A method and system for predicting building electrical load based on deep learning

Abstract

The present invention relates to the field of electrical load prediction, and in particular to a method and system for predicting building electrical load based on deep learning. The method comprises the following steps: obtaining operating parameters of electrical equipment and building traffic flow monitoring parameters; identifying electrical equipment in the building based on sensors, and marking building electrical equipment nodes; reconstructing the spatial topological plane of building electrical equipment nodes, and constructing a spatial topological network of electrical equipment; performing multi-scale time-frequency decomposition on the operating parameters of electrical equipment, and generating multi-time-frequency frequency domain segments; performing frequency energy distribution analysis on the multi-time-frequency frequency domain segments, and generating time-frequency distribution features; performing time-series voltage feature extraction processing on the operating parameters of electrical equipment, and generating a time-series voltage feature sequence; performing multi-period trend evolution on the time-series voltage feature sequence based on the time-frequency distribution features, and generating multi-period load trend feature data. The present invention realizes high-precision and highly adaptive building electrical load prediction.

Description

Technical Field

[0001] The present invention relates to the field of electrical load prediction, and in particular to a method and system for predicting building electrical load based on deep learning. Background Art

[0002] Building electrical energy consumption management plays a key role in improving energy utilization efficiency and reducing environmental pollution. Accurate prediction of building electrical load is the basis for realizing intelligent energy consumption management, which is of great significance for effectively dispatching power resources, optimizing power consumption structure and reducing power waste. However, traditional building electrical load prediction methods usually rely on manual observation, empirical models, etc., and have problems such as low prediction accuracy and poor adaptability. With the rapid development of big data and artificial intelligence technology, the use of deep learning methods for building electrical load prediction has become a new trend. Deep learning is a branch of machine learning. It automatically learns features from massive data and establishes complex prediction models. It has made breakthrough progress in various prediction and classification problems. In the field of building electrical load prediction, deep learning models use multi-source heterogeneous data such as historical electricity consumption data, meteorological data, and building characteristics to automatically mine the complex associations contained in the data and construct a load prediction model with strong adaptability and high prediction accuracy. Therefore, an intelligent building electrical load prediction method is needed to effectively solve the problems of traditional methods and provide strong technical support for building energy consumption management. Summary of the invention

[0003] In order to solve the above-mentioned technical problems, the present invention proposes a building electrical load prediction method and system based on deep learning to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above object, the present invention provides a building electrical load prediction method based on deep learning, comprising the following steps:

[0005] Step S1: Obtaining operating parameters of electrical equipment and building traffic monitoring parameters; identifying electrical equipment in the building based on sensors, marking building electrical equipment nodes; reconstructing the spatial topology of building electrical equipment nodes, and constructing a spatial topology network of electrical equipment;

[0006] Step S2: performing multi-scale time-frequency decomposition on the operating parameters of the electrical equipment to generate multiple time-frequency frequency domain segments; performing frequency energy distribution analysis on the multiple time-frequency frequency domain segments to generate time-frequency distribution features;

[0007] Step S3: extracting time-series voltage characteristics from the operating parameters of the electrical equipment to generate a time-series voltage characteristic sequence; performing multi-period trend evolution on the time-series voltage characteristic sequence based on the time-frequency distribution characteristics to generate multi-period load trend characteristic data;

[0008] Step S4: obtaining pedestrian dynamic trajectories based on building pedestrian flow monitoring parameters; calculating spatial power load demand for pedestrian dynamic trajectories based on the spatial topological network of power-consuming equipment, thereby obtaining spatial power demand data for each area;

[0009] Step S5: using the multi-period load trend characteristic data and the spatial power demand data of each region to perform dynamic load multi-step prediction on the spatial topology network of power-consuming equipment, and generate regional multi-step load prediction data;

[0010] Step S6: Perform deep load prediction learning on the regional multi-time-step load prediction data and construct a deep load prediction model to perform building electrical load prediction operations.

[0011] The present invention obtains the operating parameters of electrical equipment to understand the key parameters of each device, such as power, current, voltage, and their changes and trends, and obtains building traffic monitoring parameters to provide information about the activities and flows of people in the building, which is very important for the analysis and prediction of electrical loads. The electrical equipment in the building is identified by sensors, and they are marked as nodes to construct the topological structure of the electrical equipment, making the analysis of the power system more accurate and feasible. The spatial topological plane of the building electrical equipment nodes is reconstructed, and the connections and relationships between the electrical equipment are established to form a spatial topological network of electrical equipment. This network reflects the physical locations and connection methods between different devices, providing a basis for subsequent load prediction and optimization, and is very important for the analysis and prediction of electrical equipment. The equipment operation parameters are decomposed into different time-frequency domain segments to capture the change characteristics at different time and frequency scales. The frequency energy distribution of these time-frequency domain segments is analyzed to understand the energy distribution in different frequency ranges, which helps to identify and understand the characteristics of equipment operation. The time-series voltage feature extraction processing is performed on the operating parameters of the power equipment to extract the time-related voltage feature sequences. These time-series voltage feature sequences reflect the dynamic changes of equipment operation and provide important information for subsequent load prediction and fault detection. Based on the time-frequency distribution characteristics, the time-series voltage feature sequences are subjected to multi-period trend evolution to analyze the change trend of the equipment load in different time periods. These multi-period load trend feature data It is used to understand the periodic, seasonal and trend changes of equipment loads, provide a reference for load prediction and optimization, and use the building traffic monitoring parameters to obtain the dynamic trajectory of pedestrians to understand the movement path and distribution of people in the building. This is important for predicting the power demand in different areas and optimizing the distribution of building electrical loads. Based on the spatial topological network of electrical equipment, the spatial power load demand is calculated for the dynamic trajectory of pedestrians, and the location and connection relationship of pedestrian activities with electrical equipment are combined to calculate the power demand of each area, optimize the distribution of power loads and power supply strategies, and use multi-period load trend feature data and spatial power demand data of each region to dynamically predict the load of the spatial topological network of electrical equipment, and predict the power consumption in multiple time steps in the future. The load conditions provide a reference for the operation and planning of the power system. The generated regional multi-time-step load forecast data is used to formulate power supply strategies, load balancing and energy optimization decisions. The accuracy and timeliness of the forecast data are crucial to ensuring the stable operation and efficient utilization of the power system. Deep load forecasting learning is performed on the regional multi-time-step load forecast data, and a load forecasting model is constructed using deep learning and other technologies. This model more accurately predicts the future changes in the building's electrical load and provides more reliable forecast results. The constructed deep load forecasting model is applied to the operation management, energy scheduling and load control of the building's power system. Through accurate load forecasting, effective energy utilization, reasonable load distribution and efficient operation of the power system can be achieved.

[0012] Preferably, step S1 comprises the following steps:

[0013] Step S11: Acquire the operating parameters of the electrical equipment and the monitoring parameters of the building's human traffic; identify the electrical equipment in the building based on the sensor, and mark the building's electrical equipment nodes;

[0014] Step S12: Calculate the spatial position of the building electrical equipment nodes to obtain the node spatial coordinates;

[0015] Step S13: performing spatial distribution analysis on building electrical equipment nodes based on node spatial coordinates to obtain electrical equipment node spatial distribution data;

[0016] Step S14: mining the interactive associations between building electrical equipment nodes to generate interactive associations between nodes;

[0017] Step S15: reconstructing the spatial topology plane of the spatial distribution data of the power-consuming equipment nodes based on the interactive association relationship between the nodes, and constructing a spatial topology network of the power-consuming equipment.

[0018] The present invention provides key data by acquiring the operating parameters of electrical equipment and the building traffic monitoring parameters, and provides basic information for building electrical load prediction. These parameters include the operating parameters of the electrical equipment such as power, current, voltage, and the building traffic monitoring parameters, such as the number of people and location information. The electrical equipment in the building is identified and marked by sensors, and nodes of the electrical equipment are established to identify and distinguish different electrical equipment, and provide a basis for subsequent spatial analysis and topological reconstruction. By calculating the spatial position of the building electrical equipment nodes, the specific position of each node in the building is determined, the distribution of the electrical equipment in the building is understood, and accurate location information is provided for subsequent spatial analysis and topological reconstruction. The spatial distribution analysis of the building electrical equipment nodes based on the node spatial coordinates reveals the distribution pattern and characteristics of the electrical equipment in the building. , understand the density, concentration and spatial relationship of power equipment in different areas, provide a basis for load prediction and optimization, conduct equipment interaction mining on building power equipment nodes to discover the interaction relationship and connection method between different equipment, understand the mutual influence and dependency between equipment, and provide important information for the analysis and optimization of power systems. Based on the interaction relationship between nodes, the spatial topological plane of the spatial distribution data of power equipment nodes is reconstructed to establish a spatial topological network of power equipment. This network reflects the connection and relationship between equipment, analyzes and predicts the load distribution, transmission path and influencing factors in the power system, and constructs a spatial topological network of power equipment for analysis, optimization and fault diagnosis of the power system. By analyzing the topological structure and characteristics of the network, the rationality of the load distribution is evaluated, potential fault points are identified, and corresponding response strategies are formulated.

