Multi-mechanism cooling load prediction method and system

Abstract

The invention discloses a multi-mechanism cooling load prediction method and system, and the method comprises the steps: building a machine learning model which comprises a sub-model cooperation prediction system and is based on the physical constitution of a load, and building a mixed model system through a historical data matching model capable of customizing parameters and carrying out the self-adaption of a matching algorithm; model switching is adaptively adjusted through automatic preferential operation for actual prediction, a model self-perfection mechanism is further deployed for a machine learning model and a historical data matching model to ensure that the models have the self-perfection characteristic in the operation process, it is ensured that the two models are continuously optimized, and the model with the better effect is used all the time; and the scene coverage rate and the capability adaptability of the model after data accumulation are obviously improved.

Description

Technical Field

[0001] The present invention relates to the field of building operation and maintenance and artificial intelligence technology, and in particular to a multi-mechanism cooling load prediction method and system for large venues. Background Art

[0002] In recent years, the rapid development of urbanization has promoted the sustainable development of the construction industry, making the scale of the construction industry continue to expand. From 2001 to 2015, the total energy consumption of buildings has increased significantly. Taking 2015 as an example, the total commercial energy consumption of building operations was 864 million tce, accounting for about 20% of the total energy consumption in the country. Building energy consumption includes lighting, sockets, power, heating, ventilation and air conditioning system energy consumption (HVAC system), etc. Among them, the HVAC system energy consumption accounts for the largest proportion and is one of the current focuses of building energy conservation.

[0003] Research shows that the high energy consumption of HVAC systems is mainly due to the fact that most units are operating at a relatively low load rate, and the operating efficiency of units and water pumps is relatively low; based on experience, operation and maintenance personnel often use a system operation mode with large flow and small temperature difference, which increases the energy consumption of water pumps. Therefore, a reasonable operation and adjustment strategy is needed to improve the energy efficiency of HVAC systems. The cooling and heating loads of buildings are the purpose and adjustment basis of air-conditioning system operation. Through the research and analysis of historical building data, and the establishment of a prediction model on this basis, the building load at future moments can be accurately predicted, and the future cooling demand of the building can be determined in advance, so that the operation strategy of the HVAC system can be reasonably arranged to reduce the energy consumption of the system.

[0004] Most of the existing cooling load prediction methods are aimed at scenarios such as office buildings or commercial complexes. In such scenarios, because the working conditions are continuous and regular, it is easy to accumulate data and the load is relatively stable. Therefore, it is easier to implement load prediction by building a machine learning model based on historical data.

[0005] However, in application scenarios such as large venues, there is a huge difference in load on weekdays, event days or other different working conditions. The load is discontinuous in the time dimension and is extremely discrete, which makes data accumulation more difficult. When there is a lack of data and the working conditions fluctuate greatly, the method based on data training prediction model is generally not effective.

[0006] For example, the prediction method mentioned in CN202411117891.9 uses the concept of historical data mean and some parameters to make a relatively simple prediction, and its prediction object is commercial, office, school and other building types with continuous working conditions.

[0007] Therefore, if the existing cooling load prediction method is applied to large venues, it has the following limitations:

[0008] 1. Strong reliance on large-scale data and limited applicability. Existing cooling load forecasting methods are highly dependent on large-scale data, and often require a certain amount of historical data that meets quality requirements for model training and application. However, it is difficult to provide large-scale data in the early stages of a project or for projects that have not yet established a data system, and their applicability will be greatly limited.

[0009] 2. The model has a single structure, lacks diversity, and has poor adaptability. When using a single machine learning model or other algorithm for prediction, the more general model will inevitably show lower performance for specific tasks under limited conditions. The adaptability of a single model in different scenarios will also be greatly limited.

[0010] 3. The model cannot iterate and improve itself. After the existing prediction models or methods are initially established, they are no longer modified or upgraded, or they require a high degree of manual intervention for retraining or parameter adjustment, which greatly increases the cost of model iteration and is not conducive to the improvement and efficient iteration of the model.

[0011] 4. Unable to independently select models to run to maximize the model effect. As data accumulates or scenarios change, the effects of different models will change accordingly. If the models cannot be rotated in time, a single model may fall into the dilemma of local optimality.

[0012] It can be seen that a multi-mechanism cooling load prediction method is needed to solve the above technical problems, which is the problem to be solved by the present invention. Summary of the invention

[0013] In view of the above-mentioned technical problems existing in the existing cooling load prediction methods, the purpose of the present invention is to provide a multi-mechanism cooling load prediction method, which constructs a hybrid model system by establishing a machine learning model and a historical data matching model, and adaptively adjusts the model switching through automatic preferential operation, and deploys a model self-improvement mechanism to ensure that the model has the characteristic of self-improvement during operation, thereby greatly improving the iteration efficiency. On this basis, a prediction system capable of implementing the multi-mechanism cooling load prediction method is also provided, which effectively overcomes the problems existing in the prior art.

