Characterization and prediction update method of building energy consumption throughout the life cycle with multi-dimensional attribute coupling
A deep learning-based method for building energy prediction addresses the limitations of historical data reliance and linear models by analyzing multi-dimensional attributes and updating neural networks in real-time, enhancing prediction accuracy and adaptability.
Patent Information
- Application Number
- CN202411312943.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-20
- Publication Date
- 2024-12-24
- Estimated Expiration
- 2044-09-20
AI Technical Summary
Existing building energy management methods rely on historical data and linear models, failing to capture complex nonlinear relationships and adapt to dynamic changes in building demand throughout the life cycle, leading to poor prediction accuracy and adaptability.
A deep learning-based method that analyzes multi-dimensional attributes of buildings, constructs neural networks with neurons, trains gradient parameters using pre-training data, and updates models in real-time to predict energy consumption accurately and flexibly.
Improves prediction accuracy and adaptability by comprehensively classifying and dynamically adjusting multi-dimensional attributes, enabling real-time updates and intelligent forecasting of building energy consumption.
Smart Images
Figure CN118839869B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to data processing technology, and in particular to a method for characterizing and predicting the energy consumption of a building throughout its life cycle through multi-dimensional attribute coupling. Background Art
[0002] With the continuous development of building life cycle management and energy management, more accurate and real-time methods are needed to characterize and predict the energy characteristics of buildings throughout their life cycle. Existing building energy management methods mainly rely on historical data and experience for prediction, lack a systematic and intelligent dynamic update mechanism, and cannot fully consider the complex coupling relationship between the multi-dimensional attributes of buildings and the life cycle characteristic attributes.
[0003] Traditional building energy prediction methods have several shortcomings. First, most of these methods rely on linear models, which make it difficult to capture the complex nonlinear relationship between building energy and multi-dimensional attributes. Second, these methods are difficult to cope with the dynamic changes in building demand at different life cycle stages, resulting in poor prediction accuracy and adaptability.
[0004] Therefore, how to combine the characteristics and energy consumption data of buildings in different life cycles to achieve more accurate and flexible building energy consumption forecasts has become an urgent problem to be solved. Summary of the invention
[0005] The embodiment of the present invention provides a method for characterizing and predicting the energy consumption of a building throughout its life cycle through multi-dimensional attribute coupling, which can combine the characteristics and energy consumption data of a building in different life cycles to achieve more accurate and flexible building energy consumption prediction and management.
[0006] A first aspect of an embodiment of the present invention provides a method for characterizing and predicting building energy consumption throughout its life cycle by coupling multiple attributes, including:
[0007] Acquire multi-dimensional attributes of the target building over its entire life cycle, wherein the multi-dimensional attributes include multiple geometric characteristic attributes, environmental characteristic attributes, functional characteristic attributes, and life cycle characteristic attributes;
[0008] Based on deep learning neural network, the mapping relationship between building energy consumption and multi-dimensional attributes is analyzed and processed, a neural network with multiple neurons is constructed and the connection relationship between neurons is established;
[0009] The gradient parameters of the neural network are trained based on pre-training data of different life cycles to obtain the gradient parameters corresponding to each neuron under different life cycles;
[0010] Construct a representation network corresponding to the neural network, obtain the multi-dimensional attributes of the target building in real time, and obtain the current updated data based on the display and prediction of the life cycle of the representation network.
[0011] Optionally, in a possible implementation manner of the first aspect, the multi-dimensional attributes of the target building in the entire life cycle are obtained, and the multi-dimensional attributes include multiple ones of geometric feature attributes, environmental feature attributes, functional feature attributes and life cycle feature attributes, including:
[0012] Classify the target building according to its multi-dimensional attributes throughout its life cycle, and obtain multi-dimensional attribute nodes and node connection labels under each ontology attribute;
[0013] Based on the user's adjustment information of any multi-dimensional attribute node and node connection label, delete or modify the multi-dimensional attribute node and node connection label;
[0014] If it is determined that the user adds any multi-dimensional attribute nodes and node connection labels, the newly added multi-dimensional attribute nodes and node connection labels are saved and processed.
[0015] Optionally, in a possible implementation of the first aspect, the deep learning-based neural network analyzes and processes the mapping association relationship between building energy consumption and multi-dimensional attributes, constructs a neural network with multiple neurons and establishes a connection relationship between neurons, including:
[0016] Constructing neurons corresponding to each multi-dimensional attribute node, connecting corresponding neurons based on node connection labels, and obtaining a corresponding first neural network;
[0017] Determine the corresponding relationship information between the building's main state and the building energy consumption corresponding to each main state at each neuron;
[0018] Based on the connection relationship of the multi-dimensional attribute nodes, the neurons corresponding to the corresponding input layer and output layer are constructed, and the newly constructed neurons are connected to the first neural network to establish a second neural network.
[0019] Optionally, in a possible implementation manner of the first aspect, determining, at each neuron, the corresponding relationship information between the main state of the building and the building energy usage corresponding to each main state includes:
[0020] Determine all possible ontological states of the multidimensional attribute nodes corresponding to each neuron in the first neural network, each multidimensional attribute node having a preset number of ontological states;
[0021] Construct corresponding relationship information between each entity state and building energy consumption, wherein the corresponding relationship information at least includes a calculation algorithm.
[0022] Optionally, in a possible implementation of the first aspect, constructing neurons corresponding to the input layer and the output layer respectively based on the connection relationship of the multi-dimensional attribute nodes, and connecting the newly constructed neurons to the first neural network to establish the second neural network includes:
[0023] Construct neurons corresponding to the input layer and output layer based on the user's configuration;
[0024] Configure the corresponding input algorithm and output algorithm for each neuron in the input layer and output layer respectively;
[0025] If it is determined that the neurons in the input layer and the output layer have a connection relationship with the neurons in the first neural network, the neurons in the input layer and the output layer are connected to the neurons in the first neural network;
[0026] If the user's input configuration information and output configuration information for the neurons in the input layer and the output layer are received, the neurons in the input layer and the output layer are connected to the neurons in the first neural network based on the input configuration information and the output configuration information to obtain a second neural network.
