A building economic forecasting method and system based on the Internet of Things

Abstract

The present invention discloses a method and system for forecasting building economy based on the Internet of Things. The method comprises: collecting Internet data, Internet of Things data and enterprise economic data related to the building, dividing the data into a first training set and a first test set after preprocessing; training an ARIMA model with the first training set and forecasting the Internet of Things data with the first test set to obtain the forecast result of the ARIMA model; acquiring event data of the building, obtaining event embedding of the building with the building knowledge graph according to the event data, and using the multi-channel link vector composed of the forecast result of the ARIMA model and the event embedding as a second test set; using the multi-channel link vector composed of the first training set and the event embedding and the current time as a second training set, training an LSTM model with the second training set based on blockchain and federated learning to obtain an enterprise economic forecast model, and using the enterprise economic forecast model to forecast the future economy of the building according to the second test set.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a building economic forecasting method and system based on the Internet of Things. Background Art

[0002] In the process of enterprise economic development, forecasting and estimating its specific economic conditions and future development trends can help the corresponding managers to more clearly understand the trend of enterprise economic development, improve the accuracy of enterprise economic decision-making, and to a large extent increase the speed of enterprise development and avoid enterprise economic losses. Effective economic forecasting can provide detailed information and data support for specific economic decision-making and ensure the accuracy of economic decision-making. Secondly, effective economic forecasting can guarantee the effectiveness and accuracy of enterprise economic planning to a large extent, avoid planning errors as much as possible, and ensure the healthy development of enterprises.

[0003] The current economic forecasting method is mainly based on historical economic development and historical time series data to predict future economic development. However, this method can only predict based on historical economic development, not based on the actual operation status of the enterprise. Once the operation of the enterprise is affected by force majeure factors, the forecasting method that only uses past economic data will fail. In addition, the factors that affect the economic development of enterprises are diverse and easily affected by emergencies. The information expressed only by time series is limited. The models and algorithms learned on this basis cannot better learn the development and operation of enterprises, but are only suitable for data with strong time series rules. In the era of the Internet of Things, combining more data information to predict the economic development of enterprises has become an urgent problem to be solved. In addition, there will be data privacy and security issues in the economic data of enterprises. How to train models without obtaining the economic data of enterprises is also an urgent problem to be solved. Summary of the invention

[0004] In view of the above-mentioned technical problem that only using historical economic data to predict economic development has limitations, the present invention proposes a building economic forecasting method and system based on the Internet of Things.

[0005] In a first aspect, an embodiment of the present application provides a building economic forecasting method based on the Internet of Things, comprising:

[0006] Influencing factor data acquisition step: collecting Internet data, Internet of Things data and enterprise economic data related to the building, and calculating the correlation between the lagged Internet of Things data and the enterprise economic data, selecting the lagged data with the largest correlation, and pre-processing the Internet data, the Internet of Things data, the enterprise economic data and the lagged data into a first training set and a first test set;

[0007] ARIMA model prediction step: using the first training set to train the ARIMA model and using the first test set to perform time series prediction on the IoT data, to obtain a prediction result of each influencing factor in the IoT data as a prediction result of the ARIMA model;

[0008] Event data acquisition step: acquiring event data of a building, obtaining event embedding of the building using the building knowledge graph according to the event data, and using the multi-channel link vector composed of the prediction result of the ARIMA model and the event embedding as the second test set;

[0009] LSTM model prediction step: use the multi-channel link vector composed of the first training set and the event embedding and the current time as the second training set, train the LSTM model according to the second training set based on blockchain and federated learning, obtain the enterprise economic forecasting model, and use the enterprise economic forecasting model to predict the future economy of the building according to the second test set.

[0010] In the above-mentioned building economic forecasting method, the ARIMA model forecasting step comprises:

[0011] ARIMA model establishment step: establishing the ARIMA model, verifying the ARIMA model according to the AIC criterion based on each influencing factor data in the IoT data and obtaining the optimal parameters corresponding to each influencing factor data, and obtaining the optimal ARIMA model corresponding to each influencing factor data respectively;

[0012] The step of obtaining the prediction result of the ARIMA model is as follows: the optimal ARIMA model is trained by the first training set, and each influencing factor data is predicted respectively by the optimal ARIMA model using the first test set to obtain the prediction result of the ARIMA model corresponding to each influencing factor data.

[0013] In the above-mentioned building economic forecasting method, the event data acquisition step comprises:

[0014] Event tuple acquisition step: acquiring event data of the building, and extracting event tuples from the event data;

[0015] Event embedding acquisition step: obtaining an event subgraph by linking the event tuple to the building knowledge graph, and embedding the event subgraph to obtain the event embedding of the building.

