BIM-based medical equipment energy consumption prediction model construction method

Abstract

The application discloses a BIM-based medical equipment energy consumption prediction model construction method, relates to the technical field of medical equipment energy consumption prediction, and comprises the following steps: obtaining medical equipment in a region based on a related BIM model, collecting equipment power time series data and equipment related time series data of the medical equipment, and performing data preprocessing and data correlation analysis to obtain node time series data and node correlation data; an energy consumption prediction model is constructed based on a space-time graph neural network; the energy consumption of the medical equipment is predicted according to the energy consumption prediction model, and reliability analysis is performed to obtain a medical equipment energy consumption prediction result; the application is used to solve the problem that the existing medical equipment energy consumption prediction technology cannot establish a unified energy consumption prediction model for multiple medical equipments and simultaneously predict the energy consumption of the multiple medical equipments according to the correlation between the multiple medical equipments and related parameters.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical equipment energy consumption prediction, in particular to a medical equipment energy consumption prediction model construction method based on BIM. BACKGROUND

[0002] Medical equipment energy consumption prediction technology refers to a technology system for quantitatively estimating the future energy consumption of various medical equipment such as diagnosis and treatment equipment, life support equipment, and auxiliary equipment in the medical field based on multi-source information such as medical equipment operation data, environmental data, business data, and equipment static properties, through data preprocessing, feature engineering, and machine learning methods, while ensuring the continuity of medical services and the reliability of equipment operation.

[0003] The existing medical equipment energy consumption prediction technology often predicts the energy consumption of each medical equipment separately when predicting the energy consumption of multiple medical equipment. Technical personnel will first select relevant parameters that only adapt to the target medical equipment in combination with its own operating characteristics and energy consumption influencing factors. These parameters may include medical business data and local environmental data associated with the equipment. On this basis, a dedicated prediction algorithm and model are designed and constructed for each device. This separate modeling for each device has high modeling and maintenance costs, and technical personnel need to repeatedly invest a large amount of manpower and time, which significantly lengthens the modeling period. During maintenance, the algorithm or model of each device needs to be updated separately, and the maintenance cost increases linearly with the number of devices, making it difficult to manage on a large scale. Moreover, the energy consumption of multiple devices in the medical field is not completely independent, but has explicit or implicit correlations, such as devices in the same department sharing the same power distribution circuit and being affected by the same environmental temperature. Separate modeling ignores this correlation: it cannot use the energy consumption changes of associated devices to assist prediction, has weak anti-interference ability, and has low prediction accuracy. Therefore, the existing medical equipment energy consumption prediction technology cannot establish a unified energy consumption prediction model for multiple medical equipment based on the correlation between multiple medical equipment and related parameters when predicting the energy consumption of multiple medical equipment, and simultaneously predict the energy consumption of multiple medical equipment. SUMMARY

[0004] The present application aims to at least solve one of the technical problems in the prior art, by acquiring medical equipment in the region based on the relevant BIM model, and collecting device power time series data and device related time series data of the medical equipment; data preprocessing is performed on the device power time series data and the device related time series data, and data correlation analysis is performed to obtain node time series data and node correlation data; based on the space-time graph neural network, an energy consumption prediction model is constructed; the energy consumption of the medical equipment is predicted according to the energy consumption prediction model, and reliability analysis is performed to obtain the energy consumption prediction result of the medical equipment; to solve the problem that the existing medical equipment energy consumption prediction technology cannot establish a unified energy consumption prediction model for multiple medical equipment and simultaneously predict the energy consumption of multiple medical equipment according to the correlation between multiple medical equipment and related parameters when predicting the energy consumption of multiple medical equipment.

[0005] To achieve the above-mentioned purpose, the present application provides a BIM-based medical equipment energy consumption prediction model construction method, comprising the following steps:

[0006] Acquiring medical equipment in the region based on the relevant BIM model, and collecting device power time series data and device related time series data of the medical equipment;

[0007] Data preprocessing is performed on the device power time series data and the device related time series data, and data correlation analysis is performed to obtain node time series data and node correlation data;

[0008] Based on the space-time graph neural network, and using the node time series data and the node correlation data to construct an energy consumption prediction model;

[0009] According to the energy consumption prediction model, the energy consumption of the medical equipment is predicted, and reliability analysis is performed to obtain the energy consumption prediction result of the medical equipment.

[0010] Further, acquiring medical equipment in the region based on the relevant BIM model, and collecting device power time series data and device related time series data of the medical equipment comprises the following sub-steps:

[0011] Mark the region to be predicted as an energy consumption prediction region, obtain the BIM of the energy consumption prediction region, and mark it as a prediction region model; obtain all medical equipment in the region to be predicted according to the prediction region model, and mark them in turn as medical equipment 1-medical equipment n1; and mark any one medical equipment as a first device;

[0012] Obtain the environmental parameters and medical business parameters related to the energy consumption of the first device, and mark them as the relevant parameters corresponding to the first device; repeat the acquisition of the relevant parameters corresponding to all medical equipment, and perform deduplication to obtain a relevant parameter collection, and mark various types of parameters in the relevant parameter collection as relevant parameter 1-relevant parameter n1.

[0013] Further, the medical equipment in the region is acquired based on the relevant BIM model, and the device power time sequence data and the device related time sequence data of the medical equipment further include the following sub-steps:

[0014] The input power of each medical equipment is synchronously collected at a first time interval, and the collection time is recorded. According to the corresponding medical equipment classification, it is sorted in time sequence, and it is recorded as the time sequence data of the input power of the corresponding medical equipment. After completion, the power time sequence data is obtained, wherein the first time interval is t1;

[0015] At the same time, the parameter values of each related parameter are synchronously collected at a first time interval, and the collection time is recorded. According to the corresponding related parameter classification, it is sorted in time sequence, and it is recorded as the time sequence data of the corresponding related parameter. After completion, the device related time sequence data is obtained.

