Full-life-cycle commercial concrete demand prediction method and device and storage medium

Abstract

The invention provides a full-life-cycle commercial concrete demand prediction method and device and a storage medium, and belongs to the technical field of building information, and the method comprises the steps: obtaining BIM information of a building project, and converting the BIM information into a fixed dimension vector based on a first encoder; the first encoder is obtained by training a BiLSTM encoder based on a positive sample pair and a negative sample pair; determining the Euclidean distance between the fixed dimension vector of the target building project and the fixed dimension vector of the completed building project, and forming a similar project set by a plurality of completed building projects with the minimum distance; determining a context vector of the similar item set; and inputting the feature sequence corresponding to the target building project, the decoder state and the context vector into a decoder to obtain the predicted relative construction time of the component, and further determining the total commercial concrete demand. According to the method, the commercial concrete demand can be predicted in advance, so that the purpose of better arranging production and distribution of commercial concrete raw materials is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of building information technology, and in particular to a full life cycle commercial concrete demand prediction method and device and storage medium. BACKGROUND

[0002] With the development of the construction industry, the commercial concrete (commercial concrete) supply industry is also increasingly mature. In order to ensure that commercial concrete is supplied in sufficient quantity and in a timely manner according to orders, the supplier of commercial concrete needs to be able to predict the demand for commercial concrete of a construction project in advance, so as to arrange raw materials, labor, transportation and distribution in advance. At present, the overall demand for commercial concrete mainly relies on the signing of contracts in advance by the construction project and the commercial concrete supplier, while the specific demand for commercial concrete mainly relies on the reservation of the construction project party in advance, or even can only be confirmed when ordering immediately. This will cause certain obstacles to the arrangement of raw materials, labor and transportation and distribution of commercial concrete suppliers. Advance prediction of commercial concrete demand can assist commercial concrete suppliers to better arrange production and distribution work.

[0003] At present, there is a lack of methods for predicting commercial concrete demand in advance, which cannot better arrange the production and distribution of commercial concrete raw materials. SUMMARY

[0004] Therefore, it is necessary to provide a full life cycle commercial concrete demand prediction method, device and storage medium, which can realize the prediction of commercial concrete demand in advance, so as to better arrange the production and distribution of commercial concrete raw materials.

[0005] In order to solve the above problems, in a first aspect, the present application provides a full life cycle commercial concrete demand prediction method, comprising: obtaining BIM information of a construction project, and converting the BIM information into a fixed dimension vector based on a first encoder; the first encoder is trained based on a BiLSTM encoder for representing a positive sample pair with high building similarity and a negative sample pair with low building similarity; determining the Euclidean distance between the fixed dimension vector of the target construction project and the fixed dimension vector of the completed construction project, and selecting a number of completed construction projects with the smallest Euclidean distance to form a similar project set; determining the context vector of the similar project set; inputting the feature sequence corresponding to the component sequence of the target construction project, the decoder state and the context vector into the decoder to obtain the predicted component relative construction time; the decoder is trained based on the mean square error between the predicted component relative construction time and the real component relative construction time as a loss function; determining the total commercial concrete demand based on the predicted component relative construction time, the construction time of the target construction project and the component volume.

[0006] In a possible implementation, the converting the BIM information into a fixed-dimension vector based on the first encoder comprises: encoding, by the first encoder, the component sequence of the BIM information to obtain a hidden state of each component; performing global average pooling on the hidden state to obtain a fixed-dimension vector of the building project.

[0007] In a possible implementation, the determining the context vector of the similar project set comprises: fusing the component sequence corresponding to the BIM information of the similar project set and the relative time sequence between components to obtain a feature sequence of the completed building project; encoding the component sequence corresponding to the BIM information of the target building project to obtain a feature sequence of the target building project; determining a current state of the decoder based on the feature sequence of the target building project; determining an attention score between the current state of the decoder and the feature sequence of the completed building project; performing weighted summation on a part of the feature sequence of the completed building project that needs to be focused on based on the attention score to obtain the context vector.

[0008] In a possible implementation, the fusing the component sequence corresponding to the BIM information of the similar project set and the relative time sequence between components to obtain a feature sequence of the completed building project comprises: fusing the component sequence corresponding to the BIM information of the similar project set and the relative time sequence between components to obtain a new sequence; inputting the new sequence into a second encoder to obtain the feature sequence of the completed building project; the second encoder is trained based on a Transformer encoder.

[0009] In a possible implementation, the encoding the component sequence corresponding to the BIM information of the target building project to obtain a feature sequence of the target building project comprises: encoding, by a third encoder, the component sequence corresponding to the BIM information of the target building project to obtain the feature sequence of the target building project; the third encoder is trained based on a Transformer encoder.

