A method and system for predicting power consumption of medium-temperature chillers based on data assimilation

Abstract

The present invention discloses a method and system for predicting power consumption of a medium-temperature chiller based on data assimilation, comprising: obtaining historical operation data of the medium-temperature chiller; preprocessing the historical operation data and dividing it into a training set and a test set; constructing an LSTM-Attention medium-temperature chiller time series prediction model based on a long short-term memory network LSTM and a multi-scale attention mechanism Attention; using the training set and the test set to train and optimize hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model to obtain an optimal LSTM-Attention medium-temperature chiller time series prediction model; dynamically simulating the LSTM-Attention medium-temperature chiller time series prediction model within a preset time window to obtain a predicted value, and judging whether there is an observed value corresponding to the predicted value, if not, outputting the predicted value; if yes, performing accuracy evaluation on the predicted value and the observed value to judge whether they meet the preset standard.

Description

Technical Field

[0001] The present invention relates to the technical field of air conditioning systems, and in particular to a method and system for predicting power consumption of a medium-temperature chiller based on data assimilation. Background Art

[0002] Chillers account for the largest proportion of energy consumption in central air-conditioning systems. Adjusting the operating conditions of chillers through chiller prediction models is an important technical means to improve the energy efficiency of central air-conditioning systems. Chillers mainly include components such as condensers, evaporators, compressors, and expansion valves. The components are coupled with each other, making it difficult to establish an accurate and fast physical mechanism model. Existing medium-temperature chiller models mainly include semi-empirical and empirical models. Common semi-empirical models include the Gordon-NG model and the Lee model. Both of them simplify the heat transfer process formula of the equipment to obtain the model of the coefficient of performance COP of the medium-temperature chiller. The empirical models mainly include the DOE-2 model and the multivariate polynomial regression model (MP model). These model parameters are usually obtained by fitting a large number of manufacturer performance data using the least squares method. Changing the traditional low-temperature water supply to medium-temperature water supply is a key measure to improve the energy efficiency of chillers. Currently, medium-temperature chillers are in a rapid research and development stage. However, due to the significant differences in the operating states of medium-temperature chillers and traditional chillers, as well as the transient characteristics of actual operating conditions, the performance curve of medium-temperature chillers in actual operation often has a certain gap with the performance curve of traditional chillers, which leads to a decrease in the prediction accuracy of traditional models in actual operation.

[0003] The prediction accuracy of medium-temperature chillers has a significant impact on the operation strategy of air conditioning water systems. Through high-precision prediction models, real-time optimization of the set values ​​of the chilled water supply temperature and flow rate of medium-temperature chillers is carried out to maximize the COP of medium-temperature chillers, which is the key to improving the entire centralized air conditioning system. To this end, it is necessary to combine actual measurement data and historical operation data with the medium-temperature chiller model to provide more accurate prediction results.

[0004] In order to solve the problems of reduced prediction accuracy of existing medium-temperature chiller models during actual operation and difficulty in capturing transient characteristics of actual operating conditions, data assimilation technology can provide more accurate prediction results by combining actual measurement data with mathematical models. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for predicting power consumption of a medium-temperature chiller based on data assimilation, so as to solve the problems that the prediction accuracy of the existing medium-temperature chiller model is low during actual operation and the transient characteristics of the actual operating conditions are difficult to capture.

[0006] A first aspect of the present invention provides a method for predicting power consumption of a medium-temperature chiller based on data assimilation, comprising:

[0007] Obtain historical operation data of medium temperature chillers;

[0008] Preprocessing the historical operation data and dividing it into a training set and a test set;

[0009] Based on the long short-term memory network LSTM and the multi-scale attention mechanism Attention, a LSTM-Attention medium-temperature chiller time series prediction model is constructed;

[0010] Using the training set and the test set to train and optimize hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model to obtain an optimal LSTM-Attention medium-temperature chiller time series prediction model;

[0011] The LSTM-Attention medium-temperature chiller time series prediction model is used to dynamically simulate the prediction value within the preset time window, and determine whether there is an observation value corresponding to the prediction value. If not, the prediction value is output; if yes, the prediction value and the observation value are accurately evaluated to determine whether they meet the preset standard. If yes, the prediction value is output; if not, the LSTM-Attention medium-temperature chiller time series prediction model is adjusted through the ensemble Kalman filter data assimilation algorithm and dynamically simulated and accurately evaluated again until it meets the preset standard; wherein, the observation value is the real-time operation data of the medium-temperature chiller obtained by sensor monitoring.