[0016] In the above-mentioned building economic forecasting method, the LSTM model forecasting step includes:

[0017] Data conversion step: normalizing the second training set and the second test set;

[0018] Model structure determination step: initializing the parameters of the LSTM model;

[0019] Model training step: the enterprise economic forecasting model is obtained by training the LSTM model using the normalized second training set, and the prediction result of the enterprise economic forecasting model is obtained according to the normalized second test set.

[0020] In the above-mentioned building economic forecasting method, the LSTM model prediction step also includes: calculating the accuracy of the prediction result of the enterprise economic forecasting model by comparing the prediction result of the enterprise economic forecasting model with the true value of the enterprise economic data.

[0021] In the above-mentioned building economic forecasting method, the model training step comprises:

[0022] Local model training step: using the normalized second training set to train the local LSTM model, and uploading the trained local LSTM model parameters to the blockchain through a smart contract;

[0023] Global model parameter acquisition step: monitoring and aggregating the local LSTM model parameters through the smart contract, obtaining global model parameters and storing them in the blockchain;

[0024] Enterprise economic forecasting model acquisition step: determine whether the global model has converged based on the global model parameters. If not, send the global model parameters from the blockchain to each participant to update the local LSTM model parameters, continue iterative training until the global model converges, and use the converged global model as the enterprise economic forecasting model.

[0025] In the above-mentioned building economic forecasting method, the Internet data includes but is not limited to: the building's operating income, the building's news and public opinion information and favorable information; the Internet of Things data includes but is not limited to: parking lot vehicle entry and exit data, enterprise electricity consumption data and enterprise access control personnel information data.

[0026] In the above-mentioned building economic forecasting method, the step of acquiring the influencing factor data includes: removing abnormal points from the Internet data, the Internet of Things data, the enterprise economic data and the lagged data by using the 3 sigma principle, and pre-processing them by using the multiple interpolation method to supplement the missing values.

[0027] In the above-mentioned building economic forecasting method, the model structure determination step includes: determining the activation function of the LSTM model, determining the fully connected artificial neural network that receives the output of the LSTM model, determining the discard rate of each layer of network nodes, determining the error calculation method, determining the iterative update method of the weight parameters, and determining the epoch and batch size of the model training.

[0028] In a second aspect, the embodiment of the present application provides a building economic forecasting system based on the Internet of Things, including:

[0029] An influencing factor data acquisition unit is used to collect Internet data, Internet of Things data and enterprise economic data related to the building, and calculate the correlation between the lagged Internet of Things data and the enterprise economic data, select the lagged data with the largest correlation, and pre-process the Internet data, the Internet of Things data, the enterprise economic data and the lagged data into a first training set and a first test set;

[0030] ARIMA model prediction unit: using the first training set to train the ARIMA model and using the first test set to perform time series prediction on the IoT data, to obtain a prediction result of each influencing factor in the IoT data as a prediction result of the ARIMA model;

[0031] An event data acquisition unit: acquires event data of a building, obtains event embedding of the building using a building knowledge graph according to the event data, and uses a multi-channel link vector composed of the prediction result of the ARIMA model and the event embedding as a second test set;

[0032] LSTM model prediction unit: take the multi-channel link vector composed of the first training set and the event embedding and the current time as the second training set, train the LSTM model according to the second training set based on blockchain and federated learning, obtain the enterprise economic prediction model, and use the enterprise economic prediction model to predict the future economy of the building according to the second test set.

[0033] Compared with the prior art, the advantages and positive effects of the present invention are:

[0034] 1. The present invention provides a building economic forecasting method based on the Internet of Things. The Internet data and the Internet of Things data in the building operation are added to the model as factors affecting economic development. The time series forecast of the sequence adopts the ARIMA model with better effect in the traditional time series analysis model for forecasting, and the nonlinear part adopts the LSTM model with better long-term series forecasting effect for forecasting. The present invention uses diversified data information, combines the advantages of the two models, and realizes a more accurate forecast of the economic development of enterprises.

[0035] 2. Model emergencies in buildings through knowledge graphs to improve the sensitivity of predicting sudden economic changes. In addition, this knowledge-driven approach can encode explainable knowledge into deep prediction models, making the prediction results explainable.