[0027] Optionally, in a possible implementation of the first aspect, the training process of the gradient parameters of the neural network based on the pre-training data of different life cycles to obtain the gradient parameters corresponding to each neuron under different life cycles includes:
[0028] Obtain buildings of the same type corresponding to the target building with different life cycles, extract the multi-dimensional attributes of the same type of buildings configured by the user, and the energy consumption information corresponding to each ontology attribute;
[0029] Based on the multi-dimensional attributes of each building of the same type, the first neural network is adjusted accordingly so that the neurons in the first neural network correspond to the multi-dimensional attributes of each building of the same type, thereby obtaining a second neural network to be trained;
[0030] Input the ontology training information and energy training information of the ontology attributes of the same type of buildings in the corresponding life cycle into the second neural network to be trained, and set up training accordingly to obtain the sub-gradient parameters corresponding to each neuron under different life cycles;
[0031] The sub-gradient parameters of all buildings of the same type in the same period are fused to obtain the total gradient parameter.
[0032] Optionally, in a possible implementation of the first aspect, the sub-gradient parameters of all buildings of the same type in the same period are fused to obtain a total gradient parameter, including:
[0033] Obtain the sub-gradient parameters of all buildings of the same type under each neuron in the same period, and classify the neurons of the same multi-dimensional attribute nodes to obtain the corresponding classification set;
[0034] Obtaining the number of neurons deleted and adjusted in the second neural network corresponding to each neuron in the classification set, and the level of the deleted and adjusted neurons;
[0035] Based on the number of neurons deleted and adjusted in the second neural network and the levels of neurons, the gradient coefficients of the corresponding neurons in each classification set are calculated comprehensively, and the total gradient parameters are obtained by fusion calculation of the sub-gradient parameters based on the gradient coefficients.
[0036] Optionally, in a possible implementation manner of the first aspect, the step of comprehensively calculating the gradient coefficient of the corresponding neuron in each classification set based on the number of neurons deleted and adjusted in the second neural network and the level of the neurons, and fusing the sub-gradient parameters based on the gradient coefficient to obtain the total gradient parameter comprises:
[0037] The corresponding quantity gradient offset sub-coefficient is calculated based on the quantity of the deleted and adjusted neurons and the quantity normalization value;
[0038] Determine the level offset value of each neuron based on the level of the neuron, the level of each neuron has a preset level offset value, and calculate the corresponding level gradient offset sub-coefficient based on the sum of the level offset values and the level normalization value;
[0039] The preset reference weight is deducted from the gradient coefficient of each neuron of the quantity gradient offset sub-coefficient and the level gradient sub-coefficient, and the corresponding sub-gradient parameters are fused and calculated based on each gradient coefficient to obtain the total gradient parameter.
[0040] Optionally, in a possible implementation manner of the first aspect, the step of subtracting the gradient coefficient of each neuron of the quantity gradient offset sub-coefficient and the level gradient sub-coefficient from the preset reference weight, and fusing the corresponding sub-gradient parameters based on each gradient coefficient to obtain a total gradient parameter includes:
[0041] The gradient coefficient of each neuron is calculated by the following formula,
[0042] ,
[0043] in, For the classification set The gradient coefficient of each neuron, For the The number of neurons deleted and adjusted in the second neural network corresponds to the number of neurons deleted and adjusted in the second neural network. is the normalized value of quantity, For the The neurons in the second neural network correspond to The level offset value of the deleted adjustment neuron, For the The upper limit of the number of neurons deleted and adjusted in the second neural network corresponding to the neurons, is the level normalized value;
[0044] Based on the fusion mean calculation of the corresponding sub-gradient parameters for each gradient coefficient in the classification set, the total gradient parameter corresponding to the classification set is obtained, which is calculated by the following formula:
[0045] ,
[0046] in, For the The total gradient parameter corresponding to the classification set, For the classification set The gradient coefficient of each neuron, For the classification set The sub-gradient parameters of neurons, For the The upper limit of the neurons in the classification set, For the classification set The gradient coefficient of a neuron.
[0047] Optionally, in a possible implementation of the first aspect, constructing a representation network corresponding to the neural network, acquiring multi-dimensional attributes of a target building in real time, and obtaining current update data based on the current life cycle display and prediction of the representation network include:
[0048] Mapping the neurons of the second neural network structure of the target building to obtain the representation grid mapped by each neuron, and determining the positional relationship of the representation grid based on the positional relationship of the neurons;
[0049] Obtain the ontological attributes of the neuron corresponding to each representation grid and generate corresponding attribute tags and store them in correspondence with the representation grid to obtain a representation network;
[0050] The multi-dimensional attributes and current life cycle of the target building are input into the second neural network structure to obtain analysis data of the second neural network structure, the analysis data is decomposed based on the representation network display, and the current update data is obtained.
[0051] Optionally, in a possible implementation of the first aspect, the inputting of the multi-dimensional attributes and the current life cycle of the target building into the second neural network structure to obtain analysis data of the second neural network structure, decomposing the analysis data based on the representation network display, and obtaining current update data includes:
[0052] Based on the current life cycle, a second neural network structure corresponding to the cycle is selected, and the multi-dimensional attributes of the target building are input into the second neural network structure to obtain analysis data of the second neural network structure;
[0053] Extracting energy usage information of corresponding neurons in the analysis data, and displaying and processing the energy usage information based on corresponding representation grids;
[0054] The information of the neurons in the output layer of the analysis data is extracted to obtain the current update data.
