A model training method, a cold load prediction method and an electronic device

Abstract

Embodiments of the present application provide a model training method, a cold load prediction method and an electronic device. The model training method comprises: after obtaining first collection data, determining a first weight and a second weight based on a total number of cumulative collection data; selecting a first number of second collection data from a second collection data set based on the first weight; selecting a second number of simulation data from a simulation data set based on the second weight; training a last round of cold load prediction model based on the first collection data, the selected second collection data and the selected simulation data, to obtain a current round of cold load prediction model; wherein the first round of cold load prediction model is obtained by training based on the simulation data set. When there is no collection data, the first round of cold load prediction model is obtained by training based on the simulation data set; after there is collection data, the first weight and the second weight are determined according to the cumulative collection data, and the proportion of simulation data and second collection data used in model training is gradually reduced.

Description

Technical Field

[0001] This application relates to the field of air conditioning control technology, and in particular to a model training method, a cooling load prediction method, and an electronic device. Background Technology

[0002] With the development of technology and the increase in energy consumption, building energy consumption is a significant component of overall energy consumption. Therefore, energy-saving optimization in buildings has received widespread attention. A large portion of building energy consumption is generated by air conditioning systems, making their energy-saving optimization crucial. Energy-saving optimization strategies for air conditioning systems typically rely on cooling load; that is, appropriate energy-saving optimization strategies need to be determined based on the predicted cooling load of the air conditioning system.

[0003] In related technologies, a cooling load prediction model is trained based on a large amount of historical data, enabling the cooling load prediction model to accurately predict the cooling load.

[0004] However, in the early stages of using the air conditioning system, there is a lack of available historical data, making it impossible to train the cooling load prediction model. Furthermore, with only a small amount of historical data available later, it is difficult to accurately train the cooling load prediction model, making it difficult for the above methods to accurately predict the cooling load. Summary of the Invention

[0005] This application provides a model training method, a cooling load prediction method, and an electronic device to accurately train a cooling load prediction model even when data collection is lacking.

[0006] In a first aspect, embodiments of this application provide a first model training method, which includes: After obtaining the first collection data within the target time period, based on the total number of accumulated collection data, a first weight corresponding to the second collection dataset and a second weight corresponding to the simulation dataset are determined; wherein, the second collection dataset includes multiple second collection data outside the target time period; Based on the first quantity corresponding to the first weight, select the first quantity of second collected data from the second collected dataset; and based on the second quantity corresponding to the second weight, select the second quantity of simulation data from the simulation dataset; Based on the first collected data, the selected second collected data, and the selected simulation data, the previous round of cooling load prediction model is trained to obtain the current round of cooling load prediction model; wherein, the first round of cooling load prediction model is obtained by training based on the simulation dataset.

[0007] The above scheme, when no data is collected, first trains a cooling load prediction model based on a simulation dataset to obtain the initial cooling load prediction model. In this way, there is a corresponding cooling load prediction model to predict the cooling load in the early stage of the air conditioning system's use. Then, after data is collected, considering the insufficient amount of collected data, the first weight (the weight of historical collected data) and the second weight (the weight of simulation data) corresponding to the second collected dataset are obtained based on the total number of accumulated collected data. The second collected data is selected based on the first weight, and the simulation data is selected based on the second weight. The proportion of simulation data and second collected data used in model training is gradually reduced, thereby achieving a smooth filtering of model training from simulation data dominance to collected data dominance, so that the cooling load predicted by the cooling load prediction model is closer and closer to the current actual usage of the air conditioning system.

[0008] In some alternative implementations, the first weight is obtained in the following manner: A first attenuation parameter is obtained based on the total number of accumulated collected data and the first attenuation coefficient; wherein, the first attenuation parameter is inversely proportional to the total number of accumulated collected data; The product of the first preset ratio and the first attenuation parameter is determined as the first weight.