[0036] 3. By adopting the horizontal federated learning method, a joint training model is established based on the economic data of multiple local enterprises, thereby ensuring the privacy and security of the economic data of each enterprise. Since federated learning requires an aggregation server, the present invention uses blockchain to replace the aggregation server, avoiding the problem of single point failure caused by the easy failure of the aggregation server. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic diagram of the steps of a building economic forecasting method based on the Internet of Things provided by the present invention;

[0038] Figure 2 The present invention provides Figure 1 Schematic diagram of the process of step S2;

[0039] Figure 3 The present invention provides Figure 1 Schematic diagram of the process of step S3;

[0040] Figure 4 The present invention provides Figure 1 Schematic diagram of the process of step S4;

[0041] Figure 5 A schematic diagram of the building economic process prediction using the ARIMA-LSTM model based on IoT data provided by the present invention;

[0042] Figure 6 The enterprise event graph triple link graph provided by the present invention;

[0043] Figure 7 The blockchain and federated learning training flow chart provided by the present invention;

[0044] Figure 8 The ACF-PACF diagram of the total number of vehicles visiting the enterprise within one month provided by the present invention;

[0045] Fig. 9 The ACF-PACF diagram of the total monthly electricity consumption of the enterprise provided by the present invention;

[0046] Fig.10 The ACF-PACF diagram of the average number of active employees per month in an enterprise provided by the present invention;

[0047] Fig.11 A schematic diagram of the results of predicting building economics using the ARIMA-LSTM model based on IoT data provided by the present invention;

[0048] Fig.12 A framework diagram of a building economic forecasting system based on the Internet of Things provided by the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0050] Obviously, the drawings described below are only some examples or embodiments of the present application. For ordinary technicians in this field, the present application can also be applied to other similar scenarios based on these drawings without creative work. In addition, it can also be understood that although the efforts made in this development process may be complicated and lengthy, for ordinary technicians in this field related to the content disclosed in this application, some changes in design, manufacturing or production based on the technical content disclosed in this application are just conventional technical means, and should not be understood as insufficient content disclosed in this application.

[0051] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0052] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the" and the like involved in this application do not indicate a quantitative limitation, and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0053] The present invention is described in detail below in conjunction with the various embodiments shown in the accompanying drawings, but it should be noted that these embodiments are not limitations of the present invention, and any equivalent transformations or substitutions in functions, methods, or structures made by ordinary technicians in the field based on these embodiments are all within the scope of protection of the present invention.

[0054] Before elaborating on each embodiment of the present invention in detail, the core inventive concept of the present invention is summarized and elaborated in detail through the following several embodiments.

[0055] The present invention proposes a building economic forecasting method and system based on the Internet of Things, which collects Internet data, Internet of Things data, economic development data and building-related events of the building, and uses the knowledge graph to obtain event embedding. The ARIMA model and the LSTM model are combined through the blockchain and federated learning method to train the model based on multi-party data, thereby achieving a more accurate prediction of the economic development of the enterprise.

[0056] Embodiment 1:

[0057] Figure 1 The following is a schematic diagram of the steps of a building economic forecasting method based on the Internet of Things provided by the present invention. Figure 1 As shown, this embodiment discloses a specific implementation method of a building economic forecasting method based on the Internet of Things (hereinafter referred to as the "method").

[0058] Specifically, the method disclosed in this embodiment mainly includes the following steps:

[0059] Step S1: collecting Internet data, Internet of Things data and enterprise economic data related to the building, and calculating the correlation between the lagged Internet of Things data and the enterprise economic data, selecting the lagged data with the largest correlation, and pre-processing the Internet data, the Internet of Things data, the enterprise economic data and the lagged data into a first training set and a first test set;

[0060] Specifically, the Internet data includes but is not limited to: the operating income of the building, the news and public opinion information of the building, and favorable information; the Internet of Things data includes but is not limited to: the parking lot vehicle entry and exit data, the enterprise electricity consumption data, and the enterprise access control personnel information data. The Internet data, the Internet of Things data, the enterprise economic data and the lagged data are preprocessed by removing abnormal points using the 3sigma principle and using the multiple interpolation method to supplement missing values.

[0061] Step S2: using the first training set to train the ARIMA model and using the first test set to perform time series prediction on the IoT data, and obtaining a prediction result of each influencing factor in the IoT data as a prediction result of the ARIMA model;

[0062] like Figure 2 As shown, step S2 specifically includes the following contents:

[0063] Step S21: Establishing the ARIMA model, verifying the ARIMA model according to the AIC criterion based on each influencing factor data in the IoT data and obtaining the optimal parameters corresponding to each influencing factor data, and obtaining the optimal ARIMA model corresponding to each influencing factor data;

[0064] Specifically, the ARIMA model is recorded as ARIMA (p, d, q), where p is the number of autoregressive terms, q is the number of moving average terms, and d is the number of differences made when the time series becomes stationary.

[0065] Step S22: training the optimal ARIMA model through the first training set, and using the first test set to predict each influencing factor data through the optimal ARIMA model, to obtain a prediction result of the ARIMA model corresponding to each influencing factor data.

