Building group power load prediction method, system and equipment based on multi-modal depth forest algorithm

Abstract

The invention relates to the technical field of electric power engineering, and discloses a building group power load prediction method, system and device based on a multi-modal depth forest algorithm, and the method comprises the steps: obtaining the multi-modal data of a building group; performing feature extraction on the multi-modal data, and performing fusion processing on feature extraction results to obtain multi-modal fusion features; and inputting the multi-modal fusion features into a trained power load prediction model based on a multi-modal depth forest algorithm to carry out prediction processing so as to obtain a power load prediction result. According to the method, the multi-modal data of the building group is obtained, after feature extraction and fusion processing, the multi-modal data is input into the power load prediction model based on the multi-modal depth forest algorithm for prediction, time-space information in the multi-modal data can be fully mined, a complex relation can be effectively processed, and therefore the precision of power load prediction is improved. Adaptive adjustment is carried out according to different multi-modal data, and continuous accuracy of prediction is ensured in facing power load prediction scenes of different building groups.

Description

Technical Field

[0001] The present invention relates to the technical field of power engineering, and more specifically, to a method, system and device for predicting the power load of a building complex based on a multi-modal deep forest algorithm. Background Art

[0002] With the rapid development of smart grid and Internet of Things technologies, the prediction of the power load of building complexes plays an increasingly important role in the dispatching and management of power systems. Accurate power load prediction is crucial for optimizing energy distribution, improving energy utilization efficiency, reducing energy waste, and lowering operating costs.

[0003] Currently, traditional power load prediction methods mostly rely on single data sources and traditional algorithms, such as time series analysis (ARIMA), linear regression, etc. These methods have obvious deficiencies in dealing with complex multi-modal data and capturing spatio-temporal dependence relationships. Although edge computing technology has brought new ideas for power load prediction in recent years, and advanced algorithms such as deep learning and ensemble learning perform outstandingly in dealing with large-scale, multi-modal data, they face challenges in high computing resource requirements for deployment on edge devices. Even though the emerging deep forest algorithm combines the interpretability of decision trees and the feature extraction ability of deep learning and is suitable for running on resource-constrained edge devices, there are still limitations in dealing with multi-modal data and complex spatio-temporal dependence relationships, and there are defects in low prediction accuracy in the prediction of the power load of building complexes. Summary of the Invention

[0004] In order to improve the defect of low prediction accuracy existing in the existing power load prediction technology of building complexes, the present invention proposes the following technical solutions:

[0005] In a first aspect, the present invention proposes a method for predicting the power load of a building complex based on a multi-modal deep forest algorithm, including:

[0006] Obtain the multi-modal data of the building complex.

[0007] Extract features from the multi-modal data, and perform fusion processing on the feature extraction results to obtain multi-modal fusion features.

[0008] Input the multi-modal fusion features into a trained power load prediction model based on the multi-modal deep forest algorithm for prediction processing to obtain a power load prediction result.

[0009] As a preferred technical solution, extracting features from the multi-modal data includes:

[0010] According to a preset rule, each building in the building complex is used as a node in the graph of the multi-modal deep forest algorithm, and the geographical location and functional relationship are used as the edges of the graph.

[0011] For each node i, according to the following formula, extract the multi-modal feature vector:

[0012]

[0013] where M is the number of modalities, is the feature of the m-th modality.

[0014] As a preferred technical solution, the nodes of the graph include:

[0015] Function nodes, used to classify buildings according to the functional attributes of the buildings.

[0016] Perception nodes, used to collect real-time data inside the buildings.

[0017] Computing nodes, used to process data from the perception nodes and perform prediction processing of the power load.

[0018] Control nodes, used to manage the energy equipment of the building complex.

[0019] Top-level nodes, used as the main control center of the building complex, responsible for task scheduling and resource allocation.

[0020] Middle-level nodes, used to summarize the data of the bottom-level nodes and transfer it to the top-level nodes.

[0021] Bottom-level nodes, used to collect data of the equipment inside the buildings and transfer it to the middle-level nodes.

[0022] As a preferred technical solution, the edges of the graph include:

[0023] Spatial adjacency edges, used to connect buildings that are physically adjacent in space.

[0024] Distance weight edges, which allocate weights according to the geographical distance between buildings.

[0025] Functional coupling edges, used to connect buildings with related functions.

[0026] Task collaboration edges, used to connect functional nodes and control nodes.

[0027] Data flow direction edges, used to represent the data transfer path from the perception nodes to the computing nodes or from the computing nodes to the control nodes.

[0028] Priority resource allocation edges, used to connect nodes with matching computing capabilities.

[0029] As a preferred technical solution, perform fusion processing on the feature extraction results to obtain multi-modal fusion features, including:

[0030] Perform normalization processing on the multi-modal feature vector, and its expression is as follows:

[0031]

[0032] Among them, represents the eigenvector of the m-th mode, μ m and σ m are the mean and standard deviation of the eigenvector of the m-th mode, respectively.

[0033] Weighted feature fusion is performed on the normalized multi-modal feature vectors, and its expression is as follows:

[0034]

[0035] Among them, ω m is the weight of the eigenvector of the m-th mode, satisfying M is the number of modes.