[0010] In a possible implementation, the determining the total concrete demand based on the predicted relative construction time of the component, the construction start time of the target building project, and the component volume comprises: determining, based on the predicted relative construction time of each component and the construction start time of the target building project, a predicted absolute construction time of each component; Based on the predicted absolute construction time of each component, from the component sequence corresponding to the BIM information of the target construction project, a target component sequence satisfying a preset time requirement is selected; Based on the target component sequence and the volume of the target component, the total commercial concrete demand is determined.

[0011] In one possible implementation, the preset time requirement is satisfied:

[0012]

[0013] Among them, t c represents the timestamp at which the commercial concrete prediction demand exists, t a represents the predicted absolute construction time of one component, t b represents the predicted absolute construction time of another component, t a-1 represents t a the last time of t b-1 represents t b the last time of

[0014] In a second aspect, the present application also provides a full life cycle commercial concrete demand prediction device, comprising: An information encoding module is configured to obtain BIM information of a construction project and convert the BIM information into a fixed-dimensional vector based on a first encoder; the first encoder is trained based on a BiLSTM encoder for representing a positive sample pair with high building similarity and a negative sample pair with low building similarity; A similar project determination module is configured to determine the Euclidean distance between the fixed-dimensional vector of the target construction project and the fixed-dimensional vector of the completed construction project, and select a number of completed construction projects with the smallest Euclidean distance to form a similar project set; A context vector component module is configured to determine a context vector of the similar project set; A construction time prediction module is configured to input the feature sequence corresponding to the component sequence of the target construction project, the decoder state, and the context vector into a decoder to obtain the predicted component relative construction time; the decoder is trained based on the mean square error between the predicted component relative construction time and the real component relative construction time as a loss function; A demand determination module is configured to determine the total commercial concrete demand based on the predicted component relative construction time, the start time of the target construction project, and the component volume.

[0015] In a third aspect, the present application also provides an electronic device comprising a memory and a processor, wherein, The memory is configured to store a program. The processor is coupled to the memory and configured to execute the program stored in the memory to implement the steps of the full life cycle commercial concrete demand prediction method according to any one of the preceding aspects.

[0016] In a fourth aspect, the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the full life cycle commercial concrete demand prediction method according to any one of the preceding aspects.

[0017] The full life cycle commercial concrete demand prediction method, device and storage medium provided by the present application have the beneficial effects that: the first encoder is used to convert the BIM information of the construction project into a fixed-dimensional vector; the first encoder is trained based on a positive sample pair for representing high building similarity and a negative sample pair BiLSTM encoder for representing low building similarity; the construction project includes an unfinished construction project and a completed construction project, and the unfinished construction project is the target construction project; the encoder is trained using the positive sample pair and the negative sample pair, i.e., the training mode of the encoder is a contrast learning mode.

[0018] The encoder obtained in this way can obtain a fixed-dimensional vector from the BIM information of the construction project, determine a similar construction project set, i.e., a similar project set, based on the fixed-dimensional vector, and thus determine a context vector, which contains the context information of the input data; based on the feature sequence corresponding to the component sequence of the target construction project, the decoder state and the context vector, the relative construction time between two components can be predicted, and the total commercial concrete demand can be determined in combination with the construction start time and the construction volume, so that the total commercial concrete demand can be predicted in advance based on the BIM information of the construction project, without the need to sign a contract with the commercial concrete supplier in advance or confirm the order at the moment, thereby achieving the purpose of better arranging the production and distribution of commercial concrete raw materials. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0020] Figure 1A flow chart of one embodiment of the full life cycle commercial concrete demand prediction method provided by the present application; Figure 2 A flow chart of another embodiment of the full life cycle commercial concrete demand prediction method provided by the present application; Figure 3 A principle block diagram of one embodiment of the full life cycle commercial concrete demand prediction device provided by the present application; Figure 4 A structure schematic diagram of one embodiment of the electronic device provided by the present application. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be clearly and completely described in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0022] In the description of the embodiments of the present application, the meaning of "a plurality of" is two or more, unless otherwise specified.

[0023] In the embodiments of the present application, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, device, product or equipment comprising a series of steps or modules does not have to be limited to the clearly listed steps or modules, but can include other steps or modules that are not clearly listed or inherent to the process, method, product or equipment.

[0024] The naming or numbering of steps appearing in the embodiments of the present application does not mean that the steps in the method flow must be performed in the time / logical order indicated by the naming or numbering. The named or numbered flow steps can change the execution order according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved.

[0025] Reference to "an embodiment" in this document means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The phrase appears at various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily independent or alternative embodiments to other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0026] The present application provides a full life cycle commercial concrete demand prediction method, device and storage medium, which are described below respectively.