[0012] Furthermore, the historical operation data and the observed values ​​include: chilled water supply and return water temperature, cooling water inlet and outlet temperatures, chilled water flow rate, cooling water flow rate, power, and cooling capacity.

[0013] Furthermore, the historical operation data is preprocessed and divided into a training set and a test set, specifically:

[0014] The preprocessing includes: data cleaning, data filling and normalization processing;

[0015] The historical operation data after preprocessing is divided into: a training set is composed of several groups of output factors and input factors formed by the historical operation data before x days from the current time, and a test set is composed of several groups of output factors and input factors formed by the historical operation data within y days from the current time. The ratio of the training set to the test set is 8:2.

[0016] Furthermore, the training set and the test set are used to train and optimize the hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model to obtain the optimal LSTM-Attention medium-temperature chiller time series prediction model, specifically:

[0017] The training set and the test set are input into the LSTM-Attention medium-temperature chiller time series prediction model, and the hyperparameter values ​​are adjusted. The hyperparameters include: time window size, number of hidden layer nodes, number of layers, and learning rate; wherein the model with the best performance in the test set is selected as the optimal LSTM-Attention medium-temperature chiller time series prediction model.

[0018] Furthermore, the predicted value is obtained by dynamically simulating the LSTM-Attention medium-temperature chiller time series prediction model within the preset time window, specifically:

[0019] Analyze the historical operation data by principal component analysis to obtain the core influencing factors affecting the power consumption of the medium-temperature chiller;

[0020] Perform data cleaning, data filling and normalization on the core influencing factors;

[0021] The processed core influencing factors are input into the LSTM-Attention medium-temperature chiller time series prediction model. Within a preset observation time window, the LSTM-Attention medium-temperature chiller time series prediction model is used to simulate forward in time to the moment of the first simultaneous phase of external observation data to obtain the predicted value.

[0022] Furthermore, the predicted value and the observed value are evaluated for accuracy to determine whether they meet the preset standards. If they meet the standards, the predicted value is output; if they do not meet the standards, the LSTM-Attention medium-temperature chiller time series prediction model is adjusted through the ensemble Kalman filter data assimilation algorithm and the dynamic simulation and accuracy evaluation are re-performed until they meet the preset standards, specifically:

[0023] Step 1: Assimilate the predicted value and the observed value using an ensemble Kalman filter data assimilation algorithm to obtain an assimilated value of the predicted value and the observed value;

[0024] Step 2: Gaussian filtering is performed on the observed value to obtain the true value, and the relative error between the assimilated value and the true value is used as the objective function for accuracy evaluation.

[0025] When the relative error is less than 5%, the predicted value is output; when the relative error is greater than 5%, the initial state and hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model are updated according to the assimilation value, and the dynamic simulation is re-performed to obtain a new predicted value, and steps one and two are repeated until the relative error is less than 5%.

[0026] A second aspect of the present invention provides a medium-temperature chiller power consumption prediction system based on data assimilation, comprising:

[0027] Data acquisition module: used to obtain historical operation data of the chiller, the historical operation data includes: chilled water supply and return water temperature, cooling water inlet and outlet temperature, chilled water flow, cooling water flow, power, and cooling capacity;

[0028] LSTM-Attention medium-temperature chiller time series prediction model construction module: used to preprocess the historical operation data and divide it into a training set and a test set, and to construct an LSTM-Attention medium-temperature chiller time series prediction model, and use the training set and the test set to train and optimize the LSTM-Attention medium-temperature chiller time series prediction model to obtain the optimal LSTM-Attention medium-temperature chiller time series prediction model;