[0036] As a preferred technical solution, the power load prediction model based on the multi-modal deep forest algorithm includes several layers of random forest models, and each layer of random forest model includes several decision trees.

[0037] The multi-modal fusion features are input into the trained power load prediction model based on the multi-modal deep forest algorithm for prediction processing to obtain the power load prediction result, including:

[0038] Input the multi-modal fusion feature X i , and extract the output features of the initial layer through the first layer of random forest model. Its expression is as follows:

[0039] H (1) = RandomForest (1) (X i )

[0040] Among them, H (1) is the output feature of the first layer of random forest model, and RandomForest (1) represents the first layer of random forest model.

[0041] Concatenate the output feature H (l-1) of the previous layer with the multi-modal fusion feature X i to generate the fused feature through the intermediate layer random forest model. Its expression is as follows:

[0042] H (l) = RandomForest (l) ([X (l-1) ; X i )

[0043] Among them, [H (l-1) ; X iIt represents the output feature H of the previous layer (l-1) is concatenated with the multi-modal fusion feature X i to form RandomForest (l) which represents the random forest model of the l-th layer.

[0044] The output feature H of the last layer random forest model (L) is used for ensemble learning to generate the final power load prediction value, and its expression is as follows:

[0045]

[0046] where, represents the power load prediction value, N is the total number of decision trees in the random forest model, and T i represents the i-th decision tree in the random forest model, and H (L) is the output feature of the last layer.

[0047] As a preferred technical solution, before inputting the multi-modal fusion feature into the trained power load prediction model based on the multi-modal deep forest algorithm for prediction processing, the method further includes training the power load prediction model based on the multi-modal deep forest algorithm, including:

[0048] Integrate multi-modal data including historical power load data, weather data, and holiday information into a training set.

[0049] Through cross-validation, determine the number of random forest models and the depth of decision trees in the power load prediction model.

[0050] Use the training set to train the power load prediction model until its loss function converges to obtain the trained power load prediction model. The expression of the loss function of the power load prediction model is as follows:

[0051]

[0052] where, x represents the input training set data, θ is the parameter set of the random forest model, n is the number of training set samples, is the i-th power load prediction value, and y i is the i-th true power load label value.

[0053] As a preferred technical solution, the multi-modal data includes at least one of the building's structural data, historical power load data, electrical equipment data, environmental data, time information, holiday information, and resident activity data.

[0054] Second aspect, the present invention also provides a building complex power load prediction system based on a multi-modal deep forest algorithm, which is applied to the building complex power load prediction method based on the multi-modal deep forest algorithm described in any of the solutions in the first aspect, and includes:

[0055] An acquisition module, configured to acquire multi-modal data of the building complex.

[0056] An extraction and fusion module, configured to perform feature extraction on the multi-modal data and perform fusion processing on the feature extraction results to obtain multi-modal fusion features.

[0057] A prediction module, configured to input the multi-modal fusion features into a trained power load prediction model based on the multi-modal deep forest algorithm for prediction processing to obtain a power load prediction result.

[0058] Third aspect, the present invention also provides an electronic device, the electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the program, it implements the operations performed by the building complex power load prediction method based on the multi-modal deep forest algorithm described in any of the solutions in the first aspect.

[0059] The beneficial effects of the present invention at least include:

[0060] By acquiring multi-modal data of the building complex, after feature extraction and fusion processing, and inputting it into a power load prediction model based on the multi-modal deep forest algorithm for prediction, the present invention can fully exploit the spatio-temporal information in the multi-modal data, enable the multi-modal deep forest algorithm to effectively process complex relationships, thereby significantly improving the accuracy of power load prediction, and further more accurately allocate energy and improve resource utilization rate. At the same time, it can be adaptively adjusted according to different multi-modal data, and can efficiently handle different building complex power load prediction scenarios to ensure the continuous accuracy of prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a schematic flowchart of a building complex power load prediction method based on a multi-modal deep forest algorithm provided by an embodiment of the present invention.

[0062] Figure 2 It is an architecture diagram of a building complex power load prediction system based on a multi-modal deep forest algorithm provided by an embodiment of the present invention.

[0063] Figure 3 It is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0064] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred technical solutions. Those skilled in the art can easily understand the other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred technical solutions are only for illustrating the present invention rather than limiting the protection scope of the present invention.

[0065] It should be noted that the diagrams provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner. Therefore, only the components related to the present invention are shown in the diagrams, rather than being drawn according to the number, shape, and size of the components in actual implementation. The type, quantity, and ratio of each component in actual implementation can be arbitrarily changed, and the component layout type may also be more complex.

[0066] In the following description, a large number of details are explored to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0067] Embodiment 1

[0068] This embodiment proposes a method for predicting the power load of a building complex based on a multi-modal deep forest algorithm, as Figure 1 shown, Figure 1 is a schematic flowchart of a method for predicting the power load of a building complex based on a multi-modal deep forest algorithm provided by an embodiment of the present invention. The method includes the following steps:

[0069] S1: Obtain the multi-modal data of the building complex.