[0027] As Figure 1As shown, the present application provides a full life cycle commercial concrete demand prediction method, comprising: S101, obtaining BIM (Building Information Modeling) information of a building project, and converting the BIM information into a fixed dimension vector based on a first encoder; the first encoder is trained based on a BiLSTM (‌bidirectional long short-term memory network) encoder for representing a positive sample pair with high building similarity and a negative sample pair with low building similarity.

[0028] It can be understood that the building project in this step includes a target building project (i.e. a building project not yet started) and a completed building project. The BIM information can be regarded as an ordered list (sequence) of concrete components of a building project, which includes component id, type and volume. The order of the concrete components indicates the order of construction of each component in the process of constructing the entire building. The type of the component significantly distinguishes various components in the building.

[0029] High similarity means that the vector representation of the building is close in the vector space; low similarity means that the vector of the building with large difference, i.e. the vector representation of the building is far in the vector space.

[0030] S102, determining the Euclidean distance between the fixed dimension vector of the target building project and the fixed dimension vector of the completed building project, and selecting a number of completed building projects with the smallest Euclidean distance to form a similar project set.

[0031] It can be understood that the smaller the Euclidean distance, the closer the two vectors, and the more similar the corresponding building projects. The number in this embodiment refers to at least two.

[0032] S103, determining the context vector of the similar project set.

[0033] It can be understood that the context vector is a specific representation that contains context information about the input data (such as words or phrases).

[0034] S104, inputting the feature sequence corresponding to the component sequence of the target building project, the decoder state and the context vector into the decoder to obtain the predicted component relative construction time; the decoder is trained based on the mean square error between the predicted component relative construction time and the real component relative construction time as a loss function.

[0035] It can be understood that the decoder is trained based on the mean square error between the predicted component relative construction time and the real component relative construction time of the completed building project as a loss function. The decoder state , the corresponding feature of the target building project at the time step and the context vector is input into the decoder to predict the relative construction time of the current component and update the decoder state :

[0036] Finally, the predicted relative construction time sequence of each component of the target project is obtained .

[0037] S105, based on the predicted relative construction time of the component, the start time of the target building project and the component volume, determine the total commercial concrete demand.

[0038] It can be understood that based on the predicted relative construction time of each component, the start time of the target building project, the timestamp of each component can be obtained, and based on the timestamp of each component and the corresponding volume, the total commercial concrete demand can be obtained.

[0039] The method provided by the application can be applied to a terminal or a server, and is executed by an application program on the terminal or the server. The terminal can be a mobile phone or a computer, and the server can be an edge server or a cloud server.

[0040] The purpose of the application is to dynamically predict the total commercial concrete demand of an unstarted building project, i.e. a target building project, according to the BIM information of a historical completed building project and the construction time of each component corresponding to the BIM information. In the case of only knowing the BIM information of the building project, the construction time of each concrete component in the BIM of the target building project is predicted, and the corresponding commercial concrete demand is generated.

[0041] In some embodiments, the first encoder is used to convert the BIM information into a fixed-dimensional vector, including: The component sequence of the BIM information is encoded by the first encoder to obtain the hidden state of each component; The hidden state is globally averaged and pooled to obtain a fixed-dimensional vector of the building project.

[0042] It can be understood that the ordered list of concrete components is encoded by using the trained BiLSTM encoder. The sequence representing the BIM information is input into the trained BiLSTM encoder, and the bidirectional dependency relationship of the concrete components at different periods in the time sequence can be captured due to the bidirectional design of the BiLSTM. After inputting the sequence, the BiLSTM outputs the hidden state of each component : ​

[0043] where, and are the hidden states of the forward and backward LSTM at position i respectively. Then, the hidden states are pooled by a global average pooling layer to get a fixed dimension representation vector of the building :

[0044] In some embodiments, determining the context vector of the similar project set comprises: fusing the component sequence corresponding to the BIM information of the similar project set and the relative time sequence between components to obtain a feature sequence of the completed building project; encoding the component sequence corresponding to the BIM information of the target building project to obtain a feature sequence of the target building project; determining the current state of the decoder based on the feature sequence of the target building project; determining the attention score between the current state of the decoder and the feature sequence of the completed building project; based on the attention score, performing weighted summation on the part of the feature sequence of the completed building project that needs to be focused to obtain the context vector.

[0045] It can be understood that at each time step (from 1 to m , m is the number of components of the target project), first calculate the attention score, for each similar project , calculate the attention score of the current state of the decoder and the feature sequence of the similar project :

[0046] where W is a learnable weight matrix. Through this attention mechanism, based on the component currently needed to be built under the current state, the model can automatically focus on similar construction patterns in the historical similar projects.

