Medical institution energy consumption prediction method, equipment and medium
By classifying the energy consumption data of medical institutions by region and equipment and combining it with the CEEMD-AutoARIMA-LSTM hybrid model, the problems of insufficient prediction accuracy and generalization ability in traditional methods are solved, and more accurate energy consumption prediction is achieved.
Patent Information
- Application Number
- CN202510732960.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-04
- Publication Date
- 2025-09-12
AI Technical Summary
Traditional hospital energy consumption prediction methods have low prediction accuracy and insufficient generalization ability. They are difficult to handle the complex nonlinear relationship of hospital energy consumption data, and a single model is difficult to achieve accurate prediction.
The energy consumption data of medical institutions are classified according to regional and equipment dimensions, a multidimensional energy consumption classification system is constructed, the feature engineering strategy is dynamically adjusted, and the CEEMD-AutoARIMA-LSTM hybrid model is used for prediction, including model training and decomposition based on a multidimensional feature library, and reconstruction using intrinsic mode function components and residual components.
The prediction accuracy and generalization ability of the energy consumption data prediction model of medical institutions have been improved, which can more accurately predict energy consumption time series, adapt to the energy consumption characteristics of hospitals, and improve prediction accuracy and adaptability.
Smart Images
Figure CN120636726A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of energy consumption prediction, and more specifically, to a method, device and medium for predicting energy consumption of a medical institution. Background Art
[0002] Among the many factors influencing hospital operational efficiency, energy consumption is a crucial one. Hospital logistics consume significant amounts of energy resources to maintain various needs, including hot and cold water supply, temperature and humidity control, lighting, clinical operations, medical equipment operation, and water and natural gas supplies. Therefore, predicting hospital energy consumption can facilitate effective energy management, achieving the goals of green hospital development and refined logistics management.
[0003] However, traditional hospital energy consumption prediction methods usually use a unified feature set and algorithm parameters to process the energy consumption data of the entire hospital, and feature engineering also relies heavily on common features (such as timestamps and total energy consumption values), which makes the prediction accuracy of the prediction model insufficient and the generalization ability is not good enough.
[0004] Furthermore, traditional hospital energy consumption forecasting models often rely on time series decomposition algorithms, which struggle to handle the complex nonlinear relationships in hospital energy consumption data. Hospital energy consumption is influenced by multiple factors, including staff turnover, equipment usage, and climate, exhibiting highly nonlinear characteristics. Traditional time series algorithms, based on linear modeling assumptions, cannot effectively capture these characteristics. Nonlinear decomposition algorithms also suffer from modal aliasing and unstable results, and a single model struggles to achieve accurate predictions. Summary of the Invention
[0005] In response to at least one defect or improvement need in the prior art, the present application provides a method, device and medium for predicting energy consumption of medical institutions, with the aim of improving the prediction accuracy of the medical institution energy consumption data prediction model and enhancing the generalization ability of the model.
[0006] To achieve the above objectives, in a first aspect, the present application provides a method for predicting energy consumption of a medical institution, comprising:
[0007] Classify the energy consumption data of medical institutions by regional dimensions and equipment dimensions to construct a multidimensional energy consumption classification system; dynamically adjust the feature engineering strategy based on the multidimensional energy consumption classification system to construct a multidimensional feature library;
[0008] Performing model training based on the multidimensional feature library to obtain an intrinsic mode function component prediction model and a residual component prediction model;
[0009] Decompose the classified medical institution energy consumption data to obtain intrinsic mode function components and residual components;
[0010] Using the intrinsic mode function component prediction model to predict the intrinsic mode function component to obtain an intrinsic mode function component prediction result; using the residual component prediction model to predict the residual component to obtain a residual component prediction result;
[0011] The intrinsic mode function component prediction result and the residual component prediction result are reconstructed to obtain the medical institution energy consumption prediction result.
[0012] Furthermore, the classification of the regional dimension includes one or more of business function area, logistic public area and commercial area;
[0013] The business functional areas include one or more of outpatient areas, inpatient areas, medical technology areas, emergency areas and administrative office areas;
[0014] The logistics public area includes one or more of an equipment room area, a parking lot area, a dormitory area, and a building peripheral area;
[0015] The commercial area includes one or more of a restaurant area and a supermarket area.
[0016] Furthermore, the classification of the equipment dimension includes one or more of general equipment, large medical equipment, medical equipment belts, and emergency equipment;
[0017] The general equipment includes one or more of lighting equipment, socket equipment, power equipment and environmental equipment;
[0018] The large medical equipment includes one or more of CT, DR, MR, DSA and linear accelerator.