[0029] Sensor monitoring module: used to monitor the real-time operation data of the medium-temperature chiller to obtain observation values, including: chilled water supply and return water temperature, cooling water inlet and outlet temperature, chilled water flow, cooling water flow, power, and cooling capacity;

[0030] LSTM-Attention medium-temperature chiller time series prediction model prediction module: used to dynamically simulate the LSTM-Attention medium-temperature chiller time series prediction model constructed by the LSTM-Attention medium-temperature chiller time series prediction model construction module to obtain a prediction value

[0031] Data assimilation module: used to assimilate the predicted value and the observed value using the ensemble Kalman filter data assimilation algorithm to obtain the assimilated value of the predicted value and the observed value, and to adjust the LSTM-Attention medium-temperature chiller time series prediction model;

[0032] Output module: used to output the predicted value of the LSTM-Attention medium-temperature chiller time series prediction model;

[0033] Precision evaluation module: used to evaluate the accuracy of the predicted value and the observed value to determine whether they meet the preset standard; if they meet, the predicted value is output through the output module; if they do not meet, the data assimilation module is called to adjust the LSTM-Attention medium-temperature chiller time series prediction model and call the LSTM-Attention medium-temperature chiller time series prediction model prediction module to re-perform dynamic simulation, and perform precision evaluation again until it meets the preset standard;

[0034] Judgment module: used to judge whether there is an observed value corresponding to the predicted value within a preset time window. If not, the predicted value is output through the output module; if yes, the predicted value and the observed value are evaluated for accuracy through the accuracy evaluation module to determine whether they meet the preset standards.

[0035] The present invention has at least the following beneficial effects:

[0036] The present invention constructs a LSTM-Attention medium-temperature chiller time series prediction model based on a long short-term memory network LSTM and a multi-scale attention mechanism, takes into account the influence of historical data on the power consumption of the medium-temperature chiller at the current moment, effectively improves the dynamic adaptability of the LSTM-Attention medium-temperature chiller time series prediction model, enhances the recognition of key features of the LSTM-Attention medium-temperature chiller time series prediction model, and thus improves the prediction ability.

[0037] The present invention assimilates the observation values ​​actually measured by the sensor with the model prediction values ​​through the ensemble Kalman filter data assimilation algorithm, effectively utilizes the actual measurement data to realize the dynamic adjustment of the model, and real-time corrects and optimizes the initial state and hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model, thereby improving the prediction accuracy of the LSTM-Attention medium-temperature chiller time series prediction model, enhancing the reflection of the LSTM-Attention medium-temperature chiller time series prediction model on the current system state of the medium-temperature chiller, and also improving the accuracy and reliability of short-, medium- and long-term predictions; in addition, the method ensures that the LSTM-Attention medium-temperature chiller time series prediction model can flexibly respond to the rapid changes in the state of the medium-temperature chiller, so that the transient characteristics of the actual operating conditions can be captured, and the adaptability of the model is further enhanced.

[0038] The present invention can accurately predict the power consumption of medium-temperature chillers, and the prediction results can help decision makers understand potential system risks, thereby formulating more effective medium-temperature chiller operation strategies and control measures.

[0039] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flow chart of a method for predicting power consumption of a medium-temperature chiller based on data assimilation provided by the present invention;

[0041] Figure 2 A flow chart for constructing a time series prediction model for a medium-temperature chiller using LSTM-Attention provided by the present invention;

[0042] Figure 3 A data assimilation flow chart provided by the present invention;

[0043] Figure 4 A structural schematic diagram of a medium-temperature chiller power consumption prediction system based on data assimilation provided by the present invention. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0045] In the description of the present invention, it is necessary to understand that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship are based on the orientation or position relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0046] Example 1: Combination Figure 1-Figure 4 This embodiment is described.