[0055] Optionally, in a possible implementation of the first aspect, the selecting a second neural network structure corresponding to a cycle based on the current life cycle, inputting the multi-dimensional attributes of the target building into the second neural network structure, and obtaining analysis data of the second neural network structure includes:
[0056] Inputting the multi-dimensional attributes of the target building into the second neural network structure;
[0057] If the neuron is determined to be a neuron for calculating energy usage information, the corresponding neuron is regarded as a neuron with energy usage information;
[0058] If the neuron is determined to be a neuron that does not compute energy-consuming information, it is determined to be a neuron that does not need to be added to the representation network.
[0059] Technical Effect 1: Comprehensive Classification and Dynamic Adjustment of Multi-dimensional Attributes
[0060] The present invention generates multi-dimensional attribute nodes and node connection labels by acquiring and classifying the multi-dimensional attributes of buildings throughout their life cycle (including geometric features, environmental features, functional features, and life cycle characteristic attributes). This classification process not only comprehensively covers the characteristics of buildings, but also allows users to dynamically adjust these attributes and connection labels, and add or delete nodes. Through this dynamic adjustment mechanism, the model can be updated and maintained according to user needs and new data, improving the flexibility and adaptability of the model.
[0061] Technical Effect 2: Construction and Training of Deep Learning Models
[0062] The present invention uses deep learning technology to construct and train a neural network model to analyze the mapping relationship between multi-dimensional attributes and building energy consumption. A first neural network is formed by constructing corresponding neurons for each multi-dimensional attribute node and determining the connection relationship between neurons based on the node connection labels. On this basis, a complete second neural network is generated by constructing input layer and output layer neurons. The gradient parameters of the neural network are trained using pre-training data of different life cycles to obtain the gradient parameters corresponding to each neuron under different life cycles, which further improves the prediction accuracy and adaptability of the model.
[0063] Technical effect three: real-time update and intelligent prediction
[0064] The present invention acquires the multi-dimensional attributes and current life cycle data of the target building in real time by constructing a characterization network corresponding to the neural network, and displays and predicts based on the characterization network. The model can be dynamically updated and optimized according to real-time data to provide the latest building energy consumption forecast information. Through this real-time update mechanism, the model can adapt to the changing needs of buildings at different life cycle stages, improving the intelligence level and efficiency of energy management. In summary, the present invention significantly improves the accuracy and adaptability of building energy consumption characterization and prediction through technical means such as comprehensive classification and dynamic adjustment of multi-dimensional attributes, construction and training of deep learning models, and real-time updating and intelligent prediction, solving the problems of poor flexibility, low prediction accuracy and lack of real-time update mechanism in the prior art. Through this method, applications in the fields of energy-saving building design, building energy management and continuous performance optimization can be better supported. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] Figure 1 It is a flow chart of a method for characterizing and predicting and updating building energy consumption through a full life cycle with multi-dimensional attribute coupling provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0066] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.
[0067] See also Figure 1 , is a flow chart of a method for characterizing and predicting building energy consumption throughout the life cycle through multi-dimensional attribute coupling provided by an embodiment of the present invention, including:
[0068] S1, obtaining multi-dimensional attributes of a target building throughout its life cycle, wherein the multi-dimensional attributes include multiple types of geometric feature attributes, environmental feature attributes, functional feature attributes, and life cycle feature attributes.
[0069] This solution can build and continuously update a deep learning model that characterizes the energy characteristics of a building throughout its life cycle. The model can not only describe existing building characteristics, but can also be updated based on user adjustments and newly added information to accurately predict future trends in building energy use. This approach has a great impact on areas such as energy-saving building design, building energy management, and continuous performance optimization.
[0070] In some embodiments, the multi-dimensional attributes of the target building in the whole life cycle are obtained, and the multi-dimensional attributes include multiple geometric feature attributes, environmental feature attributes, functional feature attributes and life cycle feature attributes, including:
[0071] S11, classify the target building according to its multi-dimensional attributes throughout its life cycle, and obtain multi-dimensional attribute nodes and node connection labels under each ontology attribute.
[0072] This scheme can classify the collected multi-dimensional attributes, generate multi-dimensional attribute nodes under different ontology attributes, and determine the connection labels between nodes. These nodes and connection labels can help build a neural network that represents the characteristics and relationships of buildings.
[0073] Among them, different ontology attributes may have multi-dimensional attribute nodes. For example, for environmental characteristic attributes, the corresponding multi-dimensional attribute nodes may include nodes of attributes such as temperature and humidity. The node connection label may be a connection label between a temperature node and an environmental characteristic attribute node, and a connection label between a humidity node and an environmental characteristic attribute node.
[0074] S12, deleting or modifying the multi-dimensional attribute nodes and node connection labels based on the user's adjustment information on any multi-dimensional attribute nodes and node connection labels;
[0075] This solution takes into account the differences between different buildings, so users can adjust any multi-dimensional attribute node and its connection label, including deleting or modifying existing nodes and connections. This step allows the model to be updated and maintained according to actual conditions or new data.
[0076] S13: If it is determined that the user adds any multi-dimensional attribute nodes and node connection labels, the newly added multi-dimensional attribute nodes and node connection labels are saved and processed.
[0077] If the user decides to add new multi-dimensional attribute nodes or node connection labels, the system will save and process these new elements to further expand and improve the model.
[0078] S2, based on deep learning neural network, analyzes and processes the mapping correlation between building energy consumption and multi-dimensional attributes, constructs a neural network with multiple neurons and establishes the connection relationship between neurons.
[0079] This solution uses deep learning technology to analyze and process the mapping relationship between building energy use and multi-dimensional attributes. By building a complex neural network model, this method aims to capture and learn the complex relationship between these attributes and building energy use. These steps provide a basis for the prediction of building energy use.