[0014] In order to achieve the above object, the present invention provides a multi-mechanism cooling load prediction method, the multi-mechanism cooling load prediction method comprising the following steps:

[0015] S1: Collect data and build a database in chronological order

[0016] This step collects the original data of the project and equipment and converts the collected data into a structured entry library and stores it in a non-relational database;

[0017] S2: Select prediction model based on data size and characteristics

[0018] This step is to perform parameter integrity verification and data scale verification in conjunction with the physical model based on the parameter types and quantity scale collected in S1, and select different prediction models based on parameter integrity and data scale;

[0019] S3: Prediction model building

[0020] This step constructs a machine learning prediction model and a historical data matching model respectively according to the prediction model selected in S2;

[0021] S4: Prediction model verification and debugging

[0022] This step is used to simulate and debug the prediction model constructed in S3;

[0023] S5: Prediction model self-improvement mechanism

[0024] This step is used to construct a new data set by adding valid data to the data set and verify the prediction effect of the new model and the original model, and retain the model with better effect;

[0025] S6: Automatic optimization of prediction models

[0026] This step is used to compare the root mean square error between the load predicted by the two models and the actual load when the two models constructed in S3 are running simultaneously, and to select the best operation based on the error.

[0027] Furthermore, if the parameters in S2 are incomplete, it is considered whether there are alternative parameters for the missing parameters. If there are alternative parameters, they can be used to replace the original parameters and added to the subsequent model building process.

[0028] Furthermore, in S3, when the data scale verified in S2 meets the training requirements of the machine learning prediction model, that is, it can ensure a certain degree of accuracy after the model training is completed, a hybrid system composed of a machine learning prediction model and a historical data matching model can be used to predict the cooling load; when the parameters are incomplete and there are no alternative parameters or the data scale is small, the historical data matching model can be used.

[0029] Furthermore, the machine learning prediction model includes three sub-models, namely, the total cooling load prediction model, the hourly cooling load prediction model and the hourly cooling load correction model. The three sub-models are all trained with the same training set format, that is, a single sample in the training set contains features and labels. The labels of the total cooling load prediction model and the hourly cooling load correction model are the total cooling load and the correction value of the cooling in the next hour, both of which are single values, while the label of the hourly cooling load prediction model is the hourly cooling load, which is in list form and contains the cooling load of all hours in a day, totaling 24 data.

[0030] Furthermore, the construction of the historical data matching model includes two parts: the basic model architecture and the matching algorithm;

[0031] Basic model architecture: First, you need to establish a candidate parameter set. Select at least two parameters from the candidate parameter set and assign weight coefficients between 0 and 1. The sum of the weight coefficients assigned to all parameters should be equal to 1. After completing parameter selection and weight coefficient assignment, solidify this configuration to form the basic model architecture.

[0032] Matching algorithm: According to the corresponding parameter input selected in the basic model architecture, the parameter information of the day needs to be predicted, and the matching algorithm is executed based on this parameter information.

[0033] Furthermore, the matching algorithm is divided into 8 steps:

[0034] M2-S1: Get the structured data in the database constructed in S1;

[0035] M2-S2: Traverse the data: calculate the absolute difference between all data items and the corresponding numerical parameters in the parameter information of the forecast day;

[0036] M2-S3: According to the absolute difference calculated for each parameter, the data items are sorted in descending order and normalized and scored respectively;

[0037] M2-S4: When there is a date type in the parameters of the basic model architecture, this parameter corresponds to a score when the date type of the predicted day is consistent with the date type of the data entry;

[0038] M2-S5: The scores of relevant parameters including date type have been calculated in M2-S3 and M2-S4. According to the score information and the weight coefficient of each parameter in the basic model architecture, the total score is calculated for each data item;

[0039] M2-S6: When there is a date in the parameters of the basic model architecture, select the top 10 data items with the total score in M2-S5, calculate the similarity between the date and the date of the prediction, and sort them according to the similarity, with the most similar score, and so on until the last item;

[0040] M2-S7: According to the scores of the 10 data items, the total score is calculated for each data item according to the weight coefficient of each parameter in the basic model architecture; the total score here will cover the total score in M2-S5;

[0041] M2-S8: The cooling load information in the data item with the highest total score is selected as the predicted value, and the remaining data items will be used for the self-improvement mechanism.

[0042] Furthermore, in the above steps, M2-S4 and M2-S6 are optional steps and can only be executed when the basic model architecture contains date types and date parameters. When M2-S6 is not executed, M2-S7 does not need to be executed. The remaining steps are executed in sequence, and the steps that do not need to be executed can be skipped.