[0047] Then, the context vector is generated correspondingly to represent the weighted summation of the part of the similar project that needs to be focused using the attention score:

[0048] In some embodiments, fusing the component sequence corresponding to the BIM information of the similar project set and the relative time sequence between components to obtain a feature sequence of the completed building project comprises: ​fusing the component sequence corresponding to the BIM information of the similar project set and the relative time sequence between components, to obtain a new sequence; inputting the new sequence into a second encoder to obtain a feature sequence of the completed construction project; the second encoder is obtained based on a Transformer encoder.

[0049] It can be understood that, by fusing the component sequence and the relative time sequence , the features corresponding to the time steps are spliced, and the new sequence is input into the encoder . The encoder adopts a Transformer encoder, and outputs a feature sequence of the same length, and the feature representation of each component not only represents the features of the component, but also fuses the context information of the component in the entire construction project through the Transformer encoder.

[0050] In some embodiments, encoding the component sequence corresponding to the BIM information of the target construction project to obtain a feature sequence of the target construction project, comprises: encoding the component sequence corresponding to the BIM information of the target construction project based on a third encoder to obtain a feature sequence of the target construction project; the third encoder is obtained based on a Transformer encoder.

[0051] It can be understood that the encoding of the target construction project and the encoding of the historical similar projects are roughly the same, except that the relative time sequence (which is also the sequence to be finally predicted) is missing, and only the component sequence exists. The sequence is input into an encoder also adopting a Transformer encoder architecture, to obtain a feature sequence of the target project , wherein m is the number of components of the target project.

[0052] In some embodiments, based on the predicted relative construction time of the components, the start time of the target construction project and the volume of the components, the total concrete demand is determined, comprising: based on the predicted relative construction time of each component and the start time of the target construction project, determining the predicted absolute construction time of each component; based on the predicted absolute construction time of each component, selecting a target component sequence meeting a preset time requirement from the component sequence corresponding to the BIM information of the target construction project; based on the target component sequence and the volume of the target component, determining the total concrete demand.

[0053] It can be understood that each value in the target component sequence is multiplied by the corresponding component volume, and the corresponding product of each component in the target component sequence is accumulated to obtain the total commercial concrete demand.

[0054] In some embodiments, a preset time requirement is met:

[0055]

[0056] wherein, t c denotes the timestamp of the time when the commercial concrete prediction demand exists, t a denotes the predicted absolute construction time of one component, t b denotes the predicted absolute construction time of another component, t a-1 denotes t a the last time of t b-1 denotes t b the last time of

[0057] It can be understood that to obtain the commercial concrete prediction demand of a day (assuming the timestamp of the day is ), only the subsequence index and in the timestamp sequence need to be found, and the above formula is met.

[0058] In some embodiments, the decoder is trained based on the mean square error between the predicted component relative construction time and the real component relative construction time as the loss function.

[0059] It can be understood that the model is trained by minimizing the error between the predicted time sequence and the real time sequence , and the mean square error (MSE) is used as the loss function, that is:

[0060] wherein is the real relative construction time of the i-th component of the target project.

[0061] In some embodiments, the method provided by the present application can refer to Figure 2 , wherein the input and output are represented by a rectangular frame, and the module is represented by a rounded rectangular frame.

[0062] ​BIM information encoding module based on contrastive learning: This module can encode the BIM information of a building into a vector, and the vector representations of similar buildings are close in the vector space, while the vector representations of buildings with large differences are far apart. In order to train the encoder, positive and negative sample pairs annotated by expert knowledge are used for contrastive learning training.

[0063] BIM prior knowledge enhanced component construction time prediction module: Based on the BIM information encoding module, a component historical building project database is constructed. For the target building project to be predicted, the concrete component sequence in the BIM of the target project is encoded using the encoder to obtain a feature sequence. Through attention mechanism, the feature information of similar building projects in the database is fused, and the decoder is used to predict the relative construction time of each component step by step.

[0064] Commercial concrete demand generation module based on component construction time: Based on the relative construction time of each concrete component obtained previously, the absolute time stamp is calculated, and the commercial concrete demand of the day is calculated according to the query date.

[0065] In order to enhance the BIM information for commercial concrete demand prediction, the BIM information is regarded as an ordered list of concrete components of a building project, denoted as , where each component is composed of its id, type and volume, and the order of the concrete components indicates the order of construction of each component in the process of constructing the entire building. The type of the component significantly distinguishes various components in the building, including but not limited to the types shown in Table 1: Table 1: Component types and corresponding names

[0066] The object of the present application is to predict the expected start construction time of each component given the BIM information model (ordered list of concrete components) of a building project that has not yet started (i.e. the target building project). To this end, the data of historical completed building projects need to be used for model training. In the historical data, in addition to the corresponding BIM information model, the start construction time of each component is also included. After training with historical data, the start construction time of each component of the new building project can be predicted, and thus the concrete demand plan in each period can be predicted.