[0009] The above scheme sets a first attenuation coefficient and a first preset ratio, and obtains a first attenuation parameter based on the total number of accumulated collected data and the first attenuation coefficient, so that the first attenuation parameter decreases as the total number of accumulated collected data increases; then, based on the product of the first preset ratio and the first attenuation parameter, a first weight is determined, so that the first weight also decreases as the total number of accumulated collected data increases.

[0010] In some alternative implementations, the second weight is obtained in the following manner: A second attenuation parameter is obtained based on the total number of accumulated collected data and the second attenuation coefficient; wherein the second attenuation parameter is inversely proportional to the total number of accumulated collected data. The product of the second preset ratio and the second attenuation parameter is determined as the second weight.

[0011] The above scheme sets a second attenuation coefficient and a second preset ratio, and obtains a second attenuation parameter based on the total number of accumulated collected data and the second attenuation coefficient, so that the second attenuation parameter decreases as the total number of accumulated collected data increases; then, based on the product of the second preset ratio and the second attenuation parameter, a second weight is determined, so that the second weight also decreases as the total number of accumulated collected data increases.

[0012] In some alternative implementations, the first quantity and the second quantity are obtained in the following ways: Based on the first weight and the second weight, a third weight corresponding to the first collected data is obtained; The total number of samples is obtained based on the number of the first collected data and the third weight; The product of the total number of samples and the first weight is determined as the first quantity, and the product of the total number of samples and the second weight is determined as the second quantity.

[0013] The above scheme obtains a third weight corresponding to the first collected data based on the first weight and the second weight, and obtains the total number of samples based on the number of the first collected data and the third weight. In this way, with the number of the first collected data remaining unchanged, the total number of samples will dynamically change with the total number of accumulated collected data. After obtaining the total number of samples, a more suitable number of samples for the second collected data is obtained based on the product of the total number of samples and the first weight, and a more suitable number of samples for the simulation data is obtained based on the product of the total number of samples and the second weight.

[0014] In some optional implementations, selecting the first number of second collected data from the second collected dataset includes: From the second collected dataset, select a first number of second collected data points whose collection time is adjacent to the target time period.

[0015] The above scheme selects a second data point whose collection time is adjacent to the target time period, so that the collection time of the selected second data point is close to that of the first data point. In this way, the data distribution of the selected second data point is closer to that of the first data point.

[0016] In some optional implementations, selecting the second number of simulation data from the simulation dataset includes: Obtain the target environment parameters corresponding to the first collected data; Based on the target environment parameters, the second number of simulation data is selected from the simulation dataset.

[0017] The above scheme selects simulation data based on the target environmental parameters of the first collected data, making the selected simulation data closer to the environmental parameters of the first collected data. This makes the selected simulation data more similar to the data distribution of the first collected data.

[0018] In some optional implementations, based on the first collected data, selected second collected data, and selected simulation data, the previous round of cooling load prediction model is trained to obtain the current round of cooling load prediction model, including: Based on the first collected data, the selected second collected data, and the selected simulation data, the cooling load prediction model of the previous round is trained iteratively multiple times to obtain the cooling load prediction model of the current round; wherein each iteration includes: Data sequences are selected from each training dataset, and the target cold load corresponding to each selected data sequence is obtained; wherein, the data sequence is a preset number of consecutive data in the corresponding training dataset, and the target cold load is the cold load of the data sequence at the next moment in the corresponding training dataset; the training dataset is the first collected data, the selected second collected data, or the selected simulation data; Based on the selected data sequence, the predicted cooling load is obtained; The target loss is obtained based on the difference between the predicted cooling load and the corresponding target cooling load; The parameters of the previous round of cold load prediction model are adjusted based on the target loss.

[0019] Secondly, embodiments of this application provide a cooling load prediction method, including: After obtaining the data sequence corresponding to the target time, the data sequence is input into the current round of cold load prediction model, and the predicted cold load at the target time is obtained through the current round of cold load prediction model; wherein, the current round of cold load prediction model is trained using any of the methods described in the first aspect above.