[0043] In order to achieve the above object, the present invention provides a multi-mechanism cooling load prediction system, which includes:

[0044] A database establishment module, which is used to collect actual operation and maintenance information and equipment data of a specific project or engineering and to build an original database;

[0045] A prediction model selection module, which interacts with the database building module data and is used to select a prediction model according to the data scale and characteristics in the database building module;

[0046] A prediction model construction module, which interacts with the prediction model selection module data to construct a machine learning prediction model and a historical data matching model;

[0047] A prediction model verification and debugging module, which interacts with the prediction model construction module data to test and debug the constructed prediction model;

[0048] A prediction model self-improvement mechanism module, which interacts with the prediction model verification and debugging module data, constructs a new data set by adding valid data to the data set and verifies the prediction effect of the new model and the original model;

[0049] The prediction model automatic optimization operation module interacts with the prediction model construction module data, and is used to compare the root mean square error between the load predicted by the two models and the actual load when the two models are running simultaneously, and to perform optimization operation.

[0050] The multi-mechanism cooling load prediction method provided by the present invention has the following characteristics and advantages:

[0051] The hybrid model system no longer relies too much on data scale and supports load forecasting under low-level data. The hybrid model system is constructed by establishing a machine learning model and a historical data matching model. The machine learning model is suitable for scenarios where the data has a certain scale, is relatively complete and high-quality, and the historical data matching model can adapt to situations such as the early stage of project operation and maintenance when the amount of data accumulation is small.

[0052] A multi-model system with different capabilities and adaptability was built. A machine learning model based on the physical composition of the load, including a sub-model coordination prediction system, and a historical data matching model with customizable parameters and adaptive matching algorithms were established.

[0053] A self-improvement mechanism for the model has been established. Self-improvement mechanisms have been established for the machine learning model and the historical data matching model, which can continuously improve themselves during operation to improve the actual prediction effect, eliminating the need for manual intervention in the subsequent process.

[0054] An automated optimization operation system for hybrid models has been established. During operation, the two models will be automatically compared in real time based on the criteria, and the model with better performance will be switched to for actual prediction. In conjunction with the self-improvement mechanism, it will ensure that the two models are continuously optimized and the model with better performance is always used, significantly improving the scenario coverage and the adaptability of the model after data accumulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0056] Figure 1 This is a schematic diagram of the overall process of the multi-mechanism cooling load prediction method;

[0057] Figure 2 This is a flow chart of database construction in the multi-mechanism cooling load prediction method;

[0058] Figure 3 This is a schematic diagram of the selection process of the prediction model in the multi-mechanism cooling load prediction method;

[0059] Figure 4 This is a flow chart of the execution of M1 in the multi-mechanism cooling load prediction method;

[0060] Figure 5 This is a schematic diagram of the process flow constructed by M2 in the multi-mechanism cooling load prediction method;

[0061] Figure 6 This is a flow chart of the M1 self-improvement mechanism in the multi-mechanism cooling load forecasting method;

[0062] Figure 7 This is a flow chart of the M2 self-improvement mechanism in the multi-mechanism cooling load forecasting method;

[0063] Figure 8 It is a schematic diagram of the process of automatic optimal operation in the multi-mechanism cooling load prediction method;

[0064] Fig. 9 This is a flowchart of the M1 training implementation case in the multi-mechanism cooling load forecasting method;

[0065] Fig.10 This is a flowchart of the M2 training implementation case in the multi-mechanism cooling load forecasting method. DETAILED DESCRIPTION

[0066] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below with reference to specific diagrams.

[0067] In view of the above-mentioned technical problems existing in the existing cooling load prediction methods, the purpose of the present invention is to provide a multi-mechanism cooling load prediction method, which constructs a hybrid model system by establishing a machine learning model and a historical data matching model, and adaptively adjusts the model switching through automatic preferential operation, and deploys a model self-improvement mechanism to ensure that the model has the characteristics of self-improvement during operation, thereby greatly improving the iteration efficiency.

[0068] The multi-mechanism cooling load prediction method provided by the present invention is shown in Figure 1 , which comprises the following steps:

[0069] S1: Collect data and build a database in chronological order (e.g. Figure 2 )

[0070] This step is used to collect the original data of the project and equipment and convert the collected data into a structured entry library to build a database for subsequent retrieval and call.

[0071] S11: Collect data and organize them into raw data

[0072] This step is used to collect two parts of data: the actual operation and maintenance information of a specific project or engineering and equipment data.

[0073] The operation and maintenance information includes but is not limited to the date, the activities carried out on that day or the type of date (such as working day, activity day, weekend, etc.), the number of people, indoor and outdoor temperature, humidity, etc.

[0074] Equipment data includes but is not limited to data related to the HVAC system, such as inlet and outlet water temperature, ice storage capacity, etc. Its general source is the sensors, recording equipment or integrated BA system of the HVAC system, and is mainly used to calculate the total cooling load and hourly cooling load.