[0070] S2: Extract features from the multi-modal data and perform fusion processing on the feature extraction results to obtain multi-modal fusion features.

[0071] S3: Input the multi-modal fusion features into a trained power load prediction model based on the multi-modal deep forest algorithm for prediction processing to obtain a power load prediction result.

[0072] In this embodiment, the multi-modal data at least includes the structural data of the building, historical power load data, electrical equipment data, environmental data, time information, holiday information, and resident activity data.

[0073] As an exemplary illustration, taking an industrial park as an example, first, multi-modal data is obtained, such as structural data like the building materials and area of factory buildings, historical power load data for each period in the past year, power consumption data of various production equipment in the park, environmental data such as temperature and humidity in the park, as well as time information like weekdays and rest days, holiday arrangements, and resident activity data such as workers' commuting to and from work. Then, feature extraction is performed on this data. Features such as the scale of factory buildings are refined from the structural data, features such as peak power consumption periods are summarized from the historical data, features such as equipment power are generalized from the power consumption equipment data, features such as the impact of temperature changes on power consumption are extracted from the environmental data, features such as the power consumption differences between weekdays and rest days are found from the time information, and features such as the impact of personnel flow on power consumption are analyzed from the resident activity data. Then, these features are fused to form multi-modal fusion features. Finally, the multi-modal fusion features are input into a trained power load prediction model based on the multi-modal deep forest algorithm. Based on previous training and learning, the model predicts the future power load of the industrial park. For example, it predicts the approximate value of the power load during the production peak period tomorrow morning, providing a basis for power supply guarantee and management in the park to ensure the stability and efficiency of production power consumption.

[0074] It can be understood that by obtaining the multi-modal data of the building complex, through feature extraction and fusion processing, and then inputting it into the power load prediction model based on the multi-modal deep forest algorithm for prediction, the spatio-temporal information in the multi-modal data can be fully exploited, enabling the multi-modal deep forest algorithm to effectively handle complex relationships, thereby significantly improving the accuracy of power load prediction, and further more precisely allocating energy and improving resource utilization rate. At the same time, adaptive adjustments can be made according to different multi-modal data, enabling efficient response in the face of power load prediction scenarios for different building complexes and ensuring the continuous accuracy of prediction.

[0075] Embodiment 2

[0076] This embodiment makes improvements on the basis of the building complex power load prediction method based on the multi-modal deep forest algorithm proposed in Embodiment 1.

[0077] In this embodiment, a high-performance edge computing processor is used to process data in real time, which can reduce data transmission latency and improve the system response speed. Edge nodes undertake key tasks such as data preprocessing, feature extraction, and preliminary prediction, and only transmit necessary data and results to the cloud to relieve its pressure and optimize system performance. In real-time data processing, edge devices such as sensors collect load data in real time to ensure its timeliness and integrity. Distributed at key positions in the building complex, they can obtain first-hand data to support subsequent analysis and prediction. Moreover, preprocessing work including cleaning and denoising is carried out locally to avoid sending a large amount of raw data to the central server, reducing transmission and processing pressure, making the data more refined and valuable, and improving the efficiency and quality of system data processing. It can directly run the pre-trained model for local inference and quickly output results, achieving low-latency response to meet real-time requirements. When new data arrives at the edge device, it can be immediately calculated and predicted without waiting for cloud processing, shortening the response time to provide a decision-making basis for power dispatching management. At the same time, the edge device monitors the error between the prediction and the actual data in real time. When it exceeds the predetermined threshold, it quickly triggers the model update mechanism or alarm. Continuous monitoring can timely detect deviations and take measures such as retraining, adjusting parameters, or alarming. Based on real-time errors, the edge computing device can also directly locally adjust the model inference process or initiate local calculations such as incremental training without cloud resources, quickly adapting to new situations in the event of sudden changes or emergencies in power loads, and improving the system response speed and response ability.

[0078] In this embodiment, the nodes of the graph include:

[0079] Functional nodes divide buildings into different node categories according to the functional differences of each building in the building complex, such as office buildings, laboratories, parking lots, etc., so as to better analyze the impact of different functional buildings on power loads. There are often significant differences in power consumption time, types of electrical equipment, and power consumption among buildings with different functions. By classifying them into different functional nodes, it is possible to more specifically study their respective power load characteristics and provide a basis for accurate prediction.

[0080] Sensing nodes. Sensor devices (such as temperature sensors, load monitoring sensors, etc.) in each building or area are also important components of the nodes. They are responsible for collecting real-time data and transmitting the data to other computing units or control systems, providing real-time environmental and load information for the system. These sensor devices are distributed in every corner of the building complex, capable of real-time sensing of changes in key parameters such as temperature, humidity, and power load, providing first-hand information for subsequent analysis and prediction, and ensuring that the system can timely understand the operating state of the building complex.

[0081] Computing nodes. Edge computing nodes shoulder the important task of processing data from sensing nodes. They have a certain computing power and storage resources and are used for key tasks such as real-time prediction, data analysis, and model inference. They are the core units for system data processing and analysis. Computing nodes quickly process and analyze the data transmitted from sensing nodes, use advanced algorithms and models for prediction and decision-making, and timely feedback the processing results to the control system or transmit them to other relevant nodes to achieve the intelligent operation of the system.

