Task result prediction model construction method and device, electronic equipment and storage medium

Abstract

The application discloses a kind of construction methods of task result prediction model, device, electronic equipment and storage medium.The method comprises the following steps: determining the task result prediction model to be trained, determining the structure data of at least two kinds of objects, the structure data of at least two kinds of objects is divided into the reference structure data of at least one reference task, and the task result corresponding to each reference task is determined;According to the reference structure data of at least one reference task, the task result prediction model to be trained is trained to obtain the task prediction result of at least one reference task, and the accuracy of the trained task result prediction model to be trained is determined based on the task prediction result corresponding to each reference task and task result, the trained task result prediction model to be trained with accuracy reaching preset value is used as task result prediction model.The technical scheme of the present application solves the problem that model cannot accurately simulate physical data, and improves the accuracy of model prediction result.

Description

Technical Field

[0001] This invention relates to the field of building simulation, and more particularly to a method, apparatus, electronic device, and storage medium for constructing a task outcome prediction model. Background Technology

[0002] In the field of building simulation, common methods include physical modeling, statistical modeling, and machine learning-based modeling. Compared to physical and statistical modeling, machine learning methods have unique advantages, capable of simulating the behavior of building systems by learning complex patterns in data, and exhibiting stronger generalization ability. However, current machine learning methods build mathematical models based on physical models. Due to varying degrees of simplification of the models, their understanding of physical data is limited, restricting the types of input data and thus affecting the accuracy of the models. Summary of the Invention

[0003] This invention provides a method, apparatus, electronic device, and storage medium for constructing a task outcome prediction model, in order to solve the problem that the model cannot accurately simulate physical data and improve the accuracy of the model's prediction results.

[0004] According to one aspect of the present invention, a method for constructing a task outcome prediction model is provided, the method comprising:

[0005] A task result prediction model to be trained is determined. The task result prediction model includes an encoding layer and at least one task layer, with the task layer located after the encoding layer. The encoding layer is used to encode structural data and output object feature encoding. The task layer is used to calculate the object feature encoding to obtain the prediction result of the corresponding task. A task includes at least one type of object.

[0006] The structural data of at least two types of objects is determined, and the structural data of the at least two types of objects is divided into reference structural data for at least one reference task. The task result corresponding to each reference task is determined. The structural data is transformed from the actual data and simulated data corresponding to each object according to the data structure of the task result prediction model to be trained. The actual data is data containing building-related physical meaning. The reference task is the task corresponding to the training of the task layer in the task result prediction model to be trained. The reference task includes at least one type of object.

[0007] Based on the reference structure data of at least one reference task, the task result prediction model to be trained is trained to obtain the task prediction result of at least one reference task. Based on the task prediction result and the task result corresponding to each reference task, the accuracy of the trained task result prediction model is determined. The trained task result prediction model with an accuracy reaching a preset value is used as the task result prediction model.

[0008] According to another aspect of the present invention, an apparatus for constructing a task outcome prediction model is provided, the apparatus comprising:

[0009] The model determination module is used to determine the prediction model for the result of the task to be trained. The prediction model for the result of the task to be trained includes an encoding layer and at least one task layer, and the task layer is located after the encoding layer. The encoding layer is used to encode the structural data and output the object feature encoding. The task layer is used to calculate the object feature encoding to obtain the prediction result of the corresponding task. A task includes at least one type of object.

[0010] A data determination module is used to determine the structural data of at least two types of objects, divide the structural data of the at least two types of objects into reference structural data of at least one reference task, and determine the task result corresponding to each reference task; the structural data is transformed from the actual data and simulated data corresponding to each object according to the data structure of the task result prediction model to be trained; the actual data is data containing building-related physical meaning; the reference task is the task corresponding to the training of the task layer in the task result prediction model to be trained; the reference task includes at least one type of object.

[0011] The training module is used to train the task result prediction model to be trained based on the reference structure data of at least one reference task to obtain the task prediction result of at least one reference task. Based on the task prediction result and the task result corresponding to each reference task, the accuracy of the trained task result prediction model is determined. The trained task result prediction model with an accuracy reaching a preset value is used as the task result prediction model.

[0012] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0013] At least one processor; and

[0014] A memory communicatively connected to the at least one processor; wherein,

[0015] The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to execute the task result prediction model construction method according to any embodiment of the present invention.

[0016] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the method for constructing a task result prediction model according to any embodiment of the present invention.

