Ice storage load prediction method and system based on data and model driving

Abstract

The invention discloses an ice storage load prediction method and system based on data and model driving, and the method comprises the steps: determining the ice storage amount of an ice storage system in a power trough period through the rough prediction of the load in the next 24 hours, and carrying out the hourly prediction of the ice storage load through an ice storage load hourly prediction model, so as to control the ice release amount. The problem that in traditional hourly load prediction, due to the fact that prediction errors occur in a certain hour, other hours are affected is solved. Meanwhile, a simulation model is established through simulation software to obtain load data, and the reliability of the model is verified through a small amount of real data, so that the problem that the data for training is less when a new building is built, a few buildings are monitored or a data privacy problem is involved is solved; besides, energy-saving data are introduced into the simulation model, and the situation that the ice storage system is guided to run in a non-energy-saving state due to the fact that non-energy-saving data exist in a data set can be avoided by using the data set for prediction according to energy-saving related specifications.

Description

Technical Field

[0001] The present invention belongs to the technical field of ice storage technology, and in particular relates to a method and system for predicting ice storage load based on data and model driving. Background Art

[0002] Building cooling and heating load forecasting refers to the process of predicting and estimating the cooling load (cooling demand) and heating load (heating demand) of a building in the future. Energy consumption and carbon emissions during the operation phase of a building are in the entire life cycle of the building. Under the national goal of "carbon peak and carbon neutrality", optimizing the operation strategy of the building energy system is the main means to achieve building energy conservation and emission reduction. Building load forecasting is of great significance to the optimization of the operation strategy of the energy system.

[0003] Ice storage system is a technology that uses dual-mode refrigeration host to freeze the cold storage medium and store the cold energy during the low electricity consumption period at night when the power load is very low. During the day when the power load is high, that is, during the peak electricity consumption period, the stored cold energy will be released to meet the needs of building air conditioning or production process.

[0004] Ice storage systems need to make predictions for the next 24 hours to ensure that the ice-making and ice-melting processes are closely aligned with the actual cooling needs of the building, avoiding excessive ice-making or insufficient cooling, thereby maximizing energy efficiency. Through accurate hourly load forecasting, ice-making can be selected during low-price periods to reduce electricity consumption during peak-price periods and reduce overall electricity costs. Load forecasting enables the system to dynamically adjust operating strategies based on real-time changes in cooling demand to ensure that the system can provide a stable supply of cooling during peak load periods. Respond to sudden load changes: predict sudden load increases or decreases, adjust operating modes in advance, and avoid system overload or insufficient cooling. Machine learning is a commonly used method in building cooling and heating load prediction problems. Due to its powerful feature extraction and data fitting capabilities, it has a good effect on target buildings with rich data.

[0005] However, there are the following problems in predicting the ice storage system: 1. When faced with newly built buildings, less monitored buildings, or when data privacy issues are involved, there is less data available for training and the accuracy of the predictions will be reduced.

[0006] 2. For projects with abundant historical load data, if these historical load data are not collected and operated in accordance with energy-saving related specifications, then the load forecasting based on machine learning based only on these historical load data will reflect a load situation that is not energy-saving. The load predicted by the machine learning model will not be able to accurately reflect the actual situation under the energy-saving state. Therefore, such prediction results have limited significance for energy-saving evaluation and guidance.

[0007] 3. For buildings of different formats, their operation schedules do differ significantly. These differences are not only reflected in the normal working hours of the day, but also in the unconventional working hours, such as rest hours, overtime hours, and non-operating hours. Since the weights of the factors affecting the cooling load (outdoor weather conditions, effective energy supply area, indoor demand conditions) are different at each time, therefore, the unified processing and analysis of historical load data under the same conditions in different time periods has a greater impact on the accuracy of the prediction.

[0008] 4. For an ice storage air conditioning system, which is an important form of air conditioning system, it is often necessary to predict the load for the next 24 hours due to the influence of advance ice storage. To obtain the load for the next 24 hours, an iterative prediction method is often chosen, that is, the first prediction can obtain the load for the next hour, and then use it as an input variable in the load prediction model for the next 2 hours. Other parameters remain unchanged, and so on, the load for the next 24 hours can be obtained. In this scheme, if there is an error in the prediction of a certain hour, other hours will be affected. Summary of the invention

[0009] The object of the present invention is to provide a method and system for predicting ice storage load based on data and model driving.