[0080] In some embodiments, the deep learning-based neural network analyzes and processes the mapping association relationship between building energy consumption and multi-dimensional attributes, constructs a neural network with multiple neurons and establishes a connection relationship between neurons, including:
[0081] S21, constructing neurons corresponding to each multi-dimensional attribute node, connecting corresponding neurons based on node connection labels, and obtaining a corresponding first neural network.
[0082] For each multi-dimensional attribute node, a corresponding neuron is constructed, and the connection relationship between these neurons is determined based on the node connection label. For example, temperature corresponds to one neuron and humidity corresponds to one neuron. The neural network constructed in this way is called the first neural network, which directly reflects the structured relationship between multi-dimensional attributes.
[0083] S22, determining, at each neuron, the corresponding relationship information between the main state of the building and the building energy consumption corresponding to each main state.
[0084] This approach determines the building's ontological state at each neuron, which describes specific characteristics of the building under specific conditions, such as geometry, environmental conditions, functional use, etc. Then, a corresponding building energy use relationship is defined for each ontological state. This helps train the neural network to understand the changes in building energy use under different states.
[0085] Wherein, determining the corresponding relationship information between the main state of the building and the building energy consumption corresponding to each main state at each neuron includes:
[0086] S221, determining all possible ontological states of the multidimensional attribute nodes corresponding to each neuron in the first neural network, each multidimensional attribute node having a preset number of ontological states.
[0087] In this step, the multidimensional attribute nodes corresponding to each neuron in the first neural network are analyzed, and all possible ontological states of these nodes are determined. The ontological state here is a specific manifestation of a certain aspect of the building. For example, the temperature node, its ontological state can include low temperature, medium temperature and high temperature. Each multidimensional attribute node has a series of preset number of ontological states, which are defined at the time of design and reflect the different manifestations of the attribute under different conditions. It can be understood that the energy consumption of the building is different in different ontological states. For example, in the low temperature ontological state, the building needs heating and the energy consumption is high; in the high temperature ontological state, the building needs cooling and the energy consumption is high; in the medium temperature ontological state, the building does not need heating and cooling, and the energy consumption is low.
[0088] S222, constructing corresponding relationship information between each entity state and building energy consumption, wherein the corresponding relationship information at least includes a calculation algorithm.
[0089] This solution defines the corresponding relationship information between each entity state and the building energy consumption. This includes building a calculation algorithm that can calculate the corresponding building energy consumption based on the current entity state. The calculation algorithm can be pre-configured manually, for example, it can be a calculation algorithm between outdoor temperature and energy consumption.
[0090] S23, constructing neurons corresponding to the corresponding input layer and output layer respectively based on the connection relationship of the multi-dimensional attribute nodes, and connecting the newly constructed neurons to the first neural network to establish a second neural network.
[0091] Based on the connection relationship of the multi-dimensional attribute nodes, the input layer and output layer neurons are constructed, which will be used to receive external data and output prediction results. These newly constructed neurons are connected to the first neural network to construct a complete second neural network. This step combines the multi-dimensional attributes with the building energy prediction to form a complete deep learning model that can be trained and optimized. Through these steps, a deep neural network that can learn and predict the complex relationship between building energy use and its multi-dimensional attributes can be constructed. This deep learning-based model can identify potential patterns and regularities from large amounts of data, which is crucial for accurately predicting the energy demand of buildings and optimizing energy use. In addition, this model can continue to learn and adjust as new data is obtained, so that its adaptability and accuracy increase over time.
[0092] The step of constructing neurons corresponding to the input layer and the output layer based on the connection relationship of the multi-dimensional attribute nodes, and connecting the newly constructed neurons to the first neural network to establish a second neural network includes:
[0093] S231, construct neurons corresponding to the input layer and the output layer respectively based on the user's configuration.
[0094] This solution will determine the specific neurons in the input layer and output layer based on the user's configuration. This step involves analyzing the user's needs and settings to determine the number and type of neurons in the input and output layers. These neurons correspond to the input and output nodes of the system, reflecting the input data dimensions required by the model and the expected output results.
[0095] S232, respectively configuring a corresponding input algorithm and an output algorithm for each neuron in the input layer and the output layer.
[0096] At the same time, the corresponding input algorithm and output algorithm need to be configured for each neuron in the input layer and output layer. These algorithms define how data flows into and out of the neural network. For example, the input algorithm may include data preprocessing steps, such as data classification. Specifically, temperature data needs to be classified into neurons corresponding to temperature, and humidity data needs to be classified into neurons corresponding to humidity. The output algorithm may include data aggregation, merging calculations, etc.
[0097] S233: If it is determined that the neurons in the input layer and the output layer have a connection relationship with the neurons in the first neural network, the neurons in the input layer and the output layer are connected to the neurons in the first neural network.
[0098] This solution needs to check the connection relationship between the neurons in the input layer and the output layer and the neurons in the first neural network. If these neurons are connected, they are connected to ensure that data can flow seamlessly throughout the neural network. This step ensures that the newly constructed input layer and output layer can be correctly integrated with the first neural network.
[0099] S234: If input configuration information and output configuration information of the user for the neurons of the input layer and the output layer are received, the neurons of the input layer and the output layer are connected to the neurons in the first neural network based on the input configuration information and the output configuration information to obtain a second neural network.
[0100] If the user provides specific configuration information for the input and output layer neurons, the neurons in the input and output layers are connected to the neurons in the first neural network based on this configuration information. This step is also to ensure that the newly constructed neurons can interact and work correctly with the existing network to form a complete second neural network. Through the above steps, the construction process of the second neural network is completed. This new network not only contains the original first neural network structure, but also expands the network's functionality through the newly added input and output layer neurons, enabling it to handle more dimensional data and more complex tasks. Such a network structure design is very useful for modeling and prediction of complex systems, and can improve the accuracy and efficiency of the system.