[0016] Further, the device power time sequence data and the device related time sequence data are preprocessed, and the data correlation analysis is performed to obtain the node time sequence data and the node correlation data, which include the following sub-steps:

[0017] The input power of the medical equipment 1 to the medical equipment n1 is recorded as the feature node 1 to the feature node n1 in turn; and the related parameters 1 to the related parameters n1 are recorded as the feature nodes (n1+1) to the feature nodes (n1+n2) in turn, and are merged with the feature nodes 1 to the feature nodes n1 to be recorded as the feature nodes 1 to the feature nodes n0, wherein n0=n1+n2;

[0018] Any one of the feature nodes is recorded as a first node, the time sequence data corresponding to the first node is obtained and recorded as a first parameter sequence, the mean and the standard deviation of the first parameter sequence are calculated and recorded as AP and AB respectively;

[0019] Any one data in the first parameter sequence is recorded as a first value; taking the first value as the center, the first value and the k1 nearest data before and after the first value are taken to form a local window of the first value, wherein k1 is the number set;

[0020] The coefficient of variation of the data in the local window of the first value is calculated and recorded as CV1; and the dynamic adjustment coefficient k0 corresponding to the first value is calculated, wherein k0=k2+min{(k3-k2),[(k3-k2) / CV0]*CV1}, wherein [k2, k3] is the value range of the set dynamic adjustment coefficient, k0∈[k2, k3], and CV0 is the set coefficient of variation threshold.

[0021] Further, the device power time sequence data and the device related time sequence data are preprocessed, and the data correlation analysis is performed to obtain the node time sequence data and the node correlation data, which include the following sub-steps:

[0022] If the first value does not belong to [AP-k0*AB, AP+k0*AB], it is marked as abnormal data; otherwise, it is marked as normal data.

[0023] If the first data is abnormal, then obtain the k4 nearest normal data before and after the first value, and calculate the weighted average of all the normal data based on the positional distance between each normal data and the first data, and replace the first data, where k4 is the set number;

[0024] Repeatedly obtain all abnormal data in the first parameter sequence and replace them to obtain the denoised time series data of the first node; repeat the process to obtain the denoised time series data of all feature nodes.

[0025] Let any one of the feature nodes from feature node 1 to feature node n1 be denoised as the first power node, and obtain the denoised time series data of the first power node, which is denoised as the first power sequence.

[0026] Set the size of the first window to k5, where k5 is an odd number. Denote any data point in the first power sequence as the first power point. Using the first power point as the center of the first window, calculate the average value of all data in the first window and replace the first power. Repeat the replacement process for all data in the first power sequence. After completion, the smoothed time series data of the first power node is obtained.

[0027] Repeatedly acquire the smoothed time series data of feature node 1 to feature node n1, and merge the denoised time series data of feature node (n1+1) to feature node (n1+n2) and record them as node time series data.

[0028] Furthermore, the process of preprocessing the equipment power time-series data and related equipment time-series data, and performing data correlation analysis to obtain node time-series data and node correlation data includes the following sub-steps:

[0029] Smoothed and denoised time series data in node time series data are collectively referred to as feature time series data; any two feature nodes are referred to as node combination, and are respectively referred to as the first node and the second node; the feature time series data corresponding to the first node and the second node are obtained respectively, and are referred to as the first sequence and the second sequence respectively;

[0030] Set the sliding window size to k6 and the sliding step size to 1. Start sliding synchronously from the starting position of the first sequence and the second sequence respectively. Record the two partial sequences in the two sliding windows each time as subsequence groups, and record all subsequence groups as subsequence group 1 - subsequence group n3 in sequence.

[0031] Any one of the subsequence groups is denoted as a first sub-group, and the partial sequences of the first sequence and the second sequence in the first sub-group are denoted as a first subsequence and a second subsequence respectively;

[0032] The first subsequence and the second subsequence are standardized by using the z-score method respectively, and the Pearson correlation coefficient of the standardized first subsequence and the standardized second subsequence is calculated, denoted as a fluctuation correlation degree AR1 of the first sub-group.

[0033] Further, the device power time series data and the device related time series data are pre-processed, and data correlation analysis is performed to obtain the node time series data and the node correlation data, which further includes the following sub-steps:

[0034] The first subsequence and the second subsequence are linearly fitted respectively, and the slopes of the straight lines obtained by linear fitting are obtained respectively, denoted as KX1 and KX2 in order respectively; a trend correlation degree AR2 of the first sub-group is calculated, wherein AR2 = sign (KX1*KX2) * min (|KX1|, |KX2|) / max (|KX1|, |KX2|), and sign () is a sign function;

[0035] The coefficients of variation of the first subsequence and the second subsequence are calculated respectively, and the reciprocal of the average of the two coefficients of variation is calculated, denoted as a weight coefficient VQ of the first sub-group; the weight coefficients of all subsequence groups are repeatedly obtained and summed, denoted as a weight coefficient sum HQ;

[0036] A comprehensive correlation degree HR of the first sub-group is calculated, wherein HR = (VQ / HQ) * [ (AR1+AR2) / 2 ]; the comprehensive correlation degrees of all subsequence groups are repeatedly obtained and summed, and then the absolute values are taken, denoted as a node correlation of the first node and the second node;

[0037] The node correlations of all node combinations are repeatedly obtained, denoted as node correlation data.