[0075] S12: Converting raw data into structured entries

[0076] This step is used to convert the raw data collected in S11 into structured data entries (in the form of "label: value"), where the label is a single data name in S11 (such as "date") and the value is the specific value of the data (such as "2024-09-20").

[0077] S13: Storing structured data items in a non-relational database

[0078] This step is used to store the converted structured entries into a non-relational database (such as MongoDB, etc.) in order to maintain the designed data structure and facilitate subsequent retrieval and calling.

[0079] S2: Select a prediction model based on the data size and characteristics (e.g. Figure 3 )

[0080] S21: Extract the parameter type and quantity scale of the data collected in S1

[0081] S22: Perform parameter integrity check in conjunction with the physical model. The composition of the load physical model is as follows:

[0082] Q τ =KF1(t τ-ξ -t n )+F1X g J wτ +1000(G1+G2)(H τ -H n )+n(q s +q l )+W τ

[0083] Where K is the heat transfer coefficient of the enclosure structure, W / m 2 / K; F1 is the area of ​​the enclosure structure, m 2 ;t τ-ξ The temperature for calculating the cooling load when considering the temperature wave action, °C; t n is the indoor temperature, ℃; F2 is the area of ​​the exterior window and glass curtain wall, m 2 ;X g J is the structural correction factor of exterior windows and glass curtain walls; wτ The cooling load intensity of solar radiation through glass without sunshade facilities at the calculation time, W / m 2 ; G1 is the organized fresh air volume, kg / s; G2 is the unorganized infiltration air volume, kg / s; H τ is the outdoor air enthalpy at the time of calculation, kJ / kg; H n is the indoor air enthalpy value at the time of calculation, kJ / kg; n is the number of people in the room; qs is the sensible heat output of personnel, W; q l is the latent heat generated by personnel, W; W τ The electrical energy consumed by all equipment in a building is the heat generated by the equipment.

[0084] S23: Verification of data scale

[0085] Based on the above S22, the parameter integrity check is carried out in conjunction with the physical model. If the parameters are complete, the next step of data scale verification can be carried out; if the parameters are incomplete, consider whether there are alternative parameters for the missing parameters. If there are alternative parameters, they can be used to replace the original parameters and added to the subsequent model building process. When the parameters are complete or there are alternative parameters, the data scale verification is carried out (S23).

[0086] When the data scale meets the training requirements of the machine learning prediction model, that is, it can ensure that the model has a certain degree of accuracy after training (which can be judged according to the specific situation, usually the order of magnitude needs to be higher than one hundred), the machine learning prediction model M1 and the historical data matching model M2 can be used to form a hybrid system for cooling load prediction; when the parameters are incomplete and there are no alternative parameters or the data scale is small, the historical data matching model M2 can be used.

[0087] S3: Prediction model building

[0088] This step is used to build a prediction model. There are two main types of prediction models, namely the machine learning prediction model M1 and the historical data matching model M2. The construction methods of M1 and M2 are different and need to be constructed separately.

[0089] Among them, Figure 4 ,The machine learning prediction model M1 contains three sub-models, namely the total cooling load prediction model M11, the hourly cooling load prediction model M12, and the hourly cooling load correction model M13.

[0090] The three sub-models are all trained with the same training set format, that is, a single sample in the training set contains features and labels. The labels of M11 and M13 are the total cooling load and the correction value of the cooling in the next hour, both of which are single values, while the label of M12 is the hourly cooling load, which is in list form and contains the cooling load of all hours in a day, totaling 24 data.

[0091] M11 and M13 can be trained using methods such as linear regression, nonlinear regression, neural network, XGBoost, LightGBM, and random forest, while M12 can only use methods such as neural network, XGBoost, LightGBM, and random forest because it needs to output a list.

[0092] The training method is to extract the required data from the database, divide it into training set and test set according to 9:1, 8:2 or other ratios, and then select an algorithm to train M11, M12, and M13 respectively. After training, test it on the test set until the accuracy index requirements are met. The indicators used to evaluate the effect or accuracy of the model can be cross entropy loss, mean square error, etc. Their expressions are:

[0093] Cross Entropy Loss:

[0094]

[0095] Mean Square Error:

[0096]

[0097] After the training of each sub-model of M1 is completed, it can be used for actual load forecasting through a fixed process.

[0098] First, it is necessary to input the features required by M11 and M12, that is, the features used during training. M11 will output the total cooling load, and M12 will output the hourly cooling load. The total cooling load is compared with the sum of the hourly cooling loads. When the sum of the hourly loads is greater than the total load, the hourly load is used; when the sum of the hourly loads is less than the total load, the hourly load is calibrated with the total load, that is, the total load is redistributed according to the load ratio per hour under the original hourly load to determine the final hourly load. In the actual operation of the HVAC system, when there is the operating data of the previous hour, the cooling load of the next hour is corrected by M13, and its input features are consistent with the training set features input during training.