[0020] The above scheme, when no data is collected, first trains the initial cooling load prediction model based on a large amount of simulation data. This allows the model to predict the cooling load at the target time during the initial use of the air conditioning system. Furthermore, after data collection becomes available, considering the limited amount of available data, a first weight (for historical data) and a second weight (for simulation data) are determined based on the total number of collected data. Second collected data are selected based on the first weight, and simulation data is selected based on the second weight. This gradually reduces the proportion of simulation data and second collected data used in model training, achieving a smooth transition from simulation data dominance to collected data dominance. This makes the cooling load predicted by the model increasingly closer to the actual usage of the air conditioning system. Therefore, the cooling load prediction model trained in this way can typically predict the target cooling load with relatively high accuracy.

[0021] In some optional implementations, after obtaining the predicted cooling load at the target time, the method further includes: Obtain the error between the predicted cooling load and the actual cooling load within the target duration; When the error reaches the error threshold, the previous round's cooling load prediction model is selected to replace the current round's cooling load prediction model.

[0022] The above scheme indicates that when the error between the predicted cooling load and the actual cooling load within the target time period reaches the error threshold, it means that the current cooling load model is unable to predict the current cooling load. By replacing the current cooling load prediction model with the previous cooling load prediction model, that is, reverting to the previous cooling load prediction model to continue the cooling load prediction, even if the cooling load prediction model trained in a certain round is not accurate enough, there is still a backup cooling load prediction model from the previous round to continue the subsequent cooling load prediction.

[0023] Thirdly, embodiments of this application provide a model training apparatus, the apparatus comprising: The weight determination module is used to determine the first weight corresponding to the second collection dataset and the second weight corresponding to the simulation dataset based on the total number of accumulated collection data after obtaining the first collection data within the target time period; wherein, the second collection dataset includes multiple second collection data outside the target time period; The data selection module is used to select the first quantity of second collected data from the second collected dataset based on the first quantity corresponding to the first weight; and to select the second quantity of simulation data from the simulation dataset based on the second quantity corresponding to the second weight. The training module is used to train the previous round of cooling load prediction model based on the first collected data, the selected second collected data, and the selected simulation data to obtain the current round of cooling load prediction model; wherein, the first round of cooling load prediction model is obtained by training based on the simulation dataset.

[0024] In some optional implementations, the weight determination module is specifically used for: A first attenuation parameter is obtained based on the total number of accumulated collected data and the first attenuation coefficient; wherein, the first attenuation parameter is inversely proportional to the total number of accumulated collected data; The product of the first preset ratio and the first attenuation parameter is determined as the first weight.

[0025] In some optional implementations, the weight determination module is specifically used for: A second attenuation parameter is obtained based on the total number of accumulated collected data and the second attenuation coefficient; wherein the second attenuation parameter is inversely proportional to the total number of accumulated collected data. The product of the second preset ratio and the second attenuation parameter is determined as the second weight.

[0026] In some optional implementations, the data selection module is further configured to obtain the first quantity and the second quantity in the following manner: Based on the first weight and the second weight, a third weight corresponding to the first collected data is obtained; The total number of samples is obtained based on the number of the first collected data and the third weight; The product of the total number of samples and the first weight is determined as the first quantity, and the product of the total number of samples and the second weight is determined as the second quantity.

[0027] In some optional implementations, the data selection module is specifically used for: From the second collected dataset, select a first number of second collected data points whose collection time is adjacent to the target time period.

[0028] In some optional implementations, the data selection module is specifically used for: Obtain the target environment parameters corresponding to the first collected data; Based on the target environment parameters, the second number of simulation data is selected from the simulation dataset.