[0019] Preferably, step S2 comprises the following steps:

[0020] Step S21: performing segmented low-pass filtering on the operating parameters of the electrical equipment to obtain low-pass filtered operating parameters;

[0021] Step S22: performing non-overlapping timing window division based on the low-pass filtering operation parameters to obtain a plurality of non-overlapping timing windows;

[0022] Step S23: performing multi-scale time-frequency decomposition on the low-pass filter operating parameters using multiple non-overlapping time series windows to generate multiple time-frequency frequency domain segments;

[0023] Step S24: performing frequency energy distribution analysis on multiple time-frequency frequency domain segments to generate energy distribution data for each frequency domain segment;

[0024] Step S25: Perform time-frequency distribution fitting on the energy distribution data of each frequency domain segment to generate time-frequency distribution features.

[0025] The present invention performs segmented low-pass filtering on the operating parameters of electrical equipment to reduce the influence of high-frequency noise and instantaneous fluctuations, extracts smoother low-frequency components, eliminates instantaneous changes in operating parameters, makes them more representative and stable, and provides a more reliable data basis for subsequent time-frequency analysis and feature extraction. By performing non-overlapping time series window division on the low-pass filtered operating parameters, the long-term operating parameter sequence is divided into multiple relatively independent time series windows, the characteristics and change trends in different time periods are analyzed and captured, and more detailed and comprehensive information is provided. The low-pass filtered operating parameters are subjected to multi-scale time-frequency decomposition using multiple non-overlapping time series windows to convert time domain data into time-frequency domain data, and the operating parameter changes on different frequencies and time scales are captured. ization, providing more detailed and comprehensive time-frequency information. The generated multi-time-frequency frequency domain segments reflect the energy distribution and characteristics in different frequency ranges, providing a basis for subsequent frequency energy distribution analysis and time-frequency distribution fitting. Frequency energy distribution analysis is performed on multiple time-frequency frequency domain segments to calculate the energy distribution at different frequencies in each frequency domain segment, understand the energy contribution and importance in different frequency ranges, and provide a basis for subsequent time-frequency distribution fitting and feature extraction. Time-frequency distribution fitting is performed on the energy distribution data of each frequency domain segment to extract the time-frequency distribution characteristics. These characteristics reflect the time-varying characteristics and distribution laws of energy in different frequency ranges, identify and describe the time-frequency characteristics of electrical loads, and provide a more comprehensive and accurate feature description for building electrical load prediction.

[0026] Preferably, step S3 comprises the following steps:

[0027] Step S31: extracting the time-sequence voltage characteristics of the operating parameters of the electrical equipment to obtain the time-sequence voltage characteristics of the electrical equipment;

[0028] Step S32: reorganizing the time-series voltage characteristics of the electrical equipment into time-series hierarchical representations to generate a time-series voltage characteristic sequence;

[0029] Step S33: performing voltage load deduction on the time-series voltage characteristic sequence, thereby generating a voltage load time-series curve;

[0030] Step S34: performing load frequency domain variation analysis on the voltage-load timing curve based on the time-frequency distribution characteristics to obtain voltage-load frequency domain variation data;

[0031] Step S35: Perform multi-period trend evolution on the voltage load frequency domain variation data to generate multi-period load trend characteristic data.

[0032] The present invention extracts the time series features related to voltage by performing time series voltage feature extraction processing on the operating parameters of the power consumption equipment. These features include the amplitude, frequency, phase and other information related to the voltage waveform. The voltage features are extracted to capture the dynamic characteristics of voltage changes, and the voltage features are organized and represented according to different time scales by performing time series hierarchical characterization and reorganization on the time series voltage features of the power consumption equipment. The voltage change law and trend on different time scales are analyzed to provide multi-scale voltage information. By performing voltage load deduction on the time series voltage feature sequence, the corresponding load change situation is inferred according to the voltage feature. Understand the relationship between voltage and load, and infer the change trend and characteristics of the load. The generated voltage-load time series curve provides the time series information of the load change, and the load frequency domain change analysis is performed on the voltage-load time series curve based on the time-frequency distribution characteristics to convert the load time series curve in the time domain into the load frequency domain characteristics in the frequency domain. Analyze the energy distribution and characteristics of the load change in different frequency ranges. The generated voltage-load frequency domain change data provides the energy distribution of the load in the frequency domain, and the multi-period trend evolution analysis is performed on the voltage-load frequency domain change data to analyze the trend and evolution law of the load change in different time periods. Understand the load change patterns and trends in different time periods and provide long-term evolution characteristics of the load. The generated multi-period load trend characteristic data reflects the periodicity, seasonality and trend of load changes.

[0033] Preferably, the specific steps of step S34 are:

[0034] Perform wavelet decomposition on the voltage-load timing curve to generate voltage-load frequency domain data;

[0035] Based on the time-frequency distribution characteristics, the voltage load frequency domain data is fitted with time-frequency transformation to obtain the voltage load time-frequency matrix;

[0036] Calculate the adjacent time-frequency characteristic difference of the voltage-load time-frequency matrix to generate the load power change value at each frequency point;

[0037] The load power change value at each frequency point is analyzed for change amplitude distribution to obtain the voltage load frequency domain change data.

[0038] The present invention performs wavelet decomposition on the voltage load timing curve, decomposes the voltage load signal in the time domain into wavelet coefficients of different frequencies, extracts the load frequency domain characteristics in different frequency ranges, captures the energy distribution of the load at different frequencies, and generates voltage load frequency domain data that provides more detailed and accurate frequency domain information. By performing time-frequency transformation fitting on the voltage load frequency domain data, the load information in the frequency domain is converted into a time-frequency matrix representation, and the change characteristics of the load at different time periods and frequencies are analyzed to provide the time-frequency distribution information of the load. By performing adjacent time-frequency characteristic difference calculation on the voltage load time-frequency matrix, the change value of the load power at each frequency point is obtained, the degree of power change of the load at different frequencies is quantified, and the change information of the load at each frequency point is provided. The generated load power change value reflects the dynamic change of the load at different frequencies. By performing change amplitude distribution analysis on the load power change value at each frequency point, the change amplitude distribution of the load at different frequencies is obtained, the change amplitude and distribution characteristics of the load within different frequency ranges are understood, and more comprehensive and accurate voltage load frequency domain change data are provided.

[0039] Preferably, the specific steps of step S4 are:

[0040] Step S41: performing real-time pedestrian position tracking on building traffic monitoring parameters and extracting real-time pedestrian position sequences;

[0041] Step S42: filtering abnormal time intervals of real-time pedestrian positions to obtain a filtered pedestrian position sequence;

[0042] Step S43: performing discrete sample connection on the filtered pedestrian position sequence to obtain the pedestrian dynamic trajectory;

[0043] Step S44: mining the regional pedestrian trajectory features of the pedestrian dynamic trajectory based on the spatial topological network of the electrical equipment, thereby generating the pedestrian trajectory features in the region;

[0044] Step S45: Calculate the spatial power load demand based on the pedestrian trajectory characteristics in the area, so as to obtain the spatial power demand data of each area.

[0045] The present invention tracks the real-time pedestrian position of building traffic monitoring parameters, obtains the position information of pedestrians in the building at different time points, extracts the real-time pedestrian position sequence to obtain a series of temporally continuous pedestrian position data, and provides a basis for subsequent analysis. By filtering the real-time pedestrian position for abnormal time intervals, abnormal or inaccurate pedestrian position data are excluded, and the accuracy and reliability of the data are improved. The filtered pedestrian position sequence eliminates noise or errors and obtains more reliable pedestrian position data. By connecting discrete samples of the filtered pedestrian position sequence, the pedestrian position samples adjacent in time are connected to form the dynamic trajectory of the pedestrian. The obtained pedestrian dynamic trajectory reveals the movement path and Behavior patterns provide more detailed pedestrian activity information for subsequent analysis. Based on the spatial topological network of power-consuming equipment, regional pedestrian trajectory feature mining is performed on pedestrian dynamic trajectories, that is, the trajectory characteristics of pedestrians in different areas are analyzed. Through regional pedestrian trajectory feature mining, the activity patterns, density distribution or other related characteristics of pedestrians in different areas are discovered, which provides more refined information for subsequent load forecasting. The spatial power load demand is calculated based on the pedestrian trajectory characteristics in the region. According to the characteristics of pedestrian activity patterns and density distribution, the power load demand of each area is estimated, and the spatial power demand data of each area is obtained. The power demand information of different areas is provided, which provides a more accurate basis for the prediction and optimization of building electrical load.