[0082] Control nodes. These nodes are responsible for managing and controlling the energy or facilities of the building complex (such as HVAC systems, power load scheduling, etc.), ensuring the reasonable distribution and efficient utilization of energy. According to real-time data and prediction results, they precisely regulate the energy system of the building complex to achieve the goals of energy conservation, consumption reduction, and optimized operation.

[0083] Top-level nodes, such as the main control center or data center of the building complex, are responsible for overall coordinating task scheduling, resource allocation, and monitoring within the building complex. They usually communicate with the cloud platform to achieve the macro management and data interaction of the system. It is at the highest level of the entire system, comprehensively controlling and managing the overall operation status of the building complex, receiving data and information from each middle-level node and bottom-level node, and making global decisions and instructions based on this information. At the same time, it uploads important data and instructions to the cloud platform to achieve interaction and collaboration with external systems.

[0084] Middle-level nodes, located between buildings, are mainly responsible for summarizing, analyzing, and transmitting data from local nodes to top-level nodes. They are usually distributed in key areas of the building complex and play the role of data transfer and local coordination. These nodes are like the "bridges" in the system. On the one hand, they collect data from surrounding bottom-level nodes, conduct preliminary integration and analysis, remove some redundant information and noise, and improve the quality and effectiveness of the data. On the other hand, they transmit the processed data to the top-level nodes according to certain rules and protocols to ensure the smooth transmission of data and the accurate delivery of information. At the same time, middle-level nodes can also, according to the instructions of the top-level nodes, conduct some local coordination and management of surrounding bottom-level nodes, such as adjusting the sensor sampling frequency in certain areas, controlling local energy equipment, etc., to improve the local response speed and flexibility of the system.

[0085] The underlying nodes refer to small devices (such as sensors, home automation devices, etc.) within each building. They are responsible for collecting data and transmitting it to the intermediate layer nodes. They are the source of data collection and provide basic data support for the system. These underlying nodes are numerous and widely distributed. They continuously sense various environmental parameters and device operating states within the building, such as temperature, humidity, light intensity, power consumption, etc., and send this data to the intermediate layer nodes in a timely and accurate manner. The performance and reliability of the underlying nodes directly affect the data quality and availability of the entire system. Therefore, they need to have characteristics such as low power consumption, high stability, and high precision to ensure long-term stable operation and provide a continuous stream of basic data for the system.

[0086] As an exemplary illustration, when configuring nodes, the nodes may be equipped with computing units (such as Raspberry Pi, Edge servers, IoT gateways), storage devices (such as local storage, cache), and sensor interfaces (such as temperature and humidity sensors, ambient light sensors, load sensors) to meet the requirements of data collection, processing, and storage. The computing unit is the core of the node and is responsible for performing various computing tasks, such as data preprocessing, model inference, etc. Its performance directly affects the processing speed and efficiency of the node. The storage device is used to temporarily store the collected data and intermediate calculation results, avoiding data loss and frequent data transmission, and improving the reliability and response speed of the system. The sensor interface is used to connect various sensor devices to achieve data collection and input. Different types of sensor interfaces can support different types of sensors to meet diverse data collection needs.

[0087] The communication protocols between different nodes can be based on wireless communication (such as LoRa, Wi-Fi, Zigbee) or wired communication (such as Ethernet). Edge nodes usually communicate with each other using a local area network (LAN) to reduce latency and bandwidth consumption and ensure the efficiency and stability of data transmission. Wireless communication methods have the advantages of high flexibility and convenient deployment and are suitable for communication between some nodes that are difficult to wire or need to be mobile, but may be limited by signal interference and transmission distance. Wired communication methods have the advantages of high stability, fast transmission rate, and strong anti-interference ability and are suitable for scenarios with high requirements for communication quality and relatively fixed node positions. In practical applications, the appropriate communication method and protocol can be selected according to the specific environment and requirements of the building complex, or a combination of multiple communication methods can be used to build an efficient, stable, and reliable communication network to ensure that data between nodes can be transmitted in a timely and accurate manner.

[0088] In this embodiment, the edges of the graph include:

[0089] Spatial adjacency edge: If two buildings or areas are physically adjacent and there is a possibility of resource sharing or task coordination, an edge is established to connect these two buildings, facilitating data sharing and collaborative work. For example, adjacent office buildings can communicate and exchange data through a local area network, improving work efficiency. This way of constructing edges based on spatial adjacency can make full use of the physical proximity between buildings, reduce the distance and latency of data transmission, and improve the real-time performance and response speed of the system. At the same time, there are often certain correlations between adjacent buildings in terms of energy use, environmental impact, etc. By establishing edges, these correlations can be better analyzed and utilized to achieve more accurate power load forecasting and energy management.

