power, reactive power, current, power factor, harmonics, etc. from the meters, providing a data basis for the decomposition of residential electricity load components. This paper clusters the active and reactive power of user meters to obtain the characteristics of different types of loads, thereby achieving the purpose of decomposing residential electricity load components.

[0102] Fuzzy C-means clustering method (FCM) is a data clustering method. It divides the active and reactive data of residential power load into clusters through FCM, so that the similarity of active and reactive data within the same type of electrical appliances is maximized, while the similarity of active and reactive data of different types of electrical appliances is minimized. This paper uses FCM to analyze the composition of residential power load. for:

[0103] (8)

[0104] Where: The number of samples clustered for residential electricity load components; It is the degree of membership of different residential electricity load samples to a certain type of electrical appliance cluster center. It is an electrical appliance clustering center for different residential power loads.

[0105] After performing FCM clustering on the residential power load, the load component curves of different types of electrical appliances in the residential power load are obtained.

[0106] Step S22: Analyze external disturbances of residential power load.

[0107] External disturbance data such as social economy and meteorological environment are closely related to residential electricity load. When the social economy develops well, residential users purchase a large number of electrical appliances, and the residential electricity load increases. Conversely, the residential electricity load decreases. When the meteorological environment changes, the temperature rises, residents turn on air conditioners more, and the residential electricity load increases. Therefore, the analysis of residential electricity load caused by external disturbances in the social economy and meteorological environment is helpful to improve the accuracy of residential electricity load prediction.

[0108] The space-time conduction network refers to the process by which the model converts a type of external disturbance input data that changes over time into the impact on the change of residential electricity load. When external factors such as social economy and meteorological environment change, the residential electricity load changes over time according to certain rules, thus obtaining the change of residential electricity load.

[0109] In the process of identifying the spatiotemporal conduction network, the principle of network analysis method is used to construct a dynamic network for the conduction of external disturbances of residential electricity load, so as to capture the sensitivity of various types of external disturbances of residential electricity load on residential electricity load, and thus explore the differences in the impact of different types of external factors of residential electricity load on residential electricity load.

[0110] The network density and correlation of external disturbances to residential electricity loads are analyzed to identify the characteristics of external disturbances. The network density reflects the degree of integration between the overall spatiotemporal conduction network and residential electricity loads. As the lag order increases, the density of the spatiotemporal conduction network of residential electricity loads increases, and the network density of residential electricity loads increases. for:

[0111] (9)

[0112] Where: The number of nodes in the space-time transmission network for residential electricity load; is the actual relationship number of the space-time transmission network of residential electricity load;

[0113] The network correlation is used to measure the robustness of the spatial-temporal transmission network structure of residential power load. The higher the correlation of the spatial-temporal transmission network of residential power load, the more robust the spatial-temporal transmission network of residential power load. for:

[0114] (10)

[0115] Where: The number of unreachable points in the space-time transmission network for residential electricity load.

[0116] Step S3: Residential electricity load prediction: Call the LSTM network to predict the residential electricity load data components, calculate the weights of the predicted residential electricity load data components, and predict the residential electricity load based on the weights.

[0117] First, the model uses LSTM to predict the components of residential electricity load; second, the entropy weight method is used to calculate the weights of the components of residential electricity load and obtain the optimal value of the weights. Finally, the residential electricity load is predicted based on the different weights of the components of residential electricity load.

[0118] LSTM is an improved temporal neural network. It includes input gate, output gate and forget gate structures, which can effectively solve the time correlation problem of different scales in temporal neural networks. In the LSTM network structure, LSTM receives historical residential electricity load data, external disturbance impact data of residential electricity load in the spatiotemporal transmission network, future socio-economic forecast data released by the Statistics Bureau, and meteorological environment forecast data released by the Meteorological Bureau. The forget gate determines which input information of residential electricity load is stored; the output gate forms the forecast result of residential electricity load.

[0119] The lth residential electricity load component data predicted by LSTM for:

[0120] (11)

[0121] Where: External economic forecast data for input; The input external meteorological environment forecast data; The data passed through the LSTM forget gate; Forget gate activation function for LSTM residential electricity load prediction; is the activation function in RPL prediction. Photovoltaic power generation for households; The energy storage discharge power for residential households.

[0122] The entropy weight method is a method of assigning weights based on information entropy. The entropy weight method determines the influence weight of the residential electricity load component by calculating the entropy value of the index value. The smaller the entropy value of the residential electricity load component, the greater the impact on the residential electricity load forecast.

[0123] Joint entropy for forecasting residential electricity load components for:

[0124] (12)

[0125] Where: Predict conditional entropy for residential electricity load components; , , The information entropy is predicted for the 1st, 2nd and kth residential electricity load components respectively;

[0126] According to the different weights of the residential electricity load components, the results of the residential electricity load forecast are calculated.

[0127] In the LSTM network, the forget gate is used to predict the proportion of residents’ load from the previous moment to the current moment. for:

[0128] (13)

[0129] Where: The sigmiod activation function of the LSTM network for residential electricity load prediction; The weight of the residential electricity load forecast; Output data of the residential electricity load forecast at the previous moment; Bias data in LSTM for residential electricity load forecasting.

[0130] Working principle: This embodiment integrates multi-source heterogeneous data such as social economy and meteorological environment, adds external reference factors to the prediction of residential electricity load, and has the characteristics of comprehensive data analysis. The spatiotemporal conduction network is used to analyze the propagation mechanism of changes in residential electricity load demand under external factors, and the impact of temperature and per capita GDP on changes in residential electricity load is identified. LSTM is used to predict residential electricity load at multiple time scales, effectively correlating the impact of data at different time scales on residential electricity load prediction, and improving the accuracy of residential electricity load prediction.