[0101] S3, training the gradient parameters of the neural network based on the pre-training data of different life cycles to obtain the gradient parameters corresponding to each neuron under different life cycles.
[0102] This solution can use pre-training data of different life cycles to train the gradient parameters of the neural network to obtain the gradient parameters corresponding to each neuron under these different life cycles.
[0103] In some embodiments, the training process of the gradient parameters of the neural network based on the pre-training data of different life cycles to obtain the gradient parameters corresponding to each neuron under different life cycles includes:
[0104] S31, obtaining buildings of the same type corresponding to the target building with different life cycles, extracting multi-dimensional attributes of the buildings of the same type configured by the user, and energy usage information corresponding to each ontology attribute.
[0105] It is worth mentioning that buildings have corresponding life cycles, such as the first 20 years as a life cycle, 20-50 years as a life cycle, and so on. The energy consumption of buildings with different life cycles is different. Therefore, this plan needs to analyze the energy consumption in combination with the life cycle of the building. It should also be noted that the energy consumption corresponding to different types of buildings is also different. For example, the energy consumption of office buildings and stadiums is different. Therefore, this plan will also analyze the type of building.
[0106] First, this solution needs to obtain data on buildings of the same type as the target building at different life cycles. The multi-dimensional attributes of these buildings and the energy information corresponding to each ontological attribute (such as energy consumption data) are extracted. These data provide the basis for subsequent neural network training.
[0107] S32, based on the multi-dimensional attributes of each building of the same type, the first neural network is adjusted accordingly so that the neurons in the first neural network correspond to the multi-dimensional attributes of each building of the same type, thereby obtaining a second neural network to be trained.
[0108] This solution requires corresponding adjustments to the first neural network based on the multi-dimensional attributes of the same type of buildings, so that the neurons in the network can correspond to the multi-dimensional attributes of each building of the same type. In this way, the second neural network to be trained is obtained. This step ensures that the neural network can process multi-dimensional data features similar to the target building.
[0109] It is understandable that although the types of buildings are the same, their corresponding multi-dimensional attributes may also be different. For example, for office buildings, the corresponding multi-dimensional attributes are different in different regions. Specifically, the temperature of office buildings in region A may be lower in winter and higher in summer, while the temperature of office buildings in region B may not be very different in winter and summer, resulting in the existence of temperature nodes in some buildings and not in others. Therefore, the corresponding neurons also need to change. Among them, the corresponding adjustment can be made manually, such as deleting or adding nodes.
[0110] S33, inputting the ontology training information and energy training information of the ontology attributes of the same type of buildings in the corresponding life cycle into the second neural network to be trained, and setting up training accordingly to obtain the sub-gradient parameters corresponding to each neuron under different life cycles.
[0111] This solution will input the ontology training information and energy training information of the ontology attributes of the same type of buildings in the corresponding life cycle into the second neural network to be trained, and perform corresponding training. Through these training data, the network can learn the characteristics of each ontology attribute and energy information under different life cycles, and then obtain the sub-gradient parameters corresponding to each neuron under different life cycles. Among them, the ontology training information is, for example, historical information such as temperature and humidity, and the energy training information is the information on building energy consumption. The sub-gradient parameters are, for example, the weight information corresponding to the neuron.
[0112] S34, fusing the sub-gradient parameters of all buildings of the same type in the same period to obtain the total gradient parameter.
[0113] For each life cycle, the sub-gradient parameters of all buildings of the same type are summarized and fused to obtain the total gradient parameters. These total gradient parameters can more accurately reflect the data of the neural network under different life cycles. Through the above steps, the gradient parameter training of the neural network can be completed based on the pre-trained data of different life cycles, ensuring that the network has good performance and adaptability when processing data of different life cycles. This method is particularly suitable for data analysis and prediction scenarios such as building life cycle management that need to consider long-term changes.
[0114] The sub-gradient parameters of all buildings of the same type in the same period are fused to obtain the total gradient parameters, including:
[0115] S341, obtaining the sub-gradient parameters of all buildings of the same type under each neuron in the same period, and classifying the neurons of the same multi-dimensional attribute nodes to obtain the corresponding classification set.
[0116] This scheme needs to extract the sub-gradient parameters of each neuron in the period from all buildings of the same type. The sub-gradient parameters of neurons with the same multi-dimensional attributes are classified. For example, if the attribute corresponding to multiple neurons is the temperature of the building, the sub-gradient parameters of these neurons will be classified into the same set.
[0117] S342, obtaining the number of neurons deleted and adjusted in the second neural network corresponding to each neuron in the classification set, and the level of the deleted and adjusted neurons.
[0118] Calculate the number of neurons deleted or adjusted in each classification set in the second neural network, and obtain the ranks of these neurons. The ranks represent the positions of these neurons in the network. For example, the rank of the environmental feature attribute node is greater than the rank of the temperature node, and also greater than the rank of the humidity node.
[0119] S343, based on the number of neurons deleted and adjusted in the second neural network and the levels of the neurons, a gradient coefficient of the corresponding neuron in each classification set is obtained by comprehensive calculation, and based on the gradient coefficient, a sub-gradient parameter is fused and calculated to obtain a total gradient parameter.
[0120] This solution requires the use of the number of deleted and adjusted neurons and their levels to calculate the gradient coefficient of the neurons in each classification set. This coefficient is used to adjust the weight of the sub-gradient parameter. Using the calculated gradient coefficient, the sub-gradient parameters in the classification set are fused to finally obtain the total gradient parameter of all buildings of the same type in the period. This step ensures that the data of different buildings can be integrated under a unified framework, thereby improving the robustness and generalization ability of the model. Through these steps, the sub-gradient parameters of different buildings in the same period can be effectively fused, and the deletions and adjustments in the neural network can be taken into account to finally obtain a more accurate and effective total gradient parameter. This method can improve the performance and adaptability of neural networks when processing diverse building data.