[0010] In a first aspect, the present invention provides a method for predicting ice storage load based on data and model driving, which comprises the following steps: Step 1, obtaining load data of the ice storage system under test and environmental parameters affecting the ice storage load as a data set; Step 2: Perform cluster analysis on the samples in the data set to identify the load data and environmental parameter data corresponding to the ice storage system in different usage states of the building, and perform load prediction for different usage states of the building; Step 3: Select key parameters from the environmental parameters, filter out key parameters from the samples and update the data set; when starting cold storage, obtain key parameters of the tested ice storage system at the target time, and predict the load of the tested ice storage system at the target time; determine the ice storage capacity based on the predicted load of the tested ice storage system; Step 4: During the cooling process, the ice storage load is predicted hourly by using the ice storage load hourly prediction model. The specific process is as follows: 4-1. Construct an hourly prediction model for ice storage load; the hourly prediction model for ice storage load includes an input layer, one or more fully connected layers, and an output layer connected in sequence; 4-2. Use the dataset to train the hourly prediction model for ice storage load; 4-3. Input the key parameters at the target time, the current load data and the remaining ice volume into the hourly prediction model of the ice storage load, predict the ice storage load hourly, and control the ice storage system according to the hourly predicted ice storage load.

[0011] Preferably, in step 1, the load data obtained is energy-saving load data; the energy-saving load data includes simulated energy-saving load data, and the method for obtaining the simulated energy-saving load data is as follows: A thermodynamic model is established based on the physical property data, building material property data, usage data and energy-saving data of the building where the ice storage system under test is located. The thermodynamic model is verified through the historical load data of the ice storage system under test, and the parameters of the thermodynamic model are adjusted. The simulated energy-saving load data at different times are obtained through simulation of the adjusted thermodynamic model.

[0012] Preferably, the energy-saving load data also includes actual energy-saving load data in historical load data; the process of screening actual energy-saving load data in historical load data is: through a thermodynamic model under the same boundary conditions, the simulated energy-saving load data and the historical load data are compared and analyzed, and the data in the historical load data that meets the energy-saving requirements are screened out as actual energy-saving load data.

[0013] Preferably, in step 4, after using the predicted load to control the ice storage, if the actual ice storage load is greater than the predicted load, then check for energy-inefficient factors in the building until the actual ice storage load is equal to the predicted load.

[0014] Preferably, the environmental parameters include outdoor temperature, humidity, wind speed, solar radiation intensity and internal heat source; the key parameters selected are outdoor temperature, humidity and solar radiation intensity.

[0015] Preferably, in step 2, the cluster analysis method is adopted k Mean clustering algorithm.

[0016] Preferably, in step 2, the method for selecting key parameters is: analyzing the correlation between environmental parameters and loads respectively by using the Pearson correlation coefficient, and selecting parameters with correlation coefficients in the range of 0.7 to 1 as key parameters.

[0017] Preferably, in step 3, the specific process of predicting load is as follows: Calculate the distance between the key parameters at the target time and the key parameters in the data set; use one of the Euclidean distance, Manhattan distance and Chebyshev distance to calculate the distance; obtain the best weighted KNN load forecasting algorithm through cross validation or grid search KValue, select the key parameter in the data set that is closest to the key parameter at the target time K eigenvalues ​​as neighbors, and K The neighbors are weighted to obtain the predicted load.

[0018] As a preference, calculate the distance between the key parameters at the target time and the key parameters in the data set d ( x , y ) is:

[0019] in, x i and y i They are the key parameters at the target moment and the key parameters in the data set respectively; n is the number of key parameters.

[0020] Preferably, in step 4, the hidden layer includes three fully connected layers connected in sequence, and the three fully connected layers all use a ReLU activation function.

[0021] In the second aspect, the present invention provides an ice storage system based on hierarchical multi-objective optimization, characterized in that: it includes a dual-mode refrigeration main unit, an ice storage device, a water pump system, a cooling tower and a controller; the water pump system includes a freezing pump and a cooling pump; the controller is used to execute the above-mentioned ice storage load prediction method; in the cold storage state, the controller determines the ice storage amount according to the predicted load of the ice storage system under test at the target time, controls the freezing pump to transport the chilled water to the ice storage device, and controls the cooling pump to transport the cooling water generated by the dual-mode refrigeration main unit to the cooling tower for heat dissipation; in the cold release state, the controller controls the ice storage device to melt ice according to the hourly predicted ice storage load, and the melted cooling water is transported to the air conditioner or other cooling system through the cooling pump.