[0038] Further, based on the space-time graph neural network, and using the node time series data and the node correlation data to construct an energy consumption prediction model includes the following sub-steps:

[0039] An adjacency matrix is constructed according to the node correlation data, denoted as A = (a ij ) n0*n0 , wherein a ij is the node correlation of the feature node i and the feature node j, and a ij =1 when i=j; i∈[0,n0];

[0040] And according to the node time series data, the data of all feature nodes at the same collection time are constructed with the adjacency matrix to construct a space-time feature matrix at each collection time, and model training data is obtained;

[0041] The spatio-temporal graph neural network is denoted as an initial prediction model, the model output is set as the input power of the medical device 1-the medical device n1, the initial prediction model is trained by using the model training data, and the energy consumption prediction model is obtained after the training.

[0042] Further, the energy consumption of the medical device is predicted according to the energy consumption prediction model, and reliability analysis is performed to obtain the medical device energy consumption prediction result, including the following sub-steps:

[0043] The input power of the medical device 1-the medical device n1 at each time in the future first time period is predicted by using the energy consumption prediction model, and the predicted input power data of all medical devices in the first time period is obtained;

[0044] The date type of the date in the first time period is obtained, denoted as the first type, and the date type includes weekdays and weekends;

[0045] The input power data of the first device in the first time period is arranged in chronological order, denoted as the first prediction sequence, and any data in the first prediction sequence is denoted as the first predicted power. The time corresponding to the first predicted power is denoted as the first time;

[0046] The k7 nearest dates of the first type before the date in the first time period are obtained, denoted as reference dates, the actual input power of the first device at the first time of each reference date is obtained, denoted as a reference power set, the mean and standard deviation of the reference power set are calculated, denoted as CP and CB in order, wherein k7 is the number of settings.

[0047] Further, the energy consumption of the medical device is predicted according to the energy consumption prediction model, and reliability analysis is performed to obtain the medical device energy consumption prediction result, including the following sub-steps:

[0048] If the first predicted power is located in [CP-k8*CB, CP-k8*CB], it is marked as reliable data, otherwise it is marked as fluctuation data; all reliable data in the first prediction sequence are repeatedly obtained, and the number ratio of the reliable data is calculated, denoted as the reliable coefficient of the first device, wherein k8 is a set proportion coefficient;

[0049] The reliable coefficients of all medical devices are repeatedly obtained, and the average value is calculated, denoted as the overall reliable coefficient;

[0050] If the overall reliable coefficient is less than E1, it is judged that the predicted input power data of all medical devices in the first time period is unreliable;

[0051] If the overall reliable coefficient is not less than E1, but the reliable coefficient of the first device is less than E2, it is judged that the predicted input power data of the first device in the first time period is unreliable;

[0052] If the overall reliability coefficient is not less than E1, and the reliability coefficient of the first device is not less than E2, it is judged that the predicted input power data of the first device in the first time period is reliable;

[0053] If the predicted input power data of the first device in the first time period is reliable, the energy consumption of the first device in the first time period is calculated according to the corresponding predicted input power data;

[0054] The energy consumption of all medical devices whose predicted input power data in the first time period is reliable is repeatedly obtained, and the medical device energy consumption prediction result is obtained.

[0055] The present application has the following advantages: the present application obtains medical devices in a region based on a related BIM model, and collects device power time sequence data and device related time sequence data of the medical devices; the device power time sequence data and the device related time sequence data are preprocessed and analyzed for data correlation to obtain node time sequence data and node correlation data; an energy consumption prediction model is constructed based on a space-time graph neural network and using the node time sequence data and the node correlation data; the energy consumption of the medical devices is predicted according to the energy consumption prediction model, and a reliability analysis is performed to obtain a medical device energy consumption prediction result; when the energy consumption of multiple medical devices is predicted, a unified energy consumption prediction model of the multiple medical devices can be established according to the correlation and related parameters among the multiple medical devices, and the energy consumption of the multiple medical devices can be accurately predicted;

[0056] The present application uses the mean and standard deviation of the first parameter sequence, and dynamically adjusts the coefficient according to the coefficient of variation of the window data, so that the threshold value is self-adaptively adjusted with the sequence volatility; compared with the fixed threshold value, the high volatility data and the real abnormal data can be better distinguished, the misjudgment and excessive correction are reduced, and the true dynamic characteristics of the signal are protected; by calculating the Pearson correlation coefficient of the sub-sequence group and the trend correlation based on the linear fitting slope, and then combining the weighted comprehensive correlation degree obtained by the coefficient of variation, the correlation of different nodes is obtained, and the advantage is that the single correlation coefficient will lose the trend information; the node correlation constructed by the double indexes reflects not only the instantaneous collaborative fluctuation, but also the long-term collaborative trend, and can more truly describe the correlation degree of the nodes. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 The step flowchart of the method of the present application is shown;

[0058] Figure 2 The data correlation analysis flowchart of the present application is shown;

[0059] Figure 3 The adjacency matrix diagram of the present application is shown;

[0060] Figure 4A structural schematic diagram of an electronic device of the present application. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0062] Embodiment 1, please refer to Figure 1 As shown in the figure, the present application provides a BIM-based medical equipment energy consumption prediction model construction method, including the following steps:

[0063] Step S1, based on the related BIM model, the medical equipment in the region is obtained, and the equipment power time series data and the equipment related time series data of the medical equipment are collected; Step S1 includes the following sub-steps:

[0064] Step S101, mark the area to be predicted as an energy consumption prediction area, obtain the BIM of the energy consumption prediction area, mark it as a prediction area model; according to the prediction area model, obtain all the medical equipment in the area to be predicted, in turn mark them as medical equipment 1-medical equipment n1; and mark any one medical equipment as a first device; the range of the energy consumption prediction area can be set according to the actual application scene, which can be large or small, for example, a certain operation area or the entire outpatient area; BIM is a digital model containing device location, type, interface and attributes, which automatically obtains the device list, location and building association by using BIM, avoiding manual omission;

[0065] Step S102, obtain the environmental parameters and medical business parameters related to the energy consumption of the first device, mark them as the related parameters corresponding to the first device; for example, indoor temperature, indoor humidity, outdoor temperature, patient number, etc.; as long as the parameters related to the energy consumption of the device can be recorded and collected; repeat the acquisition of the related parameters corresponding to all medical devices, and remove the duplicates to obtain a related parameter collection, and various types of parameters in the related parameter collection are in turn marked as related parameter 1-related parameter n1; for example, the related parameters of device 1 and the related parameters of device 2 both include the indoor temperature of a certain area, only one needs to be kept, and there is no need to repeat the collection;

[0066] Step S103, synchronously collect the input power of each medical equipment at a first time interval, record the collection time, sort according to the time sequence of the corresponding medical equipment, mark it as the time series data of the input power of the corresponding medical equipment, and obtain the power time series data after completion, wherein the first time interval is t1; in this embodiment, t1=10 seconds, which can be flexibly adjusted according to the actual application scene;

[0067] Step S104, synchronously collect the parameter values corresponding to each relevant parameter at a first time interval, record the collection time, sort according to the corresponding relevant parameters in time sequence, and record as the time sequence data of the corresponding relevant parameters, and obtain the device-related time sequence data after completion;

[0068] In the specific implementation process, the environmental parameter is an external influencing factor of the energy consumption of the medical device, mainly by increasing the auxiliary energy consumption of the device, such as temperature control, dehumidification, and filtration, indirectly affecting the total energy consumption; the medical business parameter is the core driving factor of the energy consumption, which directly dominates the energy consumption level by determining the actual use intensity and length of the device; the two jointly determine the final energy consumption performance of the medical device.

[0069] Step S2, data preprocessing is performed on the device power time sequence data and the device-related time sequence data, and data correlation analysis is performed to obtain node time sequence data and node correlation data; step S2 includes the following substeps:

[0070] Step S201, the input power of the medical device 1 to the medical device n1 is sequentially recorded as the feature node 1 to the feature node n1; and the relevant parameters 1 to the relevant parameters n1 are sequentially recorded as the feature node (n1+1) to the feature node (n1+n2), and are combined with the feature node 1 to the feature node n1, recorded as the feature node 1 to the feature node n0, wherein n0=n1+n2;

[0071] Step S202, any one feature node is recorded as a first node, the time sequence data corresponding to the first node is obtained and recorded as a first parameter sequence, the mean and standard deviation of the first parameter sequence are calculated and recorded as AP and AB, respectively;

[0072] Step S203, any data in the first parameter sequence is recorded as a first value; taking the first value as the center, taking the first value and the k1 nearest data before and after the first value, and forming a local window of the first value, wherein k1 is the number of settings; in this embodiment, k1=15, that is, 31 data centered on the first value form the corresponding local window; k1 can be flexibly adjusted.

[0073] Step S204, calculate the coefficient of variation of the data in the local window of the first numerical value, denoted as CV1; and calculate the dynamic adjustment coefficient k0 corresponding to the first numerical value, wherein k0=k2+min{(k3-k2),[(k3-k2) / CV0]*CV1}, wherein [k2, k3] is the value range of the set dynamic adjustment coefficient, k0∈[k2, k3], CV0 is the set coefficient of variation threshold; in this embodiment, [k2, k3] is [2, 3.5] and CV0 is 0.4, that is, when CV1 is greater than or equal to 0.4, k0 takes the maximum value, that is, k0=k3; when CV1 is less than 0.4, k0 changes linearly with CV1;

[0074] Step S205, if the first numerical value does not belong to [AP-k0*AB, AP+k0*AB], it is marked as abnormal data, otherwise it is marked as normal data; time series often has periodic fluctuations, and if the mean and standard deviation of the whole sequence are used as the judgment standard, it will lead to the mistake of judging normal data as abnormal in the stage with small fluctuations, and the omission of abnormal data in the stage with large fluctuations; the global mean and standard deviation combine with the coefficient of variation of the local window to dynamically determine the threshold width, which can distinguish between high fluctuation sequences and low fluctuation sequences, and reduce the risk of regarding normal high fluctuation as abnormal or ignoring real abnormality.

[0075] Step S206, if the first data is abnormal data, then get the k4 nearest normal data before and after the first numerical value, and calculate the weighted average value of all normal data according to the position distance of each normal data and the first data, which can use the reciprocal of the position distance as the weight, and replace the first data, wherein k4 is the set number; in this embodiment, k4 is equal to 2, which can be flexibly set, for example, using normal data 1, data 2, data 4, data 5 to replace data 3, then the weighted average value is=[(1 / 2)*data1+data2+data4+(1 / 2)*data5] / [(1 / 2)+1+1+(1 / 2)];

[0076] Step S207, repeat to get all abnormal data in the first parameter sequence and replace it, and get the denoised time series data of the first node after completion; repeat to get the denoised time series data of all feature nodes;

[0077] Step S208, any one of the feature nodes 1-feature nodes n1 is marked as the first power node, and the denoised time series data of the first power node is obtained, denoted as the first power sequence; the feature nodes 1-feature nodes n1 are the time series data of the input power of each medical device.