[0066] Step S3: Acquire event data of the building, obtain event embedding of the building using the building knowledge graph according to the event data, and use the multi-channel link vector composed of the prediction result of the ARIMA model and the event embedding as the second test set;

[0067] like Figure 3 As shown, step S3 specifically includes the following contents:

[0068] Step S31: Acquire event data of the building, and extract event tuples from the event data;

[0069] Specifically, a structured event tuple (S, P, O) is extracted from the event information, where S represents the event generator, P represents the action that occurs, and O represents the target to which the event is executed.

[0070] Step S32: obtaining an event subgraph by linking the event tuple to the building knowledge graph, and embedding the event subgraph to obtain event embedding of the building.

[0071] Step S4: Use the multi-channel link vector composed of the first training set, the event embedding and the current time as the second training set, train the LSTM model according to the second training set based on blockchain and federated learning, obtain the enterprise economic forecasting model, and use the enterprise economic forecasting model to predict the future economy of the building according to the second test set.

[0072] like Figure 4 As shown, step S4 specifically includes the following contents:

[0073] Step S41: normalizing the second training set and the second test set;

[0074] Step S42: Initialize the parameters of the LSTM model;

[0075] Specifically, step S42 includes: determining the activation function of the LSTM model, determining the fully connected artificial neural network that receives the output of the LSTM model, determining the discard rate of each layer of network nodes, determining the error calculation method, determining the iterative update method of the weight parameters, and determining the epoch and batch size of the model training.

[0076] Step S43: The enterprise economic forecasting model is obtained by training the LSTM model using the normalized second training set, and the prediction result of the enterprise economic forecasting model is obtained according to the normalized second test set.

[0077] Specifically, step S43 includes: using the normalized second training set to train the local LSTM model, and uploading the trained local LSTM model parameters to the blockchain through a smart contract; monitoring and aggregating the local LSTM model parameters through the smart contract, obtaining global model parameters and storing them in the blockchain; judging whether the global model has converged according to the global model parameters, if not, sending the global model parameters from the blockchain to each participant to update the local LSTM model parameters, continuing iterative training until the global model converges, and using the converged global model as the enterprise economic forecasting model.

[0078] Step S4 also includes: calculating the accuracy of the prediction result of the enterprise economic prediction model by comparing the prediction result of the enterprise economic prediction model with the real value of the enterprise economic data.

[0079] This embodiment provides a building economy forecasting method based on the Internet of Things, which combines building Internet of Things operation data and economic development data, and uses a method combining knowledge-driven and deep learning to forecast building economy. The Internet of Things data generated in the operation of the building is added to the model as a factor affecting economic development. The longitudinal prediction of the sequence is predicted by the ARIMA model with better effect in the traditional time series analysis model. The building-related events are used to construct the knowledge graph structure data, and embedded processing is performed to obtain embedded events, thereby adding knowledge-driven to the prediction of the building economy. The nonlinear part uses the LSTM model with better long-term sequence prediction effect for prediction. By adopting a horizontal federated learning method, the model is jointly trained based on the economic data of multiple local enterprises, thereby ensuring the privacy security of each enterprise's economic data. Since federated learning requires an aggregation server, it is easy to have a single point failure caused by the failure of the aggregation server, and the blockchain has the characteristics of decentralization. Therefore, the present invention combines federated learning with blockchain and uses blockchain instead of aggregation servers. This model can combine the advantages of the two models to achieve a more accurate prediction of corporate economic development.

[0080] Please refer to the following Figure 5 , Figure 5 ARIMA-LSTM model based on IoT data to predict building economic process diagram provided by the present invention, combined with Figure 5 The specific application process of this method is as follows:

[0081] Collect Internet data related to the development of a certain building: including the building's operating income, news and public opinion information, and favorable information;

[0082] Collect IoT data related to the development of a building: including parking lot vehicle entry and exit data, enterprise electricity consumption data, and enterprise access control personnel information data;

[0083] The above data collection time interval is one month, and the IoT data is processed into: the total number of vehicles visiting the enterprise within one month, the total monthly electricity consumption of the enterprise, and the average monthly number of active personnel in the enterprise.

[0084] Taking into account the lag of building economic data, the correlation between the three types of IoT data with a lag of one month, two months, and three months and the enterprise economic data is calculated, and the forecast data with the best forecast effect is selected.

[0085] The above Internet data, IoT data, the most relevant lagged data, and the current month are used as the training set input of the LSTM model;

[0086] The ARIMA model is used to make time series forecasts for each type of data, and the forecast results are used as the input of the LSTM model to predict the future economy of the building.