[0046] Preferably, the specific steps of step S44 are:

[0047] Perform crowd flow fluctuation statistics on building crowd flow monitoring parameters to obtain crowd flow fluctuation data;

[0048] Performing regional segmentation on the spatial topological network of electrical equipment to obtain spatial regions of electrical equipment;

[0049] Calculate the regional human flow density of the electrical equipment space area based on human flow fluctuation data to generate multiple regional human flow density data;

[0050] Based on the pedestrian flow density data of multiple regions, the spatial intersection distribution of pedestrian dynamic trajectories is analyzed to obtain the spatial intersection distribution data of dynamic trajectories;

[0051] Calculate the dwell time of pedestrian dynamic trajectory to obtain the trajectory dwell time;

[0052] The regional pedestrian trajectory features are mined based on the dynamic trajectory spatial intersection distribution data and trajectory dwell time, thereby generating the pedestrian trajectory features in the region.

[0053] The present invention analyzes the change of the flow of people in the building, including the peak period and the trough period, through the statistics of the fluctuation of the flow of people. The flow of people fluctuation data provides the time series information of the flow of people, which is used to understand the change trend and periodic characteristics of the flow of people in different time periods, predict the future flow of people in the building, and provide data support in the time dimension for the prediction of the electrical load. The electrical equipment in the building is divided into different areas through the regional segmentation of the spatial topological network of the electrical equipment. The division of the spatial area of ​​the electrical equipment is based on the physical position relationship or functional relationship between the equipment, and the electrical equipment in the building is divided into relatively independent parts. The density of the flow of people in the spatial area of ​​the electrical equipment is calculated based on the flow of people fluctuation data, and the estimation of each area is carried out. The pedestrian flow intensity in each area is estimated by associating the pedestrian flow fluctuation data with the spatial area of ​​electrical equipment. The spatial intersection distribution analysis of pedestrian dynamic trajectories reveals the spatial intersection of pedestrians in the building. By combining the pedestrian flow density data of multiple areas, the intersection and flow paths of pedestrians between different areas are analyzed to understand the aggregation and dispersion of pedestrians in the building. The pedestrian intersection distribution data is provided for subsequent pedestrian trajectory feature mining and electrical load prediction. The residence time of pedestrian dynamic trajectories is calculated and analyzed to analyze the length of time pedestrians stay in different areas. By identifying the stop points and stay time in the trajectory, the activity intensity and stay preference of pedestrians in different areas are estimated.

[0054] Preferably, the specific steps of step S5 are:

[0055] Step S51: using multi-period load trend feature data to perform three-dimensional mapping and rendering processing on the space topology network of electrical equipment to construct a three-dimensional space load network;

[0056] Step S52: performing regional load evolution on the three-dimensional spatial load network based on the spatial power demand data of each region to obtain regional load evolution data;

[0057] Step S53: Perform dynamic load multi-time-step prediction on the regional load evolution data to generate regional multi-time-step load prediction data.

[0058] The present invention displays the spatial distribution of electrical equipment in a visual manner through a three-dimensional spatial load network, providing an intuitive understanding of the layout of electrical equipment in a building.

[0059] Through 3D mapping and rendering, the multi-period load trend characteristic data is combined with the spatial topological network of power equipment, providing visual data support for subsequent regional load evolution and load forecasting. The regional load evolution data reflects the changes in electricity demand in different regions and reveals the load distribution and load change trend of each area in the building.

[0060] By performing regional load evolution on the three-dimensional spatial load network based on the spatial power demand data of each region, the power demand is associated with the spatial location, providing data support for the spatial distribution of subsequent load forecasts. The regional multi-time-step load forecast data provides the load conditions of each area in the building in multiple future time steps, which is used to evaluate the load conditions of the electrical system and plan the power supply.

[0061] Dynamic load multi-time-step forecasting is performed based on regional load evolution data. Historical load data and trend information are used in combination with forecasting models and algorithms to predict and speculate future loads.

[0062] Preferably, the specific steps of step S6 are:

[0063] Step S61: performing loss error calculation on the regional multi-time-step load forecast data to generate a load forecast error parameter;

[0064] Step S62: performing back propagation iterative optimization on the regional multi-time-step load forecast data based on the load forecast error parameter to obtain error-optimized load forecast data;

[0065] Step S63: Perform deep load prediction learning on the error-optimized load prediction data and construct a deep load prediction model to perform building electrical load prediction operations.

[0066] The present invention evaluates the accuracy and reliability of the load prediction model by calculating the loss error between the predicted data and the actual data.

[0067] The generated load forecast error parameters provide a measure and feedback of the forecast results for subsequent error optimization and model improvement. Through back-propagation iterative optimization, the load forecast error parameters are used to correct and adjust the forecast data, reduce the forecast error, and improve the forecast accuracy.

[0068] Error-optimized load forecasting data reflects optimized and adjusted load forecasting results, which are closer to the actual load situation and provide a more accurate data basis for the subsequent load forecasting model construction. Deep load forecasting learning uses error-optimized load forecasting data for model training and learning, and can learn load characteristics, changing trends and laws from a large amount of data, improving the accuracy and robustness of load forecasting.

[0069] The constructed deep load prediction model can predict the building electrical load and provide prediction results with high accuracy and reliability, providing decision support for the planning and regulation of the power system.

[0070] In this specification, a building electrical load prediction system based on deep learning is provided, which is used to execute the building electrical load prediction method based on deep learning as described above, including:

[0071] The topological network module is used to obtain the operating parameters of electrical equipment and the monitoring parameters of building traffic flow; identify electrical equipment in the building based on sensors and mark the nodes of building electrical equipment; reconstruct the spatial topological plane of the nodes of building electrical equipment and construct the spatial topological network of electrical equipment;

[0072] The time-frequency distribution module is used to perform multi-scale time-frequency decomposition on the operating parameters of the power-consuming equipment to generate multiple time-frequency frequency domain segments; perform frequency energy distribution analysis on multiple time-frequency frequency domain segments to generate time-frequency distribution features;

[0073] The load trend module is used to extract the time-series voltage characteristics of the operating parameters of the power-consuming equipment to generate a time-series voltage characteristic sequence; based on the time-frequency distribution characteristics, the time-series voltage characteristic sequence is subjected to multi-period trend evolution to generate multi-period load trend characteristic data;

[0074] The power load demand module is used to obtain the dynamic trajectory of pedestrians based on the building traffic monitoring parameters; the spatial power load demand of pedestrian dynamic trajectories is calculated based on the spatial topological network of power-consuming equipment, so as to obtain the spatial power demand data of each area;

[0075] The multi-time-step prediction module is used to perform dynamic load multi-time-step prediction on the spatial topology network of power-consuming equipment using multi-period load trend characteristic data and spatial power demand data of each region, and generate regional multi-time-step load prediction data;

[0076] The load forecasting learning module is used to perform deep load forecasting learning on regional multi-time-step load forecasting data and build a deep load forecasting model to perform building electrical load forecasting operations.