[0019] Furthermore, dynamically adjusting the feature engineering strategy based on the multi-dimensional energy consumption classification system includes:
[0020] For the regional dimension, regional-specific features are introduced, including: features of increased number of registered patients in the outpatient area and / or features of increased bed occupancy rate and operating status of large medical equipment in the inpatient area;
[0021] For the equipment dimension, equipment-specific features are introduced, including: additional diagnosis times and power curve features for large medical equipment and / or additional maintenance cycle and power parameter features for environmental equipment;
[0022] Dynamically switch feature sets based on seasonality strategies and / or device abnormality awareness strategies.
[0023] Furthermore, the classified medical institution energy consumption data is decomposed to obtain the intrinsic mode function components and residual components including:
[0024] Add complementary white noise to the normalized energy consumption time series data;
[0025] The energy consumption time series data with complementary white noise added thereto is subjected to empirical mode decomposition and averaging processing to obtain the intrinsic mode function component and the residual component.
[0026] Furthermore, the method further includes the step of processing the intrinsic mode function component using the intrinsic mode function component prediction model, specifically including:
[0027] Integrating the decomposed intrinsic mode function components with multi-source features as input to the LSTM model; the multi-source features include one or more of time features, weather features, and business features;
[0028] Use a multi-layer LSTM structure to capture long-range dependencies;
[0029] The mean square error is used as the loss function and the Adam optimizer is used to train the LSTM model.
[0030] Furthermore, the method further includes the step of processing the residual component using the residual component prediction model, specifically including:
[0031] The residual components obtained by decomposition are used as inputs of the AutoARIMA model, and features that may affect the long-term trend are added;
[0032] The AutoARIMA model is expanded into the ARIMAX model, the optimal parameters are searched to meet the AIC information criterion optimal principle, and exogenous variables are added to the ARIMAX model;
[0033] The exogenous variables include the total rated power of the energy-consuming equipment of the medical institution and / or the area of the department of the medical institution.
[0034] Furthermore, it also includes:
[0035] The energy consumption prediction result of the medical institution is subjected to denormalization processing to obtain the actual energy consumption prediction result of the medical institution.
[0036] In a second aspect, the present application provides an electronic device comprising at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the computer program is executed by the processing unit, the processing unit is enabled to perform the steps of any of the aforementioned methods for predicting energy consumption of medical institutions.
[0037] In a third aspect, the present application provides a storage medium storing a computer program executable by an access authentication device. When the computer program runs on the access authentication device, the access authentication device can execute the steps of any of the aforementioned methods for predicting energy consumption of medical institutions.
[0038] In general, the above technical solutions conceived by this application can achieve the following beneficial effects compared with the existing technology:
[0039] (1) This application divides the energy consumption data of medical institutions into two dimensions: region and equipment, and constructs a more refined prediction model. At the same time, it can automatically match feature sets based on classification labels and support dynamic adjustment of features. In this way, the prediction accuracy of the medical institution energy consumption data prediction model can be improved and the generalization ability of the model can be enhanced.
[0040] (2) This application uses a hybrid model based on CEEMD-AutoARIMA-LSTM for training and prediction, combining the CEEMD algorithm with the time series prediction algorithm, making the model more suitable for the energy consumption characteristics of the hospital, and more accurately predicting the energy consumption time series, thereby improving the prediction accuracy and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0042] Figure 1 This is a core flow chart of a method for predicting energy consumption in a medical institution provided in an embodiment of the present application;
[0043] Figure 2 A flowchart of a method for predicting energy consumption in a medical institution based on CEEMD-AutoARIMA-LSTM provided in an embodiment of the present application;
[0044] Figure 3 An energy consumption curve diagram of a hospital outpatient department provided in an embodiment of the present application;
[0045] Figure 4 A schematic diagram of the standardized CEEMD decomposition provided in the embodiments of the present application;
[0046] Figure 5 A block diagram of an electronic device suitable for implementing the above-described method for predicting energy consumption in medical institutions provided in an embodiment of the present application. DETAILED DESCRIPTION
[0047] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely for the purpose of explaining this application and are not intended to limit this application. In addition, the technical features involved in the various embodiments of this application described below may be combined with each other as long as they do not conflict with each other.
[0048] The terms "first," "second," or "nth" in the specification, claims, or drawings of this application may be used to distinguish different objects or to describe a specific order, depending on the specific scenario. In addition, the terms "including" or "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products, or devices.
[0049] As described in the background technology section of the specification, traditional hospital energy consumption prediction methods usually use a unified feature set and algorithm parameters to process the energy consumption data of the entire hospital, and feature engineering also relies heavily on common features (such as timestamps and total energy consumption values), which makes the prediction accuracy of the prediction model not high enough and the generalization ability is not good enough. In addition, traditional hospital energy consumption prediction models mostly use time series decomposition algorithms, which are difficult to handle the complex nonlinear relationships of hospital energy consumption data. Hospital energy consumption is affected by multiple factors such as personnel flow, equipment use, and climate, and exhibits highly nonlinear characteristics. Traditional time series algorithms are based on linear assumptions and cannot effectively capture them; nonlinear decomposition algorithms also have problems such as modal aliasing and unstable results, and a single model is difficult to achieve accurate predictions. In view of this, the present application provides a medical institution energy consumption prediction method, equipment, and medium, the purpose of which is to improve the prediction accuracy of the medical institution energy consumption data prediction model and enhance the generalization ability of the model.