[0017] The technical solution of this invention involves determining a task result prediction model to be trained, determining structural data for at least two types of objects, dividing the structural data of the at least two types of objects into reference structural data for at least one reference task, and determining the task result corresponding to each reference task. This invention divides the structural data into reference structural data corresponding to different reference tasks, facilitating subsequent data analysis and training. Furthermore, based on the reference structural data of at least one reference task, the task result prediction model to be trained is trained to obtain task prediction results for at least one reference task. Based on the task prediction results corresponding to each reference task and the task results, the accuracy of the trained task result prediction model is determined. The trained task result prediction model with an accuracy reaching a preset value is used as the task result prediction model. The structural data of this invention is derived from the actual data and simulated data corresponding to each object based on the data structure of the task result prediction model to be trained. The actual data contains data with physical meaning related to architecture. Therefore, the task result prediction model obtained by training the task result prediction model using structural data can accurately predict physical data, solving the problem that the model cannot accurately simulate physical data and improving the accuracy of the model's prediction results.

[0018] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a method for constructing a task result prediction model according to an embodiment of the present invention;

[0021] Figure 2 This is a flowchart of another method for constructing a task result prediction model according to an embodiment of the present invention;

[0022] Figure 3 This is an architecture diagram of a task result prediction model applicable to embodiments of the present invention;

[0023] Figure 4 This is a schematic diagram of a device for constructing a task result prediction model according to an embodiment of the present invention;

[0024] Figure 5This is a schematic diagram of the structure of an electronic device that implements the task result prediction model construction method of the present invention according to an embodiment of the present invention. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "reference" and others in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] Example 1

[0028] Figure 1 This is a flowchart of a method for constructing a task result prediction model according to an embodiment of the present invention. This embodiment is applicable to the construction of a task result prediction model for task result prediction. The method can be executed by a task result prediction model construction device, which can be implemented in hardware and / or software and can be configured in any electronic device with network communication function.

[0029] like Figure 1 As shown, the method for constructing the task result prediction model of the present invention includes the following steps:

[0030] S110. Determine the prediction model for the task results to be trained.

[0031] The training task result prediction model includes an encoding layer and at least one task layer, with the task layer located after the encoding layer. The encoding layer is used to encode the structural data and output object feature encoding. The task layer is used to calculate the object feature encoding to obtain the prediction result of the corresponding task. A task includes at least one type of object.

[0032] Specifically, the task result prediction model to be trained is a neural network model. The combination of the encoding layer and at least one task layer is used to construct the task result prediction model to be trained. One task layer corresponds to one task. The task can be understood as a task related to the building that reflects the changing state of objects in the project. For example, the target task can be a task such as predicting building energy consumption, predicting cooling load, HVAC control actions, system equipment fault diagnosis (classification), and the impact of equipment operation and maintenance frequency on energy consumption prediction.

[0033] S120. Determine the structural data of at least two types of objects, divide the structural data of at least two types of objects into reference structural data of at least one reference task, and determine the task result corresponding to each reference task.

[0034] The structured data is derived from the actual and simulated data corresponding to each object, based on the data structure of the prediction model for the task results to be trained. The actual data contains data with physical meaning related to the building. The structured data includes various conditions, strategies, and working conditions, covering various scenarios, so that the task result prediction model obtained after training the prediction model for the task results using the structured data can more accurately understand the physical world.

[0035] The reference task is the training task corresponding to the task layer in the prediction model for the task result to be trained, and the reference task includes at least one type of object. The task result can be understood as data reflecting the result of task execution.

[0036] Specifically, the process involves acquiring actual and simulated data for at least two types of objects, and then transforming this data into structured data for at least two types of objects based on data transformation relationships. These data transformation relationships are the correspondences between actual and simulated data and structured data, used to represent the data in the structural form of a prediction model for the task to be trained, without affecting the physical nature of the data.

[0037] The model for predicting the results of the task to be trained contains multiple task layers, and each task layer corresponds to the execution of one task. Therefore, it is necessary to divide the structural data of at least two types of objects into reference structural data of at least one reference task to facilitate model training. Furthermore, each reference task will correspond to an actual task result. Determining the task result corresponding to each reference task will facilitate the subsequent adjustment of model training parameters.

[0038] As an optional but non-limiting approach, determining the structural data of at least two types of objects may include the following steps A1-A4:

[0039] Step A1: Obtain multi-dimensional data for each type of object, including actual data and simulated data.

[0040] Specifically, simulated data can be understood as the simulation data of various simulation software related to each type of object for the target task; actual data can be understood as the data of the actual project related to each type of object, including physically meaningful data of interest to each type of object, and the physically meaningful data is of great significance to the implementation of tasks of different combinations of object types. Therefore, the model training process needs to be able to analyze the impact of relevant physical data well.

[0041] Step A2: Extract the static and dynamic data corresponding to each object from the multi-dimensional data of each object.