[0067] The problem is defined as follows: given an ordered list of concrete components representing the building information model (in construction order), , predict a time stamp sequence of the same length , where each time stamp represents the construction time of component .

[0068] A BIM information encoding module based on contrastive learning: in order to obtain the BIM of a new building project, the information thereof can be encoded and compared with vectors in historical completed projects, and those similar projects are referred to, therefore, a BIM information encoding module based on contrastive learning is proposed.

[0069] Given a plurality of building projects, the BIM of each building project is represented as an ordered list of components sorted by construction time , wherein each component is represented as a feature vector , wherein is a one-hot encoding vector of the component type, is a volume value of the component.

[0070] In order to obtain positive and negative sample pairs for contrastive learning, the data is labeled using expert knowledge, and the buildings are divided into a plurality of sets . The buildings in each set are very similar in building type (for example, hospital) and building size (for example, same number of floors), while the buildings between different sets are quite different. The goal is to extract a representation vector for each building through BIM information through contrastive learning, so that the buildings in the same set are closer in the embedding space, and the buildings in different sets are farther away.

[0071] The ordered list of concrete components is encoded using a trained BiLSTM (bidirectional long short-term memory network) encoder. The sequence representing the BIM information is input into the trained BiLSTM encoder. Due to the bidirectional design of the BiLSTM encoder, it can capture the bidirectional dependency relationship of the concrete components at different times in the time series. After inputting the sequence, the BiLSTM encoder outputs the hidden state of each component :

[0072] wherein and are the hidden states of the forward and reverse LSTM at position i, respectively. Then, the hidden state is pooled by a global average pooling layer to obtain a fixed-dimensional representation vector of the building :

[0073] After obtaining the encoding of the BIM, we need to obtain the positive and negative sample pairs for training. The positive sample pair is selected from the same expert set, representing that the buildings are very similar; the negative sample pair is selected from different expert sets. The contrastive loss function is used to make the representation vectors of the positive sample pair closer and the representation vectors of the negative sample pair farther away. Let the number of positive and negative sample pairs be N, then the loss function is as follows: The contrastive learning loss function based on sample pairs can be defined as:

[0074] wherein, is the number of sample pairs, denotes the loss of the i th positive sample pair, denotes the loss of the i th negative sample pair.

[0075] For a positive sample pair:

[0076] wherein, and are the representation vectors of the two samples in the positive sample pair, respectively. The loss function target of the positive sample pair is to make the representation vectors of the positive sample pair (two similar buildings) as close as possible, so the square of the Euclidean distance is used as the loss.

[0077] For a negative sample pair:

[0078] wherein, and are the representation vectors of the two samples in the negative sample pair, respectively, is a predefined Margin value, representing the minimum allowed distance between negative sample pairs. The goal of the negative sample loss function is to make the representation vectors of the negative sample pair (two buildings with large differences) as far apart as possible, at least more than the Margin value.

[0079] In this way, the contrastive learning loss function based on sample pairs can simultaneously perceive similar buildings and buildings with large differences, and optimize the BiLSTM encoder to learn an effective BIM information representation.

[0080] For a new building BIM, input its component sequence into the trained model to obtain the representation vector of the entire building . This BIM representation vector will be used for subsequent component construction time prediction.

[0081] BIM prior knowledge enhanced component construction time prediction module: Based on the above description, an encoder of BIM information has been obtained, which can obtain the vector representation of the building through the building BIM information (concrete component sequence). Similar buildings have similar vector representations. Next, how to integrate similar building information to enhance the construction time prediction of each concrete component of the target building will be described.

[0082] Top-k similar buildings found: Since there is a historical information database containing BIM information of multiple completed projects, it is encoded and converted into vector representation. For the target building project, its BIM is encoded into a vector Afterwards, calculate The Euclidean distance between the BIM vectors of all completed projects in the database is used to find the top k projects with the smallest distance, which are recorded as the similar project set. .

[0083] Historically similar project codes: For each similar item Since it is a historical completed project, there is not only BIM information (component sequence) (in It is a project i The original time series is a timestamp sequence, with each element representing the absolute construction time of the corresponding component. However, absolute timestamps are meaningless; a more meaningful representation is the relative construction time between components. For example, the foundation cushion layer was built first, and the foundation slab began to be built two days later. To this end, the present invention converts the original timestamp sequence into a relative time series. , the relative time corresponding to each component Indicates how many days after the previous component the component was built .