[0121] The method of comprehensively calculating the number of neurons deleted and adjusted in the second neural network and the level of neurons to obtain the gradient coefficient of the corresponding neurons in each classification set, and fusion calculating the sub-gradient parameters based on the gradient coefficient to obtain the total gradient parameter includes:
[0122] The corresponding quantity gradient offset sub-coefficient is calculated based on the number of deleted adjustment neurons and the quantity normalization value. It can be understood that the more the number of deleted adjustment neurons is, the greater the difference is, and the larger the obtained quantity gradient offset sub-coefficient will be.
[0123] The level offset value of each neuron is determined based on the level of the neuron. The level of each neuron has a preset level offset value. The corresponding level gradient offset sub-coefficient is calculated based on the sum of the level offset values and the level normalization value. It can be understood that the higher the level, the larger the preset level offset value can be. Finally, the level offset values are summarized to obtain the level gradient offset sub-coefficient.
[0124] The gradient coefficient of each neuron of the quantity gradient offset sub-coefficient and the level gradient sub-coefficient is deducted from the preset reference weight, and the total gradient parameter is obtained by fusion calculation of the corresponding sub-gradient parameters based on each gradient coefficient. In this solution, the gradient coefficient of each neuron is obtained by deducting the quantity gradient offset sub-coefficient and the level gradient sub-coefficient from the preset reference weight (usually a constant).
[0125] In some embodiments, the step of subtracting the gradient coefficient of each neuron of the quantity gradient offset sub-coefficient and the level gradient sub-coefficient from the preset reference weight, and fusing the corresponding sub-gradient parameters based on each gradient coefficient to obtain the total gradient parameter includes:
[0126] The gradient coefficient of each neuron is calculated by the following formula,
[0127] ,
[0128] in, For the classification set The gradient coefficient of each neuron, For the The number of neurons deleted and adjusted in the second neural network corresponds to the number of neurons deleted and adjusted in the second neural network. is the normalized value of quantity, For the The neurons in the second neural network correspond to The level offset value of the deleted adjustment neuron, For the The upper limit of the number of neurons deleted and adjusted in the second neural network corresponding to the neurons, is the level normalized value.
[0129] In the above formula, Represents the quantity gradient offset sub-coefficient. The more neurons are deleted and adjusted, the greater the difference is, and the larger the obtained quantity gradient offset sub-coefficient will be. The quantity normalization value can be preset manually. It is used to reflect the impact of deleting and adjusting the neuron level on the neuron gradient, where: The larger the value, the greater the difference. The smaller the gradient coefficient of each neuron, the smaller the level normalization value can be preset manually.
[0130] Based on the fusion mean calculation of the corresponding sub-gradient parameters for each gradient coefficient in the classification set, the total gradient parameter corresponding to the classification set is obtained, which is calculated by the following formula:
[0131] ,
[0132] in, For the The total gradient parameter corresponding to the classification set, For the classification set The gradient coefficient of each neuron, For the classification set The sub-gradient parameters of neurons, For the The upper limit of the neurons in the classification set, For the classification set The gradient coefficient of a neuron.
[0133] It is understandable that the closer it is to this neural network, The larger it is, the total gradient parameter of the classification set is calculated by taking the weighted average of the sub-gradient parameters of each neuron in the classification set.
[0134] S4, constructing a representation network corresponding to the neural network, obtaining multi-dimensional attributes of the target building in real time, and obtaining current update data based on the life cycle display and prediction of the representation network.
[0135] This solution requires the construction of a representation network to acquire and display the multi-dimensional attributes and life cycle of the target building in real time, and make predictions based on this to obtain the current update data.
[0136] In some embodiments, the constructing of a representation network corresponding to the neural network, obtaining multi-dimensional attributes of the target building in real time, and obtaining current update data based on the current life cycle display and prediction of the representation network include:
[0137] S41, mapping processing is performed on the neurons of the second neural network structure of the target building to obtain the representation grid mapped by each neuron, and the positional relationship of the representation grid is determined based on the positional relationship of the neurons.
[0138] This solution will map each neuron in the second neural network of the target building to obtain the representation grid corresponding to each neuron. The purpose of this step is to map the neurons in the neural network to a visual representation grid. The positional relationship of these representation grids is determined based on the positional relationship of the neurons. This step ensures that the mapped representation grid retains the relative positional relationship between neurons, so that the structure of the representation network is consistent with the structure of the neural network.
[0139] S42, obtaining the ontological attributes of the neurons corresponding to each representation grid and generating corresponding attribute tags and storing them in correspondence with the representation grid to obtain a representation network.
[0140] This solution obtains the ontological attributes of the neuron corresponding to each representation grid. This step obtains the specific attribute information of each neuron through the mapping between the neuron and its corresponding representation grid. These attributes are generated into corresponding attribute tags and stored with the corresponding representation grid. This step ensures that each representation grid not only has location information, but also contains specific attribute information.
[0141] S43, input the multi-dimensional attributes and current life cycle of the target building into the second neural network structure, obtain analysis data of the second neural network structure, decompose the analysis data based on the representation network display, and obtain current update data.
[0142] At the same time, this solution requires the multi-dimensional attributes and current life cycle of the target building to be input into the second neural network structure. This step is to input the current state and attribute information of the building into the neural network for analysis. Through the calculation of the second neural network, the analysis data is obtained. In this step, the neural network uses the input data for calculation and analysis to obtain relevant analysis results. The analysis data is decomposed and displayed based on the representation network. In this step, the analysis data of the neural network is mapped back to the representation network to visualize the analysis results. Finally, the current updated data is obtained through the display and prediction of the representation network. This step visualizes the analysis results and provides the latest updated data based on the analysis and prediction results.