[0029] In some alternative implementations, the training module is specifically used for: Based on the first collected data, the selected second collected data, and the selected simulation data, the cooling load prediction model of the previous round is trained iteratively multiple times to obtain the cooling load prediction model of the current round; wherein each iteration includes: Data sequences are selected from each training dataset, and the target cold load corresponding to each selected data sequence is obtained; wherein, the data sequence is a preset number of consecutive data in the corresponding training dataset, and the target cold load is the cold load of the data sequence at the next moment in the corresponding training dataset; the training dataset is the first collected data, the selected second collected data, or the selected simulation data; Based on the selected data sequence, the predicted cooling load is obtained; The target loss is obtained based on the difference between the predicted cooling load and the corresponding target cooling load; The parameters of the previous round of cold load prediction model are adjusted based on the target loss.

[0030] Fourthly, embodiments of this application provide a cooling load prediction device, which includes: The prediction module is used to input the data sequence corresponding to the target time into the current round of the cold load prediction model after obtaining the data sequence, and obtain the predicted cold load at the target time through the current round of the cold load prediction model; wherein, the current round of the cold load prediction model is trained using any of the methods described in the first aspect above.

[0031] Some alternative implementations also include a fallback module for: Obtain the error between the predicted cooling load and the actual cooling load within the target duration; When the error reaches the error threshold, the previous round's cooling load prediction model is selected to replace the current round's cooling load prediction model.

[0032] Fifthly, embodiments of this application provide an electronic device, including at least one processor and at least one memory; wherein the memory stores a computer program, and when the program is executed by the processor, the processor performs the model training method as described in any of the first aspects above, or the cold load prediction method as described in any of the second aspects above.

[0033] Sixthly, embodiments of this application provide a computer-readable storage medium storing a computer program executable by a processor, which, when run on the processor, causes the processor to perform the model training method described in any of the first aspects or the cold load prediction method described in any of the second aspects. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 A flowchart illustrating the first model training method provided in this application embodiment; Figure 2 This is a schematic diagram of the training process of the cooling load prediction model provided in the embodiments of this application; Figure 3 A flowchart illustrating the first weight acquisition method provided in an embodiment of this application; Figure 4 A flowchart illustrating the second weight acquisition method provided in this application embodiment; Figure 5 A flowchart illustrating the first quantity and the second quantity acquisition method provided in the embodiments of this application; Figure 6A flowchart illustrating the second data selection method provided in this application embodiment; Figure 7 A flowchart illustrating the simulation data selection method provided in the embodiments of this application; Figure 8 A flowchart illustrating the second model training method provided in this application embodiment; Figure 9 This is a schematic diagram of a sliding window provided in an embodiment of this application; Figure 10 A diagram illustrating the architecture of the cooling load prediction model provided in this application embodiment; Figure 11 A flowchart illustrating a cooling load prediction method provided in an embodiment of this application; Figure 12 This is a schematic diagram of the structure of a model training device provided in an embodiment of this application; Figure 13 This is a schematic diagram of the structure of a cooling load prediction device provided in an embodiment of this application; Figure 14 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0037] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0038] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the term "connection" should be interpreted broadly. For example, it can refer to a direct connection, an indirect connection through an intermediate medium, or a connection within two devices. Those skilled in the art can understand the specific meaning of the above term in this application based on the specific circumstances.

[0039] With the development of technology and the increase in energy consumption, building energy consumption is a significant component of overall energy consumption. Therefore, energy-saving optimization in buildings has received widespread attention. A large portion of building energy consumption is generated by air conditioning systems, making their energy-saving optimization crucial. Energy-saving optimization strategies for air conditioning systems typically rely on cooling load; that is, appropriate energy-saving optimization strategies need to be determined based on the predicted cooling load of the air conditioning system.

[0040] In related technologies, a cooling load prediction model is trained based on a large amount of historical data, enabling the cooling load prediction model to accurately predict the cooling load.

[0041] However, in the early stages of using an air conditioning system, there is a lack of available historical data, making it impossible to train a cooling load prediction model. Furthermore, with only a small amount of historical data available later, it is difficult to accurately train the cooling load prediction model, making it difficult to accurately predict the cooling load using the above methods.