[0099] like Figure 5 ,The construction of the historical data matching model M2 includes two parts: the ,basic model architecture and the matching algorithm.

[0100] First: The establishment of the basic model architecture first requires the establishment of a candidate parameter set, which includes but is not limited to parameters such as date, date type, maximum outdoor temperature, maximum outdoor humidity, average outdoor temperature, and average indoor temperature.

[0101] The above parameters are divided into two categories, namely special parameters and value parameters. The special parameters include date and date type, and the remaining parameters are value parameters.

[0102] Then select at least 2 parameters from the candidate parameter set and assign them weight coefficients between 0 and 1. The sum of the weight coefficients assigned to all parameters should be equal to 1. After completing the parameter selection and weight coefficient assignment, solidify this configuration to form the basic model architecture.

[0103] According to the corresponding parameters selected in the basic model architecture, the parameter information of the day to be predicted is input, and the matching algorithm is executed based on this parameter information. The matching algorithm is divided into 8 steps, namely M2-S1 to M2-S8:

[0104] M2-S1: To obtain the structured data in the database constructed in S1, it is necessary to call all valid data from the database through a program. If the accumulated data has been more than 3 years, the data of the last 3 years can be obtained.

[0105] M2-S2: Traverse the data and calculate the absolute difference between all data entries and the corresponding numerical parameters in the parameter information of the predicted day. For example, the basic model architecture includes the parameter "average outdoor temperature". There is a data entry of "2024-09-19" in the database, and its average outdoor temperature is 30.2. The average outdoor temperature of the day needs to be predicted to be 30.5, so its absolute difference is 0.3. The absolute difference between all parameters in all entries and the parameters of the predicted day is calculated accordingly.

[0106] M2-S3: According to the absolute difference calculated for each parameter, the data items are sorted in descending order and normalized and scored respectively. The scores of the data items in each sorting are:

[0107]

[0108] Among them, si is the score of the data entry corresponding to the i-th parameter, ni is the ranking of the data entry among all data entries sorted according to the absolute difference of the i-th parameter, and L is the number of data entries.

[0109] M2-S4: When the parameters of the basic model architecture include date type, when the date type of the predicted day is consistent with the date type of the data entry, the score corresponding to this parameter is 10 points; when custom similarity rules are used, such as when the date type is a rehearsal day or a general activity, they are considered similar, and the score for the date type parameter is 5 points; in other cases, the score for the data entry date type is 0 points.

[0110] M2-S5: The scores corresponding to the relevant parameters including date type have been calculated in M2-S3 and M2-S4. According to the score information and the weight coefficient of each parameter in the basic model architecture, the total score is calculated for each data item.

[0111] M2-S6: When there is a date in the parameters of the basic model architecture, select the top 10 data items in M2-S5, calculate the similarity between the date and the predicted date, and sort them according to the similarity. The most similar score is 10 points, and so on until the last item is 1 point. The similarity is calculated by giving priority to the distance between the month and the day, and then considering the distance between the year.

[0112] M2-S7: According to the scores of the 10 data items, the total score is calculated for each data item according to the weight coefficient of each parameter in the basic model architecture; the total score here will cover the total score in M2-S5.

[0113] M2-S8: The cooling load information in the data item with the highest total score is selected as the predicted value, and the remaining data items will be used for the self-improvement mechanism.

[0114] In the above steps, M2-S4 and M2-S6 are optional steps and can only be executed when the basic model architecture contains date types and date parameters. When M2-S6 is not executed, M2-S7 does not need to be executed. The remaining steps are executed in sequence, and unnecessary steps can be skipped.

[0115] S4: Prediction model verification and debugging

[0116] This step is used to test and debug the prediction model built in S3.

[0117] After the model in S3 is built, it needs to be put into specific operation for testing. The simulation prediction is carried out in the first 72 hours, and the root mean square error between the actual hourly cooling load and the predicted hourly cooling load within 72 hours is calculated by comparing the actual cooling load value. A threshold is set, and the model is considered to have passed verification and debugging only when it is less than the threshold.

[0118] S5: Prediction model self-improvement mechanism

[0119] This step is used to improve the self-improvement of the prediction model and the prediction effect. The machine learning prediction model M1 and the historical data matching model M2 follow different self-improvement mechanisms.

[0120] like Figure 6 ,The core of the self-improvement mechanism of the machine learning prediction model M1 is to add new valid data to the data set as the data gradually accumulates, train a new model with the newly constructed data set, and verify the prediction effect of the new model and the original model, and retain the model with better effect.

[0121] First, it is necessary to set the training set increment threshold Δm, and then regularly check the amount of new valid data in the database Δn. When Δn≥Δm, the self-improvement mechanism process of M1 is triggered. Add new samples to the original training set to form a new training set, train the new model according to the original mechanism, and obtain a new M1 (including 3 sub-models).