[0143] The multi-dimensional attributes and current life cycle of the target building are input into the second neural network structure to obtain analysis data of the second neural network structure, the analysis data is decomposed based on the representation network display, and the current update data is obtained, including:
[0144] S431, based on the current life cycle selection cycle corresponding to the second neural network structure, the multi-dimensional attributes of the target building are input into the second neural network structure to obtain analysis data of the second neural network structure.
[0145] This solution will select a second neural network structure suitable for the current life cycle. This step ensures that the selected neural network can effectively process the building data of the current cycle. The multi-dimensional attributes of the target building are input into the selected second neural network structure for analysis
[0146] The second neural network structure corresponding to the cycle is selected based on the current life cycle, and the multi-dimensional attributes of the target building are input into the second neural network structure to obtain the analysis data of the second neural network structure, including:
[0147] Inputting the multi-dimensional attributes of the target building into the second neural network structure;
[0148] If the neuron is determined to be a neuron for calculating energy usage information, the corresponding neuron is regarded as a neuron with energy usage information;
[0149] If the neuron is determined to be a neuron that does not compute energy-consuming information, it is determined to be a neuron that does not need to be added to the representation network.
[0150] It is understandable that some neurons are used for data classification, and some neurons are used for calculating energy information. This solution will determine that neurons are neurons that calculate energy information, and treat these neurons as neurons with energy information. Neurons are determined to be neurons that do not calculate energy information, and these neurons are excluded from the representation network.
[0151] S432, extracting energy usage information of corresponding neurons in the analysis data, and displaying and processing the energy usage information based on corresponding representation grids.
[0152] This solution extracts the energy usage information of the corresponding neurons from the analysis data. The extracted energy usage information is displayed based on the corresponding representation grid. This step ensures that the energy usage information is visualized on the representation network, allowing users to intuitively view and analyze the energy usage information.
[0153] S433, extracting information of neurons in the output layer from the analysis data to obtain current update data.
[0154] Finally, the information of the output layer neurons is extracted from the analyzed data, and the currently updated data is obtained based on the information of the output layer neurons. This step provides the latest update information by parsing and processing the output layer data.
[0155] The present invention also provides a storage medium, in which a computer program is stored. When the computer program is executed by a processor, it is used to implement the methods provided by the various embodiments described above.
[0156] Among them, the storage medium can be a computer storage medium or a communication medium. The communication medium includes any medium that facilitates the transmission of a computer program from one place to another. The computer storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer. For example, the storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application-specific integrated circuit (Application Specific Integrated Circuits, referred to as: ASIC). In addition, the ASIC can be located in a user device. Of course, the processor and the storage medium can also exist in a communication device as discrete components. The storage medium can be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.
[0157] The present invention also provides a program product, which includes an execution instruction, which is stored in a storage medium. At least one processor of a device can read the execution instruction from the storage medium, and at least one processor executes the execution instruction so that the device implements the methods provided in the above various embodiments.
[0158] In the above-mentioned terminal or server embodiments, it should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly implemented as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor.
[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A multi-dimensional attribute coupling full life cycle building energy characterization and prediction update method, characterized by: include: Acquire multi-dimensional attributes of the target building over its entire life cycle, wherein the multi-dimensional attributes include multiple geometric characteristic attributes, environmental characteristic attributes, functional characteristic attributes, and life cycle characteristic attributes; Classify the target building according to its multi-dimensional attributes throughout its life cycle, and obtain multi-dimensional attribute nodes and node connection labels under each ontology attribute; Based on the user's adjustment information of any multi-dimensional attribute node and node connection label, delete or modify the multi-dimensional attribute node and node connection label; If it is determined that the user adds any multi-dimensional attribute nodes and node connection labels, the newly added multi-dimensional attribute nodes and node connection labels are saved and processed; Based on deep learning neural network, the mapping relationship between building energy consumption and multi-dimensional attributes is analyzed and processed, a neural network with multiple neurons is constructed and the connection relationship between neurons is established; The gradient parameters of the neural network are trained based on pre-training data of different life cycles to obtain the gradient parameters corresponding to each neuron under different life cycles; Constructing a representation network corresponding to the neural network, obtaining multi-dimensional attributes of the target building in real time, and obtaining current update data based on the display and prediction of the life cycle of the representation network; The deep learning-based neural network analyzes and processes the mapping association relationship between building energy consumption and multi-dimensional attributes, constructs a neural network with multiple neurons and establishes a connection relationship between neurons, including: Constructing neurons corresponding to each multi-dimensional attribute node, connecting corresponding neurons based on node connection labels, and obtaining a corresponding first neural network; Determine the corresponding relationship information between the building's main state and the building energy consumption corresponding to each main state at each neuron; Based on the connection relationship of the multi-dimensional attribute nodes, the neurons corresponding to the corresponding input layer and output layer are constructed, and the newly constructed neurons are connected to the first neural network to establish a second neural network; The training process of the gradient parameters of the neural network based on the pre-training data of different life cycles to obtain the gradient parameters corresponding to each neuron under different life cycles includes: Obtain buildings of the same type corresponding to the target building with different life cycles, extract the multi-dimensional attributes of the same type of buildings configured by the user, and the energy consumption information corresponding to each ontology attribute; Based on the multi-dimensional attributes of each building of the same type, the first neural network is adjusted accordingly so that the neurons in the first neural network correspond to the multi-dimensional attributes of each building of the same type, thereby obtaining a second neural network to be trained; Input the ontology training information and energy training information of the ontology attributes of the same type of buildings in the corresponding life cycle into the second neural network to be trained, and set up training accordingly to obtain the sub-gradient parameters corresponding to each neuron under different life cycles; The sub-gradient parameters of all buildings of the same type in the same period are merged to obtain the total gradient parameters; The sub-gradient parameters of all buildings of the same type in the same period are fused to obtain the total gradient parameters, including: Obtain the sub-gradient parameters of all buildings of the same type under each neuron in the same period, and classify the neurons of the same multi-dimensional attribute nodes to obtain the corresponding classification set; Obtaining the number of neurons deleted and adjusted in the second neural network corresponding to each neuron in the classification set, and the level of the deleted and adjusted neurons; Based on the number of neurons deleted and adjusted in the second neural network and the levels of neurons, the gradient coefficients of the corresponding neurons in each classification set are calculated comprehensively, and the total gradient parameters are obtained by fusion calculation of the sub-gradient parameters based on the gradient coefficients.