[0090] Distance-weighted edge: The weight of the connection is determined based on the geographical distance between nodes. Edges are preferentially established between buildings that are closer in distance, which can effectively reduce communication latency and energy consumption. For example, multiple floors within a building can be connected through an internal network, while buildings that are farther apart may use long-distance wireless communication (such as Wi-Fi or 5G) to ensure the timeliness and reliability of data transmission. In this way, the connection relationship between nodes can be reasonably planned according to the actual geographical distribution, optimizing the topological structure of the communication network, and reducing the overall energy consumption and cost of the system while ensuring the quality of data transmission.

[0091] Function-coupling edge: When two buildings or areas are closely related in function and there is a need for data flow, an edge is constructed based on the functional relationship. For example, between a data center and an office area, due to the frequent need for data exchange and computing resource sharing, a high-bandwidth edge can be established to ensure the smoothness of data transmission. This strategy of constructing edges based on function coupling can highlight the interdependent and collaborative relationship between buildings in terms of function, providing a basis for the division of functional modules and resource allocation in the system. For example, in a large scientific research park, there may be a large amount of data interaction between laboratories and data analysis centers. By establishing a high-bandwidth edge connection, it can ensure that experimental data can be quickly and accurately transmitted to the data analysis center for processing and analysis, improving the efficiency and quality of scientific research work.

[0092] Task collaboration edge: In an energy scheduling system, frequent interactions are required between the control node and each building. Nodes in areas such as office buildings, parking lots, and public facilities need to establish connections to achieve collaborative execution of tasks due to functions such as power load scheduling and lighting control. For example, in an intelligent building complex, when the power load of an office building reaches a certain threshold, the control node can communicate with the nodes in the parking lot to coordinate the operating states of the lighting and charging equipment in the parking lot, so as to achieve the energy balance and optimal utilization of the entire building complex. This method of constructing edges based on task collaboration can organically integrate each functional area in the building complex into a whole that works collaboratively, improving the intelligence level and energy management efficiency of the system.

[0093] Data flow direction edge: The direction of the edge is determined according to the data flow direction to clarify the transmission path of data in the network. For example, data is collected from the sensing nodes and sent to the computing nodes, or the processing results are sent from the computing nodes to the control nodes to ensure the orderliness and accuracy of data transmission. This method of constructing edges based on the data flow direction can clearly depict the flow trajectory of data in the system, facilitating the monitoring and management of the data transmission process. For example, in a power load forecasting system, the power load data collected by the sensor nodes will be transmitted along specific edges to the edge computing nodes for preprocessing and forecasting, and then the forecasting results will be transmitted to the control nodes through the corresponding edges for guiding power scheduling and energy management decisions. By clarifying the direction of the edges, chaos and errors in the data transmission process can be avoided, improving the reliability and stability of the system.

[0094] Priority resource allocation edge: Optimize the connection of the edge according to the task priorities and resource requirements between the nodes. For example, dedicated and high-priority edges can be established between computing nodes with high loads to ensure efficient computing and data transmission, improving the overall performance of the system. In practical applications, different nodes may undertake different tasks and functions. Some tasks may have high requirements for real-time performance and computing resources, while some tasks are relatively relaxed. By optimizing the connection of the edges according to the task priorities and resource requirements, the communication resources and computing resources of the system can be reasonably allocated to ensure that critical tasks can be processed and responded to in a timely manner, improving the overall performance and efficiency of the system. For example, in a large data center, high-priority edge connections can be established between the computing nodes responsible for core business processing to ensure the fast transmission and processing of data, while some auxiliary nodes can use lower-priority edge connections to reduce communication costs and resource consumption without affecting the overall performance of the system.

[0095] In this embodiment, in view of the different distributions and scales of different modalities of data in the feature space, independent preprocessing is required, including operations such as normalization, missing value filling, and noise filtering.

[0096] Fuse the feature extraction results to obtain multi-modal fusion features, including:

[0097] Normalize the multi-modal feature vectors, and its expression is as follows:

[0098]

[0099] Wherein, represents the feature vector of the m-th modality, μ m and σ m are the mean and standard deviation of the feature vector of the m-th modality, respectively.

[0100] Perform weighted feature fusion on the multi-modal feature vectors after normalization, and its expression is as follows:

[0101]

[0102] Wherein, ω m is the weight of the feature vector of the m-th modality, satisfying M is the number of modalities. By reasonably setting the weights, the importance of different modality data in prediction can be highlighted, and the fusion effect can be improved. For example, in power load prediction, historical power load data may have a greater impact on the prediction results, so a higher weight can be assigned to it. While weather data and the like may have a relatively smaller impact, the weight can be appropriately reduced, but it cannot be completely ignored, because weather changes may also have a certain impact on power load. In this way, through weighted fusion, the advantages of each modality data can be fully utilized to improve the accuracy and reliability of the overall prediction model.