[0022] The present invention has the following beneficial effects: 1. The present invention determines the ice storage capacity of the ice storage system during the power off-peak period by roughly predicting the load in the next 24 hours, and controls the ice release capacity by predicting the ice storage load hourly according to the actual load several hours before the target time through the ice storage load hourly prediction model, thus overcoming the problem in traditional hourly load prediction that errors in the prediction of one hour will affect other hours.

[0023] 2. The present invention acquires load data by establishing a simulation model through simulation software, verifies the reliability of the model through a small amount of real data, enriches the data set through the data of the model, improves the prediction accuracy, and solves the problem of less data for training in newly built buildings, buildings with less monitoring, or involving data privacy issues; at the same time, the present invention introduces energy-saving data into the simulation model, which can make the data set used for prediction comply with energy-saving related specifications, and can filter out energy-saving data in actual data, avoiding the ice storage system being guided to operate in a non-energy-saving state due to the presence of non-energy-saving data in the data set.

[0024] 3. The present invention uses cluster analysis to distinguish between normal working time periods and irregular working time periods, such as rest time periods, overtime time periods, and non-operating time periods. The working status of different working time periods is clustered to establish a database for classification decision-making and improve prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of the overall process of the present invention.

[0026] Figure 2 It is a schematic diagram of the weighted KNN load forecasting process in the present invention.

[0027] Figure 3 It is a schematic diagram of the hourly prediction model of ice storage load constructed in the present invention. DETAILED DESCRIPTION

[0028] The present invention will be further described below in conjunction with the accompanying drawings.

[0029] like Figure 1 As shown, a data and model-driven ice storage load prediction method includes the following steps:

[0030] Step 1: Obtain the physical property data, building material property data, usage data and energy-saving data of the building where the ice storage system is located; the physical property data include length, width, height and number of floors; the building material property data include the thickness and U value (thermal conductivity) of the building material; the usage data include room functions (such as office, conference room, toilet, etc.), indoor equipment usage time and room usage schedule. Energy-saving data include indoor temperature and humidity, indoor fresh air (indoor ventilation system) and lighting equipment switching time, as shown in Table 1, Table 2 and Table 3.

[0031] Table 1 Indoor temperature at different times in the heating and air-conditioning area

[0032] For different types of rooms, the requirements of Table 1 of the energy-saving specification should be followed to prevent the indoor temperature from being seriously lower than the design requirements and to avoid turning on the air conditioner when not in operation, which would cause the cooling load to be greater than the design. At the same time, the load forecast based on this method can help on-site operation and maintenance personnel to troubleshoot problems.

[0033] Table 2 Lighting switch duration

[0034] The opening of indoor fresh air and equipment should comply with the requirements of Table 2 and Table 3. Different types of rooms should meet the requirements of energy-saving specifications and avoid opening equipment and fresh air systems during non-working hours to prevent energy waste.

[0035] Table 3 Fresh air operation status

[0036] Step 2: Use TRNSYS or EnergyPlus software to build a thermodynamic model based on the building data obtained in step 1. Verify the thermodynamic model by collecting historical load data of the ice storage system and adjust the relevant parameters of the thermodynamic model to make the thermodynamic model more accurate and reliable. The heat load and cooling demand of the building at different times are simulated by the adjusted thermodynamic model, and then the load of the ice storage system under test at different times is obtained to make up for the problem that the historical load data of the ice storage system under test is small or there is non-energy-saving data; at the same time, the thermodynamic model is used to compare and analyze the real-time data on site under the same boundary conditions to determine whether the on-site data is non-energy-saving data.

[0037] Step 3: Data processing A data set is established based on the energy-saving load data, remaining ice volume, and environmental parameters that affect load prediction of the tested ice storage system. A set of energy-saving load data and environmental parameters is used as a sample in the data set. Environmental parameters that affect load prediction include outdoor temperature, humidity, wind speed, solar radiation intensity, and internal heat source. The data set is marked with timestamps, and the sensors of the building air conditioning system (temperature and humidity sensors, solar radiation sensors, and calorimeters) are used to record in real time the type of building where the tested ice storage system is located, including office buildings, hotels, shopping malls, hospitals, and factories. The k-means algorithm ( k The mean clustering algorithm) clusters the energy-saving load data and environmental parameters of the building in different usage states (working day operation state, working day non-operation state, non-working day state). The specific implementation process is as follows: 3-1. Preprocess the data set; if there are data segments missing for more than 2 hours in a row in the data set, it is considered as faulty data and removed. For data segments missing for less than 2 hours, interpolation is used to supplement the corresponding data. The Laida criterion is used to identify and remove abnormal data.