[0078] Step S209, set the first window size as k5, k5 is an odd number, take any one data in the first power sequence as the first power point; take the first power point as the center of the first window, calculate the average value of all data in the first window, and replace the first power; repeat the replacement of all data in the first power sequence, and the smooth time sequence data of the first power point is obtained after completion; in this embodiment, k5=5; because the power consumption will be suddenly changed when the medical device starts and stops, the smoothing processing of the time sequence data of the input power can smooth the sudden energy consumption of the medical device start and stop, and avoid the prediction shock;

[0079] Step S210, repeat the smooth time sequence data of the feature node 1-feature node n1, and combine the de-noising time sequence data of the feature node (n1+1)-feature node (n1+n2) to be recorded as node time sequence data.

[0080] Step S211, the smooth time sequence data and the de-noising time sequence data in the node time sequence data are uniformly recorded as feature time sequence data; it is convenient for subsequent description; any two feature nodes are recorded as node combination, and are recorded as first node and second node respectively; the feature time sequence data corresponding to the first node and the second node is obtained respectively, and is recorded as first sequence and second sequence respectively;

[0081] Step S212, please refer to Figure 2 The size of the sliding window is set to k6, and the sliding step is 1. The two parts of the sequence in the two sliding windows are recorded as sub-sequence groups, and all the sub-sequence groups are recorded as sub-sequence groups 1-sub-sequence groups n3 in turn; in this embodiment, k6=20, which can be flexibly adjusted;

[0082] Step S213, take any one sub-sequence group as the first sub-group, and record the part of the first sequence and the second sequence in the first sub-group as the first sub-sequence and the second sub-sequence respectively;

[0083] Step S214, the first sub-sequence and the second sub-sequence are standardized by using z-score method respectively, the data is converted into dimensionless relative fluctuation value, and the fluctuation amplitude of different orders of magnitude parameters is comparable; and the Pearson correlation coefficient of the standardized first sub-sequence and the standardized second sub-sequence is calculated, recorded as the fluctuation correlation degree AR1 of the first sub-group, which is used to capture local and immediate correlation;

[0084] Step S215, linear fitting is performed on the first subsequence and the second subsequence respectively, and the slopes of the straight lines obtained by linear fitting are respectively denoted as KX1 and KX2 in order; the trend correlation AR2 of the first sub-group is calculated, where AR2 = sign (KX1*KX2) * min (|KX1|, |KX2|) / max (|KX1|, |KX2|), and sign () is a sign function; the traditional linear correlation method, such as Pearson coefficient, has weak ability to describe the correlation of nonlinear trend, and AR2 is used to capture the local trend correlation and focus on the long-term trend characteristics in the local interval, and accurately describe the trend correlation between parameters.

[0085] Step S216, the coefficients of variation of the first subsequence and the second subsequence are calculated respectively, and the inverse of the average of the two coefficients of variation is calculated, denoted as the weight coefficient VQ of the first sub-group; the weight coefficients of all subsequence groups are repeatedly obtained and summed, denoted as the weight coefficient sum HQ; the smaller the coefficient of variation, the more stable the data, and the greater the weight;

[0086] Step S217, the comprehensive correlation HR of the first sub-group is calculated, where HR = (VQ / HQ) * [ (AR1+AR2) / 2 ]; the comprehensive correlation of all subsequence groups is repeatedly obtained and summed, and the absolute value is taken, denoted as the node correlation of the first node and the second node; HR combines the short-term fluctuation consistency and the long-term trend consistency, which can realize more comprehensive and more robust correlation evaluation; the greater the comprehensive correlation, the stronger the correlation between the two feature nodes, and some feature nodes obviously have no correlation, for example, the device input power of room 1 and the indoor humidity of room 5, which can not be calculated node correlation, and the node correlation is directly set to zero;

[0087] Step S218, the node correlation of all node combinations is repeatedly obtained and denoted as node correlation data;

[0088] In the specific implementation process, the traditional correlation analysis, such as Pearson coefficient, calculates a single global correlation value using the full sequence data, which can easily hide the situation that the local strong correlation is weakly correlated globally or the early positive correlation is negatively correlated later. The multiple subsequence groups obtained by sliding window splitting are to split the full sequence into local fragments to obtain different correlation degrees in different stages, so as to avoid that the key correlation information is hidden by the global correlation value; the correlation between different parameters is accurately captured; and the accuracy of subsequent prediction is improved.

[0089] Step S3, based on the space-time graph neural network, and using the node time series data and the node correlation data to construct an energy consumption prediction model; step S3 includes the following sub-steps:

[0090] Step S301, please refer to Figure 3 shown in Figure 3The values in the table represent the node relevance of the corresponding feature nodes. An adjacency matrix is constructed according to the node relevance data, denoted as A = (a ij ) n0*n0 where a ij is the node relevance of feature node i and feature node j, a ij = 1 when i = j, that is, the node relevance of a feature node to itself; i ∈ [0, n0]; the node relevance represents the influence strength or connection strength between any two feature nodes; organizing these values into an adjacency matrix encodes the originally dispersed mutual relationship into a graph, which is the basic structure that can be utilized by the spatio-temporal graph neural network;

[0091] In step S302, according to the node time sequence data, the data of all feature nodes at the same collection time are combined with the adjacency matrix to construct a spatio-temporal feature matrix at each collection time, and model training data are obtained; at a time, the feature values of all feature nodes at the same collection time form a spatial slice, and the adjacency matrix gives the spatial topology of the slice; connecting these slices in time gives a spatio-temporal feature matrix, which is used for the spatio-temporal graph neural network to learn the coupling relationship between time sequence evolution and graph propagation; a long sequence can be cut into multiple groups of samples through a sliding window, which are used for subsequent training of the method.