[0077] The present invention obtains the operating parameters of power equipment and the monitoring parameters of building traffic flow, provides the correlation information of power equipment and human flow in the building, provides basic data for subsequent analysis and prediction, marks the nodes of building power equipment, establishes the topological structure of power equipment, facilitates the subsequent spatial analysis and load prediction, performs multi-scale time-frequency decomposition, separates and extracts the time domain and frequency domain information of the operating parameters of power equipment, reveals the operating characteristics and changing laws of power equipment, and frequency energy distribution analysis provides energy distribution in different frequency bands to help identify the frequency characteristics of power equipment, provides a basis for subsequent feature extraction and analysis, extracts the time series voltage feature sequence, captures the time series changes of the operating parameters of power equipment, and provides time dimension information for load trend analysis. Multi-period trend evolution analysis reveals the changing trend of the load of power equipment in different time periods, understands the load characteristics and laws of power equipment, and is based on the monitoring parameters of building traffic flow and the spatial topological network of power equipment. The network is used to calculate the spatial electricity demand of each area, and the distribution and demand of electricity load in different areas are provided. The calculation of the spatial electricity load demand considering the dynamic trajectory of pedestrians makes the electricity demand closer to the actual situation, which provides a basis for accurately predicting the electrical load of buildings. The dynamic load multi-step prediction of the spatial topological network of power equipment is carried out by using the multi-period load trend characteristic data and the spatial electricity demand data of each region, and the load prediction results for multiple time steps in the future are provided. The multi-time-step load prediction data provides a decision-making basis for the planning and regulation of the power system, evaluates the load status, optimizes the power supply and resource allocation, and constructs a deep load prediction model through deep load prediction learning of the regional multi-time-step load prediction data to improve the accuracy and robustness of the load prediction. The deep load prediction model can predict the electrical load of the building, provide accurate prediction results for the operation and regulation of the power system, and support the rational allocation of power resources and the optimization of energy utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 A schematic diagram of the steps of a method for predicting building electrical load based on deep learning according to the present invention;

[0079] Figure 2 Detailed implementation flow chart of step S1;

[0080] Figure 3 Detailed implementation flow chart of step S2;

[0081] Figure 4 Detailed implementation flow chart of step S3. DETAILED DESCRIPTION

[0082] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0083] The present application example provides a method and system for forecasting building electrical load based on deep learning. The execution subjects of the method and system for forecasting building electrical load based on deep learning include but are not limited to: mechanical equipment, data processing platform, cloud server node, network upload device, etc. equipped with the system can be regarded as the general computing node of the present application, and the data processing platform includes but is not limited to: at least one of audio and image management system, information management system, and cloud data management system.

[0084] See also Figures 1 to 4 The present invention provides a building electrical load prediction method based on deep learning, comprising the following steps:

[0085] Step S1: Obtaining operating parameters of electrical equipment and building traffic monitoring parameters; identifying electrical equipment in the building based on sensors, marking building electrical equipment nodes; reconstructing the spatial topology of building electrical equipment nodes, and constructing a spatial topology network of electrical equipment;

[0086] Step S2: performing multi-scale time-frequency decomposition on the operating parameters of the electrical equipment to generate multiple time-frequency frequency domain segments; performing frequency energy distribution analysis on the multiple time-frequency frequency domain segments to generate time-frequency distribution features;

[0087] Step S3: extracting time-series voltage characteristics from the operating parameters of the electrical equipment to generate a time-series voltage characteristic sequence; performing multi-period trend evolution on the time-series voltage characteristic sequence based on the time-frequency distribution characteristics to generate multi-period load trend characteristic data;

[0088] Step S4: obtaining pedestrian dynamic trajectories based on building pedestrian flow monitoring parameters; calculating spatial power load demand for pedestrian dynamic trajectories based on the spatial topological network of power-consuming equipment, thereby obtaining spatial power demand data for each area;

[0089] Step S5: using the multi-period load trend characteristic data and the spatial power demand data of each region to perform dynamic load multi-step prediction on the spatial topology network of power-consuming equipment, and generate regional multi-step load prediction data;

[0090] Step S6: Perform deep load prediction learning on the regional multi-time-step load prediction data and construct a deep load prediction model to perform building electrical load prediction operations.

[0091] The present invention obtains the operating parameters of electrical equipment to understand the key parameters of each device, such as power, current, voltage, and their changes and trends, and obtains building traffic monitoring parameters to provide information about the activities and flows of people in the building, which is very important for the analysis and prediction of electrical loads. The electrical equipment in the building is identified by sensors, and they are marked as nodes to construct the topological structure of the electrical equipment, making the analysis of the power system more accurate and feasible. The spatial topological plane of the building electrical equipment nodes is reconstructed, and the connections and relationships between the electrical equipment are established to form a spatial topological network of electrical equipment. This network reflects the physical locations and connection methods between different devices, providing a basis for subsequent load prediction and optimization, and is very important for the analysis and prediction of electrical equipment. The equipment operation parameters are decomposed into different time-frequency domain segments to capture the change characteristics at different time and frequency scales. The frequency energy distribution of these time-frequency domain segments is analyzed to understand the energy distribution in different frequency ranges, which helps to identify and understand the characteristics of equipment operation. The time-series voltage feature extraction processing is performed on the operating parameters of the power equipment to extract the time-related voltage feature sequences. These time-series voltage feature sequences reflect the dynamic changes of equipment operation and provide important information for subsequent load prediction and fault detection. Based on the time-frequency distribution characteristics, the time-series voltage feature sequences are subjected to multi-period trend evolution to analyze the change trend of the equipment load in different time periods. These multi-period load trend feature data It is used to understand the periodic, seasonal and trend changes of equipment loads, provide a reference for load prediction and optimization, and use the building traffic monitoring parameters to obtain the dynamic trajectory of pedestrians to understand the movement path and distribution of people in the building. This is important for predicting the power demand in different areas and optimizing the distribution of building electrical loads. Based on the spatial topological network of electrical equipment, the spatial power load demand is calculated for the dynamic trajectory of pedestrians, and the location and connection relationship of pedestrian activities with electrical equipment are combined to calculate the power demand of each area, optimize the distribution of power loads and power supply strategies, and use multi-period load trend feature data and spatial power demand data of each region to dynamically predict the load of the spatial topological network of electrical equipment, and predict the power consumption in multiple time steps in the future. The load conditions provide a reference for the operation and planning of the power system. The generated regional multi-time-step load forecast data is used to formulate power supply strategies, load balancing and energy optimization decisions. The accuracy and timeliness of the forecast data are crucial to ensuring the stable operation and efficient utilization of the power system. Deep load forecasting learning is performed on the regional multi-time-step load forecast data, and a load forecasting model is constructed using deep learning and other technologies. This model more accurately predicts the future changes in the building's electrical load and provides more reliable forecast results. The constructed deep load forecasting model is applied to the operation management, energy scheduling and load control of the building's power system. Through accurate load forecasting, effective energy utilization, reasonable load distribution and efficient operation of the power system can be achieved.

[0092] In the embodiment of the present invention, refer to Figure 1 , is a schematic diagram of the steps of a building electrical load prediction method based on deep learning of the present invention. In this example, the steps of the building electrical load prediction method based on deep learning include:

[0093] Step S1: Obtaining operating parameters of electrical equipment and building traffic monitoring parameters; identifying electrical equipment in the building based on sensors, marking building electrical equipment nodes; reconstructing the spatial topology of building electrical equipment nodes, and constructing a spatial topology network of electrical equipment;

[0094] In this embodiment, the operating parameters of each electrical equipment in the building are collected, including information such as current, voltage, and power. These parameters are obtained through power monitoring equipment or sensors. After obtaining user authorization, the crowd flow monitoring parameters in the building are collected, such as using equipment such as crowd flow sensors or surveillance cameras to obtain information on people entering and leaving the building or a specific area. Sensor technology, such as non-invasive current sensors, power monitoring equipment, or smart meters, is used to identify and monitor the electrical equipment in the building. By connecting sensors to electrical equipment or installing sensors around electrical equipment, the operating data of the electrical equipment is obtained in real time, and each electrical equipment is marked for subsequent spatial topological reconstruction and network construction. According to the structure and layout of the building, the marked building electrical equipment nodes are reconstructed in spatial topological planes, that is, the positions and relative relationships of the nodes in the building are determined, and the physical connections and power transmission paths between the electrical equipment are considered. The electrical equipment nodes are connected to form a spatial topological network of the electrical equipment, and the connection methods between the electrical equipment are determined, such as direct connection, connection through a distribution box, etc., as well as the connected power transmission paths.

[0095] Step S2: performing multi-scale time-frequency decomposition on the operating parameters of the electrical equipment to generate multiple time-frequency frequency domain segments; performing frequency energy distribution analysis on the multiple time-frequency frequency domain segments to generate time-frequency distribution features;

[0096] In this embodiment, an appropriate time-frequency analysis method is selected, such as wavelet transform, short-time Fourier transform (STFT), etc., to convert the operating parameters of the electrical equipment from the time domain to the time-frequency domain, decompose the operating parameters of the electrical equipment according to different scales, and use a multi-scale analysis method, such as wavelet scale decomposition or multi-resolution analysis, to obtain time-frequency domain segments of the operating parameters of the electrical equipment at different scales and frequencies after time-frequency decomposition. A frequency energy distribution analysis is performed on each time-frequency domain segment to determine the energy distribution at different frequencies. The frequency energy distribution analysis is performed by calculating the energy value or power spectrum density at each frequency in each time-frequency domain segment. For each time-frequency domain segment, a frequency energy distribution feature, that is, the energy distribution at different frequencies, is obtained. The frequency energy distribution features of multiple time-frequency domain segments are comprehensively considered to generate the time-frequency distribution features of the electrical equipment. Statistical analysis methods, such as mean, variance, peak, etc., are used to integrate and describe the frequency energy distribution features of multiple time-frequency domain segments. The generated time-frequency distribution features are used to analyze the working status, energy consumption mode, abnormal behavior, etc. of the electrical equipment.