[0038] 3-2. The data set is normalized using the min-max method. The normalization formula is:

[0039] in, X ՚ is the processed data; X min and X max are the minimum and maximum values ​​in the data; X is the original data.

[0040] 3-3. Use k-means (k-means clustering algorithm) to perform cluster analysis on the normalized data set to identify the energy-saving load data corresponding to different usage states of the building. The specific process is as follows: Select relevant features for cluster analysis, including time (hour, day of the week), temperature, humidity, solar radiation, and energy consumption. Determine the optimal number of clusters k by the "elbow method" or "silhouette coefficient". Randomly select k initial center points. Assign each data point to the cluster to which the center point closest to it belongs. Recalculate the center point of each cluster, which is the average value of the data points in the cluster. Repeat the process of assigning data points and updating center points until the center point no longer changes significantly or reaches the preset number of iterations. Based on the clustering results, identify different operating time periods.

[0041] 3-4. Use the statistical method person correlation coefficient (Pearson correlation coefficient) to analyze the correlation between all environmental parameters and cold and hot loads in each cluster of data after clustering in step three, and select parameters with correlation coefficients ranging from 0.7 to 1 as key parameters, and screen out the key parameters in all samples as well as the energy-saving load data and the remaining ice volume update data set.

[0042] In this embodiment, the key parameters selected are outdoor temperature, humidity and solar radiation intensity.

[0043] Step 4: Rough Forecast like Figure 2 As shown, when cold storage begins, the key parameters of the next 24 hours are used as the predicted characteristic values, and the weighted KNN load prediction algorithm is called; the distance between the predicted characteristic value and the key parameter in the data set is calculated; the method for calculating the distance adopts one of the Euclidean distance, Manhattan distance and Chebyshev distance; in this embodiment, the Euclidean distance is used to calculate the distance between the predicted characteristic value and the remaining characteristic values ​​in the data set. d ( x , y ), and its calculation formula is:

[0044] in, x i and y i are the predicted feature values ​​and the key parameters in the data set respectively; n is the number of key parameters.

[0045] Obtain the best weighted KNN load forecasting algorithm through cross validation or grid search K The K value is selected to balance the classification accuracy and computational efficiency; the closest feature value to the predicted feature value is selected in the data set. K key parameters as neighbors and K The neighbors are weighted to obtain the predicted load of the ice storage system. The closer the neighbors are, the greater the weight is, so as to improve the prediction accuracy. The ice storage capacity of the ice storage system the night before is determined by predicting the load of the ice storage system.

[0046] Step 5: Hourly Forecast like Figure 3 As shown in the figure, the accuracy of the direct load forecast for the next 24 hours can meet the ice-making demand of the ice storage system the night before, but it cannot accurately control the operation of the ice storage system to meet the cooling demand in a specific time period. Since DNN (deep neural network) has powerful nonlinear fitting and feature extraction capabilities, the hourly load forecast for the next 24 hours through DNN can be refined and corrected hour by hour for the initial forecast value, thereby improving the accuracy and reliability of the forecast.

[0047] 5-1. Construct an hourly prediction model for ice storage load; the hourly prediction model for ice storage load includes an input layer, a hidden layer, and an output layer connected in sequence. The input data of the input layer are key parameter data and energy-saving load data; the hidden layer includes three fully connected layers, each with 14 nodes, and all use the ReLU activation function.

[0048] 5-2. Divide the data set into a training set and a test set; use the training set to train the hourly prediction model of ice storage load; set the learning rate to 0.1, which determines the update step size of the model parameters during the training process. Select stochastic gradient descent (SGD) as the optimization algorithm to minimize the loss function and update the model weights.

[0049] 5-3. Input the test set into the trained ice storage load hourly prediction model for prediction and compare it with the true value. The performance of the trained deep neural network is evaluated by R2 square value, mean absolute error and mean absolute percentage error.

[0050] 5-4. During the cooling process, the real-time key parameter data, the remaining ice volume, and the energy-saving load data at t-1, t-2, and t-3 are respectively input into the trained ice storage load hourly prediction model to predict t The ice storage load is monitored at all times, and the ice storage system is controlled according to the hourly predicted ice storage load. If the actual ice storage load is greater than the predicted load, it means that the indoor temperature of the building is set too low, some rooms have air conditioners turned on when not in use, doors and windows are not closed, and the fresh air volume is too large. The building should be checked until the actual ice storage load is equal to the predicted load.