[0050] refer to Figure 1 and Figure 2 An embodiment of the present application provides a method for predicting energy consumption of a medical institution, which may specifically include the following steps.
[0051] Step 1: Classify the energy consumption data of medical institutions (such as hospitals at all levels, community health service centers, etc.) by regional and equipment dimensions to build a multi-dimensional energy consumption classification system.
[0052] The energy consumption of the outpatient area is strongly correlated with the flow of people and the number of outpatient visits, while the energy consumption of the equipment room is mainly affected by the equipment load rate. However, traditional methods generally do not distinguish regional characteristics, resulting in unreasonable distribution of characteristic factors. Large medical equipment (such as CT, MR) exhibits intermittent high energy consumption characteristics, while environmental equipment (such as air conditioning) has periodic operating rules. However, traditional methods generally do not adjust model parameters according to equipment types, making it difficult to capture differentiated energy consumption patterns. Based on the above analysis, an embodiment of the present application classifies hospital energy consumption data by regional dimension and equipment dimension. The specific classification is as follows:
[0053] 1. Regional dimension classification:
[0054] (1) Business functional areas (outpatient, inpatient, medical technology, emergency, administrative offices, etc.): The main energy consumption comes from lighting, air conditioning and medical equipment in areas with intensive human activities.
[0055] (2) Logistics public areas (equipment rooms, parking lots, dormitories, building perimeters, etc.): The main energy consumption is concentrated in equipment operation (such as refrigeration units, water pumps) and infrastructure maintenance.
[0056] (3) Commercial areas (restaurants, supermarkets, etc.): The main energy consumption is significantly affected by fluctuations in business hours and passenger flow.
[0057] 2. Equipment dimension classification:
[0058] (1) General equipment (lighting, sockets, power, environment, etc.): Energy consumption patterns are strongly related to time, temperature and humidity.
[0059] (2) Large medical equipment (CT, DR, MR, DSA, linear accelerator, etc.): Intermittent high energy consumption requires modeling in combination with equipment usage logs.
[0060] (3) Medical equipment and emergency equipment: Energy consumption is driven by sudden events, and an abnormality detection mechanism needs to be introduced.
[0061] Then, based on the above two-dimensional classification results, a multidimensional energy consumption classification system is constructed. Step 2: Dynamically adjust the feature engineering strategy based on the multidimensional energy consumption classification system and construct a multidimensional feature library. That is, dynamically adjust the feature engineering and model according to the scenario, and then, considering the impact of multidimensional features on energy consumption, build a multidimensional feature library to store daily average temperature and humidity data, hospital traffic data, outpatient volume data, bed data and other data. In some embodiments, dynamically adjusting the feature engineering strategy and constructing the multidimensional feature library specifically include the following contents.
[0062] 1. Establishment of multidimensional feature library
[0063] (1) Data acquisition
[0064] Energy consumption data: Through smart electricity meters, gas meters, water meters and other IoT devices, categorized energy consumption data (electricity, natural gas, water, etc.) of each area (outpatient department, inpatient department, logistics public areas, etc.) and each device (large medical equipment, air conditioning, lighting, etc.) is automatically collected every day.
[0065] Environmental data: Deploy temperature and humidity sensors (such as in clinics, wards, and computer rooms) to monitor environmental parameters in real time.
[0066] Operational data: Synchronize operational indicators such as outpatient volume, emergency volume, bed occupancy rate, and surgical schedule from the hospital information system (HIS).
[0067] Equipment log: Integrates the operation logs of major electrical equipment (such as air conditioner host start and stop, equipment power curve, etc.).
[0068] (2) Data standardization and storage
[0069] Data alignment: All data are aligned by timestamp and stored in a time series format. Fields include timestamp, location, device_type, energy_type, value, etc.
[0070] Classification storage: Create sub-tables to store characteristic data of different areas such as outpatient areas, inpatient areas, and equipment rooms; create independent tables for equipment types (CT, MR, air conditioning) to record parameters such as power, operating status, and maintenance cycle.
[0071] (3) Data cleaning
[0072] Missing value processing: Linear interpolation is used to fill short-term missing data, and long-term missing data are marked as abnormal and trigger manual verification.
[0073] Outlier detection: Use the Z-score algorithm (threshold ±3σ) to identify outliers and combine it with business rule corrections (for example, a sudden drop in inpatient lighting energy consumption to zero during the day is considered an equipment anomaly, and the relevant data needs to be reported, and energy consumption data repaired through tools).