[0122] Compare the performance of the new model and the original model on the same test set. If the new model is inferior to the original model, you can choose to discard the amount of data Δn in that part, or add an additional data review mechanism to further improve the data quality, and return to the step of regularly checking the amount of new valid data Δn in the database to continue the cycle;

[0123] When the new model is better than the original model, the new model is added to the workflow and runs collaboratively in "shadow mode" for 72 hours. "Shadow mode" means that the new model also predicts the cooling load at the same time, but its predicted data does not participate in the actual control of the equipment.

[0124] After 72 hours, compare the mean deviations of the new model and the original model from the actual cooling load. If the new model is still better than the original model, the new model will be used to replace the original model, but the original model needs to be backed up and saved in case of subsequent emergencies that require version rollback. However, if the new model is worse than the original model, it will be handled according to the method for handling the test set performance being worse than the original model.

[0125] like Figure 7 ,The core of the self-improvement mechanism of the historical data matching model M2 is to ,ensure that the data entries that are most similar to the actual ,cooling load conditions can obtain a higher total score as much possible.

[0126] First, the M2 model runs for 72 hours, with 24 hours as a cycle. The root mean square error (MSE) between the hourly cooling load and the actual hourly cooling load of the 10 data items retained in M2-S8 (including the data item with the highest total score) in each cycle is counted, and the average value of the root mean square error of each cycle is taken as the final comparison index. The data item with the smallest cycle MSE mean is selected. When the data item with the smallest cycle MSE mean is the original item with the highest total score, no change is made and the next 72-hour cycle is continued.

[0127] When the data entry with the smallest period MSE mean is another entry, compare the parameter scores of the original data entry with the highest total score and the data entry with the smallest period MSE mean, and increase the parameter weight coefficient with the lower score in the data entry with the smallest period MSE mean, for example, by 0.1 or 0.05. After completion, normalize the weights of all parameters to ensure that their sum is equal to 1. At this point, the weight coefficient update of the basic model architecture is completed, and the next cycle can be executed.

[0128] S6: Automatic optimization of prediction models (such as Figure 8 )

[0129] Considering that the accuracy of the machine learning prediction model M1 and the historical data matching model M2 may fluctuate in different scenarios and different data scales, a mechanism for automatically selecting a better model is needed to ensure that a model with better prediction effect can be used in a timely manner.

[0130] When both models are built, they are run simultaneously, and the root mean square error (MSE) between the predicted load and the actual load of the M1 and M2 models is compared based on the 72-hour operation data. Subsequently, the model with the smaller MSE is switched to the actual load forecasting model, and the other model is run synchronously and continuously self-improved. This process is repeated to complete the optimal operation. When only M2 is used, no optimal selection is required.

[0131] The multi-mechanism cooling load prediction method given in this example solution can be used as a corresponding software program to form a corresponding multi-mechanism cooling load prediction system when it is applied. When the software program is running, it will execute the multi-mechanism cooling load prediction method and store it in a corresponding storage medium for the processor to call and execute.

[0132] The multi-mechanism cooling load prediction system thus formed mainly includes the following functions: database establishment module, prediction model selection module, prediction model construction module, prediction model verification and debugging module, prediction model self-improvement mechanism module and prediction model automatic optimization operation module.

[0133] The database building module in this system is used to collect data and build a database in chronological order. The database module includes a collection unit, a conversion unit and a storage unit.

[0134] The collection unit is used to collect the actual operation and maintenance information and equipment data of a specific project or engineering and organize them into raw data.

[0135] The conversion unit interacts with the data of the collection unit and is used to convert the original data collected by the collection unit into structured entries.

[0136] The storage unit interacts with the conversion unit data to store the structured data items converted by the conversion unit into a non-relational database.

[0137] The database establishment module is configured to execute the above database establishment steps to realize corresponding functions.

[0138] The prediction model selection module in this system interacts with the database construction module data, and is used to select a prediction model according to the data scale and characteristics. The prediction model selection module includes an extraction unit, a verification unit and a data scale verification unit.

[0139] The extraction unit is used to extract the parameter types and quantity scales collected in the database establishment module.

[0140] The verification unit interacts with the extraction unit data, and verifies the integrity of the parameters according to the type and quantity of the extracted parameters in conjunction with the physical model.

[0141] The data scale verification unit interacts with the verification unit data to verify whether the relevant parameters in the physical model verification collected data are complete.

[0142] The prediction model selection module is configured to execute the above prediction model selection steps to achieve corresponding functions.

[0143] The prediction model construction module in this system interacts with the prediction model selection module data to build a machine learning prediction model and a historical data matching model.

[0144] The prediction model building module is configured to execute the above prediction model building steps to realize corresponding functions.

[0145] The prediction model verification and debugging module in this system interacts with the prediction model construction module data to test and debug the constructed prediction model.