[0084] By fusion construct sequence and relative time series , concatenate the features of the corresponding time steps, and input the new sequence into the encoder The encoder uses the Transformer encoder to output feature sequences of the same length , the characteristic representation of each component Not only does it characterize the features of the component, but it also integrates its contextual information in the entire construction project through the Transformer encoder.

[0085] Target building project code: The coding of the target construction project is roughly the same as that of similar historical projects, except that there is a lack of relative time series. (This is also the sequence that needs to be predicted in the end), only the component sequence exists This sequence is input to the encoder which also uses the Transformer encoder architecture. In the example above, we get the feature sequence of the target item. ,in m is the target project's artifact quantity.

[0086] Construction time prediction with similar project attention: The relative construction time of each component (relative to the previous component) needs to be generated step by step, using an RNN (Recurrent Neural Network) decoder implemented by an MLP (Multi-Layer Perceptron), with the input at each time step being the state and the remaining features, and the output being the relative construction time of the component. The initial state is set to the global features of the target project encoding, and the target project feature sequence is the average pooling result:

[0087] At each time step (from 1 to m , m is the number of components of the target project), first calculate the attention score, for each similar project , calculate the attention score of the feature sequence of the similar project and the current state of the decoder:

[0088] where W is a learnable weight matrix. Through this attention mechanism, based on the current component that needs to be built under the current state, the model can automatically focus on similar construction patterns in the historical similar projects.

[0089] Then, the context vector is generated correspondingly, which represents the weighted sum of the part of the similar project that needs to be focused on using the attention score:

[0090] The construction time of the component corresponding to this time step is predicted by the decoder. The decoder state , the corresponding feature of the target project at this time step and the context vector are input into the decoder to predict the relative construction time of the current component , and update the decoder state :

[0091] Finally, the predicted relative construction time sequence of each component of the target project is obtained.

[0092] The predicted time sequence Error between real time series Model training is carried out between the error, and mean square error (MSE) is used as the loss function, that is:

[0093] Among them is the real relative construction time of the target project th component.

[0094] Based on the component construction time, the commercial concrete demand generation module: In the previous part, the construction time of each component of the target building project has been predicted through BIM prior knowledge enhancement, and the sequence is obtained, which is briefly denoted as , and then the relative time sequence is converted into absolute timestamp sequence .

[0095] Let the starting time of the target project be , then the predicted construction time (timestamp) of the i th component is:

[0096] The predicted construction time of the i th component is known, and for each component, there is a volume . To obtain the commercial concrete prediction demand on a certain day (assuming that the timestamp of a day is ), only the subsequence index and in the timestamp sequence need to be found, which satisfy:

[0097]

[0098] The total commercial concrete demand on the corresponding day can be obtained:

[0099] The present application has the following advantages: The present application can predict the commercial concrete demand of the project at different times through the BIM information (concrete component information) of the project before the project starts. Existing schemes are mostly based on time series prediction or market research, and there is no research on commercial concrete demand prediction based on BIM information. In addition, the present application also expands the application of BIM information.

[0100] The key points of the present application are: The BIM information encoding module based on contrast learning can encode the BIM information of the building, and similar building vectors are close in distance, and dissimilar building vectors are far in distance.

[0101] The component construction time prediction module enhanced by BIM prior knowledge can fuse the construction process information of similar building projects to assist the target project in predicting the construction time of each concrete component.

[0102] As shown in Figure 3 The application further provides a full-life-cycle commercial concrete demand prediction device 300, which comprises: An information encoding module 301 is configured to acquire BIM information of a building project and convert the BIM information into a fixed-dimension vector based on a first encoder; the first encoder is trained based on a BiLSTM encoder for representing a positive sample pair with high building similarity and a negative sample pair with low building similarity; A similar project determination module 302 is configured to determine the Euclidean distance between the fixed-dimension vector of the target building project and the fixed-dimension vector of a completed building project, and select a number of completed building projects with the smallest Euclidean distance to form a similar project set; A context vector construction module 303 is configured to determine a context vector of the similar project set; A construction time prediction module 304 is configured to input a feature sequence corresponding to a component sequence of the target building project, a decoder state and the context vector into a decoder to obtain a predicted component relative construction time; the decoder is trained based on the mean square error between the predicted component relative construction time and the real component relative construction time as a loss function; A demand determination module 305 is configured to determine the total commercial concrete demand based on the predicted component relative construction time, the start time of the target building project and the component volume.

[0103] The full-life-cycle commercial concrete demand prediction device provided by the above embodiment can implement the technical solutions described in the full-life-cycle commercial concrete demand prediction method embodiment, and the principles of the implementation of the above modules or units can be referred to the corresponding content in the full-life-cycle commercial concrete demand prediction method embodiment, which will not be described here.