[0074] 2. Feature Engineering Processing by Scenario
[0075] Traditional feature engineering has the limitation of "one size fits all" and relies on general features (such as timestamps, total energy consumption values, etc.), without combining hospital-specific parameters (such as bed occupancy rate, emergency volume, equipment maintenance cycle, etc.) to dynamically screen features for different scenarios. For example:
[0076] The energy consumption of the outpatient area is strongly correlated with the flow of people and the lighting status of the clinic, but traditional methods may forcibly add irrelevant features such as "equipment room load rate" into the outpatient area model, thereby introducing noise.
[0077] The energy consumption of large medical devices (such as CT) is driven by the device power curve, but traditional feature sets may only contain general timestamps and total energy consumption, while ignoring device-specific parameters.
[0078] (1) Regional dimension example: outpatient area
[0079] Time characteristics: whether it is a weekend or a holiday.
[0080] Weather characteristics: daily average temperature, daily rainfall.
[0081] Outpatient business characteristics: number of outpatient registrations on the day, number of consultations with large medical equipment.
[0082] (2) Equipment dimension example: lighting system
[0083] Time characteristics: whether it is a weekday, season / month.
[0084] Weather characteristics: average daily light intensity (sufficient natural light reduces the need for artificial lighting), whether it is cloudy (additional lighting is required), and duration of daylight (short-day days require longer lighting hours).
[0085] Equipment business characteristics: total lighting on-time of the day, active hours of the lighting area, special event markers, and equipment aging coefficient.
[0086] Examples of feature engineering processing for different scenarios are as follows:
[0087] Characteristics of energy consumption in the outpatient area:
[0088] (1) Long energy consumption time and uneven distribution: Energy demand varies greatly on different days. For example, weekdays are usually the peak period for medical consultations, with higher energy consumption, while weekends are relatively lower. In addition, energy consumption time varies among departments. For example, the laboratory and radiology departments may require equipment to run for longer periods of time, and their energy consumption distribution is different from that of ordinary outpatient departments.
[0089] (2) High energy density: The outpatient area is one of the most densely populated areas in the hospital, and the energy consumption per unit area is relatively high. According to statistics, the energy consumption intensity of outpatient buildings is usually higher than that of ordinary public buildings.
[0090] (3) Energy consumption is greatly affected by the number of patients: The energy consumption of the outpatient area is closely related to the number of patients. The greater the number of patients, the higher the energy consumption. When the number of patients increases, the frequency and time of use of lighting, air conditioning, medical equipment, etc. will increase accordingly.
[0091] (4) The types of energy-consuming equipment are diverse and complex: The main energy-consuming equipment in the outpatient area includes HVAC systems, lighting systems, and medical equipment. Among them, HVAC systems are the largest energy consumers, accounting for a large proportion of the total energy consumption in the outpatient area. In addition, large medical equipment such as CT and MRI equipment has high energy consumption.
[0092] 3. Dynamic feature adjustment method
[0093] Given that traditional feature engineering relies heavily on general features and has the limitation of a "one-size-fits-all" approach, in some embodiments, dynamically adjusting feature engineering strategies includes the following.
[0094] For the regional dimension, region-specific characteristics are introduced, including: the increase in the number of registered patients in the outpatient area, the increase in the bed occupancy rate in the inpatient area, and the operating status of large medical equipment.
[0095] In terms of equipment dimensions, equipment-specific features are introduced, including: increased diagnosis times and power curve features for large medical equipment, and increased maintenance cycles and power parameter features for environmental equipment.
[0096] Seasonal strategy: Based on the hospital's cooling and heating source configuration, energy consumption and operation data of the corresponding environmentally controlled cooling and heating sources are introduced into the feature set during the cooling, heating, and transition seasons.
[0097] Device abnormal status perception strategy: When abnormal sensor temperature and humidity data is detected during data preprocessing, the backup rule is triggered and external meteorological data is used instead.
[0098] Step 3: Perform model training based on the multidimensional feature library to obtain an intrinsic mode function component prediction model and a residual component prediction model.
[0099] Before training a model, it's usually necessary to standardize the data and use it for data analysis. Different features often have different dimensions and units, which can affect the model's predictions. To eliminate the effects of dimensionality, the data needs to be standardized. This process, known as normalization, limits the data to a specific range.
[0100] In some embodiments, the maximum-minimum normalization method is used to normalize the sample data. The maximum-minimum normalization is a linear transformation of the original data, and the calculation method is:
[0101]
[0102] In the formula, Represents the original value of the feature column, minA and maxA represent the minimum and maximum values of the feature column respectively. After processing, they will be mapped to [0, 1]. .
[0103] After preprocessing the feature data in the multidimensional feature library, model training can be performed based on the multidimensional feature library to obtain the intrinsic mode function component prediction model and the residual component prediction model. The specific technical details of the training model can be found in steps 4 and 5.
[0104] Step 4: Decompose the classified medical institution energy consumption data to obtain intrinsic mode function components and residual components.