[0103] In this embodiment, the power load prediction model based on the multi-modal deep forest algorithm includes several layers of random forest models, and each layer of random forest model includes several decision trees. Input the multi-modal fusion features into the trained power load prediction model based on the multi-modal deep forest algorithm for prediction processing to obtain the power load prediction results, including:

[0104] Input the multi-modal fusion feature X i , and extract the output features of the initial layer through the first layer of random forest model, and its expression is as follows:

[0105] H (1) = RandomForest (1) (X i )

[0106] Wherein, H (1) is the output feature of the first layer of random forest model, and RandomForest (1)Represents the random forest model of the first layer. It comes from multiple fields or measurement indicators in the dataset. During the training process, each tree randomly selects a subset of the original features for training. This randomness helps reduce overfitting and improve the generalization ability of the model. Each decision tree splits nodes based on a certain value of the feature, and finally the leaf nodes generate class or regression prediction values, providing the basic data for the subsequent layer. For example, for power load forecasting, the original features may include the power load value at a historical moment, the current weather temperature, humidity, whether it is a working day, etc. After these features are input into the first-layer random forest and trained and split by each tree, the output features of the first layer are obtained. These output features are a preliminary refinement and abstraction of the original features.

[0107] Concatenate the output features H of the previous layer (l-1) with the multi-modal fusion feature X i to generate the fused feature through the intermediate-layer random forest model. The expression is as follows:

[0108] H (l) = RandomForest (l) ([H (l-1) ; X i )

[0109] where, [H (l-1) ; X i represents concatenating the output features H of the previous layer (l-1) with the multi-modal fusion feature X i , and RandomForest (l) represents the random forest model of the l-th layer. In the intermediate layer, the random forest of each layer uses the output features of the previous layer as input and continues to extract and transform features. As the number of layers increases, the model can gradually discover more complex and abstract features, which can better capture the potential patterns and rules in the data. For example, after several layers of processing, some deep features related to the power load change trend, periodicity, etc. may be obtained, and these features are of great significance for accurately predicting the power load.

[0110] Integrate the output features H of the last-layer random forest model (L) to generate the final power load prediction value. The expression is as follows:

[0111]

[0112] where, represents the power load prediction value, N is the total number of decision trees in the random forest model, T i represents the i-th decision tree in the random forest model, and H (L) is the output feature of the last layer.

[0113] In this embodiment, before inputting the multi-modal fusion features into the trained power load prediction model based on the multi-modal deep forest algorithm for prediction processing, the method further includes training the power load prediction model based on the multi-modal deep forest algorithm, including:

[0114] Integrate multi-modal data including historical power load data, weather data, and holiday information into a training set.

[0115] Determine the number of random forest models and the depth of decision trees in the power load prediction model through cross-validation.

[0116] Use the training set to train the power load prediction model until its loss function converges to obtain the trained power load prediction model. The expression of the loss function of the power load prediction model is as follows:

[0117]

[0118] where x represents the input training set data, θ is the parameter set of the random forest model, n is the number of training set samples, is the i-th power load prediction value, t i is the i-th true power load label value.

[0119] In the specific implementation process, use historical data to train the power load prediction model, and adopt methods such as cross-validation and grid search to optimize model parameters (such as the depth of decision trees, the number of random forests, the strategy of feature selection, etc.), effectively improving the prediction accuracy and the generalization ability of the model, so that it can adapt to the power load prediction requirements in different scenarios. During the training process, a large amount of historical data is divided into a training set, a validation set, and a test set according to a certain ratio. By training the model on the training set and adjusting and optimizing the parameters on the validation set, continuously try different parameter combinations to find the optimal model configuration. Finally, evaluate the optimized model on the test set to ensure that the model can also show good performance on unseen data, thereby improving the generalization ability of the model and enabling it to handle various complex situations and changes in practical applications.

[0120] Run the trained model on the edge computing processor to receive the latest multi-modal data in real time and quickly output the power load prediction results. The system has the ability to automatically monitor changes in data distribution and can update and optimize the model regularly or according to actual needs to ensure the continuous accuracy of the prediction results and effectively respond to the dynamic changes in power load. Through the error monitoring and evaluation mechanism, calculate the error between the current predicted value and the actual value regularly. Once the mean square error (MSE) exceeds the set error threshold, immediately trigger the model update process. During the actual operation process, the system continuously monitors the changes in the input data. When it is found that the data distribution has changed significantly, such as special situations like holidays and extreme weather leading to changes in the power load pattern, the model update mechanism is started in a timely manner. By retraining the model or adjusting the model parameters, the model can quickly adapt to the new data pattern and ensure the reliability and timeliness of the prediction results.

[0121] Embodiment 3

[0122] To comprehensively verify the effectiveness and superiority of the power load prediction method for building complexes based on the multi-modal deep forest algorithm of the present invention, the following three different experimental scenarios are designed. Each scenario covers steps such as data collection, system deployment, model training and prediction, and comparative analysis, and corresponding experimental data is collected to evaluate the method performance.

[0123] Experimental scenario 1: Application in a commercial complex

[0124]

[0125]

[0126] Experimental scenario 2: Application in a residential area

[0127]

[0128] Experimental scenario 3: Application in an industrial park

[0129]

[0130]

[0131] In terms of prediction accuracy, by comparing the performance of each model in different scenarios, with the mean square error (MSE) and mean absolute error (MAE) as the main evaluation indicators, the present invention is significantly superior to the traditional ARIMA model, single-modal graph convolutional network, and pure random forest model in all scenarios, and its error is also lower compared to the prediction model based on the deep neural network. This fully shows that when dealing with the power load prediction of building complexes, the present invention can more accurately capture the laws and characteristics in the data and thus provide more accurate prediction results.