[0104] As shown in Figure 4 The application further correspondingly provides an electronic device 400. The electronic device 400 comprises a processor 401, a memory 402 and a display 403. Figure 4 Only part of the components of the electronic device 400 are shown, but it should be understood that all the shown components are not required to be implemented, and more or fewer components can be alternatively implemented.

[0105] The memory 402 can be an internal storage unit of the electronic device 400, such as a hard disk or a memory of the electronic device 400 in some embodiments. The memory 402 can also be an external storage device of the electronic device 400, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, and the like equipped on the electronic device 400 in other embodiments.

[0106] Further, the memory 402 can include both an internal storage unit and an external storage device of the electronic device 400. The memory 402 is used to store application software installed on the electronic device 400 and various types of data.

[0107] The processor 401 can be a central processing unit (CPU), a microprocessor, or other data processing chip in some embodiments, used to run program codes or process data stored in the memory 402, such as the full-life-cycle commercial concrete demand prediction method in the present application.

[0108] The display 403 can be an LED display, a liquid crystal display, a touch liquid crystal display, an OLED (Organic Light-Emitting Diode) touch, and the like in some embodiments. The display 403 is used to display information of the electronic device 400 and to display a visualized user interface. The components 401-403 of the electronic device 400 communicate with each other through a system bus.

[0109] In some embodiments of the present application, when the processor 401 executes the full-life-cycle commercial concrete demand prediction program in the memory 402, the following steps can be implemented: BIM information of a construction project is obtained, and the BIM information is converted into a fixed-dimensional vector based on a first encoder; the first encoder is trained based on a BiLSTM encoder for a positive sample pair representing high building similarity and a negative sample pair representing low building similarity; The Euclidean distance between the fixed-dimensional vector of a target construction project and the fixed-dimensional vector of a completed construction project is determined, and a number of completed construction projects with the smallest Euclidean distance are selected to form a similar project set; A context vector of the similar project set is determined; A feature sequence corresponding to a component sequence of the target construction project, a decoder state, and the context vector are input into a decoder to obtain a predicted component relative construction time; the decoder is trained based on a mean square error between the predicted component relative construction time and the actual component relative construction time as a loss function; determine the total commercial concrete demand based on the predicted component relative construction time, the start time of the target construction project and the component volume.

[0110] It should be understood that, in addition to the above functions, the processor 401 can also implement other functions when executing the full-life-cycle commercial concrete demand prediction program in the memory 402. For details, refer to the description of the corresponding method embodiments.

[0111] Further, the type of the electronic device 400 referred to in the embodiments of the present application is not specifically limited, and the electronic device 400 can be a portable electronic device such as a mobile phone, a tablet computer, a personal digital assistant (PDA), a wearable device, a laptop computer, etc. Exemplary embodiments of the portable electronic device include, but are not limited to, a portable electronic device running an IOS, android, microsoft or other operating system. The above-mentioned portable electronic device can also be other portable electronic devices, such as a laptop computer having a touch-sensitive surface (e.g., a touch panel), etc. It should also be understood that, in some other embodiments of the present application, the electronic device 400 can also not be a portable electronic device, but a desktop computer having a touch-sensitive surface (e.g., a touch panel).

[0112] In yet another aspect, the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the full-life-cycle commercial concrete demand prediction method provided by the above-mentioned methods. The method comprises: obtaining BIM information of a construction project, and converting the BIM information into a fixed-dimensional vector based on a first encoder; the first encoder is trained based on a BiLSTM encoder for a positive sample pair representing high building similarity and a negative sample pair representing low building similarity; determining the Euclidean distance between the fixed-dimensional vector of the target construction project and the fixed-dimensional vectors of the completed construction projects, and selecting a number of completed construction projects with the smallest Euclidean distance to form a similar project set; determining the context vector of the similar project set; inputting the feature sequence corresponding to the component sequence of the target construction project, the decoder state and the context vector into the decoder to obtain the predicted component relative construction time; the decoder is trained based on the mean square error between the predicted component relative construction time and the real component relative construction time as a loss function; determine the total commercial concrete demand based on the predicted component relative construction time, the start time of the target construction project and the component volume.

[0113] Those skilled in the art can understand that all or part of the processes of the above-mentioned embodiment methods can be completed by instructing the relevant hardware by a computer program, and the program can be stored in a computer readable storage medium. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory, etc.