[0105] Specifically, refer to Figure 3 and Figure 4 After normalizing the hospital energy consumption data and characteristics, CEEMD was used to decompose the hospital energy consumption time series into multiple intrinsic mode function (IMF) components and residual components. Each IMF component represents the characteristics of different frequencies and time scales in the data, while the residual component contains the long-term trend of the data and possible complex seasonal patterns.
[0106] The core concept of CEEMD is to add paired positive and negative white noise to the original signal, perform EMD decomposition (Empirical Mode Decomposition) on each of these noise-added signals, and then average the decomposition results. This method can significantly offset the interference of random noise, thereby significantly reducing modal aliasing and reconstruction error. Compared with traditional EMD, CEEMD effectively reduces modal aliasing and noise interference by adding paired positive and negative noise and performing averaging, while also improving computational efficiency and being more suitable for the adaptive decomposition of nonlinear and non-stationary signals. Compared with trend-cycle-residual decomposition, CEEMD does not require a preset period, can flexibly decompose complex signals, has stronger noise handling capabilities, and has a wider range of applications. The specific CEEMD decomposition methods are as follows.
[0107] 1. Initialization
[0108] Input original signal , determine the number of times N to add noise, and the amplitude of the noise .
[0109] 2. Add complementary noise
[0110] For each experiment of adding noise , generates a white noise sequence , construct two sets of signals:
[0111]
[0112]
[0113] 3. EMD decomposition
[0114] For each signal and Perform EMD decomposition separately to obtain their respective intrinsic mode function (IMFs) sets:
[0115] :
[0116] :
[0117] 4. Calculate complementary average
[0118] For each pair of IMF components of complementary signals, calculate their average values to obtain the final IMF components:
[0119]
[0120] 5. Ensemble Average
[0121] The average IMF components of all N experiments are further averaged to obtain the final IMF component set:
[0122]
[0123] 6. Result reconstruction
[0124] The final reconstructed signal can be expressed as:
[0125]
[0126] Where M is the number of adaptive modes, is the jth IMF component, is the residual after averaging.
[0127] Step 5: Use the intrinsic mode function component prediction model to predict the intrinsic mode function component to obtain an intrinsic mode function component prediction result; use the residual component prediction model to predict the residual component to obtain a residual component prediction result.
[0128] Although CEEMD can decompose the signal into multiple IMF components and residuals, the residuals The data may still contain complex seasonal patterns. The AutoARIMA algorithm can further process residuals. Its core concept is to model time series by automatically searching for the optimal ARIMA model parameters (including the differencing order, the order of the autoregressive term, and the order of the moving average term). AutoARIMA can automatically identify trend, seasonality, and noise components in the time series and select the most appropriate parameter combination to fit the data. Furthermore, ARIMA can be expanded to an ARIMAX model, allowing the inclusion of exogenous variables (such as the total rated power of energy-consuming equipment in medical institutions and the area of medical department departments) to better capture factors that may influence time series trends. In this way, AutoARIMA can effectively handle complex time series data, improving the model's adaptability and forecasting accuracy. It can more effectively handle noise and complex seasonal patterns in hospital energy consumption data, thereby improving the accuracy and reliability of energy consumption forecasts.
[0129] Specifically, the residual components obtained from the CEEMD decomposition are used as input to the AutoARIMA algorithm, and features that may affect long-term trends are added. ARIMA is expanded to ARIMAX, long-term low-frequency features are added, and the optimal parameters are adaptively found to meet the optimal principle of the AIC information criterion.
[0130] Decomposed by CEEMD IMF data often exhibit complex nonlinear characteristics, making them difficult to effectively model with traditional linear models. As a nonlinear model, LSTM, through its gating mechanism and complex connectivity between neurons, possesses powerful nonlinear mapping capabilities. It can learn the complex inherent laws and patterns in IMF data, adapt to nonlinear changes in the data, and thus achieve more accurate IMF predictions.
[0131] Secondly, IMFs are components of different characteristic scales decomposed from the original signal, and their sequence data often exhibits long-range dependencies. The forget gate, input gate, and output gate of the LSTM can selectively retain or forget historical information, enabling long-term memorization of important data and effectively capturing long-term dependencies in the IMF sequence. This makes it advantageous for predicting future IMF values.
[0132] Furthermore, different IMFs may have different statistical characteristics and change patterns. LSTM has strong flexibility and adaptability. Through large-scale data training, it can automatically learn and adapt to the unique data characteristics of each IMF, and establish appropriate prediction models for different IMF components, thereby meeting diverse prediction needs.