[0097] Step S3: extracting time-series voltage characteristics from the operating parameters of the electrical equipment to generate a time-series voltage characteristic sequence; performing multi-period trend evolution on the time-series voltage characteristic sequence based on the time-frequency distribution characteristics to generate multi-period load trend characteristic data;

[0098] In this embodiment, voltage is selected as a key feature from the operating parameters of the electrical equipment for processing, and the voltage parameters are extracted in time series, that is, the voltage data on the time series is sampled and sorted to form a time series voltage feature sequence, different sampling rates and time intervals are selected, and preprocessing operations such as data smoothing, interpolation or downsampling are performed as needed. The time series voltage feature sequence is subjected to multi-time period trend evolution analysis, and the time series voltage feature sequence is divided into different time periods, such as by hour, by day or by week, and trend analysis is performed on the time series voltage feature sequence in each time period. Trend prediction or fitting is performed using statistical methods, regression analysis, time series models, etc. In each time period, multiple load trend feature data are extracted, such as average value, maximum value, minimum value, change rate, etc. The load trend feature data in multiple time periods are comprehensively considered to generate multi-time period load trend feature data of the electrical equipment, and statistical analysis methods, such as mean, variance, trend change rate, etc., are used to integrate and describe the load trend features in multiple time periods. The generated multi-time period load trend feature data is used to analyze the load change trend, load fluctuation, load peak, etc. of the electrical equipment.

[0099] Step S4: obtaining pedestrian dynamic trajectories based on building pedestrian flow monitoring parameters; calculating spatial power load demand for pedestrian dynamic trajectories based on the spatial topological network of power-consuming equipment, thereby obtaining spatial power demand data for each area;

[0100] In this embodiment, the pedestrian flow monitoring parameters in the building are used to obtain the location and movement information of pedestrians. According to the pedestrian flow monitoring data, the distribution of personnel in each area at each moment is obtained. Combined with the time information, the dynamic trajectory of pedestrians in the building is obtained, that is, the movement path of pedestrians between different areas. The dynamic trajectory of pedestrians is mapped to the nodes in the topological network. For the dynamic trajectory of each pedestrian, the electrical equipment nodes passed by the pedestrian are determined. According to the electrical equipment nodes passed by the pedestrians, the spatial power load demand of each area is derived. According to the pedestrian's residence time and the power parameters of the electrical equipment, the spatial power demand data of each area is calculated.

[0101] Step S5: using the multi-period load trend characteristic data and the spatial power demand data of each region to perform dynamic load multi-step prediction on the spatial topology network of power-consuming equipment, and generate regional multi-step load prediction data;

[0102] In this embodiment, multi-time period load trend characteristic data and spatial electricity demand data of each region are used to perform dynamic load multi-time step prediction on the spatial topological network of power equipment, and time series analysis, machine learning, deep learning and other methods are used to perform load prediction, such as ARIMA, neural network and other models. Multi-time period load trend characteristic data and spatial electricity demand data of each region are used as input to predict the load conditions of power equipment in multiple time steps in the future, and the predicted load conditions of power equipment are associated with the corresponding regions to form regional multi-time step load prediction data, which is divided by time step and region to generate power equipment load prediction data for each region in each time step.

[0103] Step S6: Perform deep load prediction learning on the regional multi-time-step load prediction data and construct a deep load prediction model to perform building electrical load prediction operations.

[0104] In this embodiment, a deep load prediction model is constructed based on a deep learning method, and a model suitable for processing time series data such as a recurrent neural network (RNN) or a convolutional neural network (CNN) is selected and used. The regional multi-time-step load prediction data is used as a training set and divided into an input sequence and a target sequence, wherein the input sequence contains past load data and the target sequence contains future load data. The deep load prediction model is trained, and the model parameters are iteratively optimized to minimize the error between the prediction result and the actual load data. The trained deep load prediction model is verified, and the model performance is evaluated using the verification set. The error index between the prediction result and the actual load data, such as the root mean square error (RMSE) or the mean absolute error (MAE), is calculated. The verified deep load prediction model is tested, and the test set is used for load prediction. By inputting past load data, the model predicts the future load situation. According to the output of the deep load prediction model, the building electrical load prediction operation is performed to obtain the electrical load prediction results of each area in the future time period.

[0105] In this embodiment, refer to Figure 2 , is a flowchart of detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include:

[0106] Step S11: Acquire the operating parameters of the electrical equipment and the monitoring parameters of the building's human traffic; identify the electrical equipment in the building based on the sensor, and mark the building's electrical equipment nodes;

[0107] Step S12: Calculate the spatial position of the building electrical equipment nodes to obtain the node spatial coordinates;

[0108] Step S13: performing spatial distribution analysis on building electrical equipment nodes based on node spatial coordinates to obtain electrical equipment node spatial distribution data;

[0109] Step S14: mining the interactive associations between building electrical equipment nodes to generate interactive associations between nodes;

[0110] Step S15: reconstructing the spatial topology plane of the spatial distribution data of the power-consuming equipment nodes based on the interactive association relationship between the nodes, and constructing a spatial topology network of the power-consuming equipment.

[0111] In this embodiment, the operating parameters of each electrical device in the building are collected, such as power, energy consumption, switch status, etc. These parameters are obtained through a building management system, power monitoring equipment or sensors, etc. The pedestrian flow monitoring parameters in the building, such as the number of people, location information, etc., are collected. These parameters are obtained through a pedestrian flow monitoring system, cameras, sensors, etc. Sensors or non-invasive methods are used to identify and monitor the electrical equipment in the building. This is achieved through current sensors, electricity meters, smart sockets and other devices. The purpose of identification and monitoring is to associate each electrical device with the corresponding sensor and obtain operating parameters. A node is created for each identified electrical device and marked as a building electrical device node. Each node represents an independent electrical device, such as lamps, air conditioners, elevators, etc. Based on the marked building electrical equipment nodes, spatial position calculation is performed to determine the spatial coordinates of each node. This is achieved by using the building structure and layout information, combined with the floor and location information of the node, and using the spatial coordinates of the node The spatial distribution of electrical equipment nodes in the building is analyzed. The spatial distribution of electrical equipment nodes is evaluated according to indicators such as the distance, density and relative position between nodes. The analysis results include distribution maps, heat maps or other visualization forms of electrical equipment nodes to show the distribution of nodes in the building space. Based on the data interaction or interaction relationship between electrical equipment nodes, the interactive association mining between devices is carried out. The interaction mode and association relationship between nodes are analyzed by using sensor data, communication records between devices or other relevant information. The mining results are expressed as connection relationships between nodes, adjacency matrices or other forms of interactive relationship representations. The spatial distribution information of electrical equipment nodes is mapped to a topological network using the method of spatial topological plane reconstruction. In the topological network, electrical equipment nodes represent nodes in the network, and the interaction relationships between nodes represent connection relationships in the network. The constructed spatial topological network of electrical equipment is used for subsequent applications such as dynamic load prediction, energy management and optimization decision-making.

[0112] In this embodiment, refer to Figure 3 , is a flowchart of detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include:

[0113] Step S21: performing segmented low-pass filtering on the operating parameters of the electrical equipment to obtain low-pass filtered operating parameters;

[0114] Step S22: performing non-overlapping timing window division based on the low-pass filtering operation parameters to obtain a plurality of non-overlapping timing windows;

[0115] Step S23: performing multi-scale time-frequency decomposition on the low-pass filter operating parameters using multiple non-overlapping time series windows to generate multiple time-frequency frequency domain segments;

[0116] Step S24: performing frequency energy distribution analysis on multiple time-frequency frequency domain segments to generate energy distribution data for each frequency domain segment;

[0117] Step S25: Perform time-frequency distribution fitting on the energy distribution data of each frequency domain segment to generate time-frequency distribution features.