[0047] The present invention provides a method for predicting power consumption of a medium-temperature chiller based on data assimilation, comprising:

[0048] Obtain historical operation data of medium temperature chillers;

[0049] Preprocessing the historical operation data and dividing it into a training set and a test set;

[0050] Based on the long short-term memory network LSTM and the multi-scale attention mechanism Attention, a LSTM-Attention medium-temperature chiller time series prediction model is constructed;

[0051] The long short-term memory network LSTM architecture is adopted as the basic framework. The LSTM architecture can identify the dynamic characteristics of the medium-temperature chiller system. The multi-scale attention mechanism Attention is applied to the hidden layer output of the LSTM architecture. The multi-scale attention mechanism Attention can capture features at different scales and give higher weights to important features, thereby constructing an LSTM-Attention medium-temperature chiller time series prediction model for medium-temperature chillers.

[0052] Using the training set and the test set to train and optimize hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model to obtain an optimal LSTM-Attention medium-temperature chiller time series prediction model;

[0053] The LSTM-Attention medium-temperature chiller time series prediction model is used to dynamically simulate the prediction value within the preset time window, and determine whether there is an observation value corresponding to the prediction value. If not, the prediction value is output; if yes, the prediction value and the observation value are accurately evaluated to determine whether they meet the preset standard. If yes, the prediction value is output; if not, the LSTM-Attention medium-temperature chiller time series prediction model is adjusted through the ensemble Kalman filter data assimilation algorithm and dynamically simulated and accurately evaluated again until it meets the preset standard; wherein, the observation value is the real-time operation data of the medium-temperature chiller obtained by sensor monitoring.

[0054] The historical operation data of the medium-temperature chiller is obtained, including the chilled water supply and return water temperature, cooling water inlet and outlet temperature, chilled water flow rate, cooling water flow rate, power, cooling capacity, etc. The historical operation data is preprocessed and divided into training sets and test sets to train and optimize the LSTM-Attention medium-temperature chiller time series prediction model based on the long short-term memory network LSTM and the multi-scale attention mechanism Attention. The intrinsic connection between the historical operation data of the power consumption of the medium-temperature chiller and its corresponding influencing factors is solidified in the LSTM-Attention medium-temperature chiller time series prediction model, and the pre-training of the LSTM-Attention medium-temperature chiller time series prediction model is completed to simulate and predict the power consumption of the medium-temperature chiller; the historical operation data is analyzed by principal component analysis to obtain the core influencing factors affecting the power consumption of the medium-temperature chiller, such as chilled water supply and return water temperature, cooling water inlet and outlet temperature, chilled water flow rate, cooling water flow rate, etc. Based on the core influencing factors, the LSTM-Attention medium-temperature chiller time series prediction model is used to simulate the prediction of the power consumption of the medium-temperature chiller in the pre-set observation time window. The prediction value is obtained by simulating the time of the external observation data of the first simultaneous phase, and the observation value is obtained by real-time monitoring of the medium-temperature chiller operation data through sensors; since the time step of the prediction model may be one per minute, and the observation value may have a longer time step, such as one every 20 minutes, it is necessary to determine whether there is an observation value corresponding to the prediction value within the preset time window. If not, the prediction value is directly output. If it exists, the prediction value and the observation value are evaluated for accuracy to determine whether they meet the preset standard, and then the accuracy and reliability of the LSTM-Attention medium-temperature chiller time series prediction model can be evaluated. The prediction value and the observation value are evaluated for accuracy to determine whether they meet the preset standard. If they meet the preset standard, the prediction value is output; if they do not meet the preset standard, the ensemble Kalman filter data assimilation algorithm is used to fuse the observation value into the LSTM-Attention medium-temperature chiller time series prediction model to correct the initial state and hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model, and the dynamic simulation and accuracy evaluation are re-performed until the preset standard is met to complete the prediction of the medium-temperature chiller power consumption.