[0114] The full life cycle commercial concrete demand prediction method, device and storage medium provided by the present application are described in detail above. The principles and implementation modes of the present application are described by applying specific examples. The above example is only used to help understand the method and core idea of the present application. Meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed. In summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A full life cycle commercial concrete demand prediction method, characterized in that, The method comprises the following steps: obtaining BIM information of a building project and converting the BIM information into a fixed-dimensional vector based on a first encoder; the first encoder is trained based on a BiLSTM encoder for representing positive sample pairs with high building similarity and negative sample pairs with low building similarity; determining the Euclidean distance between the fixed-dimensional vector of the target building project and the fixed-dimensional vectors of the completed building projects, and selecting a number of completed building projects with the smallest Euclidean distance to form a similar project set; determining a context vector of the similar project set; inputting the feature sequence corresponding to the component sequence of the target building project, the decoder state and the context vector into a decoder to obtain a predicted component relative construction time, and the decoder is trained based on the mean square error between the predicted component relative construction time and the actual component relative construction time as a loss function; determining the total commercial concrete demand based on the predicted component relative construction time, the construction time of the target building project and the component volume.

2. The full life cycle commercial concrete demand prediction method of claim 1, wherein, The method comprises the following steps: encoding the component sequence of the BIM information based on the first encoder to obtain the hidden state of each component; performing global average pooling on the hidden state to obtain the fixed-dimensional vector of the building project.

3. The full life cycle commercial concrete demand prediction method of claim 1, wherein, The method comprises the following steps: fusing the component sequence corresponding to the BIM information of the similar project set and the relative time sequence between components to obtain a feature sequence of the completed building project; encoding the component sequence corresponding to the BIM information of the target building project to obtain a feature sequence of the target building project; determining the current state of the decoder based on the feature sequence of the target building project; determining the attention score between the current state of the decoder and the feature sequence of the completed building project; based on the attention score, performing weighted summation on the part of the feature sequence of the completed building project that needs to be focused to obtain a context vector.

4. The full life cycle commercial concrete demand prediction method of claim 3, wherein, The method comprises the following steps: fusing the component sequence corresponding to the BIM information of the similar project set and the relative time sequence between components to obtain a new sequence; inputting the new sequence into a second encoder to obtain a feature sequence of the completed building project; the second encoder is trained based on a Transformer encoder.

5. The full life cycle commercial concrete demand prediction method of claim 3, wherein, The method comprises the following steps: encoding the component sequence corresponding to the BIM information of the target building project based on a third encoder to obtain a feature sequence of the target building project; the third encoder is trained based on a Transformer encoder.

6. The full life cycle commercial concrete demand prediction method of claim 1, wherein, The method comprises the following steps: determining the total commercial concrete demand based on the predicted component relative construction time, the construction time of the target building project and the component volume. determine a predicted absolute construction time of each component based on the predicted relative construction time of each component and a start time of the target construction project; select a target component sequence that meets a preset time requirement from the component sequence corresponding to the BIM information of the target construction project based on the predicted absolute construction time of each component; determine a total commercial concrete demand based on the target component sequence and a volume of the target component.

7. The full life cycle commercial concrete demand prediction method of claim 6, wherein, The preset time requirement meets: wherein, t c a timestamp indicating a forecasted demand for ready-mix concrete, t a a forecasted absolute construction time for one component, t b a forecasted absolute construction time for another component, t a-1 a timestamp indicating a forecasted demand for ready-mix concrete, t a a last time of t b-1 a timestamp indicating a forecasted demand for ready-mix concrete, t b a last time of 8. A device for predicting the demand for ready-mix concrete throughout its life cycle, characterized by The method comprises: an information encoding module configured to obtain BIM information of a construction project and convert the BIM information into a fixed-dimension vector based on a first encoder; the first encoder is trained based on a BiLSTM encoder for positive sample pairs representing high building similarity and negative sample pairs representing low building similarity; a similar project determining module configured to determine a Euclidean distance between a fixed-dimension vector of a target construction project and fixed-dimension vectors of completed construction projects, and select a number of completed construction projects with the smallest Euclidean distance to form a similar project set; a context vector component module configured to determine a context vector of the similar project set; a construction time predicting module configured to input a feature sequence corresponding to a component sequence of the target construction project, a decoder state and the context vector into a decoder to obtain a predicted relative construction time of the component; the decoder is trained based on a mean square error between the predicted relative construction time of the component and an actual relative construction time of the component as a loss function; a demand determining module configured to determine a total commercial concrete demand based on the predicted relative construction time of the component, a start time of the target construction project and a volume of the component.

9. An electronic device, comprising: comprise a memory and a processor, wherein the memory is configured to store a program; the processor is coupled to the memory and configured to execute the program stored in the memory to implement the steps of the full-life-cycle commercial concrete demand prediction method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the full-life-cycle commercial concrete demand prediction method according to any one of claims 1 to 7.

Images