[0118] In this embodiment, the operating parameter data is divided into a series of segments, each segment contains continuous data points, and a low-pass filter is applied to the data of each segment, the filter allows low-frequency signals to pass through and blocks high-frequency signals to obtain low-pass filtered operating parameter data, and the low-pass filtered operating parameter data is divided into multiple non-overlapping time series windows, and the non-overlapping time series windows are continuous time periods, and there is no overlapping data between each window. Multi-scale time-frequency decomposition is performed on the low-pass filtered operating parameter data of each non-overlapping time series window. Multi-scale time-frequency decomposition is a technology that decomposes a signal into different frequency ranges and time ranges to obtain multiple time-frequency frequency domain segments, and frequency energy distribution analysis is performed on each time-frequency frequency domain segment. The frequency energy distribution analysis is used to calculate the energy distribution of different frequencies in each frequency domain segment to obtain energy distribution data of each frequency domain segment, and time-frequency distribution fitting is performed on the energy distribution data of each frequency domain segment. Time-frequency distribution fitting is a method of fitting energy distribution data into time-frequency distribution characteristics, which is used to describe the law of energy change over time and frequency in different frequency domain segments.

[0119] In this embodiment, reference Figure 4 The above is a schematic flow chart of the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include:

[0120] Step S31: extracting the time-sequence voltage characteristics of the operating parameters of the electrical equipment to obtain the time-sequence voltage characteristics of the electrical equipment;

[0121] Step S32: reorganizing the time-series voltage characteristics of the electrical equipment into time-series hierarchical representations to generate a time-series voltage characteristic sequence;

[0122] Step S33: performing voltage load deduction on the time-series voltage characteristic sequence, thereby generating a voltage load time-series curve;

[0123] Step S34: performing load frequency domain variation analysis on the voltage-load timing curve based on the time-frequency distribution characteristics to obtain voltage-load frequency domain variation data;

[0124] Step S35: Perform multi-period trend evolution on the voltage load frequency domain variation data to generate multi-period load trend characteristic data.

[0125] In this embodiment, the voltage data is subjected to time-series feature extraction processing to extract time-related voltage features, such as maximum value, minimum value, mean value, volatility, etc. The time-series voltage features are hierarchically characterized and reorganized, and the voltage features in adjacent time periods are combined. By combining the features in adjacent time periods, a series of time-series voltage feature sequences are generated, each sequence represents the voltage features in a period of time, and the time-series voltage feature sequence is used to perform voltage load deduction to infer the corresponding voltage load condition. Through the deduction, a voltage load time-series curve is generated to describe the voltage load condition that changes with time. The voltage load time-series curve is subjected to time-frequency distribution analysis to obtain its frequency domain features, and the energy distribution of the voltage load time-series curve in different frequency ranges is analyzed to generate voltage load frequency domain change data to describe the load change in different frequency bands. The voltage load frequency domain change data is subjected to multi-time period trend evolution to observe the load change trend in different time periods, and load trend feature data for multiple time periods is generated to describe the load change in different time periods.

[0126] In this embodiment, the specific steps of step S34 are:

[0127] Perform wavelet decomposition on the voltage-load timing curve to generate voltage-load frequency domain data;

[0128] Based on the time-frequency distribution characteristics, the voltage load frequency domain data is fitted with time-frequency transformation to obtain the voltage load time-frequency matrix;

[0129] Calculate the adjacent time-frequency characteristic difference of the voltage-load time-frequency matrix to generate the load power change value at each frequency point;

[0130] The load power change value at each frequency point is analyzed for change amplitude distribution to obtain the voltage load frequency domain change data.

[0131] In this embodiment, wavelet decomposition is performed on the timing curve data, and the time domain data is converted into frequency domain data by wavelet transform. The wavelet decomposition decomposes the timing curve into components in different frequency ranges to generate frequency domain data of the voltage load. The frequency domain data of the voltage load is subjected to time-frequency transformation fitting by using the time-frequency distribution characteristics. The time-frequency transformation fitting is a method of fitting the frequency domain data into a time-frequency matrix, which is used to describe the law of the change of different frequency components over time. The time-frequency matrix of the voltage load is generated by time-frequency transformation fitting, in which each element represents the energy value of a specific frequency at different time points. The time-frequency matrix of the voltage load is subjected to adjacent time-frequency characteristic difference calculation, which is used to calculate the difference between adjacent time points and frequency points to obtain the change value of the load power at each frequency point. The change amplitude distribution analysis is performed on the load power change value at each frequency point. The change amplitude distribution analysis is used to calculate the distribution of the load power change value to obtain the voltage load frequency domain change data. The voltage load frequency domain change data describes the change amplitude and distribution of the load at different frequency points.

[0132] In this embodiment, step S4 includes the following steps:

[0133] Step S41: performing real-time pedestrian position tracking on building traffic monitoring parameters and extracting real-time pedestrian position sequences;

[0134] Step S42: filtering abnormal time intervals of real-time pedestrian positions to obtain a filtered pedestrian position sequence;

[0135] Step S43: performing discrete sample connection on the filtered pedestrian position sequence to obtain the pedestrian dynamic trajectory;

[0136] Step S44: mining the regional pedestrian trajectory features of the pedestrian dynamic trajectory based on the spatial topological network of the electrical equipment, thereby generating the pedestrian trajectory features in the region;

[0137] Step S45: Calculate the spatial power load demand based on the pedestrian trajectory characteristics in the area, so as to obtain the spatial power demand data of each area.

[0138] In this embodiment, a building pedestrian flow monitoring device is used to obtain real-time pedestrian flow monitoring parameters, including pedestrian position information, and the pedestrian flow monitoring parameters are processed to extract a real-time pedestrian position sequence, record the pedestrian position at each moment, perform abnormal time gap filtering on the real-time pedestrian position sequence, remove position data that does not conform to the normal pedestrian activity pattern, and filter out position data that conform to the pedestrian activity pattern by setting a time gap threshold to obtain a filtered pedestrian position sequence, perform discrete sample connection on the filtered pedestrian position sequence, connect the position data with a short continuous time interval, and obtain the dynamic trajectory of the pedestrian by connecting the discrete samples to describe the movement path of the pedestrian in the building, and build a pedestrian position sequence. A spatial topological network of electrical equipment is established to describe the association and connection between electrical equipment in the building. The dynamic trajectories of pedestrians are analyzed and mined using the established spatial topological network to extract the pedestrian trajectory features related to the electrical equipment area. By mining the pedestrian trajectory features, pedestrian trajectory feature data in the area is generated to describe the activities of pedestrians in different areas. According to the pedestrian trajectory features and the spatial topological network of electrical equipment in the area, the spatial power load demand in each area is calculated. By analyzing the association between pedestrian trajectory features and electrical equipment, the power demand situation in each area is inferred, the spatial power demand data of each area is obtained, and the power load demand situation in different areas in the building is described.

[0139] In this embodiment, the specific steps of step S44 are:

[0140] Perform crowd flow fluctuation statistics on building crowd flow monitoring parameters to obtain crowd flow fluctuation data;

[0141] Performing regional segmentation on the spatial topological network of electrical equipment to obtain spatial regions of electrical equipment;

[0142] Calculate the regional human flow density of the electrical equipment space area based on human flow fluctuation data to generate multiple regional human flow density data;

[0143] Based on the pedestrian flow density data of multiple regions, the spatial intersection distribution of pedestrian dynamic trajectories is analyzed to obtain the spatial intersection distribution data of dynamic trajectories;

[0144] Calculate the dwell time of pedestrian dynamic trajectory to obtain the trajectory dwell time;

[0145] The regional pedestrian trajectory features are mined based on the dynamic trajectory spatial intersection distribution data and trajectory dwell time, thereby generating the pedestrian trajectory features in the region.