[0087] Event tuples are extracted from the event data of the building, and the event subgraph is obtained by linking the building knowledge graph. The knowledge graph is embedded to obtain the event embedding of the building. The event embedding obtained from the event graph and the prediction data of each category obtained by the ARIMA model are linked through a multi-channel method. The prediction results and event embedding are used as the input of the prediction model to predict the future economy of the building.

[0088] Among them, the process of forecasting using the ARIMA model is as follows:

[0089] 1) Data processing: Use the 3 sigma principle to remove outliers from each column of data, use multiple interpolation to fill in missing values, and divide the data into training and test sets.

[0090] 2) According to whether the original sequence is stationary and the difference in the regression part, an ARIMA model is established, denoted as ARIMA (p, d, q), where p is the number of autoregressive terms, q is the number of moving average terms, and d is the number of differences made when the time series becomes stationary.

[0091] 3) Use the AIC criterion to verify the established ARIMA model and find the most suitable parameters for each column of time series data.

[0092] 4) Train each column of time series data parameters according to the optimal ARIMA model and predict the data for the next year to obtain the prediction results of each influencing factor.

[0093] The process of embedding building events in the knowledge graph is as follows:

[0094] 1) Obtain event text information of the building;

[0095] 2) Extract structured event tuples (S, P, O) from event information, where S represents the event generator, P represents the action that occurs, and O represents the target of the event execution.

[0096] 3) Link the obtained event tuple to the knowledge graph (KG) to obtain the event subgraph of the event in the knowledge graph;

[0097] 4) Using the TransE method to embed the event subgraph and obtain event embedding;

[0098] The event embedding and the original historical data are combined into a multi-channel link vector for LSTM model training. The prediction results obtained by ARIMA and the multi-channel link vector composed of event embedding are used for LSTM model prediction to predict the economy of the building in the next year. The process is as follows:

[0099] 1) Original data conversion: normalize all influencing factor data and create a new data set.

[0100] 2) Determine and adjust the network model structure: Initialize LSTM parameters, including determining the activation function of the LSTM module, determining the fully connected artificial neural network that receives the LSTM output, determining the discard rate of each layer of network nodes to prevent overfitting, determining the error calculation method, determining the iterative update method of weight parameters, and determining the epoch and batchsize of model training.

[0101] 3) Model training and result prediction: The event embedding and original data are used as the training set, and the ARIMA prediction results are used as the test set. The accuracy of the model's prediction results is calculated by comparing them with the actual values ​​of the enterprise economic data for the predicted year.

[0102] The model training process based on blockchain and federated learning is as follows:

[0103] 1) Each participant registers on the blockchain network and generates a node for each participant;

[0104] 2) Each participant uses local data to train the enterprise economic forecasting model and uploads the trained local model parameters to the blockchain through smart contracts;

[0105] 3) Set up a smart contract on the blockchain to automatically monitor local model uploads and aggregation. When the local model parameters of all participants are uploaded, all local model parameters are aggregated based on the weighted average method to obtain the global model parameters and store them on the blockchain through the smart contract;

[0106] 4) Determine whether the global model has converged. If not, send the global model parameters from the blockchain to each participant to update the local model and continue iterative training until the global model converges;

[0107] 5) The converged global model is used as the final enterprise economic forecasting model.

[0108] The building economic forecasting method based on the Internet of Things proposed by the present invention is further described in detail below in conjunction with specific embodiments.

[0109] The building economy is predicted by combining the building IoT operation data and economic development data. The IoT data generated during building operation is added to the model as a factor affecting economic development, including:

[0110] Collect Internet data related to the development of a certain building: including the building's operating income, news and public opinion information, and favorable information;

[0111] Collect IoT data related to the development of a building: including parking lot vehicle entry and exit data, enterprise electricity consumption data, and enterprise access control personnel information data;

[0112] The above data collection time interval is one month, and the IoT data is processed into: the total number of vehicles visiting the enterprise within one month, the total monthly electricity consumption of the enterprise, and the average monthly number of active personnel in the enterprise.

[0113] Taking into account the lag of building economic data, the correlation between the three types of IoT data with the enterprise economic forecast is calculated for the data with a lag of one month, two months, and three months, and the forecast data with the best forecast effect is selected. The results of the correlation analysis using Excel's CORREL function are shown in Table 1.

[0114] Table 1 Correlation between lag time and three IoT data

[0115] Parking relevance Number of people correlation Power Correlation Raw data 0.61827322 0.192350094 0.383395 One month delay 0.911581803 0.192932 0.712994 Two months delay 0.424093532 0.16808765 0.032653 Three months behind 0.151069058 0.138717899 -0.24489

[0116] It can be seen from the above table that the parking correlation, personnel number correlation, and power correlation with a one-month lag are the highest. Therefore, the data with a one-month lag are selected to be added to the influencing factors of the training data.