[0042] Static data refers to data that does not change over time. Dynamic data refers to data that changes over time. Dynamic data can be understood as data that changes over time; it is a dynamic process that effectively reflects the changes in each type of object.

[0043] Step A3: Based on the data structure of the prediction model for the training task results, integrate the static and dynamic data corresponding to each type of object to determine the target relation data, target static data, and target dynamic data corresponding to each type of object. The relation data is used to describe the degree of association between different types of objects that are related.

[0044] Specifically, a data structure transformation table is determined for different types of objects. Based on this table, the static and dynamic data for each type of object are transformed to obtain the target static and target dynamic data for each object type. A relationship list for different types of objects is then determined. Based on this relationship list, the static and dynamic data for each type of object, the target relationship data for each object type is determined. This process, by using data structure transformation tables and relationship lists to transform the static and dynamic data for each type of object, obtains the target static, target dynamic, and target relationship data for each object type. This ensures that the data contains sufficient physical data and that the data can be well integrated with the prediction model for the training task.

[0045] Among them, the data structure transformation table establishes the data structure transformation relationship between different types of objects and their corresponding static and dynamic data according to the data structure in the task result prediction model.

[0046] The relationship list describes different types of objects that are related. For example, it can be used to calculate spatial connectivity based on coordinate data or to calculate connectivity between spaces based on the presence or absence of doors or windows. The relationship data describes the degree of association between different types of objects, including changes over time.

[0047] Step A4: Use the target static data, target dynamic data, and target relation data as the structure data for each type of object.

[0048] This embodiment's technical solution acquires multi-dimensional data for each type of object, and extracts static and dynamic data corresponding to each type of object from the multi-dimensional data. Static data is data that does not change over time, while dynamic data is data that changes over time. This embodiment's extraction of static and dynamic data can effectively distinguish the trend of data change, facilitating rapid subsequent data processing. Furthermore, based on the data structure of the task result prediction model to be trained, the static and dynamic data corresponding to each type of object are integrated to determine the target relationship data, target static data, and target dynamic data for each type of object. The relationship data is used to describe the degree of association between different types of objects that are related. The target static data, target dynamic data, and target relationship data are used as the structural data for each type of object. This invention transforms the data into data that conforms to the processing of the task result prediction model, which can ensure the accuracy of the task result prediction model's prediction.

[0049] S130. Based on the reference structure data of at least one reference task, train the prediction model for the result of the task to be trained to obtain the task prediction result of at least one reference task. Based on the task prediction result and the task result corresponding to each reference task, determine the accuracy of the prediction model for the result of the task to be trained after training. Use the prediction model for the result of the task to be trained after training with an accuracy that reaches a preset value as the prediction model for the result of the task.

[0050] The task prediction result can be understood as the prediction result of the reference task obtained by the task result prediction model to be trained by processing the reference structure data of the reference task. It can be used to compare with the task result corresponding to the reference task to determine the accuracy of the task result prediction model to be trained.

[0051] Specifically, at least one reference structure data for a reference task is obtained. This reference structure data is then input into the task result prediction model to be trained, yielding task prediction results for at least one reference task. Only when the task prediction results are nearly identical to the actual task results can the accuracy of the task result prediction model be guaranteed. The similarity values ​​between the task prediction results corresponding to the reference tasks and the actual task results can be compared. The higher the similarity value, the higher the accuracy of the task result prediction model. A target correlation relationship between similarity value and accuracy is established. Based on this target correlation relationship, the accuracy of the trained task result prediction model is determined. Furthermore, the trained task result prediction model with an accuracy reaching a preset value is used as the task result prediction model.

[0052] The technical solution of this invention involves determining a task result prediction model to be trained, determining structural data for at least two types of objects, dividing the structural data of the at least two types of objects into reference structural data for at least one reference task, and determining the task result corresponding to each reference task. This invention divides the structural data into reference structural data corresponding to different reference tasks, facilitating subsequent data analysis and training. Furthermore, based on the reference structural data of at least one reference task, the task result prediction model to be trained is trained to obtain task prediction results for at least one reference task. Based on the task prediction results corresponding to each reference task and the task results, the accuracy of the trained task result prediction model is determined. The trained task result prediction model with an accuracy reaching a preset value is used as the task result prediction model. The structural data of this invention is derived from the actual data and simulated data corresponding to each object based on the data structure of the task result prediction model to be trained. The actual data contains data with physical meaning related to architecture. Therefore, the task result prediction model obtained by training the task result prediction model using structural data can accurately predict physical data, solving the problem that the model cannot accurately simulate physical data and improving the accuracy of the model's prediction results.