[0133] Finally, hospital energy consumption data comes from a wide range of sources, encompassing not only the energy consumption time series itself but also multi-dimensional external information, such as meteorological data (temperature, humidity, wind speed, etc.), hospital operational data (daily patient visits, department schedules, equipment usage hours), and holiday schedules. This multi-source data contains key factors influencing energy consumption fluctuations. A single LSTM prediction based solely on the IMF series fails to fully utilize this rich information. However, by incorporating this multi-source data as input features into the model, the multi-LSTM model can construct a more comprehensive prediction system. Its internal structure allows for adaptive learning based on different types of feature data. It not only captures long-range dependencies and complex nonlinear patterns in the IMF series, but also explores potential correlations between external features and energy consumption changes. This allows for a more accurate reflection of the true trends in hospital energy consumption, effectively addressing data fluctuations and unexpected situations, significantly improving prediction accuracy and robustness, and providing more reliable support for hospital energy management and optimization decisions.
[0134] Specifically, the IMF components obtained by CEEMD decomposition are used as feature inputs. In addition, other features that may affect the hospital's energy consumption trend can also be added, such as weather data, holiday information, and hospital operation conditions.
[0135] According to the characteristics of time series data, historical data and features are used as input features, and the energy consumption value at a certain time point in the future is used as the target value to generate a sample data set for LSTM model training.
[0136] Build an LSTM neural network model, including an input layer, multiple LSTM layers, and an output layer. Select appropriate activation functions, loss functions, and optimizers, using ReLU as the activation function, mean squared error (MSE) as the loss function, and the Adam optimizer to optimize model parameters.
[0137] After CEEMD decomposition, IMF components and residuals are obtained. IMF components generally represent high-frequency feature fluctuations. In one embodiment of the present application, LSTM is used to predict them, and the following features are selected:
[0138] 1. Time characteristics
[0139] Day of the week: Hospital outpatient clinics are busier on different days of the week. Generally speaking, Mondays are more crowded with patients, leading to higher electricity consumption, while weekends may be relatively lower. This reflects the cyclical fluctuations in outpatient electricity consumption, which aligns with the short-term fluctuations in the IMF component.
[0140] Holidays: During holidays, hospital outpatient clinic schedules and patient traffic change, leading to different power consumption patterns. For example, statutory holidays typically see fewer outpatient visits, leading to a corresponding decrease in power consumption, which is a significant factor influencing the daily granularity of outpatient power consumption.
[0141] 2. Weather characteristics
[0142] Daily average temperature: Temperature significantly impacts air conditioning usage in hospital outpatient clinics. In hot weather, cooling demand increases, leading to higher electricity consumption. In cold weather, heating with air conditioning or other heating equipment also causes fluctuations in electricity consumption, which is highly correlated with short-term fluctuations in outpatient electricity consumption.
[0143] Daily precipitation: Precipitation may affect patients' willingness to travel, which in turn affects the number of outpatient visits. It may also affect the use of some equipment in the hospital, such as ventilation equipment, and have a certain impact on the daily granularity fluctuation of outpatient electricity consumption.
[0144] 3. Characteristics of outpatient services
[0145] Number of outpatient registrations on the same day: The number of registrations directly reflects the patient flow in the outpatient clinic on that day. The more people there are, the more frequently various devices are used, and the higher the power consumption. It is a key factor affecting the daily fluctuation of outpatient power consumption.
[0146] Number of consultations with large medical equipment: refers to the total number of outpatient visits in departments related to large medical equipment (such as CT, MRI, etc.) on that day.
[0147] Generally speaking, residuals represent long-term trends, so one embodiment of this application uses AutoARIMA to predict residuals, and the following features are selected:
[0148] Hospital operation characteristics, including:
[0149] Total rated power of hospital energy-consuming equipment: If large medical equipment is put into use in a hospital, it may increase outpatient electricity consumption for a period of time, affecting the long-term trend represented by the residual. This is because the power and frequency of use of the new equipment will change the overall electricity consumption pattern.
[0150] Hospital Department Area: Department adjustments may lead to the reallocation of personnel and equipment, affecting power consumption in different areas. For example, relocating a department from one floor to another may alter power consumption on that floor, in turn affecting the long-term trend of outpatient power consumption and the variation pattern of the residual. For detailed characteristic data, please refer to Tables 1-3.
[0151] Table 1: Full feature data example table
[0152]
[0153] Table 2 List of features used for residual prediction
[0154]
[0155] Table 3. List of features used for IMF component prediction
[0156]
[0157] Step 6: reconstruct the intrinsic mode function component prediction result and the residual component prediction result to obtain the medical institution energy consumption prediction result.
[0158] In some embodiments, the LSTM model is used to predict each IMF component separately to obtain a predicted value for each component. The AutoARIMA model is then used to predict the residual to obtain a predicted residual value. All the predicted values obtained above are then added together to obtain the final hospital energy consumption forecast.
[0159] The final hospital energy consumption prediction results are denormalized, that is, converted from the normalized range back to the actual range of the original data, so as to be compared and evaluated with the actual energy consumption values.
[0160] Then, the error indicators between the hospital energy consumption prediction results after denormalization and the actual energy consumption values of the hospital, such as the mean square error (MSE) and mean absolute error (MAE), were calculated to evaluate the prediction performance and accuracy of the model.