[0117] The above Internet data, IoT data, the most relevant lagged data, and the current month data are used as the training set input of the LSTM model;

[0118] The longitudinal prediction of the sequence adopts the ARIMA model, which is a better traditional time series analysis model, to make time series predictions for each type of data. The prediction results are used as the input of the LSTM model to predict the future economy of the building.

[0119] Among them, the process of forecasting using the ARIMA model is as follows:

[0120] 1) Data processing: First, collect enterprise-related data from relevant departments for a total of 97 months from January 2013 to December 2020, organize possible related IoT data, Internet data and enterprise economic data, and pre-process the data. Use the 3sigma principle to remove outliers in each column of data and use multiple interpolation to supplement missing values. The data is divided into a training set and a test set, with the data from January 2013 to December 2019 as the training set and the data from January 2020 to December 2020 as the test set.

[0121] 2) Taking parking lot data as an example, an ARIMA model is established based on whether the original sequence is stationary and the difference in the regression part, denoted as ARIMA(p,d,q), where p is the number of autoregressive terms, q is the number of moving average terms, and d is the number of differences made when the time series becomes stationary. The ACF and PACF diagrams of the original sequence are as follows: Figure 8 , Fig. 9 , Fig.10 As shown in the figure, there is no truncation, which means that the sequence is not stable and needs to be differentiated. After one difference, the sequence is stable and the parameter d=1 is determined.

[0122] 3) Use the AIC criterion to verify the established ARIMA model and find the most suitable parameters for the model. Taking the parking lot data as an example, the regular period of the parking lot data time series is 12 months, and the parameter q = 12 is determined. According to the AIC result, the p value that minimizes the AIC is selected, and the parameter p = 0 is determined. The same method is used to determine the most suitable parameters for the model for other time series data.

[0123] 4) According to the optimal ARIMA model ARIMA(0,1,12) for parking lot data, the ARIMA model is trained with the parking lot data time series data parameters and the data for the next year is predicted to obtain the prediction results of the parking lot data.

[0124] 5) Use the same method to predict other time series data and obtain the prediction results of each influencing factor for the next 12 months.

[0125] The process of embedding building events in the knowledge graph is as follows:

[0126] 1) Obtain event text information of the building;

[0127] 2) First, collect relevant business event data of the building in recent years from relevant departments, such as business operation anomalies, bidding and investment, corporate recruitment information, etc., extract the main content, and form event information.

[0128] 3) Extract structured event tuples (S, P, O) from the event information, where S represents the event generator, P represents the action that occurs, and O represents the target of the event execution.

[0129] Structured event triples are extracted from the content of each event information. For example, from multiple events, we can get "lithium battery processing technology has been improved", "YY capital has invested in XX enterprise", "YY stock has risen by nearly 5%". This gives us the event triplet information of (lithium battery, improvement, technology), (YY capital, investment, XX enterprise), (YY stock, rise, 5%).

[0130] 4) Link the obtained event tuple to the knowledge graph (KG) to obtain the event subgraph of the event in the knowledge graph, such as Figure 6 As shown;

[0131] The event tuple is linked to the knowledge graph through entity linking to obtain the event subgraph in the knowledge graph. Since the subject s, predicate p, and object o in the event tuple may not always have links in the KG. In addition, the information in a single event tuple may be sparse and lack diversity. Therefore, here we enrich the subgraph in a way that utilizes neighbor information to associate discrete event tuples with each other.

[0132] 5) Get event tuples and knowledge Figure 3 After the tuples are formed, the TransE method is used to train the event subgraph to obtain the embedding of the knowledge event; the event features are parameterized into different channels, that is, V1 represents the link channel of KG, V2 represents the context channel of KG, and V3 represents the word vector channel. The final event embedding V = [V1V2V3] is obtained by linking these three channels.

[0133] The nonlinear part uses the LSTM model with good long-term sequence prediction effect for prediction. The event embedding and the original historical data form a multi-channel link vector for LSTM model training. The multi-channel link vector composed of the prediction results obtained by ARIMA and event embedding is used for LSTM model prediction to predict the future economic trend of the building. The process is as follows:

[0134] 1) Raw data conversion: Normalize the training set, prediction results and enterprise economic data to create a new data set.

[0135] 2) Determination and adjustment of the network model structure: Use Python's keras library to initialize LSTM parameters. The activation function of the LSTM module is tanh, the fully connected artificial neural network that receives the LSTM output is set to linear, the discard rate of each layer of network nodes is set to 0.2, the mean square error is used, the iterative update method of the weight parameters is the RMSprop algorithm suitable for RNN, the epoch interval of the model training is set to 100 and the batch size is set to 500.