2. The method according to claim 1, characterized in that: The determining of the corresponding relationship information between the main state of the building and the building energy consumption corresponding to each main state at each neuron includes: Determine all possible ontological states of the multidimensional attribute nodes corresponding to each neuron in the first neural network, each multidimensional attribute node having a preset number of ontological states; Construct corresponding relationship information between each entity state and building energy consumption, wherein the corresponding relationship information at least includes a calculation algorithm.
3. The method according to claim 1, characterized in that The method constructs neurons corresponding to the input layer and the output layer based on the connection relationship of the multi-dimensional attribute nodes, and connects the newly constructed neurons to the first neural network to establish a second neural network, including: Construct neurons corresponding to the input layer and output layer based on the user's configuration; Configure the corresponding input algorithm and output algorithm for each neuron in the input layer and output layer respectively; If it is determined that the neurons in the input layer and the output layer have a connection relationship with the neurons in the first neural network, the neurons in the input layer and the output layer are connected to the neurons in the first neural network; If the user's input configuration information and output configuration information for the neurons in the input layer and the output layer are received, the neurons in the input layer and the output layer are connected to the neurons in the first neural network based on the input configuration information and the output configuration information to obtain a second neural network.
4. The method according to claim 1, characterized in that: The method of comprehensively calculating the number of neurons deleted and adjusted in the second neural network and the levels of the neurons to obtain the gradient coefficient of the corresponding neurons in each classification set, and fusing the sub-gradient parameters based on the gradient coefficient to obtain the total gradient parameter includes: The corresponding quantity gradient offset sub-coefficient is calculated based on the quantity of the deleted and adjusted neurons and the quantity normalization value; Determine the level offset value of each neuron based on the level of the neuron, the level of each neuron has a preset level offset value, and calculate the corresponding level gradient offset sub-coefficient based on the sum of the level offset values and the level normalization value; The preset reference weight is deducted from the gradient coefficient of each neuron of the quantity gradient offset sub-coefficient and the level gradient sub-coefficient, and the corresponding sub-gradient parameters are fused and calculated based on each gradient coefficient to obtain the total gradient parameter.
5. The method according to claim 4, characterized in that The method of subtracting the gradient coefficient of each neuron of the quantity gradient offset sub-coefficient and the level gradient sub-coefficient from the preset reference weight and fusing the corresponding sub-gradient parameters based on each gradient coefficient to obtain the total gradient parameter comprises: The gradient coefficient of each neuron is calculated by the following formula, ， in, For the classification set The gradient coefficient of each neuron, For the The number of neurons deleted and adjusted in the second neural network corresponds to the number of neurons deleted and adjusted in the second neural network. is the normalized value of quantity, For the The neurons in the second neural network correspond to The level offset value of the deleted adjustment neuron, For the The upper limit of the number of neurons deleted and adjusted in the second neural network corresponding to the neurons, is the level normalized value; Based on the fusion mean calculation of the corresponding sub-gradient parameters for each gradient coefficient in the classification set, the total gradient parameter corresponding to the classification set is obtained, which is calculated by the following formula: ， in, For the The total gradient parameter corresponding to the classification set, For the classification set The gradient coefficient of each neuron, For the classification set The sub-gradient parameters of neurons, For the The upper limit of the neurons in the classification set, For the classification set The gradient coefficient of a neuron.
6. The method according to claim 1, characterized in that The constructing of a representation network corresponding to the neural network, obtaining multi-dimensional attributes of the target building in real time, and obtaining current update data based on the current life cycle display and prediction of the representation network include: Mapping the neurons of the second neural network structure of the target building to obtain the representation grid mapped by each neuron, and determining the positional relationship of the representation grid based on the positional relationship of the neurons; Obtain the ontological attributes of the neuron corresponding to each representation grid and generate corresponding attribute tags and store them in correspondence with the representation grid to obtain a representation network; The multi-dimensional attributes and current life cycle of the target building are input into the second neural network structure to obtain analysis data of the second neural network structure, the analysis data is decomposed based on the representation network display, and the current update data is obtained.
7. The method according to claim 6, characterized in that The multi-dimensional attributes and current life cycle of the target building are input into the second neural network structure to obtain analysis data of the second neural network structure, decompose the analysis data based on the representation network display, and obtain current update data, including: Based on the current life cycle, a second neural network structure corresponding to the cycle is selected, and the multi-dimensional attributes of the target building are input into the second neural network structure to obtain analysis data of the second neural network structure; Extracting energy usage information of corresponding neurons in the analysis data, and displaying and processing the energy usage information based on corresponding representation grids; The information of the neurons in the output layer of the analysis data is extracted to obtain the current update data.
8. The method according to claim 7, characterized in that The second neural network structure corresponding to the cycle is selected based on the current life cycle, and the multi-dimensional attributes of the target building are input into the second neural network structure to obtain the analysis data of the second neural network structure, including: Inputting the multi-dimensional attributes of the target building into the second neural network structure; If the neuron is determined to be a neuron for calculating energy usage information, the corresponding neuron is regarded as a neuron with energy usage information; If the neuron is determined to be a neuron that does not compute energy-consuming information, it is determined to be a neuron that does not need to be added to the representation network.