[0146] In this embodiment, a building pedestrian flow monitoring device is used to obtain pedestrian flow monitoring parameters, including pedestrian flow information, and a statistical analysis is performed on the pedestrian flow parameters to calculate the fluctuation of pedestrian flow in different time periods. Through statistical analysis, pedestrian flow fluctuation data is obtained to describe the change of pedestrian flow in different time periods. The spatial topological network of electrical equipment is segmented into regions, and the electrical equipment in the building is divided into multiple spatial regions. Through regional segmentation, the spatial regions of electrical equipment are obtained to describe the distribution of electrical equipment in different regions in the building. The pedestrian flow fluctuation data and the spatial regions of electrical equipment are used to calculate the pedestrian flow density in each region. By combining the pedestrian flow with the spatial region, the pedestrian flow density in different regions is calculated to reflect the concentration of pedestrian flow in different regions. The pedestrian flow density data of multiple regions are obtained to describe the concentration of pedestrian flow in different regions in the building. The pedestrian flow density data and dynamic trajectories of pedestrians are used to analyze the intersection of pedestrian trajectories in different areas. By analyzing the intersection distribution of pedestrian trajectories, the spatial intersection distribution data of dynamic trajectories in different areas are obtained. The dynamic trajectory spatial intersection distribution data describes the intersection and distribution of pedestrian trajectories between different areas in the building. The dynamic trajectories of pedestrians are analyzed to calculate the time pedestrians stay in different areas. By analyzing the time interval and area switching of pedestrian trajectories, the residence time of the trajectory in each area is calculated to obtain trajectory residence time data to describe the length of time pedestrians stay in different areas. Based on the dynamic trajectory spatial intersection distribution data and trajectory residence time data, the characteristics of pedestrian trajectories in different areas are mined. By analyzing the intersection distribution and residence time of the trajectories, the characteristics of pedestrian trajectories in the area are extracted, such as pedestrian activity hotspots, pedestrian mobility, etc.

[0147] In this embodiment, step S5 includes the following steps:

[0148] Step S51: using multi-period load trend feature data to perform three-dimensional mapping and rendering processing on the space topology network of electrical equipment to construct a three-dimensional space load network;

[0149] Step S52: performing regional load evolution on the three-dimensional spatial load network based on the spatial power demand data of each region to obtain regional load evolution data;

[0150] Step S53: Perform dynamic load multi-time-step prediction on the regional load evolution data to generate regional multi-time-step load prediction data.

[0151] In this embodiment, a three-dimensional mapping is performed on the spatial topology network of the electrical equipment, and each node is mapped to a corresponding three-dimensional spatial position. This is achieved using three-dimensional modeling software or a visualization library. Based on the collected multi-period load trend characteristic data, the three-dimensional spatial load network is rendered, and color, height or other visualization attributes are used to represent the load in different time periods. The purpose of the rendering process is to intuitively display the load conditions of the electrical equipment in three-dimensional space. Based on the collected spatial electricity demand data of each region, these data are mapped to the corresponding regions in the three-dimensional spatial load network. Different colors, heights or other visualization attributes are assigned to each region according to the size of the load. By performing regional load evolution on the three-dimensional spatial load network, the load evolution of each region in different time periods is obtained, and visualization data on load changes in different regions is provided. The regional load evolution data is predicted using a suitable prediction model or algorithm. Dynamic load multi-step prediction refers to predicting changes in regional loads in multiple future time steps. Based on the output of the prediction model or algorithm, load prediction data for the region in multiple future time steps is generated.

[0152] In this embodiment, step S6 includes the following steps:

[0153] Step S61: performing loss error calculation on the regional multi-time-step load forecast data to generate a load forecast error parameter;

[0154] Step S62: performing back propagation iterative optimization on the regional multi-time-step load forecast data based on the load forecast error parameter to obtain error-optimized load forecast data;

[0155] Step S63: Perform deep load prediction learning on the error-optimized load prediction data and construct a deep load prediction model to perform building electrical load prediction operations.

[0156] In this embodiment, real regional multi-time-step load data and predicted regional multi-time-step load data are prepared. These data are load numerical sequences in time steps, such as hourly load data. The predicted data and the real data are compared to calculate the load prediction error. Various common error calculation methods, such as mean squared error (MSE) or mean absolute error (MAE), are used to calculate the error of each time step and obtain the load prediction error parameter. The load prediction error parameter is used as the optimization target and is used as part of the loss function. The current regional multi-time-step load prediction data is used as the starting point. An optimization algorithm, such as gradient descent or other iterative optimization algorithms, is used to perform back propagation on the predicted data and update the data. In each iteration, a gradient is calculated according to the load prediction error parameter and the gradient is applied to the predicted data to gradually reduce the error. The iterative optimization steps are repeated until a convergence condition is reached or a predetermined number of iterations is reached to obtain error-optimized regional multi-time-step load prediction data. The error-optimized regional multi-time-step load prediction data is prepared as a training data set. An appropriate deep learning model, such as a recurrent neural network, is selected according to the requirements. A deep load prediction model is established by using a RNN or LSTM network. The data set is divided into a training set and a validation set. The training set is used to train the deep load prediction model, and the validation set is used to evaluate the performance of the model and adjust the hyperparameters. During the training process, the error-optimized load prediction data is used as the target to learn the deep load prediction model and the prediction error parameters. The parameters of the model are optimized to minimize the prediction error through optimization algorithms such as back propagation and gradient descent. The performance of the model is evaluated using the validation set. The model is adjusted and optimized according to the performance indicators on the validation set (such as root mean square error, mean absolute error, etc.).

[0157] In this embodiment, a building electrical load prediction system based on deep learning is provided, which is used to execute the building electrical load prediction method based on deep learning as described above, including:

[0158] The topological network module is used to obtain the operating parameters of electrical equipment and the monitoring parameters of building traffic flow; identify electrical equipment in the building based on sensors and mark the nodes of building electrical equipment; reconstruct the spatial topological plane of the nodes of building electrical equipment and construct the spatial topological network of electrical equipment;

[0159] The time-frequency distribution module is used to perform multi-scale time-frequency decomposition on the operating parameters of the power-consuming equipment to generate multiple time-frequency frequency domain segments; perform frequency energy distribution analysis on multiple time-frequency frequency domain segments to generate time-frequency distribution features;

[0160] The load trend module is used to extract the time-series voltage characteristics of the operating parameters of the power-consuming equipment to generate a time-series voltage characteristic sequence; based on the time-frequency distribution characteristics, the time-series voltage characteristic sequence is subjected to multi-period trend evolution to generate multi-period load trend characteristic data;

[0161] The power load demand module is used to obtain the dynamic trajectory of pedestrians based on the building traffic monitoring parameters; the spatial power load demand of pedestrian dynamic trajectories is calculated based on the spatial topological network of power-consuming equipment, so as to obtain the spatial power demand data of each area;

[0162] The multi-time-step prediction module is used to perform dynamic load multi-time-step prediction on the spatial topology network of power-consuming equipment using multi-period load trend characteristic data and spatial power demand data of each region, and generate regional multi-time-step load prediction data;

[0163] The load forecasting learning module is used to perform deep load forecasting learning on regional multi-time-step load forecasting data and build a deep load forecasting model to perform building electrical load forecasting operations.

[0164] The present invention obtains the operating parameters of power equipment and the monitoring parameters of building traffic flow, provides the correlation information of power equipment and human flow in the building, provides basic data for subsequent analysis and prediction, marks the nodes of building power equipment, establishes the topological structure of power equipment, facilitates the subsequent spatial analysis and load prediction, performs multi-scale time-frequency decomposition, separates and extracts the time domain and frequency domain information of the operating parameters of power equipment, reveals the operating characteristics and changing laws of power equipment, and frequency energy distribution analysis provides energy distribution in different frequency bands to help identify the frequency characteristics of power equipment, provides a basis for subsequent feature extraction and analysis, extracts the time series voltage feature sequence, captures the time series changes of the operating parameters of power equipment, and provides time dimension information for load trend analysis. Multi-period trend evolution analysis reveals the changing trend of the load of power equipment in different time periods, understands the load characteristics and laws of power equipment, and is based on the monitoring parameters of building traffic flow and the spatial topological network of power equipment. The network is used to calculate the spatial electricity demand of each area, and the distribution and demand of electricity load in different areas are provided. The calculation of the spatial electricity load demand considering the dynamic trajectory of pedestrians makes the electricity demand closer to the actual situation, which provides a basis for accurately predicting the electrical load of buildings. The dynamic load multi-step prediction of the spatial topological network of power equipment is carried out by using the multi-period load trend characteristic data and the spatial electricity demand data of each region, and the load prediction results for multiple time steps in the future are provided. The multi-time-step load prediction data provides a decision-making basis for the planning and regulation of the power system, evaluates the load status, optimizes the power supply and resource allocation, and constructs a deep load prediction model through deep load prediction learning of the regional multi-time-step load prediction data to improve the accuracy and robustness of the load prediction. The deep load prediction model can predict the electrical load of the building, provide accurate prediction results for the operation and regulation of the power system, and support the rational allocation of power resources and the optimization of energy utilization.