[0055] The present invention constructs an LSTM-Attention medium-temperature chiller time series prediction model based on a long short-term memory network LSTM and a multi-scale attention mechanism Attention, which takes into account the influence of historical operation data on the power consumption of the medium-temperature chiller at the current moment, effectively improves the dynamic adaptability of the LSTM-Attention medium-temperature chiller time series prediction model, enhances the recognition of key features of the LSTM-Attention medium-temperature chiller time series prediction model, and thus improves the prediction ability.

[0056] The present invention assimilates the observation values ​​actually measured by the sensor with the model prediction values ​​through the ensemble Kalman filter data assimilation method, effectively utilizes the actual measurement data to realize the dynamic adjustment of the model, and real-time corrects and optimizes the initial state and hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model, thereby improving the prediction accuracy of the LSTM-Attention medium-temperature chiller time series prediction model, enhancing the reflection of the LSTM-Attention medium-temperature chiller time series prediction model on the current system state of the medium-temperature chiller, and also improving the accuracy and reliability of short-, medium- and long-term predictions; in addition, the method ensures that the LSTM-Attention medium-temperature chiller time series prediction model can flexibly respond to the rapid changes in the state of the medium-temperature chiller, so that the transient characteristics of the actual operating conditions can be captured, and the adaptability of the model is further enhanced.

[0057] The present invention can accurately predict the power consumption of medium-temperature chillers, and the prediction results can help decision makers understand potential system risks, thereby formulating more effective medium-temperature chiller operation strategies and control measures.

[0058] Furthermore, the historical operation data and the observed values ​​include: chilled water supply and return water temperature, cooling water inlet and outlet temperatures, chilled water flow rate, cooling water flow rate, power, and cooling capacity.

[0059] Furthermore, the historical operation data is preprocessed and divided into a training set and a test set, specifically:

[0060] The preprocessing includes: data cleaning, data filling and normalization processing;

[0061] The preprocessed historical operation data is divided into data: a training set is composed of several groups of output factors and input factors formed by the historical operation data before x days from the current time, and a test set is composed of several groups of output factors and input factors formed by the data within y days from the current time. The ratio of the training set to the test set is 8:2.

[0062] Furthermore, the training set and the test set are used to train and optimize the hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model to obtain the optimal LSTM-Attention medium-temperature chiller time series prediction model, specifically:

[0063] The training set and the test set are input into the LSTM-Attention medium-temperature chiller time series prediction model, and the hyperparameter values ​​are adjusted. The hyperparameters include: time window size, number of hidden layer nodes, number of layers, and learning rate; wherein the model with the best performance in the test set is selected as the optimal LSTM-Attention medium-temperature chiller time series prediction model.

[0064] The training set and test set were respectively input into the LSTM-Attention medium-temperature chiller time series prediction model, and the hyperparameter values ​​were adjusted. The LSTM-Attention medium-temperature chiller time series prediction model with different time window size, number of hidden layer nodes, number of layers, learning rate and other hyperparameters was trained and tested multiple times. The model with the best performance in the test set was selected as the optimal LSTM-Attention medium-temperature chiller time series prediction model. The prediction ability of the LSTM-Attention medium-temperature chiller time series prediction model was improved by continuously optimizing the hyperparameters.

[0065] Furthermore, the predicted value is obtained by dynamically simulating the LSTM-Attention medium-temperature chiller time series prediction model within the preset time window, specifically:

[0066] Analyze the historical operation data by principal component analysis to obtain the core influencing factors affecting the power consumption of the medium-temperature chiller;

[0067] Perform data cleaning, data filling and normalization on the core influencing factors;

[0068] The processed core influencing factors are input into the LSTM-Attention medium-temperature chiller time series prediction model. Within a preset observation time window, the LSTM-Attention medium-temperature chiller time series prediction model is used to simulate forward in time to the moment of the first simultaneous phase of external observation data to obtain the predicted value.