[0053] Example 2

[0054] Figure 2 This is a flowchart of another method for constructing a task result prediction model provided by an embodiment of the present invention. The technical solution of this embodiment further optimizes the process of S130 in the above embodiment based on the above embodiment. This embodiment can be combined with various optional solutions in one or more of the above embodiments.

[0055] like Figure 2 As shown, the method for constructing the task result prediction model of the present invention includes the following steps:

[0056] S210. Determine the prediction model for the task result to be trained, determine the structural data of at least two types of objects, divide the structural data of at least two types of objects into reference structural data of at least one reference task, and determine the task result corresponding to each reference task.

[0057] The training task result prediction model includes an encoding layer and at least one task layer, with the task layer located after the encoding layer. The encoding layer is used to encode the structural data and output object feature encoding. The task layer is used to calculate the object feature encoding to obtain the prediction result of the corresponding task. A task includes at least one type of object.

[0058] Among them, the structural data is transformed from the actual data and simulated data corresponding to each object based on the data structure of the prediction model of the result of the task to be trained; the actual data is data containing physical meaning related to the building; the reference task is the task corresponding to the training in the task layer of the prediction model of the result of the task to be trained; the reference task includes at least one type of object.

[0059] The encoding layer in the prediction model for the task to be trained includes an information encoding layer, a graph encoding layer, and a time-series encoding layer. The encoding layers perform encoding operations on the data sequentially, following the order of information encoding, graph encoding, and time-series encoding. The information encoding layer encodes data at different times, the graph encoding layer encodes data based on relationships, and the time-series encoding layer encodes data within a time period; for example... Figure 3 As shown, static variables are the target static data of this invention, dynamic variables are the target dynamic data of this invention, heterogeneous graph relationships are the target relationship data of this invention, static information encoding and dynamic information encoding correspond to form the information encoding layer of this invention, heterogeneous graph encoding corresponds to the graph encoding layer of this invention, and time series encoding corresponds to the time series encoding layer of this invention.

[0060] Each object type corresponds to an information encoding layer. Each object type corresponds to a time-series encoding layer. All objects correspond to a graph encoding layer, which ensures encoding consistency and efficiency. The graph encoding layer is constructed according to the relationships between different object types, and the relationship data is used to describe the degree of association between different object types that are related.

[0061] The information encoding layer can be constructed using one of the following: Recurrent Neural Network (RNN), Convolutional Neural Network (CNN), Fully Connected Layer (Linear), and Gated Recurrent Unit (GRU). The graph encoding layer can be constructed using one of the following: Graph Convolutional Layer (GCN and CNN), Graph Pooling Layer, Attention Layer (attention, transformer, GAT), Homogeneous Relationship, and Heterogeneous Relationship (intra-inter). The time series encoding layer can be constructed using one of the following: Recurrent Neural Network (RNN), Convolutional Neural Network (CNN), and Attention Mechanism (attention, transformer).

[0062] The encoding layer of the task result prediction model of this invention includes an information encoding layer, a graph encoding layer, and a time series encoding layer. This allows for encoding of structural data at three levels, making the final object feature encoding more reflective of the physical meaning inherent in the structural data itself. Different types of neural networks play different roles in the encoding process, extracting key information from the data based on their respective structural characteristics. Within the overall architecture of the task result prediction model, these neural networks can be flexibly combined according to the needs of the task.

[0063] S220. The information encoding layer in the prediction model for the result of the task to be trained encodes the first type of data of the reference task to obtain the first object feature encoding corresponding to different times of the reference task.

[0064] The first type of data consists of the target static data and target dynamic data in the reference structure data of the reference task. The first object feature encoding includes the object feature encoding of each type of object in the reference task.

[0065] Specifically, the information encoding layer is the basic encoding of structured data, that is, it transforms the structured data into a specified dimensional structure compatible with the actual training and validation processes. The specified dimension is generally high-dimensional; for example, if a 64-dimensional structure is required, the structured data is transformed into a 64-dimensional object feature encoding form. Each object class corresponds to one information encoding layer. The reference task contains at least one object class, and each object class has corresponding first-class data. The first-class data of the reference task is input into the prediction model for the training task result. Then, the first-class data corresponding to each object class can be assigned to the corresponding information encoding layer. The information encoding layer encodes both the target static data and the target dynamic data to output the first object feature encoding corresponding to different times of the reference task. The encoding of the target dynamic data involves encoding the data at each time step. Furthermore, the target dynamic data includes data from historical times and data from future times. Encoding the historical data in the target dynamic data yields multiple first-class encodings corresponding to historical times, and encoding the future data in the target dynamic data yields multiple second-class encodings corresponding to future times. The first and second encoding parameters are not shared to ensure encoding accuracy.