[0146] The prediction model verification and debugging module is configured to execute the above prediction model verification and debugging steps to realize corresponding functions.

[0147] The prediction model self-improvement mechanism module in this system interacts with the prediction model verification and debugging module data. It constructs a new data set by adding valid data to the data set and verifies the prediction effect of the new model and the original model, retaining the model with better effect.

[0148] The prediction model self-improvement mechanism module is configured to execute the above prediction model self-improvement mechanism steps to achieve corresponding functions.

[0149] The prediction model automatic optimization operation module in this system interacts with the prediction model construction module data. It is used to compare the root mean square error (MSE) between the load predicted by the two models and the actual load when the two models are running at the same time, and then switch to the model with a smaller MSE as the actual load prediction model.

[0150] The prediction model automatic optimization operation module is configured to execute the above prediction model automatic optimization operation steps to realize the corresponding functions.

[0151] The multi-mechanism cooling load prediction method composed of the above scheme is illustrated below with examples of its working process in specific applications. It should be noted here that the working process here is only an example and does not constitute a limitation to this scheme.

[0152] The implementation scheme of the present invention is described by taking the training of the machine learning prediction model M1 and the construction of the historical data matching model M2 as the objects.

[0153] like Fig. 9, M1 includes three sub-models: M11, M12 and M13. During training, it is necessary to construct training sets and test sets with different features and labels according to the original data.

[0154] One implementation scheme is that the training set and test set of M11 include weighted parameters whose features are date type, outdoor average temperature, outdoor maximum temperature, outdoor average humidity, outdoor maximum humidity, outdoor minimum humidity, average temperature of n indoor measuring points, number of people indoors, number of fresh air units turned on and opening degree, and its label is the total cooling load.

[0155] The training set and test set of M12 include features such as date type, outdoor hourly temperature, outdoor hourly humidity, outdoor main wind direction hourly wind speed, indoor hourly temperature of n measuring points, indoor hourly number of people, and indoor hourly air intake. The parameters with hourly parameters are in list form, containing data for each hour, and the label is the cooling load per hour, also in list form.

[0156] The training set and test set of M13 include the features of actual cooling load in the previous hour, average outdoor temperature in the previous hour, average outdoor humidity in the previous hour, average wind speed in the main outdoor wind direction in the previous hour, average temperature of n indoor measurement points in the previous hour, average number of people indoor in the previous hour, predicted cooling load in the next hour (optional), average outdoor temperature in the next hour, average outdoor humidity in the next hour, average wind speed in the main outdoor wind direction in the next hour, average number of people indoor in the next hour, and the label is the correction value of cooling load in the next hour.

[0157] The input and output of the three models are somewhat different, and the dimensions are not consistent. Algorithms such as neural networks can be used to build models with consistent usage and properties.

[0158] like Fig.10 ,One implementation scheme for constructing the M2 model is to establish a candidate parameter set including date, date type, outdoor average temperature, outdoor maximum temperature, outdoor average humidity, outdoor maximum humidity, and indoor average temperature, from which 5 parameters including date, date type, outdoor average temperature, indoor average temperature, and outdoor maximum temperature are selected, and weight coefficients of 0.10, 0.40, 0.15, 0.15 and 0.20 are assigned, respectively, to construct the basic model framework.

[0159] After entering the information for the forecast day, the matching algorithm is executed on the data stored in the database to obtain the final matching result.

[0160] In summary, the multi-mechanism cooling load forecasting method and system provided by this solution has a hybrid model system that is no longer overly dependent on data scale. It has adaptable solutions under different data scales and is also suitable for load forecasting at low levels, providing a comprehensive load forecasting solution for the entire project operation and maintenance cycle.

[0161] At the same time, a machine learning model based on the physical composition of the load, including a sub-model coordination prediction system, and a historical data matching model with customizable parameters and adaptive matching algorithms were established. It has significant universality and can be applied in a variety of scenarios with only partial adjustments.

[0162] Secondly, a self-improvement mechanism for the model was established, which can automatically optimize the model's prediction effect without human intervention after deployment.

[0163] In addition, an automatic optimization operation system for hybrid models has been established. Different models are applicable to different scenarios, ensuring that the more effective model can be used for actual predictions in real time.

[0164] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A multi-mechanism cooling load prediction method, characterized in that: The multi-mechanism cooling load prediction method comprises the following steps: S1: Collect data and build a database in chronological order This step collects the original data of the project and equipment and converts the collected data into a structured entry library and stores it in a non-relational database; S2: Select prediction model based on data size and characteristics This step is to perform parameter integrity verification and data scale verification in conjunction with the physical model based on the parameter types and quantity scale collected in S1, and select different prediction models based on parameter integrity and data scale; S3: Prediction model building This step constructs a machine learning prediction model and a historical data matching model respectively according to the prediction model selected in S2; S4: Prediction model verification and debugging This step is used to simulate and debug the prediction model constructed in S3; S5: Prediction model self-improvement mechanism This step is used to construct a new data set by adding valid data to the data set and verify the prediction effect of the new model and the original model, and retain the model with better effect; S6: Automatic optimization of prediction models This step is used to compare the root mean square error between the load predicted by the two models and the actual load when the two models constructed in S3 are running simultaneously, and to select the best operation based on the error.