[0161] In the model evaluation phase, some embodiments of the present application adopt a time series partitioning strategy, using the 2024 annual energy consumption data and hospital characteristic data as a training set for optimizing and learning model parameters; data from January 2025 is selected to construct a test set to test the model's generalization ability on new data.
[0162] The evaluation index system covers mean absolute percentage error (MAPE), mean square error (MSE) and coefficient of determination (R²).
[0163] MAPE intuitively reflects the relative error between the predicted value and the true value in the form of a percentage, and is suitable for evaluating the accuracy of energy consumption forecasts.
[0164] MSE quantifies the mean square of the prediction error and is highly sensitive to outliers.
[0165] R² measures the extent to which the model explains data variation and is used to assess model goodness of fit. Comprehensive evaluation using multiple indicators ensures the scientific and comprehensive nature of model performance evaluation.
[0166] This application divides the energy consumption data of medical institutions into two dimensions, namely, region and equipment, and constructs a more refined prediction model. At the same time, it can automatically match feature sets based on classification labels and support dynamic adjustment of features. In this way, the prediction accuracy of the medical institution energy consumption data prediction model can be improved and the generalization ability of the model can be enhanced. This application uses a hybrid model based on CEEMD-AutoARIMA-LSTM for training and prediction, combining the CEEMD algorithm with the time series prediction algorithm, so that the model is more adapted to the energy consumption characteristics of the hospital, and can more accurately predict the energy consumption time series, thereby improving the prediction accuracy and adaptability.
[0167] Figure 5 The block diagram of an electronic device suitable for implementing the above-described method for predicting energy consumption of medical institutions according to an embodiment of the present application is schematically shown. Figure 5 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.
[0168] like Figure 5As shown, the electronic device 1000 described in this embodiment includes: a processor 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage part 1008 into a random access memory (RAM) 1003. The processor 1001 may, for example, include a general-purpose microprocessor (such as a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (for example, an application-specific integrated circuit (ASIC)), etc. The processor 1001 may also include on-board memory for caching purposes. The processor 1001 may include a single processing unit or multiple processing units for executing different actions of the process of the medical institution energy consumption prediction method according to an embodiment of the present application.
[0169] Various programs and data required for the operation of the electronic device 1000 are stored in RAM 1003. The processor 1001, ROM 1002 and RAM 1003 are connected to each other via a bus 1004. The processor 1001 performs various operations of the process of the method for predicting energy consumption of medical institutions according to the embodiment of the present application by executing the programs in ROM 1002 and / or RAM 1003. It should be noted that the program can also be stored in one or more memories other than ROM 1002 and RAM 1003. The processor 1001 can also perform various operations of the process of the method for predicting energy consumption of medical institutions according to the embodiment of the present application by executing the programs stored in the one or more memories.
[0170] According to an embodiment of the present application, electronic device 1000 may further include an input / output (I / O) interface 1005, which is also connected to bus 1004. Electronic device 1000 may further include one or more of the following components connected to I / O interface 1005: an input portion 1006 including a keyboard, mouse, etc.; an output portion 1007 including devices such as a cathode ray tube (CRT), liquid crystal display (LCD), and speakers; a storage portion 1008 including a hard disk; and a communication portion 1009 including a network interface card such as a LAN card or modem. Communication portion 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to I / O interface 1005 as needed. Removable media 1011, such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed in drive 1010 as needed, so that computer programs read from the removable media can be installed into storage portion 1008 as needed.
[0171] According to the process of the method for predicting energy consumption of medical institutions according to an embodiment of the present application, the process can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program contains program code for executing the method for predicting energy consumption of medical institutions shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1009, and / or installed from the removable medium 1011. When the computer program is executed by the processor 1001, the above-mentioned functions defined in the system of the embodiment of the present application are executed. According to an embodiment of the present application, the systems, devices, means, modules and / or units described above can be implemented by computer program modules.
[0172] The embodiments of the present application also provide a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not be incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, and when the one or more programs are executed, the steps of the method for predicting energy consumption of a medical institution according to the embodiments of the present application can be implemented.
[0173] According to an embodiment of the present application, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, it may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In an embodiment of the present application, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present application, the computer-readable storage medium may include one or more memories other than the ROM 1002 and / or RAM 1003 described above.
[0174] It should be noted that the functional modules in the various embodiments of the present application can be integrated into a processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above-mentioned integrated modules can be implemented in the form of hardware or in the form of software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product.
[0175] The flowcharts and / or block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flowchart and / or block diagram can represent a module, a program segment or a part of code, and the part of the above-mentioned module, program segment or code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. It should also be noted that each box in the block diagram or flowchart, and the combination of the boxes in the block diagram or flowchart, can be implemented by a special hardware-based system that performs the specified function or operation, or can be implemented by a combination of special hardware and computer instructions.