[0066] As an optional but non-limiting embodiment, before the information encoding layer in the prediction model for the result of the task to be trained encodes the first type of data of the reference task, the first type of data can be filtered. The specific process includes:

[0067] The first category of data for the reference task is filtered according to the importance of its data features. Data whose importance reaches a preset value is selected as the first category of filtered data. This allows the information encoding layer in the prediction model for the task to be trained to encode the first category of filtered data from the reference task. This filtering of the first data in this embodiment reduces irrelevant noise input and ensures data accuracy.

[0068] S230. Based on the target relationship data in the reference structure data of the reference task, the graph coding layer in the prediction model of the result of the task to be trained encodes the first object feature code to obtain the second object feature code corresponding to different times of the reference task.

[0069] Specifically, the graph coding layer encodes the relationships between structural data. Different types of objects have different degrees of correlation. In order to accurately predict the task, it is necessary to reflect the relationship between the first object feature codes corresponding to each type of target object at different times and share them in the same graph coding layer. Therefore, all objects correspond to one graph coding layer, which can ensure the consistency and efficiency of the encoding. The target relationship data in the reference structural data of the reference task is introduced. By using the target relationship data and the first object feature codes of each type of object in the reference task, the prediction model of the result of the training task is encoded and trained to obtain the second object feature codes corresponding to different times of the reference task, which are used for encoding in the time series coding layer.

[0070] S240. The time series encoding layer in the prediction model for the result of the task to be trained encodes the second object feature code within a preset time period to obtain the third object feature code corresponding to the reference task within the preset time period.

[0071] The preset time period is the time period between the current moment and the preset time after the current moment;

[0072] The time series encoding layer can be understood as a time series model. Existing time series models can usually only process one or two types of data and cannot integrate all types of data at the same time. This limits the model's ability to explore the relationships between variables and graph structures, thus affecting the prediction effect. However, the time series encoding layer of this invention can reflect the influence between structural data of different types of objects in different time periods and the degree of influence on the prediction results, which can greatly improve the accuracy of model prediction.

[0073] Specifically, each type of object corresponds to a time series coding layer. The reference task contains at least one type of object. The time series coding layer can integrate the structural data of each type of object in the reference task with the second object feature code to obtain the third object feature code of the reference task in a preset time period, which can be used for task prediction in subsequent task layers.

[0074] S250. The task layer in the prediction model for the result of the task to be trained calculates the feature encoding of the third object to obtain the task prediction result of the reference task.

[0075] Specifically, different tasks correspond to corresponding task layers. The third object feature encoding is then transmitted to the task layer corresponding to the reference task in the task result prediction model to be trained. The task prediction result of the reference task is obtained by calculating the third object feature encoding.

[0076] S260. Based on the task prediction results and task results corresponding to each reference task, determine the accuracy of the trained task result prediction model, and use the trained task result prediction model with an accuracy reaching a preset value as the task result prediction model.

[0077] Specifically, once the task result prediction model is obtained, it will be put into use. However, during long-term use, the performance of the task result prediction model may degrade. Therefore, it is possible to periodically obtain structural data of historical time periods before the current moment and use it as training data to update the task result prediction model and ensure its accuracy.

[0078] As an optional but non-limiting embodiment, after using the trained task result prediction model with a preset accuracy as the task result prediction model, the method further includes:

[0079] If the task layer of the task result prediction model is updated, the encoding layer of the task result prediction model is frozen, the updated data is obtained, and the task layer of the task result prediction model is updated based on the updated data. The update includes fine-tuning the parameters of the original task layer, replacing the original task layer, or adding a new task layer. The updated data includes the structural data and task results of a specific task.

[0080] The solution in this embodiment updates the task layer by freezing the encoding layer of the task result prediction model, which can save more time and resources, and the model has stronger generalization ability and flexibility.

[0081] The technical solution of this invention involves determining a task result prediction model to be trained, determining structural data of at least two types of objects, dividing the structural data of at least two types of objects into reference structural data for at least one reference task, and determining the task result corresponding to each reference task. The structural data of this invention includes simulated data and actual data, and contains various conditions, strategies, and working conditions, covering various scenarios and improving data richness. Further, the information encoding layer in the task result prediction model encodes the first type of data of the reference task to obtain the first object feature encoding corresponding to different times of the reference task. The first type of data consists of target static data and target dynamic data in the reference result data of the reference task. Based on the target relationship data in the reference structural data of the reference task, the graph encoding layer in the task result prediction model encodes the first object feature encoding to obtain the second object feature encoding corresponding to different times of the reference task. The time series encoding layer in the task result prediction model encodes the second object feature encoding within a preset time period to obtain the third object feature encoding corresponding to the reference task within the preset time period. The task layer in the task result prediction model calculates the third object feature encoding to obtain the reference result. The task prediction results are presented by introducing three encoding layers with different characteristics and training each layer with structured data. This allows the model to understand the physical world through various data types. Then, based on the task prediction results and actual results for each reference task, the accuracy of the trained task result prediction model is determined. The trained task result prediction model with an accuracy reaching a preset value is used as the task result prediction model. The introduction of accuracy further ensures the accuracy of the task result prediction model, guaranteeing that the task result prediction model trained with structured data can accurately predict physical data. This solves the problem that the model cannot accurately simulate physical data, improves the accuracy of the model's prediction results, and makes the model balance interpretability, robustness, and accuracy.