[0165] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0166] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A building electrical load prediction method based on deep learning, characterized in that: The following steps are involved: Step S1: Obtaining operating parameters of electrical equipment and building traffic monitoring parameters; identifying electrical equipment in the building based on sensors and marking building electrical equipment nodes; Reconstruct the spatial topology of the building electrical equipment nodes and build a spatial topology network of electrical equipment; Step S2: Perform multi-scale time-frequency decomposition on the operating parameters of the electrical equipment to generate multiple time-frequency domain segments; Perform frequency energy distribution analysis on multiple time-frequency domain segments to generate time-frequency distribution features; Step S3: extracting the time-series voltage characteristics of the operating parameters of the electrical equipment to generate a time-series voltage characteristic sequence; Based on the time-frequency distribution characteristics, the time-series voltage characteristic sequence is subjected to multi-period trend evolution to generate multi-period load trend characteristic data; Step S4: obtaining pedestrian dynamic trajectories based on building pedestrian flow monitoring parameters; Based on the spatial topological network of power-consuming equipment, the spatial power load demand of pedestrians is calculated based on their dynamic trajectory, thus obtaining the spatial power demand data of each area. Step S5: using the multi-period load trend characteristic data and the spatial power demand data of each region to perform dynamic load multi-time-step prediction on the spatial topology network of power-consuming equipment, and generate regional multi-time-step load prediction data; Step S6: Perform deep load prediction learning on the regional multi-time-step load prediction data and construct a deep load prediction model to perform building electrical load prediction operations.

2. The building electrical load prediction method based on deep learning according to claim 1 is characterized in that: The specific steps of step S1 are: Step S11: Acquire the operating parameters of the electrical equipment and the monitoring parameters of the building's human traffic; identify the electrical equipment in the building based on the sensor, and mark the building's electrical equipment nodes; Step S12: Calculate the spatial position of the building electrical equipment nodes to obtain the node spatial coordinates; Step S13: performing spatial distribution analysis on building electrical equipment nodes based on node spatial coordinates to obtain electrical equipment node spatial distribution data; Step S14: mining the interactive associations between building electrical equipment nodes to generate interactive associations between nodes; Step S15: reconstructing the spatial topology plane of the spatial distribution data of the power-consuming equipment nodes based on the interactive association relationship between the nodes, and constructing a spatial topology network of the power-consuming equipment.

3. The building electrical load prediction method based on deep learning according to claim 1 is characterized in that: The specific steps of step S2 are: Step S21: performing segmented low-pass filtering on the operating parameters of the electrical equipment to obtain low-pass filtered operating parameters; Step S22: performing non-overlapping timing window division based on the low-pass filtering operation parameters to obtain a plurality of non-overlapping timing windows; Step S23: performing multi-scale time-frequency decomposition on the low-pass filter operating parameters using multiple non-overlapping time series windows to generate multiple time-frequency frequency domain segments; Step S24: performing frequency energy distribution analysis on multiple time-frequency frequency domain segments to generate energy distribution data for each frequency domain segment; Step S25: Perform time-frequency distribution fitting on the energy distribution data of each frequency domain segment to generate time-frequency distribution features.

4. The building electrical load prediction method based on deep learning according to claim 1 is characterized in that: The specific steps of step S3 are: Step S31: extracting the time-sequence voltage characteristics of the operating parameters of the electrical equipment to obtain the time-sequence voltage characteristics of the electrical equipment; Step S32: reorganizing the time-series voltage characteristics of the electrical equipment into time-series hierarchical representations to generate a time-series voltage characteristic sequence; Step S33: performing voltage load deduction on the time-series voltage characteristic sequence, thereby generating a voltage load time-series curve; Step S34: performing load frequency domain variation analysis on the voltage-load timing curve based on the time-frequency distribution characteristics to obtain voltage-load frequency domain variation data; Step S35: Perform multi-period trend evolution on the voltage load frequency domain variation data to generate multi-period load trend characteristic data.

5. The building electrical load prediction method based on deep learning according to claim 4 is characterized in that: The specific steps of step S34 are: Perform wavelet decomposition on the voltage-load timing curve to generate voltage-load frequency domain data; Based on the time-frequency distribution characteristics, the voltage load frequency domain data is fitted with time-frequency transformation to obtain the voltage load time-frequency matrix; Calculate the adjacent time-frequency characteristic difference of the voltage-load time-frequency matrix to generate the load power change value at each frequency point; The load power change value at each frequency point is analyzed for change amplitude distribution to obtain the voltage load frequency domain change data.

6. The method according to claim 1, characterized in that The specific steps of step S4 are: Step S41: performing real-time pedestrian position tracking on building traffic monitoring parameters and extracting real-time pedestrian position sequences; Step S42: filtering abnormal time intervals of real-time pedestrian positions to obtain a filtered pedestrian position sequence; Step S43: performing discrete sample connection on the filtered pedestrian position sequence to obtain the pedestrian dynamic trajectory; Step S44: mining the regional pedestrian trajectory features of the pedestrian dynamic trajectory based on the spatial topological network of the electrical equipment, thereby generating the pedestrian trajectory features in the region; Step S45: Calculate the spatial power load demand based on the pedestrian trajectory characteristics in the area, so as to obtain the spatial power demand data of each area.

7. The building electrical load prediction method based on deep learning according to claim 6 is characterized in that: The specific steps of step S44 are: Perform crowd flow fluctuation statistics on building crowd flow monitoring parameters to obtain crowd flow fluctuation data; Performing regional segmentation on the spatial topological network of electrical equipment to obtain spatial regions of electrical equipment; Calculate the regional human flow density of the electrical equipment space area based on human flow fluctuation data to generate multiple regional human flow density data; Based on the pedestrian flow density data of multiple regions, the spatial intersection distribution of pedestrian dynamic trajectories is analyzed to obtain the spatial intersection distribution data of dynamic trajectories; Calculate the dwell time of pedestrian dynamic trajectory to obtain the trajectory dwell time; The regional pedestrian trajectory features are mined based on the dynamic trajectory spatial intersection distribution data and trajectory dwell time, thereby generating the pedestrian trajectory features in the region.

8. The building electrical load prediction method based on deep learning according to claim 1 is characterized in that: The specific steps of step S5 are: Step S51: using multi-period load trend feature data to perform three-dimensional mapping and rendering processing on the space topology network of electrical equipment to construct a three-dimensional space load network; Step S52: performing regional load evolution on the three-dimensional spatial load network based on the spatial power demand data of each region to obtain regional load evolution data; Step S53: Perform dynamic load multi-time-step prediction on the regional load evolution data to generate regional multi-time-step load prediction data.

9. The building electrical load prediction method based on deep learning according to claim 1 is characterized in that: The specific steps of step S6 are: Step S61: performing loss error calculation on the regional multi-time-step load forecast data to generate a load forecast error parameter; Step S62: performing back propagation iterative optimization on the regional multi-time-step load forecast data based on the load forecast error parameter to obtain error-optimized load forecast data; Step S63: Perform deep load prediction learning on the error-optimized load prediction data and construct a deep load prediction model to perform building electrical load prediction operations.

10. A building electrical load prediction system based on deep learning, characterized in that: The method for predicting building electrical load based on deep learning according to claim 1 comprises: The topological network module is used to obtain the operating parameters of electrical equipment and the monitoring parameters of building traffic flow; identify electrical equipment in the building based on sensors and mark the nodes of building electrical equipment; reconstruct the spatial topological plane of the nodes of building electrical equipment and construct the spatial topological network of electrical equipment; The time-frequency distribution module is used to perform multi-scale time-frequency decomposition on the operating parameters of the power-consuming equipment to generate multiple time-frequency frequency domain segments; perform frequency energy distribution analysis on multiple time-frequency frequency domain segments to generate time-frequency distribution features; The load trend module is used to extract the time-series voltage characteristics of the operating parameters of the power-consuming equipment to generate a time-series voltage characteristic sequence; based on the time-frequency distribution characteristics, the time-series voltage characteristic sequence is subjected to multi-period trend evolution to generate multi-period load trend characteristic data; The power load demand module is used to obtain the dynamic trajectory of pedestrians based on the building traffic monitoring parameters; the spatial power load demand of pedestrian dynamic trajectories is calculated based on the spatial topological network of power-consuming equipment, so as to obtain the spatial power demand data of each area; The multi-time-step prediction module is used to perform dynamic load multi-time-step prediction on the spatial topology network of power-consuming equipment using multi-period load trend characteristic data and spatial power demand data of each region, and generate regional multi-time-step load prediction data; The load forecasting learning module is used to perform deep load forecasting learning on regional multi-time-step load forecasting data and build a deep load forecasting model to perform building electrical load forecasting operations.

Images