[0136] 3) Model training: Figure 7 As shown, three local participants and a blockchain network are set up. The data and event embeddings of each participant from January 2013 to December 2019 form a multi-channel link vector as a training set, and the multi-channel link vector composed of the data and event embeddings predicted by ARIMA from January to December 2020 is used as the input of the local LSTM model of each participant. The original data of each participant from January to December 2020 is used as the test set. First, each participant uses local data to train the local model for 50 rounds, and then each participant uploads the local model parameters to the blockchain by calling the smart contract through each blockchain node. The smart contract in the blockchain monitors whether the local model parameters of the three participants are uploaded. After all participants have uploaded, the three local model parameters are aggregated by the weighted average method to obtain the global model parameters, and sent to each participant for model verification. If the model converges in the test set of each participant, the model training is terminated. Otherwise, the next round of global model training is carried out and the above training process is repeated.

[0137] 4) Result prediction: The prediction result is as follows Fig.11 As shown in the figure, the accuracy of the model's prediction results is calculated, and the final mean square error (MSE) is 1468.65.

[0138] This embodiment provides a building economic forecasting method based on the Internet of Things, which combines the building Internet of Things operation data and economic development data to predict the building economy. The Internet of Things data generated during the operation of the building is added to the model as a factor affecting economic development, and the sudden events occurring in the building are modeled through the knowledge graph, thereby improving the sensitivity of predicting economic mutations. In addition, this knowledge-driven approach can encode explainable knowledge into the deep prediction model, making the prediction results explainable. The time series prediction of the sequence adopts the ARIMA model with better effect in the traditional time series analysis model for prediction, and the nonlinear part adopts the LSTM model with better long-term series prediction effect for prediction. In addition, under the condition of satisfying data privacy security, the model is trained based on the economic data of multiple enterprises through blockchain and federated learning methods. The model can combine the advantages of the two models to achieve a more accurate prediction of the economic development of enterprises.

[0139] Embodiment 2:

[0140] In combination with the building economic forecasting method based on the Internet of Things disclosed in Example 1, this embodiment discloses a specific implementation example of a building economic forecasting system based on the Internet of Things (hereinafter referred to as the "system").

[0141] Reference Fig.12 As shown, the system comprises:

[0142] Influencing factor data acquisition unit 1: collects Internet data, Internet of Things data and enterprise economic data related to the building, calculates the correlation between the lagged Internet of Things data and the enterprise economic data, selects the lagged data with the largest correlation, and pre-processes the Internet data, the Internet of Things data, the enterprise economic data and the lagged data into a first training set and a first test set;

[0143] ARIMA model prediction unit 2: using the first training set to train the ARIMA model and using the first test set to perform time series prediction on the IoT data, and obtaining a prediction result of each influencing factor in the IoT data as a prediction result of the ARIMA model;

[0144] Event data acquisition unit 3: acquires event data of the building, obtains event embedding of the building using the building knowledge graph according to the event data, and uses the multi-channel link vector composed of the prediction result of the ARIMA model and the event embedding as the second test set;

[0145] LSTM model prediction unit 4: Use the multi-channel link vector composed of the first training set and the event embedding and the current time as the second training set, train the LSTM model according to the second training set based on blockchain and federated learning, obtain the enterprise economic prediction model, and use the enterprise economic prediction model to predict the future economy of the building according to the second test set.

[0146] For the technical solutions of the other identical parts of the building economic forecasting system based on the Internet of Things disclosed in this embodiment and the building economic forecasting method based on the Internet of Things disclosed in the first embodiment, please refer to the description in the first embodiment, which will not be repeated here.

[0147] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0148] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A building economic forecasting method based on the Internet of Things, characterized in that: include: Influencing factor data acquisition step: collecting Internet data, Internet of Things data and enterprise economic data related to the building, and calculating the correlation between the lagged Internet of Things data and the enterprise economic data, selecting the lagged data with the largest correlation, and pre-processing the Internet data, the Internet of Things data, the enterprise economic data and the lagged data into a first training set and a first test set; ARIMA model prediction step: using the first training set to train the ARIMA model and using the first test set to perform time series prediction on the IoT data, and obtaining a prediction result of each influencing factor data in the IoT data as a prediction result of the ARIMA model; Event data acquisition step: acquiring event data of a building, obtaining event embedding of the building using the building knowledge graph according to the event data, and using the multi-channel link vector composed of the prediction result of the ARIMA model and the event embedding as the second test set; LSTM model prediction step: taking the multi-channel link vector composed of the first training set and the event embedding and the current time as the second training set, training the LSTM model according to the second training set based on blockchain and federated learning, obtaining an enterprise economic prediction model, and using the enterprise economic prediction model to predict the future economy of the building according to the second test set; Wherein, the event data acquisition step includes: Event tuple acquisition step: acquiring event data of the building, and extracting event tuple (S, P, O) from the event data, where S represents the event generator, P represents the action that occurs, and O represents the target to which the event is executed; Event embedding acquisition step: The event subgraph is obtained by linking the event tuple to the building knowledge graph, and the event subgraph is embedded using the TransE method to obtain the event embedding of the building. The event features of the event subgraph are parameterized into different channels. V1 represents the link channel of KG, V2 represents the context channel of KG, and V3 represents the word vector channel. The final event embedding V=[V1V2V3] is obtained by linking these three channels.