[0082] Example 3

[0083] Figure 4 This is a schematic diagram of a device for constructing a task result prediction model according to an embodiment of the present invention. This embodiment is applicable to the construction of a task result prediction model for task result prediction. The device for constructing the task result prediction model can be implemented in hardware and / or software, and can be configured in any electronic device with network communication capabilities. Figure 4 As shown, the apparatus for constructing the task outcome prediction model includes:

[0084] The model determination module 310 is used to determine the prediction model for the result of the task to be trained. The prediction model for the result of the task to be trained includes an encoding layer and at least one task layer, and the task layer is located after the encoding layer. The encoding layer is used to encode the structural data and output the object feature encoding. The task layer is used to calculate the object feature encoding to obtain the prediction result of the corresponding task. A task includes at least one type of object.

[0085] The data determination module 320 is used to determine the structural data of at least two types of objects, divide the structural data of the at least two types of objects into reference structural data of at least one reference task, and determine the task result corresponding to each reference task; the structural data is transformed from the actual data and simulated data corresponding to each object according to the data structure of the task result prediction model to be trained; the actual data is data containing building-related physical meaning; the reference task is the task corresponding to the training of the task layer in the task result prediction model to be trained; the reference task includes at least one type of object.

[0086] The training module 330 is used to train the task result prediction model to be trained based on the reference structure data of at least one reference task to obtain the task prediction result of at least one reference task, determine the accuracy of the trained task result prediction model based on the task prediction result and the task result corresponding to each reference task, and use the trained task result prediction model with the accuracy reaching a preset value as the task result prediction model.

[0087] Based on the above embodiments, optionally, the data determination module includes a structure data determination unit, used for:

[0088] Obtain multi-dimensional data for each type of object, including actual data and simulated data;

[0089] Extract static and dynamic data corresponding to each object from the multi-dimensional data of each object category; the static data is data that does not change over time, and the dynamic data is data that changes over time;

[0090] Based on the data structure of the prediction model for the result of the task to be trained, the static data and dynamic data corresponding to each type of object are integrated to determine the target relation data, target static data and target dynamic data corresponding to each type of object. The relation data is used to describe the degree of association between different types of objects that are related.

[0091] The target static data, the target dynamic data, and the target relationship data are used as the structure data for each type of object.

[0092] Based on the above embodiments, optionally, the encoding layer in the prediction model for the result of the task to be trained includes an information encoding layer, a graph encoding layer, and a time series encoding layer; the encoding layer performs encoding operations on the data in sequence according to the information encoding layer, the graph encoding layer, and the time series encoding layer; the information encoding layer is used to encode data at different times, the graph encoding layer is used to encode based on relational data, and the time series encoding layer is used to encode data within a time period;

[0093] Each type of object corresponds to an information encoding layer, each type of object corresponds to a time series encoding layer, and all objects correspond to a graph encoding layer. The graph encoding layer is constructed according to the association relationship between different types of objects, and the relationship data is used to describe the degree of association between different types of objects that are related.

[0094] Based on the above embodiments, optionally, the training module includes:

[0095] The first training unit is used to encode the first type of data of the reference task in the information encoding layer of the task result prediction model to be trained, and to obtain the first object feature encoding corresponding to the reference task at different times; the first type of data is the target static data and target dynamic data in the reference result data of the reference task.

[0096] The second training unit is used to encode the first object feature code based on the target relationship data in the reference structure data of the reference task and the graph coding layer in the prediction model of the result of the task to be trained, so as to obtain the second object feature code corresponding to different times of the reference task.

[0097] The third training unit is used to encode the second object feature code within a preset time period in the time series encoding layer of the task result prediction model to be trained, so as to obtain the third object feature code of the reference task in the preset time period, wherein the preset time period is the time period between the current time and the preset time after the current time.

[0098] The fourth training unit is used to calculate the feature encoding of the third object in the task layer of the task result prediction model to be trained, so as to obtain the task prediction result of the reference task.