[0092] In step S303, the spatio-temporal graph neural network is denoted as an initial prediction model, the model output is set as the input power of the medical device 1 to the medical device n1, the initial prediction model is subjected to model training using the model training data, and after completion, an energy consumption prediction model is obtained;

[0093] In the specific implementation process, the node relevance constitutes the topology of the graph, which is represented by an adjacency matrix; the node time sequence data give the observation values of all nodes at each time point, which are used as the feature matrix at the time level and the time sequence, that is, the spatial dimension and the time dimension; the spatio-temporal graph neural network learns the interaction between nodes and the time dynamics by performing spatial convolution on the graph structure and combining a time modeling module, thereby predicting the power of the medical device.

[0094] In step S4, the energy consumption of the medical device is predicted according to the energy consumption prediction model, and reliability analysis is performed, and a medical device energy consumption prediction result is obtained; step S4 includes the following sub-steps:

[0095] In step S401, the input power of the medical device 1 to the medical device n1 at each time in the future first time period is predicted using the energy consumption prediction model, and prediction input power data of all medical devices in the first time period are obtained; the first time period can be set according to the actual application scenario;

[0096] Step S402, a date type of a date in which the first time period is located is obtained, denoted as a first type, and the date type includes weekdays and rest days; holidays and weekends are rest days, and weekdays from Monday to Friday in a normal state are weekdays, and medical institutions usually have significantly different equipment usage modes and load curves on weekdays and rest days; taking the date type as a grouping is helpful for reference statistics with more comparable historical data;

[0097] Step S403, input power data of the first device in the first time period is arranged in time sequence, denoted as a first prediction sequence, and for any data in the first prediction sequence, denoted as a first prediction power; a time corresponding to the first prediction power is denoted as a first time;

[0098] Step S404, the most recent k7 dates of the first type before the date in which the first time period is located are obtained, denoted as reference dates, actual input power of the first device at the first time of each reference date is obtained, denoted as a reference power set, a mean value and a standard deviation of the reference power set are calculated, and are denoted as CP and CB in order, wherein k7 is a set number. In this embodiment, k7=30; that is, data in the past month is taken as a reference;

[0099] Step S405, if the first prediction power is located in [CP-k8*CB, CP+k8*CB], it is marked as reliable data, otherwise it is marked as fluctuation data; all reliable data in the first prediction sequence are repeatedly obtained, and a number ratio of the reliable data is calculated, denoted as a reliable coefficient of the first device, wherein k8 is a set proportion coefficient; in this embodiment, k8=4, which can be flexibly set; the same type of date, that is, historical observation data at the same time of the same date type, is used to estimate a typical value and fluctuation of normal power at the time; the same time is compared because most of the loads of medical devices have clear periodic characteristics; [CP-k8*CB, CP+k8*CB] is equivalent to an acceptable interval based on historical average and fluctuation; if the prediction value falls within the interval, it means that the prediction value is consistent with the historical normality, and the reliability is high; if it exceeds, it belongs to an atypical situation, and is marked as fluctuation data.

[0100] Step S406, the reliable coefficients of all medical devices are repeatedly obtained, and an average value is calculated, denoted as an overall reliable coefficient; the overall reliable coefficient indicates the overall reliability of the model in predicting the entire device group in the first time period; if the overall is very low, it means that the model may be generally inaccurate in this time period, such as a sudden event or systematic error, and the prediction of any single device is not trustworthy;

[0101] Step S407, if the overall reliable coefficient is less than E1, it is judged that the prediction input power data of all medical devices in the first time period is unreliable; in this embodiment, E1=0.8, which can be flexibly set;

[0102] Step S408, if the overall reliability coefficient is not less than E1, but the reliability coefficient of the first device is less than E2, it is judged that the predicted input power data of the first device in the first time period is unreliable; in the embodiment, E1 = 0.7, which can be flexibly set;

[0103] Step S409, if the overall reliability coefficient is not less than E1, and the reliability coefficient of the first device is not less than E2, it is judged that the predicted input power data of the first device in the first time period is reliable; first, the overall threshold E1 is used to judge whether the model is reliable in the time period, and then the device threshold E2 is used for device-level screening; such design can quickly eliminate all outputs of the model in extreme cases, and can also retain reliable predictions of single devices in detail;

[0104] Step S410, if the predicted input power data of the first device in the first time period is reliable, the energy consumption of the first device in the first time period is calculated according to the corresponding predicted input power data;

[0105] Step S411, repeat the energy consumption of all medical devices whose predicted input power data in the first time period is reliable to obtain the medical device energy consumption prediction result;

[0106] In the specific implementation process, the unreliable prediction is excluded from the energy consumption calculation, which can avoid false energy consumption estimation. If there is an unreliable prediction, the prediction can be re-performed, or manual operation can be triggered for debugging.

[0107] Embodiment 2, please refer to Figure 4 As shown in the figure, Figure 4 An example of a schematic diagram of an electronic device, which can include: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface, the memory complete mutual communication through the communication bus. The memory stores computer readable instructions, and the processor can call the instructions in the memory, when the computer readable instructions are executed by the processor, run the steps in the BIM-based medical device energy consumption prediction model construction method, to realize the following functions: based on the related BIM model, the medical devices in the region are obtained, and the device power time series data and the device related time series data of the medical devices are collected; the device power time series data and the device related time series data are preprocessed, and the data correlation analysis is performed, to obtain node time series data and node correlation data; based on the space-time graph neural network, and using the node time series data and the node correlation data, an energy consumption prediction model is constructed; according to the energy consumption prediction model, the energy consumption of the medical devices is predicted, and the reliability analysis is performed, to obtain the medical device energy consumption prediction result.