[0069] The principal component analysis method is used to analyze the historical operating data that has not been preprocessed to obtain the core influencing factors that affect the power consumption of the medium-temperature chiller. The core influencing factors are then cleaned, filled and normalized. The processed core influencing factors are then input into the LSTM-Attention time series prediction model of the medium-temperature chiller. Within the pre-set observation time window, the LSTM-Attention time series prediction model of the medium-temperature chiller is run for dynamic simulation, continuously simulating the changes in the energy consumption and energy efficiency of the medium-temperature chiller, and collecting the predicted values ​​at the end of each observation window.

[0070] Furthermore, the predicted value and the observed value are evaluated for accuracy to determine whether they meet the preset standards. If they meet the standards, the predicted value is output; if they do not meet the standards, the LSTM-Attention medium-temperature chiller time series prediction model is adjusted through the ensemble Kalman filter data assimilation algorithm and the dynamic simulation and accuracy evaluation are re-performed until they meet the preset standards, specifically:

[0071] Step 1: Assimilate the predicted value and the observed value using an ensemble Kalman filter data assimilation algorithm to obtain an assimilated value of the predicted value and the observed value;

[0072] Step 2: Gaussian filtering is performed on the observed value to obtain the true value, and the relative error between the assimilated value and the true value is used as the objective function for accuracy evaluation.

[0073] When the relative error is less than 5%, the predicted value is output; when the relative error is greater than 5%, the initial state and hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model are updated according to the assimilation value, and the dynamic simulation is re-performed to obtain a new predicted value, and steps one and two are repeated until the relative error is less than 5%.

[0074] Analyze the statistical characteristics of the prediction error and observation error of the LSTM-Attention medium-temperature chiller time series prediction model, including the mean and covariance of the error. The statistical characteristics are used to weigh the trust of the model prediction and observation data in the assimilation process.

[0075] The observations are cleaned, filled and normalized, and then the processed observations, observation operators and error statistics are input into the ensemble Kalman filter data assimilation algorithm. The predicted values ​​and observed values ​​are then assimilated by the ensemble Kalman filter data assimilation algorithm to obtain the assimilated values ​​of the predicted values ​​and observed values.

[0076] The new parameter value obtained by Gaussian filtering the observation value measured by the sensor is taken as the true value. Based on the true value, the relative error between the assimilated value and the true value is used as the objective function to evaluate the accuracy. When the relative error is less than 5%, the LSTM-Attention medium-temperature chiller time series prediction model is considered to have high accuracy and the prediction result, i.e., the prediction value, can be output. When the relative error is greater than 5%, the model accuracy is considered to be low, and it is necessary to update the initial state and hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model and re-perform dynamic simulation to obtain a new prediction value. The data assimilation and accuracy evaluation steps are repeated until the relative error is less than 5%, and finally the new prediction value is output.

[0077] The present invention adopts a data assimilation method to combine actual measurement data with model prediction data, and adjusts the model hyperparameters based on the assimilated data, ultimately achieving accurate prediction of the power consumption of medium-temperature chillers, solving the problem of low prediction accuracy of existing medium-temperature chillers in actual engineering operation.

[0078] Embodiment 2:

[0079] The present invention also provides a medium-temperature chiller power consumption prediction system based on data assimilation, comprising:

[0080] Data acquisition module 10: used to acquire historical operation data of the medium temperature chiller, the historical operation data including: chilled water supply and return water temperature, cooling water inlet and outlet temperature, chilled water flow rate, cooling water flow rate, power, and cooling capacity;

[0081] LSTM-Attention medium-temperature chiller time series prediction model construction module 20: used to preprocess the historical operation data and divide it into a training set and a test set, and to construct an LSTM-Attention medium-temperature chiller time series prediction model, and use the training set and the test set to train and optimize the LSTM-Attention medium-temperature chiller time series prediction model to obtain the optimal LSTM-Attention medium-temperature chiller time series prediction model;

[0082] Sensor monitoring module 30: used to monitor the real-time operation data of the medium-temperature chiller to obtain observation values, which include: chilled water supply and return water temperature, cooling water inlet and outlet temperature, chilled water flow rate, cooling water flow rate, power, and cooling capacity;