[0132] In terms of real-time performance, the method of the present invention utilizes edge computing for processing, and its average response time can be less than 120 milliseconds. Compared with other traditional methods and deep learning models, the speed advantage is extremely obvious, as the response times of the latter all exceed 300 milliseconds. Such a rapid response speed is of crucial significance in practical applications, enabling timely adjustment according to changes in power load, thereby effectively ensuring the stable operation of the power system.

[0133] Since the deep forest algorithm adopts a multi-level decision tree integration method, this model has strong stability and generalization ability. Whether it is changes in different time periods, weather conditions, or production activities, the prediction error fluctuations of the method of the present invention are all within a relatively small range, that is, between ±0.005 and ±0.009, which is better than the error fluctuation ranges of other models (±0.007 to ±0.010). This characteristic demonstrates that this method can achieve stable prediction in various complex environments and will not produce large deviations due to changes in external factors, providing a solid guarantee for the reliable operation of the power system.

[0134] In addition, in terms of adaptability, the model update and self-learning mechanism in the present invention can automatically detect changes in data distribution and timely update and optimize the model. In all experimental scenarios, this method can excellently adapt to sudden changes and abnormal conditions in power load, and the prediction performance always remains stable, meeting the requirements of long-term operation. This strong adaptability enables this method to continuously self-adjust and optimize over time and with environmental changes. Regardless of the type of building complex or how the power load changes, it can achieve efficient and accurate prediction.

[0135] The present invention is superior to other comparative methods in evaluation indicators such as prediction accuracy, real-time performance, and stability, which strongly proves that it has significant advantages and high practical value in practical applications. The present invention combines the deep forest and multi-modal data processing technologies, which can effectively process complex spatio-temporal relationships and multi-modal data, significantly improving the accuracy and real-time performance of power load prediction, and is particularly suitable for smart grids and building complex power management systems.

[0136] Embodiment 4

[0137] As Figure 2 shown, this embodiment proposes a building complex power load prediction system based on the multi-modal deep forest algorithm, which is applied to the building complex power load prediction method based on the multi-modal deep forest algorithm as described in the above embodiment, and includes: an acquisition module 100, an extraction and fusion module 200, and a prediction module 300.

[0138] Among them, the acquisition module 100 is used to acquire multimodal data of the building complex. The extraction and fusion module 200 is used to extract features from the multimodal data and perform fusion processing on the feature extraction results to obtain multimodal fusion features. The prediction module 300 is used to input the multimodal fusion features into a trained power load prediction model based on the multimodal deep forest algorithm for prediction processing to obtain a power load prediction result.

[0139] It should be noted that the foregoing explanation of the embodiment of the building complex power load prediction method based on the multimodal deep forest algorithm is also applicable to the building complex power load prediction system based on the multimodal deep forest algorithm of this embodiment, and will not be elaborated here.

[0140] Embodiment 5

[0141] Figure 3 The following is a schematic structural diagram of the electronic device 400 provided in this embodiment. The electronic device 400 includes: a memory 401, a processor 402, and a computer program stored on the memory 401 and executable on the processor 402.

[0142] When the processor 402 executes the program, it implements the building complex power load prediction method based on the multimodal deep forest algorithm provided in the foregoing embodiment.

[0143] Furthermore, the electronic device 400 further includes: a communication interface 403, which is used for communication between the memory 401 and the processor 402.

[0144] The memory 401 may include a high-speed RAM (Random Access Memory) memory, and may also include a non-volatile memory, such as at least one disk memory.

[0145] If the memory 401, the processor 402, and the communication interface 403 are implemented independently, the communication interface 403, the memory 401, and the processor 402 can be interconnected through a bus and communicate with each other. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of representation, Figure 3 only a thick line is shown in the figure, but it does not mean that there is only one bus or one type of bus.

[0146] Optionally, in a specific implementation, if the memory 401, the processor 402, and the communication interface 403 are integrated on a single chip, the memory 401, the processor 402, and the communication interface 403 can communicate with each other through an internal interface.

[0147] The processor 402 may be a CPU (Central Processing Unit), or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present invention.

[0148] In the description of this specification, the descriptions with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms are not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0149] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "N" is at least two, such as two, three, etc., unless otherwise specifically defined.

[0150] Any process or method description in the flowchart or described in other ways herein may be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a customized logic function or process, and the scope of the preferred embodiments of the present invention includes additional implementations, where the functions may be executed in a substantially simultaneous manner or in a reverse order according to the functions involved, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0151] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays, field programmable gate arrays, and the like.

[0152] Those of ordinary skill in the art can understand that all or part of the steps carried by the method of the above embodiments can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0153] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limitations on the embodiments of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all the embodiments here. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the claims of the present invention.

Claims

1. A building group power load prediction method based on a multimodal deep forest algorithm, characterized in that: include: Obtain multimodal data of building complexes; Extracting features from the multimodal data and fusing the feature extraction results to obtain multimodal fusion features; The multimodal fusion features are input into a trained power load prediction model based on a multimodal deep forest algorithm for prediction processing to obtain a power load prediction result.