2. A multi-mechanism cooling load prediction method according to claim 1, characterized in that: If the parameters in S2 are incomplete, consider whether there are alternative parameters for the missing parameters. If there are alternative parameters, they can be used to replace the original parameters and added to the subsequent model building process.

3. A multi-mechanism cooling load prediction method according to claim 2, characterized in that: In S3, when the data scale verified in S2 meets the training requirements of the machine learning prediction model, that is, it can ensure a certain degree of accuracy after the model training is completed, a hybrid system composed of a machine learning prediction model and a historical data matching model can be used to predict the cooling load; when the parameters are incomplete and there are no alternative parameters or the data scale is small, the historical data matching model can be used.

4. A multi-mechanism cooling load prediction method according to claim 3, characterized in that: The machine learning prediction model contains three sub-models, namely the total cooling load prediction model, the hourly cooling load prediction model and the hourly cooling load correction model. The three sub-models are trained with the same training set format, that is, a single sample in the training set contains features and labels. The labels of the total cooling load prediction model and the hourly cooling load correction model are the total cooling load and the correction value of the cooling in the next hour, both of which are single values, while the label of the hourly cooling load prediction model is the hourly cooling load, which is in list form and contains the cooling load of all hours in a day, totaling 24 data.

5. A multi-mechanism cooling load prediction method according to claim 3, characterized in that: The construction of the historical data matching model includes two parts: the basic model architecture and the matching algorithm; Basic model architecture: First, you need to establish a candidate parameter set. Select at least two parameters from the candidate parameter set and assign weight coefficients between 0 and 1. The sum of the weight coefficients assigned to all parameters should be equal to 1. After completing parameter selection and weight coefficient assignment, solidify this configuration to form the basic model architecture. Matching algorithm: According to the corresponding parameter input selected in the basic model architecture, the parameter information of the day needs to be predicted, and the matching algorithm is executed based on this parameter information.

6. A multi-mechanism cooling load prediction method according to claim 5, characterized in that; The matching algorithm is divided into 8 steps: M2-S1: Get the structured data in the database constructed in S1; M2-S2: Traverse the data: calculate the absolute difference between all data items and the corresponding numerical parameters in the parameter information of the forecast day; M2-S3: According to the absolute difference calculated for each parameter, the data items are sorted in descending order and normalized and scored respectively; M2-S4: When there is a date type in the parameters of the basic model architecture, this parameter corresponds to a score when the date type of the predicted day is consistent with the date type of the data entry; M2-S5: The scores of relevant parameters including date type have been calculated in M2-S3 and M2-S4. According to the score information and the weight coefficient of each parameter in the basic model architecture, the total score is calculated for each data item; M2-S6: When there is a date in the parameters of the basic model architecture, select the top 10 data items with the total score in M2-S5, calculate the similarity between the date and the date of the prediction, and sort them according to the similarity, with the most similar score, and so on until the last item; M2-S7: According to the scores of the 10 data items, the total score is calculated for each data item according to the weight coefficient of each parameter in the basic model architecture; the total score here will cover the total score in M2-S5; M2-S8: The cooling load information in the data item with the highest total score is selected as the predicted value, and the remaining data items will be used for the self-improvement mechanism.

7. A multi-mechanism cooling load prediction method according to claim 6, characterized in that: In the above steps, M2-S4 and M2-S6 are optional steps and can only be executed when the basic model architecture contains date types and date parameters. When M2-S6 is not executed, M2-S7 does not need to be executed. The remaining steps are executed in sequence, and unnecessary steps can be skipped.

8. A multi-mechanism cooling load prediction system, characterized in that: It includes: A database establishment module, which is used to collect actual operation and maintenance information and equipment data of a specific project or engineering and to build an original database; A prediction model selection module, which interacts with the database building module data and is used to select a prediction model according to the data scale and characteristics in the database building module; A prediction model construction module, which interacts with the prediction model selection module data to construct a machine learning prediction model and a historical data matching model; A prediction model verification and debugging module, which interacts with the prediction model construction module data to test and debug the constructed prediction model; A prediction model self-improvement mechanism module, which interacts with the prediction model verification and debugging module data, constructs a new data set by adding valid data to the data set and verifies the prediction effect of the new model and the original model; The prediction model automatic optimization operation module interacts with the prediction model construction module data, and is used to compare the root mean square error between the load predicted by the two models and the actual load when the two models are running simultaneously, and to perform optimization operation.

Images