[0108] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0109] In embodiment 3, the present application also provides a computer readable storage medium, and the present application provides a storage medium having a computer program stored thereon, and the computer program is executed by a processor to run the steps in the above BIM-based medical device energy consumption prediction model construction method to realize the following functions: obtaining medical devices in a region based on a related BIM model, and collecting device power time series data and device related time series data of the medical devices; performing data preprocessing on the device power time series data and the device related time series data, and performing data correlation analysis to obtain node time series data and node correlation data; constructing an energy consumption prediction model based on a space-time graph neural network and using the node time series data and the node correlation data; performing energy consumption prediction on the medical devices according to the energy consumption prediction model, and performing reliability analysis to obtain a medical device energy consumption prediction result.

[0110] Through the description of the above embodiments, the embodiments of the present application can be provided as a method, a system or a computer program product. Based on such understanding, the above technical solutions essentially or the parts that contribute to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the method described in each embodiment or some parts of the embodiment.

[0111] In the embodiments of the present application, it should be understood that the disclosed system or method can be implemented in other ways. The embodiments described above are only illustrative, for example, the division of modules or units is only a logical function division, and other division manners can be used in actual implementation, for example, a plurality of modules or units can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be through some communication interface, the indirect coupling or communication connection between the system, the module and the unit can be electrical, mechanical or other forms.