2. The building economic forecasting method according to claim 1, characterized in that: The ARIMA model prediction steps include: ARIMA model establishment step: establishing the ARIMA model, verifying the ARIMA model according to the AIC criterion based on each influencing factor data in the IoT data and obtaining the optimal parameters corresponding to each influencing factor data, and obtaining the optimal ARIMA model corresponding to each influencing factor data respectively; The step of obtaining the prediction result of the ARIMA model is as follows: the optimal ARIMA model is trained by the first training set, and each influencing factor data is predicted respectively by the optimal ARIMA model using the first test set to obtain the prediction result of the ARIMA model corresponding to each influencing factor data.

3. The building economic forecasting method according to claim 2, characterized in that: The LSTM model prediction step includes: Data conversion step: normalizing the second training set and the second test set; Model structure determination step: initializing the parameters of the LSTM model; Model training step: the enterprise economic forecasting model is obtained by training the LSTM model using the normalized second training set, and the prediction result of the enterprise economic forecasting model is obtained according to the normalized second test set.

4. The building economic forecasting method according to claim 3, characterized in that: The LSTM model prediction step also includes: calculating the accuracy of the prediction result of the enterprise economic prediction model by comparing the prediction result of the enterprise economic prediction model with the true value of the enterprise economic data.

5. The building economic forecasting method according to claim 3, characterized in that: The model training step includes: Local model training step: using the normalized second training set to train the local LSTM model, and uploading the trained local LSTM model parameters to the blockchain through a smart contract; Global model parameter acquisition step: monitoring and aggregating the local LSTM model parameters through the smart contract, obtaining global model parameters and storing them in the blockchain; Enterprise economic forecasting model acquisition step: determine whether the global model has converged based on the global model parameters. If not, send the global model parameters from the blockchain to each participant to update the local LSTM model parameters, continue iterative training until the global model converges, and use the converged global model as the enterprise economic forecasting model.

6. The building economic forecasting method according to claim 3, characterized in that: The Internet data includes: the building's operating income, the building's news and public opinion information and favorable information; the Internet of Things data includes: parking lot vehicle entry and exit data, enterprise electricity consumption data and enterprise access control personnel information data.

7. The building economic forecasting method according to claim 1, characterized in that: The step of acquiring the influencing factor data includes: removing abnormal points from the Internet data, the Internet of Things data, the enterprise economic data and the lagged data by using the 3 sigma principle, and pre-processing them by using the multiple interpolation method to supplement missing values.

8. The building economic forecasting method according to claim 6, characterized in that: The model structure determination step includes: determining the activation function of the LSTM model, determining the fully connected artificial neural network that receives the output of the LSTM model, determining the discard rate of each layer of network nodes, determining the error calculation method, determining the iterative update method of the weight parameters, and determining the epoch and batch size of the model training.

9. A building economic forecasting system based on the Internet of Things, characterized in that: Applying the building economic forecasting method according to any one of claims 1 to 8, the building economic forecasting system comprises: An influencing factor data acquisition unit is used to collect Internet data, Internet of Things data and enterprise economic data related to the building, and calculate the correlation between the lagged Internet of Things data and the enterprise economic data, select the lagged data with the largest correlation, and pre-process the Internet data, the Internet of Things data, the enterprise economic data and the lagged data into a first training set and a first test set; ARIMA model prediction unit: using the first training set to train the ARIMA model and using the first test set to perform time series prediction on the IoT data, to obtain a prediction result of each influencing factor data in the IoT data as a prediction result of the ARIMA model; An event data acquisition unit: acquires event data of a building, obtains event embedding of the building using a building knowledge graph according to the event data, and uses a multi-channel link vector composed of the prediction result of the ARIMA model and the event embedding as a second test set; LSTM model prediction unit: take the multi-channel link vector composed of the first training set and the event embedding and the current time as the second training set, train the LSTM model according to the second training set based on blockchain and federated learning, obtain the enterprise economic prediction model, and use the enterprise economic prediction model to predict the future economy of the building according to the second test set.

Images