[0099] Based on the above embodiments, optionally, the first training unit includes a data filtering unit, used for:

[0100] The first type of data for the reference task is filtered according to the importance of its data features. Data whose importance reaches a preset value is selected as the first type of data after filtering, so that the information encoding layer in the prediction model for the result of the task to be trained can encode the first type of data of the reference task after filtering.

[0101] Based on the above embodiments, optionally, the training module includes an update unit, used for:

[0102] If the task layer of the task result prediction model is updated, the encoding layer of the task result prediction model is frozen, updated data is obtained, and the task layer of the task result prediction model is updated based on the updated data; the update includes fine-tuning of the original task layer parameters, replacing the original task layer, or adding a new task layer; the updated data includes the structure data and task results of a specific task.

[0103] Based on the above embodiments, the information encoding layer may optionally be constructed using one of the following: a recurrent neural network, a convolutional neural network, a fully connected layer, and a gated recurrent unit;

[0104] The graph encoding layer is constructed using one of the following: graph convolutional layer, graph pooling layer, attention layer, homogeneous relation, and heterogeneous relation.

[0105] The time series encoding layer is constructed using one of the following: recurrent neural network, convolutional neural network, or attention mechanism.

[0106] The task result prediction model construction apparatus provided in the embodiments of the present invention can execute the task result prediction model construction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0107] Example 4

[0108] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0109] Figure 5 A schematic diagram of an electronic device is shown, which can be used to implement the method for constructing a task outcome prediction model according to embodiments of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0110] like Figure 5As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 may also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0111] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0112] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as methods for constructing task outcome prediction models.

[0113] In some embodiments, the method for constructing a task outcome prediction model may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the method for constructing a task outcome prediction model described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the method for constructing a task outcome prediction model by any other suitable means (e.g., by means of firmware).

[0114] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0115] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0116] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0117] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0118] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0119] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0120] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and no limitation is imposed herein.

[0121] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for constructing a task outcome prediction model, characterized in that, For building simulation, the method includes: A task result prediction model to be trained is determined. The task result prediction model includes an encoding layer and at least one task layer, with the task layer located after the encoding layer. The encoding layer is used to encode structural data and output object feature encoding. The task layer is used to calculate the object feature encoding to obtain the prediction result of the corresponding task. A task includes at least one type of object. The structural data of at least two types of objects is determined, and the structural data of the at least two types of objects is divided into reference structural data for at least one reference task. The task result corresponding to each reference task is determined. The structural data is transformed from the actual data and simulated data corresponding to each object according to the data structure of the task result prediction model to be trained. The actual data is data containing building-related physical meaning. The reference task is the task corresponding to the training of the task layer in the task result prediction model to be trained. The reference task includes at least one type of object. Based on the reference structure data of at least one reference task, the task result prediction model to be trained is trained to obtain the task prediction result of at least one reference task. Based on the task prediction result and the task result corresponding to each reference task, the accuracy of the trained task result prediction model is determined. The trained task result prediction model with an accuracy reaching a preset value is used as the task result prediction model. Specifically, the structural data of at least two types of objects is determined, including: Obtain multi-dimensional data for each type of object, including actual data and simulated data; Extract static and dynamic data corresponding to each object from the multi-dimensional data of each object category; the static data is data that does not change over time, and the dynamic data is data that changes over time; Based on the data structure of the prediction model for the result of the task to be trained, the static data and dynamic data corresponding to each type of object are integrated to determine the target relation data, target static data and target dynamic data corresponding to each type of object. The relation data is used to describe the degree of association between different types of objects that are related. The target static data, the target dynamic data, and the target relationship data are used as the structural data for each type of object; The encoding layer in the prediction model for the task to be trained includes an information encoding layer, a graph encoding layer, and a time-series encoding layer. The encoding layer performs encoding operations on the data sequentially according to the information encoding layer, the graph encoding layer, and the time-series encoding layer. The information encoding layer is used to encode data at different times, the graph encoding layer is used to encode data based on relationships, and the time-series encoding layer is used to encode data within a time period. The information encoding layer includes static information encoding and dynamic information encoding. Each type of object corresponds to an information encoding layer, each type of object corresponds to a time series encoding layer, and all objects correspond to a graph encoding layer. The graph encoding layer is constructed according to the association relationship between different types of objects, and the relationship data is used to describe the degree of association between different types of objects that are related.