[0112] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A BIM-based medical equipment energy consumption prediction model construction method, characterized in that, The method comprises the following steps: Based on the relevant BIM model, the medical equipment in the region is obtained, and the equipment power time series data and the equipment related time series data are collected; The device power time series data and the device related time series data are preprocessed and analyzed for data correlation to obtain node time series data and node correlation data; Based on the space-time graph neural network, the node time series data and the node correlation data are used to construct an energy consumption prediction model; According to the energy consumption prediction model, the energy consumption of the medical equipment is predicted, and the reliability is analyzed to obtain the energy consumption prediction result of the medical equipment; The device power time series data and the device related time series data are preprocessed and analyzed for data correlation to obtain node time series data and node correlation data, which comprises the following sub-steps: The input power of the medical equipment 1 to the medical equipment n1 is sequentially recorded as the feature node 1 to the feature node n1; and the related parameters 1 to the related parameters n1 are sequentially recorded as the feature node (n1+1) to the feature node (n1+n2), and are combined with the feature node 1 to the feature node n1 to be recorded as the feature node 1 to the feature node n0, wherein n0=n1+n2; Any one of the feature nodes is recorded as a first node, the time series data corresponding to the first node is obtained and recorded as a first parameter sequence, the mean and the standard deviation of the first parameter sequence are calculated and recorded as AP and AB respectively; Any one data in the first parameter sequence is recorded as a first value; taking the first value as the center, the first value and the k1 nearest data before and after the first value are taken to form a local window of the first value, wherein k1 is the number set; The coefficient of variation of the data in the local window of the first value is calculated and recorded as CV1; and the dynamic adjustment coefficient k0 corresponding to the first value is calculated, wherein k0=k2+min{(k3-k2),[(k3-k2) / CV0]*CV1}, wherein [k2, k3] is the value range of the set dynamic adjustment coefficient, k0∈[k2, k3], and CV0 is the set coefficient of variation threshold; If the first value does not belong to [AP-k0*AB, AP+k0*AB], it is marked as abnormal data, otherwise it is marked as normal data; If the first data is abnormal data, the k4 nearest normal data before and after the first data are obtained, and the weighted average value of all normal data is calculated according to the position distance of each normal data and the first data, and the first data is replaced, wherein k4 is the number set; Repeat the above steps to obtain the denoising time series data of all feature nodes; Any one of the feature nodes 1 to n1 is recorded as a first power node, and the denoising time series data of the first power node is obtained and recorded as a first power sequence; The first window size is set as k5, k5 is an odd number, any data in the first power sequence is recorded as a first power point, the average value of all data of the first window is calculated with the first power point as the center of the first window, and the first power is replaced; the replacement is repeated for all data in the first power sequence, and the smoothed time sequence data of the first power node is obtained after completion; The smoothed time sequence data of the feature node 1-feature node n1 is repeatedly obtained, and the denoised time sequence data of the feature node (n1+1)-feature node (n1+n2) is combined and recorded as node time sequence data; The smoothed time sequence data and the denoised time sequence data in the node time sequence data are uniformly recorded as feature time sequence data; any two feature nodes are recorded as a node combination, and are recorded as a first node and a second node respectively; the feature time sequence data corresponding to the first node and the second node is obtained respectively, and is recorded as a first sequence and a second sequence respectively; The size of the sliding window is set as k6, the sliding step is 1, and the starting positions of the first sequence and the second sequence are synchronously slid respectively, and the two partial sequences in the two sliding windows of each sliding are recorded as a subsequence group, and all subsequence groups are sequentially recorded as subsequence group 1-subsequence group n3; Any one subsequence group is recorded as a first sub-group, and the partial sequences of the first sequence and the second sequence in the first sub-group are recorded as a first subsequence and a second subsequence respectively; The first subsequence and the second subsequence are standardized by using the z-score method respectively, and the Pearson correlation coefficient of the standardized first subsequence and the standardized second subsequence is calculated, which is recorded as the fluctuation correlation degree AR1 of the first sub-group; The first subsequence and the second subsequence are linearly fitted respectively, and the slopes of the straight lines obtained by linear fitting are obtained in sequence and are recorded as KX1 and KX2 respectively; the trend correlation degree AR2 of the first sub-group is calculated, wherein AR2=sign(KX1*KX2)*min(|KX1|,|KX2|) / max(|KX1|,|KX2|), sign() is a sign function; The coefficients of variation of the first subsequence and the second subsequence are calculated respectively, and the reciprocal of the average value of the two coefficients of variation is calculated, which is recorded as the weight coefficient VQ of the first sub-group; the weight coefficients of all subsequence groups are repeatedly obtained and summed, which is recorded as the weight coefficient sum HQ; The comprehensive correlation degree HR of the first sub-group is calculated, wherein HR=(VQ / HQ)*[(AR1+AR2) / 2]; the comprehensive correlation degrees of all subsequence groups are repeatedly obtained and summed, and the absolute value is taken, which is recorded as the node correlation of the first node and the second node; The node correlation of all node combinations is repeatedly obtained and recorded as node correlation data; Based on the space-time graph neural network, and using the node time sequence data and the node correlation data, an energy consumption prediction model is constructed, including the following sub-steps: According to the node correlation data, an adjacency matrix is constructed, denoted as A=(a ij ) n0*n0 , wherein a ij is the node correlation of feature nodes i and j, and a ij =1 when i=j; i∈[0, n0] According to the node time sequence data, the data of all feature nodes at the same collection time are combined with the adjacency matrix to construct a space-time feature matrix at each collection time, and model training data is obtained; The spatio-temporal graph neural network is recorded as an initial prediction model, the model output is set as the input power of the medical device 1-the medical device n1, the initial prediction model is trained by using the model training data, and after the training, an energy consumption prediction model is obtained. 2.The BIM-based medical device energy consumption prediction model construction method of claim 1, wherein, Based on the related BIM model, the medical devices in the region are obtained, and the device power time series data and the device related time series data of the medical devices are collected, including the following sub-steps: Record the region to be predicted as an energy consumption prediction region, obtain the BIM of the energy consumption prediction region, and record it as a prediction region model; all medical devices in the prediction region are obtained according to the prediction region model, and are sequentially recorded as medical device 1-medical device n1; and any one medical device is recorded as a first device; Obtain the environmental parameters and medical business parameters related to the device energy consumption of the first device, and record them as the related parameters corresponding to the first device; repeat the acquisition of the related parameters corresponding to all medical devices, and perform de-duplication to obtain a related parameter collection, and record various types of parameters in the related parameter collection as related parameter 1-related parameter n1. 3.The BIM-based medical device energy consumption prediction model construction method of claim 2, wherein, Based on the related BIM model, the medical devices in the region are obtained, and the device power time series data and the device related time series data of the medical devices are collected, including the following sub-steps: Synchronously collect the input power of each medical device at a first time interval, and record the collection time; according to the corresponding medical device classification, sort in time sequence, and record the time series data of the input power of the corresponding medical device; after completion, the power time series data is obtained, wherein the first time interval is t1; At the same time, the parameter values of each related parameter are synchronously collected at a first time interval, and the collection time is recorded; according to the corresponding related parameter classification, sort in time sequence, and record the time series data of the corresponding related parameter; after completion, the device related time series data is obtained. 4.The BIM-based medical device energy consumption prediction model construction method of claim 3, wherein, According to the energy consumption prediction model, the energy consumption of the medical device is predicted, and reliability analysis is performed, to obtain the medical device energy consumption prediction result, including the following sub-steps: The input power of the medical device 1-medical device n1 at each time in the future first time period is predicted by using the energy consumption prediction model, to obtain the predicted input power data of all medical devices in the first time period; Obtain the date type of the date in the first time period, and record it as the first type; the date type includes weekdays and weekends; Arrange the input power data of the first device in the first time period in time sequence, and record it as the first prediction sequence; for any data in the first prediction sequence, record it as the first predicted power; the time corresponding to the first predicted power is recorded as the first time; Obtain the nearest k7 dates of the first type before the date in the first time period, and record them as reference dates; obtain the actual input power of the first device at the first time of each reference date, and record it as a reference power collection; calculate the mean and standard deviation of the reference power collection, and record them in order as CP and CB respectively, wherein k7 is the number of settings. 5.The BIM-based medical device energy consumption prediction model construction method of claim 4, wherein, According to the energy consumption prediction model, the energy consumption of the medical device is predicted, and reliability analysis is performed, to obtain the medical device energy consumption prediction result, including the following sub-steps: If the first predicted power is located in [CP-k8*CB, CP-k8*CB], it is marked as reliable data, otherwise it is marked as fluctuant data; repeat the process of obtaining all reliable data in the first predicted sequence and calculating the proportion of the number of reliable data, and record it as the reliable coefficient of the first device, wherein k8 is a set proportion coefficient; Repeat the process of obtaining the reliable coefficient of all medical devices and calculating the average value, and record it as the overall reliable coefficient; If the overall reliable coefficient is less than E1, it is determined that the predicted input power data of all medical devices in the first time period is unreliable; If the overall reliable coefficient is not less than E1, but the reliable coefficient of the first device is less than E2, it is determined that the predicted input power data of the first device in the first time period is unreliable; If the overall reliable coefficient is not less than E1, and the reliable coefficient of the first device is not less than E2, it is determined that the predicted input power data of the first device in the first time period is reliable; If the predicted input power data of the first device in the first time period is reliable, the energy consumption of the first device in the first time period is calculated according to the corresponding predicted input power data; Repeat the process of obtaining the energy consumption of all medical devices whose predicted input power data in the first time period is reliable to obtain the medical device energy consumption prediction result.

Images