[0083] LSTM-Attention medium-temperature chiller time series prediction model prediction module 40: used to dynamically simulate the LSTM-Attention medium-temperature chiller time series prediction model constructed by the LSTM-Attention medium-temperature chiller time series prediction model construction module to obtain a prediction value;

[0084] Data assimilation module 50: used to assimilate the predicted value and the observed value using the ensemble Kalman filter data assimilation algorithm to obtain the assimilated value of the predicted value and the observed value, and to adjust the LSTM-Attention medium-temperature chiller time series prediction model;

[0085] Output module 60: used to output the predicted value of the LSTM-Attention medium-temperature chiller time series prediction model;

[0086] Precision evaluation module 70: used to evaluate the accuracy of the predicted value and the observed value to determine whether they meet the preset standard; if they meet, the predicted value is output through the output module; if they do not meet, the data assimilation module is called to adjust the LSTM-Attention medium-temperature chiller time series prediction model and call the LSTM-Attention medium-temperature chiller time series prediction model prediction module to re-perform dynamic simulation, and perform precision evaluation again until it meets the preset standard;

[0087] Judgment module 80: used to judge whether there is an observed value corresponding to the predicted value within a preset time window. If not, the predicted value is output through the output module; if yes, the predicted value and the observed value are evaluated for accuracy through the accuracy evaluation module to judge whether they meet the preset standards.

[0088] It should also be noted that the term "includes", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements. The words "first", "second" and the like are used to indicate names, but do not indicate any specific order. The above schematically describes the invention and its implementation methods, which is not restrictive. The invention can be implemented in other specific forms without departing from the spirit or basic features of the invention. What is shown in the accompanying drawings is only one of the implementation methods of the invention, and the actual structure is not limited thereto. Any figure mark in the claims should not limit the claims involved. Therefore, if a person of ordinary skill in the art is inspired by it, and does not deviate from the purpose of the invention, and designs a structural mode and an embodiment similar to the technical solution without creativity, they should all belong to the protection scope of the invention.

Claims

1. A method for predicting power consumption of a medium-temperature chiller based on data assimilation, characterized in that: include: Obtain historical operation data of medium temperature chillers; Preprocessing the historical operation data and dividing it into a training set and a test set; Based on the long short-term memory network LSTM and the multi-scale attention mechanism Attention, a LSTM-Attention medium-temperature chiller time series prediction model is constructed; Using the training set and the test set to train and optimize hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model to obtain an optimal LSTM-Attention medium-temperature chiller time series prediction model; The predicted value is obtained by dynamically simulating the LSTM-Attention medium-temperature chiller time series prediction model within the preset time window, and it is determined whether there is an observed value corresponding to the predicted value. If not, the predicted value is output; if so, the predicted value and the observed value are evaluated for accuracy to determine whether they meet the preset standard. If they meet, the predicted value is output; if not, the LSTM-Attention medium-temperature chiller time series prediction model is adjusted by the ensemble Kalman filter data assimilation algorithm and the dynamic simulation and accuracy evaluation are re-performed until the preset standard is met, specifically: Step 1: Assimilate the predicted value and the observed value using an ensemble Kalman filter data assimilation algorithm to obtain an assimilated value of the predicted value and the observed value; Step 2: Gaussian filtering is performed on the observed value to obtain the true value, and the relative error between the assimilated value and the true value is used as the objective function for accuracy evaluation. When the relative error is less than 5%, the predicted value is output; when the relative error is greater than 5%, the initial state and hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model are updated according to the assimilated value, and the dynamic simulation is re-performed to obtain a new predicted value, and steps one and two are repeated until the relative error is less than 5%; The observed value is the real-time operating data of the medium-temperature chiller obtained by sensor monitoring.

2. The method for predicting power consumption of a medium-temperature chiller based on data assimilation according to claim 1, characterized in that: The historical operation data and the observed values ​​include: chilled water supply and return water temperature, cooling water inlet and outlet temperatures, chilled water flow rate, cooling water flow rate, power, and cooling capacity.