2. The building complex power load prediction method based on the multimodal deep forest algorithm according to claim 1 is characterized in that: Extracting features from the multimodal data includes: According to the preset rules, each building in the building complex is used as a node in the multimodal deep forest algorithm, and the geographical location and functional relationship are used as the edges of the graph; For each node i, the multimodal feature vector is extracted according to the following formula: Where M is the number of modes, is the feature of the mth mode.

3. The building complex power load prediction method based on the multimodal deep forest algorithm according to claim 2 is characterized in that: The nodes of the graph include: Functional nodes, used to classify buildings according to their functional attributes; Sensing nodes, used to collect real-time data within the building; Computing nodes, used to process data from sensing nodes and perform prediction processing of power load; Control nodes, used to manage energy equipment in the building complex; The top-level node is used as the main control center of the building complex and is responsible for task scheduling and resource allocation; The middle layer nodes are used to aggregate the data of the bottom nodes and pass them to the top nodes; The bottom layer nodes are used to collect data from equipment in the building and transmit it to the middle layer nodes.

4. The building complex power load prediction method based on the multimodal deep forest algorithm according to claim 3 is characterized in that: The edges of the graph include: Spatial adjacency edges are used to connect buildings that are adjacent in physical space; distance-weighted edges, which assign weights based on the geographic distance between buildings; Functional coupling edges are used to connect functionally related buildings; Task coordination edge, used to connect functional nodes and control nodes; Data flow direction edge, used to represent the data transmission path from the sensing node to the computing node or from the computing node to the control node; Priority resource allocation edges are used to connect nodes with matching computing capabilities.

5. The building complex power load prediction method based on multimodal deep forest algorithm according to claim 2 is characterized in that: The feature extraction results are fused to obtain multimodal fusion features, including: The multimodal feature vector is normalized and its expression is as follows: in, represents the eigenvector of the mth mode, μ m and σ m are the mean and standard deviation of the eigenvector of the mth mode respectively; The normalized multimodal feature vector is subjected to weighted feature fusion, and its expression is as follows: Among them, ω m is the weight of the eigenvector of the mth mode, satisfying M is the number of modes.

6. The building complex power load prediction method based on the multimodal deep forest algorithm according to claim 5 is characterized in that: The power load prediction model based on the multimodal deep forest algorithm includes several layers of random forest models, and each layer of random forest model includes several decision trees; The multimodal fusion features are input into the trained power load prediction model based on the multimodal deep forest algorithm for prediction processing to obtain the power load prediction result, including: Input multimodal fusion feature X i , the output features of the initial layer are extracted through the first layer random forest model, and its expression is as follows: H (1) =RandomForest (1) (X i ) Among them, H (1) is the output feature of the random forest model of the first layer, RandomForest (1) Represents the random forest model of the first layer; The output feature H of the previous layer (l-1) With multimodal fusion feature X i After splicing, the fused features are generated through the intermediate layer random forest model, and the expression is as follows: H (l) =RandomForest (l) ([H (l-1) ;X i ]) Among them, [H (l-1) ;X i ] means that the output feature H of the previous layer (l-1) With multimodal fusion feature X i For splicing, RandomForest (l) represents the random forest model of the lth layer; The output feature H of the last layer of random forest model (L) Perform integrated learning to generate the final power load forecast value, which is expressed as follows: in, represents the power load forecast value, N is the total number of decision trees in the random forest model, T i represents the i-th decision tree in the random forest model, H (L) is the output feature of the last layer.

7. The building complex power load prediction method based on the multimodal deep forest algorithm according to claim 6 is characterized in that: Before inputting the multimodal fusion features into the trained power load prediction model based on the multimodal deep forest algorithm for prediction processing, the method further includes training the power load prediction model based on the multimodal deep forest algorithm, including: Integrate multimodal data including historical power load data, weather data, and holiday information into a training set; Determine the number of random forest models and the depth of decision trees in the power load forecasting model through cross validation; The power load forecasting model is trained using the training set until its loss function converges to obtain a trained power load forecasting model; the expression of the loss function of the power load forecasting model is as follows: Among them, x represents the input training set data, θ is the parameter set of the random forest model, and n is the number of training set samples. is the predicted value of the ith power load, y i is the real power load label value of the ith one.

8. The building complex power load prediction method based on the multimodal deep forest algorithm according to any one of claims 1 to 7, characterized in that: The multimodal data includes at least one of building structural data, historical power load data, power equipment data, environmental data, time information, holiday information and resident activity data.

9. A building complex power load forecasting system based on a multimodal deep forest algorithm, characterized in that: include: An acquisition module, used to acquire multimodal data of building complexes; An extraction and fusion module is used to extract features from the multimodal data and fuse the feature extraction results to obtain multimodal fusion features; The prediction module is used to input the multimodal fusion features into a trained power load prediction model based on a multimodal deep forest algorithm for prediction processing to obtain a power load prediction result.

10. An electronic device, characterized in that: The control device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the operations performed by the building complex power load prediction method based on a multimodal deep forest algorithm as described in any one of claims 1 to 8 are implemented.

Images