2. The method according to claim 1, characterized in that, Based on reference structure data of at least one reference task, the task result prediction model to be trained is trained to obtain task prediction results for at least one reference task, including: The information encoding layer in the prediction model for the result of the task to be trained encodes the first type of data of the reference task to obtain the first object feature encoding corresponding to the reference task at different times; the first type of data is the target static data and target dynamic data in the reference structure data of the reference task. Based on the target relationship data in the reference structure data of the reference task, the graph coding layer in the prediction model of the task to be trained encodes the first object feature code to obtain the second object feature code corresponding to different times of the reference task. The time series encoding layer in the prediction model for the result of the task to be trained encodes the second object feature code within a preset time period to obtain the third object feature code of the reference task in the preset time period, wherein the preset time period is the time period between the current time and a preset time after the current time. The task layer in the prediction model for the result of the task to be trained calculates the feature encoding of the third object to obtain the task prediction result of the reference task.

3. The method according to claim 2, characterized in that, Before the information encoding layer in the prediction model for the result of the task to be trained encodes the first type of data of the reference task, it includes: The first type of data for the reference task is filtered according to the importance of its data features. Data whose importance reaches a preset value is selected as the first type of data after filtering, so that the information encoding layer in the prediction model for the result of the task to be trained can encode the first type of data of the reference task after filtering.

4. The method according to claim 1, characterized in that, After using the trained task result prediction model with an accuracy reaching a preset value as the task result prediction model, the method includes: If the task layer of the task result prediction model is updated, the encoding layer of the task result prediction model is frozen, updated data is obtained, and the task layer of the task result prediction model is updated based on the updated data; the update includes fine-tuning of the original task layer parameters, replacing the original task layer, or adding a new task layer; the updated data includes the structure data and task results of a specific task.

5. The method according to claim 1, characterized in that, The information coding layer is constructed using one of the following: recurrent neural network, convolutional neural network, fully connected layer, and gated recurrent unit. The graph encoding layer is constructed using one of the following: graph convolutional layer, graph pooling layer, attention layer, homogeneous relation, and heterogeneous relation. The time series encoding layer is constructed using one of the following: recurrent neural network, convolutional neural network, or attention mechanism.

6. An apparatus for constructing a task outcome prediction model, characterized in that, For building simulation, the device includes: The model determination module is used to determine the prediction model for the result of the task to be trained. The prediction model for the result of the task to be trained includes an encoding layer and at least one task layer, and the task layer is located after the encoding layer. The encoding layer is used to encode the structural data and output the object feature encoding. The task layer is used to calculate the object feature encoding to obtain the prediction result of the corresponding task. A task includes at least one type of object. A data determination module is used to determine the structural data of at least two types of objects, divide the structural data of the at least two types of objects into reference structural data of at least one reference task, and determine the task result corresponding to each reference task; the structural data is transformed from the actual data and simulated data corresponding to each object according to the data structure of the task result prediction model to be trained; the actual data is data containing building-related physical meaning; the reference task is the task corresponding to the training of the task layer in the task result prediction model to be trained; the reference task includes at least one type of object. The training module is used to train the task result prediction model to be trained based on the reference structure data of at least one reference task to obtain the task prediction result of at least one reference task. Based on the task prediction result and the task result corresponding to each reference task, the accuracy of the trained task result prediction model is determined. The trained task result prediction model with an accuracy of a preset value is used as the task result prediction model. The data determination module includes a structure data determination unit, configured to: acquire multi-dimensional data for each type of object, the multi-dimensional data including actual data and simulated data; extract static and dynamic data corresponding to each type of object from the multi-dimensional data of each type of object; the static data is data that does not change over time, and the dynamic data is data that changes over time; based on the data structure of the prediction model for the result of the task to be trained, integrate the static and dynamic data corresponding to each type of object to determine the target relationship data, target static data, and target dynamic data corresponding to each type of object, the relationship data being used to describe the degree of association between different types of objects that are related; and use the target static data, the target dynamic data, and the target relationship data as the structure data of each type of object. The encoding layer in the prediction model for the task to be trained includes an information encoding layer, a graph encoding layer, and a time-series encoding layer. The encoding layer performs encoding operations on the data sequentially according to the information encoding layer, the graph encoding layer, and the time-series encoding layer. The information encoding layer is used to encode data at different times, the graph encoding layer is used to encode data based on relationships, and the time-series encoding layer is used to encode data within a time period. The information encoding layer includes static information encoding and dynamic information encoding. Each type of object corresponds to an information encoding layer, each type of object corresponds to a time series encoding layer, and all objects correspond to a graph encoding layer. The graph encoding layer is constructed according to the association relationship between different types of objects, and the relationship data is used to describe the degree of association between different types of objects that are related.

7. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the method for constructing a task outcome prediction model according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the method for constructing the task result prediction model according to any one of claims 1-5.

Images