3. The method for predicting power consumption of a medium-temperature chiller based on data assimilation according to claim 1 is characterized in that: The historical operation data is preprocessed and divided into a training set and a test set, specifically: The preprocessing includes: data cleaning, data filling and normalization processing; The preprocessed historical operation data is divided into data: a training set is composed of several groups of output factors and input factors formed by the historical operation data before x days from the current time, and a test set is composed of several groups of output factors and input factors formed by the data within y days from the current time. The ratio of the training set to the test set is 8:

2.

4. The method for predicting power consumption of a medium-temperature chiller based on data assimilation according to claim 1 is characterized in that: The training set and the test set are used to train and optimize the hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model to obtain the optimal LSTM-Attention medium-temperature chiller time series prediction model, specifically: The training set and the test set are input into the LSTM-Attention medium-temperature chiller time series prediction model, and the hyperparameter values ​​are adjusted. The hyperparameters include: time window size, number of hidden layer nodes, number of layers, and learning rate; wherein the model with the best performance in the test set is selected as the optimal LSTM-Attention medium-temperature chiller time series prediction model.

5. The method for predicting power consumption of a medium-temperature chiller based on data assimilation according to claim 1, characterized in that: The predicted value is obtained by dynamically simulating the LSTM-Attention medium-temperature chiller time series prediction model within the preset time window, specifically: Analyze the historical operation data by principal component analysis to obtain the core influencing factors affecting the power consumption of the medium-temperature chiller; Perform data cleaning, data filling and normalization on the core influencing factors; The processed core influencing factors are input into the LSTM-Attention medium-temperature chiller time series prediction model. Within a preset observation time window, the LSTM-Attention medium-temperature chiller time series prediction model is used to simulate forward in time to the moment of the first simultaneous phase of external observation data to obtain the predicted value.

6. A medium temperature chiller power consumption prediction system based on data assimilation, characterized in that: include: Data acquisition module: used to obtain historical operation data of the chiller, the historical operation data includes: chilled water supply and return water temperature, cooling water inlet and outlet temperature, chilled water flow, cooling water flow, power, and cooling capacity; LSTM-Attention medium-temperature chiller time series prediction model construction module: used to preprocess the historical operation data and divide it into a training set and a test set, and to construct an LSTM-Attention medium-temperature chiller time series prediction model, and use the training set and the test set to train and optimize the hyperparameters of the LSTM-Attention medium-temperature chiller time series prediction model to obtain the optimal LSTM-Attention medium-temperature chiller time series prediction model; Sensor monitoring module: used to monitor the real-time operation data of the medium-temperature chiller to obtain observation values, including: chilled water supply and return water temperature, cooling water inlet and outlet temperature, chilled water flow, cooling water flow, power, and cooling capacity; LSTM-Attention medium-temperature chiller time series prediction model prediction module: used to dynamically simulate the LSTM-Attention medium-temperature chiller time series prediction model constructed by the LSTM-Attention medium-temperature chiller time series prediction model construction module to obtain a prediction value; Data assimilation module: used to assimilate the predicted value and the observed value using the ensemble Kalman filter data assimilation algorithm to obtain the assimilated value of the predicted value and the observed value, and to adjust the LSTM-Attention medium-temperature chiller time series prediction model; Output module: used to output the predicted value of the LSTM-Attention medium-temperature chiller time series prediction model; Precision evaluation module: used to evaluate the accuracy of the predicted value and the observed value to determine whether they meet the preset standard; if they meet, the predicted value is output through the output module; if they do not meet, the data assimilation module is called to adjust the LSTM-Attention medium-temperature chiller time series prediction model and call the LSTM-Attention medium-temperature chiller time series prediction model prediction module to re-perform dynamic simulation, and perform precision evaluation again until it meets the preset standard; Judgment module: used to judge whether there is an observed value corresponding to the predicted value within a preset time window. If not, the predicted value is output through the output module; if yes, the predicted value and the observed value are evaluated for accuracy through the accuracy evaluation module to determine whether they meet the preset standards.

Images