[0131] The other parts of this embodiment are the same as those of the above-mentioned embodiment 1, and thus will not be described in detail.

[0132] Embodiment 3:

[0133] This embodiment, based on any one of the above-mentioned embodiments 1-2, proposes a residential electricity load prediction system based on a time-space conduction network, which is used to execute the above-mentioned residential electricity load prediction method based on a time-space conduction network, including a fusion unit, an analysis unit, and a prediction unit;

[0134] The fusion unit is used to fuse the collected residential electricity load data and external multi-source data of residential electricity load to obtain fused residential electricity load data;

[0135] The analysis unit is used to analyze the residential power load fusion data to obtain the residential power load data components, and analyze the impact of external disturbances according to the residential power load data components;

[0136] The prediction unit is used to call the LSTM network to predict the residential electricity load data components, calculate the weights of the predicted residential electricity load data components, and obtain the residential electricity load according to the weight prediction.

[0137] This embodiment also proposes an electronic device, including a memory and a processor; the memory stores a computer program; when the computer program is executed on the processor, the above-mentioned residential electricity load prediction method based on the space-time conduction network is implemented.

[0138] This embodiment also proposes a computer-readable storage medium, characterized in that computer instructions are stored on the computer-readable storage medium; when the computer instructions are executed on the above-mentioned electronic device, the above-mentioned residential electricity load prediction method based on the space-time conduction network is implemented.

[0139] The other parts of this embodiment are the same as any one of the above-mentioned embodiments 1-2, so they will not be repeated here.

[0140] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any simple modification or equivalent change made to the above embodiment based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for predicting residential electricity load based on a spatiotemporal conduction network, characterized in that: The specific steps include: Step S1: integrating the collected residential electricity load data and external multi-source data of residential electricity load to obtain residential electricity load integrated data; Step S2: Decomposing the residential power load fusion data to obtain residential power load data components, and analyzing the impact of external disturbances based on the residential power load data components; Step S3: calling the LSTM network to predict the residential electricity load data components, and calculating the weights of the predicted residential electricity load data components, and predicting the residential electricity load according to the weights; The step S2 specifically includes the following steps: Step S21: calling the FCM clustering method to analyze the residential power load fusion data components and obtain the membership of different types of electrical appliances in the residential power load fusion data; Step S22: Calling the network analysis method to construct a dynamic network for external disturbance conduction of residential power loads, and obtaining the network density O of residential power loads RPL , the spatial-temporal transmission network correlation of residential power load O RT ; Where: n NET is the number of nodes in the space-time transmission network for residential electricity load; m RS is the actual relationship number of the space-time transmission network of residential electricity load; Where: r RS The number of inaccessible points in the space-time transmission network for residential electricity load; Step S23: Obtain external disturbance impact data based on the membership of different types of electrical appliances in the residential power load fusion data, the network density of the residential power load, and the spatial-temporal conduction network correlation of the residential power load.

2. The method for predicting residential power load based on a time-space conduction network according to claim 1, characterized in that: The step S1 specifically includes the following steps: Step S11: Acquire historical data of residential electricity load from the power consumption information collection system of the power supply company to obtain residential electricity load data; Step S12: calling the OCR algorithm, calculating the connection weights between the two convolutional layers of the external multi-source data image of the residential power load, outputting the recognized characters, and obtaining the external multi-source data of the residential power load; Step S13: Calling linear regression to repair the external multi-source data of residential power load to obtain the repaired external multi-source data of residential power load; Step S14: extracting multi-source data features from the external multi-source data of the repaired residential power load, and calling a decision function to fuse the external multi-source data of the repaired residential power load and the residential power load data to obtain residential power load fusion data.

3. The method for predicting residential electricity load based on a time-space conduction network according to claim 1, characterized in that: The step S3 specifically comprises the following steps: Step S31: calling the LSTM network to predict the components of the residential electricity load data; Step S32: using the entropy weight method to calculate the impact weight of the residential electricity load data component to obtain the optimal impact weight value; Step S33: Predict the residential electricity load based on the impact weight.

4. The method for predicting residential power load based on a time-space conduction network according to claim 3, characterized in that: The specific operation of step S31 is: calling the LSTM network, and predicting the lth residential electricity load component data based on the residential electricity load fusion data, external disturbance impact data, the set activation function, the photovoltaic power generation power of the residential households, and the energy storage discharge power of the residential households.

5. The method for predicting residential power load based on time-space conduction network according to claim 3 is characterized in that: The specific operation of step S32 is: calling the entropy weight method, calculating the residential electricity load component prediction conditional entropy and the residential electricity load component prediction information entropy, obtaining the joint entropy of the residential electricity load component prediction, obtaining the influence weight of the residential electricity load data component according to the joint entropy, and calculating the optimal value of the influence weight.

6. A residential electricity load forecasting system based on a time-space conduction network, used to execute the residential electricity load forecasting method based on a time-space conduction network as claimed in claim 1, characterized in that: It includes fusion unit, analysis unit and prediction unit; The fusion unit is used to fuse the collected residential electricity load data and external multi-source data of residential electricity load to obtain fused residential electricity load data; The analysis unit is used to analyze the residential power load fusion data to obtain the residential power load data components, and analyze the impact of external disturbances according to the residential power load data components; The prediction unit is used to call the LSTM network to predict the residential electricity load data components, calculate the weights of the predicted residential electricity load data components, and obtain the residential electricity load according to the weight prediction.

7. An electronic device, characterized in that: It comprises a memory and a processor; the memory stores a computer program; when the computer program is executed on the processor, the residential electricity load prediction method based on the space-time conduction network as described in any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions; when the computer instructions are executed on the electronic device as described in claim 7, the residential electricity load prediction method based on the space-time conduction network as described in any one of claims 1-5 is implemented.

Images