[0176] Those skilled in the art will appreciate that the features described in the various embodiments and / or claims of this application may be combined and / or coupled in various ways, even if such combinations and / or couplings are not explicitly described in this application. In particular, without departing from the spirit and teachings of this application, the technical features described in the various embodiments and / or claims of this application may be combined and / or coupled in various ways, and all such combinations and / or couplings fall within the scope of this application.
[0177] Although the present application has been shown and described with reference to certain exemplary embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made to the present application without departing from the spirit and scope of the present application as defined by the appended claims and their equivalents. Therefore, the scope of the present application should not be limited to the above-described embodiments, but should be determined not only by the appended claims but also by the equivalents of the appended claims.
Claims
1. A method for predicting energy consumption of a medical institution, characterized in that: include: Classify the energy consumption data of medical institutions by region and equipment, and build a multi-dimensional energy consumption classification system; Dynamically adjust the feature engineering strategy based on the multidimensional energy consumption classification system to build a multidimensional feature library; Performing model training based on the multidimensional feature library to obtain an intrinsic mode function component prediction model and a residual component prediction model; Decompose the classified medical institution energy consumption data to obtain intrinsic mode function components and residual components; Using the intrinsic mode function component prediction model to predict the intrinsic mode function component to obtain an intrinsic mode function component prediction result; Predicting the residual component using the residual component prediction model to obtain a residual component prediction result; The intrinsic mode function component prediction result and the residual component prediction result are reconstructed to obtain the medical institution energy consumption prediction result.
2. The method for predicting energy consumption of a medical institution according to claim 1, wherein: The classification of the regional dimension includes one or more of business function area, logistic public area and commercial area; The business functional areas include one or more of outpatient areas, inpatient areas, medical technology areas, emergency areas and administrative office areas; The logistics public area includes one or more of the equipment room area, parking area, dormitory area and building peripheral area; The commercial area includes one or more of a restaurant area and a supermarket area.
3. The method for predicting energy consumption of a medical institution according to claim 1, wherein: The classification of the equipment dimension includes one or more of general equipment, large medical equipment, medical equipment belts and emergency equipment; The general equipment includes one or more of lighting equipment, socket equipment, power equipment and environmental equipment; The large medical equipment includes one or more of CT, DR, MR, DSA and linear accelerator.
4. The method for predicting energy consumption of a medical institution according to claim 1, wherein: Dynamically adjusting feature engineering strategies based on the multi-dimensional energy consumption classification system includes: For the regional dimension, regional-specific features are introduced, including: features of increased number of registered patients in the outpatient area and / or features of increased bed occupancy rate and operating status of large medical equipment in the inpatient area; For the equipment dimension, equipment-specific features are introduced, including: additional diagnosis times and power curve features for large medical equipment and / or additional maintenance cycle and power parameter features for environmental equipment; Dynamically switch feature sets based on seasonality strategies and / or device abnormality awareness strategies.
5. The method for predicting energy consumption of a medical institution according to claim 1, wherein: The classified medical institution energy consumption data is decomposed to obtain the intrinsic mode function components and residual components including: Add complementary white noise to the normalized energy consumption time series data; The energy consumption time series data with complementary white noise added thereto is subjected to empirical mode decomposition and averaging processing to obtain the intrinsic mode function component and the residual component.
6. The method for predicting energy consumption of a medical institution according to claim 1, wherein: The method further includes the step of processing the intrinsic mode function component using the intrinsic mode function component prediction model, specifically including: Integrating the decomposed intrinsic mode function components with multi-source features as input to the LSTM model; the multi-source features include one or more of time features, weather features, and business features; Use a multi-layer LSTM structure to capture long-range dependencies; The mean square error is used as the loss function and the Adam optimizer is used to train the LSTM model.
7. The method for predicting energy consumption of a medical institution according to claim 1, wherein: The method further includes the step of processing the residual component using the residual component prediction model, specifically including: The residual components obtained by decomposition are used as inputs of the AutoARIMA model, and features that may affect the long-term trend are added; The AutoARIMA model is expanded into the ARIMAX model, the optimal parameters are searched to meet the AIC information criterion optimal principle, and exogenous variables are added to the ARIMAX model; The exogenous variables include the total rated power of the energy-consuming equipment of the medical institution and / or the area of the department of the medical institution.
8. The method for predicting energy consumption of a medical institution according to claim 1, wherein: Also includes: The energy consumption prediction result of the medical institution is subjected to denormalization processing to obtain the actual energy consumption prediction result of the medical institution.
9. An electronic device, characterized in that: It includes at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the computer program is executed by the processing unit, the processing unit is able to perform the steps of the medical institution energy consumption prediction method described in any one of claims 1-8.
10. A storage medium, characterized in that: It stores a computer program that can be executed by an access authentication device. When the computer program runs on the access authentication device, the access authentication device can execute the steps of the medical institution energy consumption prediction method according to any one of claims 1 to 8.