Claims

1. A data and model driven ice storage load prediction method, characterized by: The following steps are involved: Step 1, obtaining load data of the ice storage system under test and environmental parameters affecting the ice storage load as a data set; Step 2: Perform cluster analysis on the samples in the data set to identify the load data and environmental parameter data corresponding to the ice storage system in different usage states of the building, and perform load prediction for different usage states of the building; Step 3: Select key parameters from the environmental parameters, filter out the key parameters in the sample and update the data set; obtain the key parameters of the tested ice storage system at the target time, and predict the load of the tested ice storage system at the target time; The load predicted by the ice storage system under test determines the amount of ice storage; Step 4: During the cooling process, the ice storage load is predicted hourly by using the ice storage load hourly prediction model. The specific process is as follows: 4-1. Construct an hourly prediction model for ice storage load; the hourly prediction model for ice storage load includes an input layer, one or more fully connected layers, and an output layer connected in sequence; 4-2. Use the dataset to train the hourly prediction model for ice storage load; 4-3. Input the key parameters at the target time, the current load data and the remaining ice volume into the hourly prediction model of the ice storage load, predict the ice storage load hourly, and control the ice storage system according to the hourly predicted ice storage load.

2. The ice storage load prediction method based on data and model drive according to claim 1 is characterized in that: In the step 1, the load data obtained is energy-saving load data; the energy-saving load data includes simulated energy-saving load data, and the method for obtaining the simulated energy-saving load data is as follows: A thermodynamic model is established based on the physical property data, building material property data, usage data and energy-saving data of the building where the ice storage system under test is located. The thermodynamic model is verified through the historical load data of the ice storage system under test, and the parameters of the thermodynamic model are adjusted. The simulated energy-saving load data at different times are obtained through simulation of the adjusted thermodynamic model.

3. The ice storage load prediction method based on data and model drive according to claim 2 is characterized by: The energy-saving load data also includes actual energy-saving load data in historical load data; the process of screening actual energy-saving load data in historical load data is: through the thermodynamic model under the same boundary conditions, the simulated energy-saving load data and the historical load data are compared and analyzed, and the data that meets the energy-saving needs in the historical load data are screened out as the actual energy-saving load data.

4. The ice storage load prediction method based on data and model drive according to claim 2 is characterized in that: In the step 4, after using the predicted load to control the ice storage, if the actual ice storage load is greater than the predicted load, then check the non-energy-saving factors in the building until the actual ice storage load is equal to the predicted load.

5. The ice storage load prediction method based on data and model drive according to claim 1 is characterized in that: The environmental parameters include outdoor temperature, humidity, wind speed, solar radiation intensity and internal heat source; the key parameters selected are outdoor temperature, humidity and solar radiation intensity.

6. The ice storage load prediction method based on data and model drive according to claim 1 is characterized by: In the step 2, the cluster analysis method is adopted k Mean clustering algorithm.

7. The ice storage load prediction method based on data and model drive according to claim 1 is characterized by: In the step 2, the method for selecting key parameters is: analyzing the correlation between environmental parameters and loads respectively through the Pearson correlation coefficient, and selecting parameters with correlation coefficients in the range of 0.7 to 1 as key parameters.

8. The ice storage load prediction method based on data and model drive according to claim 1 is characterized by: In step 3, the specific process of predicting load is as follows: Calculate the distance between the key parameters at the target time and the key parameters in the data set; use one of the Euclidean distance, Manhattan distance and Chebyshev distance to calculate the distance; obtain the best weighted KNN load forecasting algorithm through cross validation or grid search K Value, select the key parameter in the data set that is closest to the key parameter at the target time K eigenvalues ​​as neighbors, and K The neighbors are weighted to obtain the predicted load.

9. The ice storage load prediction method based on data and model drive according to claim 1 is characterized by: In the step 4, the hidden layer includes three fully connected layers connected in sequence, and the three fully connected layers all use the ReLU activation function.

10. An ice storage system based on hierarchical multi-objective optimization, characterized in that: It includes a dual-mode refrigeration host, an ice storage device, a water pump system, a cooling tower and a controller; the water pump system includes a freezing pump and a cooling pump; the controller is used to execute the data and model-driven ice storage load prediction method described in claim 1; in the cold storage state, the controller determines the ice storage amount according to the predicted load of the tested ice storage system at the target time, controls the freezing pump to transport the chilled water to the ice storage device, and controls the cooling pump to transport the cooling water generated by the dual-mode refrigeration host to the cooling tower for heat dissipation; in the cold release state, the controller controls the ice storage device to melt the ice according to the hourly predicted ice storage load, and the melted cooling water is transported to the air conditioner or other cooling system through the cooling pump.

Images