[0042] In view of this, embodiments of this application propose a model training method, a cooling load prediction method, and an electronic device to accurately train a cooling load prediction model even when there is a lack of collected data. The model training method includes: after obtaining first collected data within a target time period, determining a first weight corresponding to a second collected dataset and a second weight corresponding to a simulation dataset based on the total number of accumulated collected data; wherein the second collected dataset includes multiple second collected data outside the target time period; selecting the first number of second collected data from the second collected dataset based on a first number corresponding to the first weight; and selecting the second number of simulation data from the simulation dataset based on a second number corresponding to the second weight; training the previous round of cooling load prediction model based on the first collected data, the selected second collected data, and the selected simulation data to obtain the current round of cooling load prediction model; wherein the first round of cooling load prediction model is obtained based on the simulation dataset.

[0043] By first training a cooling load prediction model based on a simulation dataset when no data is collected, a corresponding cooling load prediction model is available for the initial use of the air conditioning system. Then, after data is collected, considering the insufficient amount of available data, a first weight (the weight of historical data) and a second weight (the weight of simulation data) are obtained for the second dataset based on the total number of collected data. The second dataset is selected based on the first weight, and the simulation data is selected based on the second weight. The proportion of simulation data and second dataset used in model training is gradually reduced, thereby achieving a smooth filtering from simulation data dominance to data collection data dominance in model training. This makes the cooling load predicted by the cooling load prediction model increasingly closer to the current actual usage of the air conditioning system.

[0044] The technical solution of this application and how it solves the above-mentioned technical problems will be described in detail below with reference to the accompanying drawings and specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0045] Figure 1 A flowchart illustrating the first model training method provided in this application embodiment is shown below. Figure 1 As shown, it includes the following steps: Step S101: After obtaining the first collection data within the target time period, determine the first weight corresponding to the second collection dataset and the second weight corresponding to the simulation dataset based on the total number of accumulated collection data; wherein, the second collection dataset includes multiple second collection data outside the target time period.

[0046] In practice, even if data is collected, if the amount of collected data is small, it is difficult to accurately train the cold load prediction model. Based on this, in the embodiments of this application, after obtaining the first collected data within the target time period, the first weight (weight of historical collected data) and the second weight (weight of simulation data) corresponding to the second collected dataset are obtained according to the total number of accumulated collected data. This allows the second collected data (historical collected data) and simulation data to be combined and used in the training process of the cold load prediction model based on the first collected data.

[0047] This application does not limit the specific implementation of the first collected data, the second collected data, and the simulation data. For example, each of the three includes multiple sub-data, including environmental sub-data (such as outdoor temperature, outdoor humidity, wind speed, wind direction, and solar radiation), time sub-data (the corresponding collection time / simulation time, and whether it is a holiday, etc.), and target cooling load (for the collected data, the target cooling load is the actual cooling load; for the simulation data, the target cooling load is the simulated cooling load).

[0048] This application does not specifically limit the method of obtaining the simulated cooling load in the above simulation data. For example, the simulation parameters are input into the modeling simulation software, and the modeling simulation software generates the simulated cooling load at each time point (e.g., hourly). The simulation parameters may include information such as building parameters, equipment parameters, meteorological parameters, and operating strategies.

[0049] For example, building parameters include thermal parameters of the building envelope (such as wall materials, window and door glass types, roof skylights, etc.) and spatial attributes (such as space division, lighting power density, etc.). Equipment parameters include cold source system parameters (such as equipment and performance attributes of chillers, chilled water pumps, cooling pumps and cooling towers), as well as terminal fan coil unit parameters and combined air conditioning parameters. Meteorological parameters include information such as project location and typical meteorological year data; The operational strategy includes information such as schedule patterns, equipment control logic, and personnel activity.

[0050] The simulation parameters described above are merely illustrative. In practice, any parameter that affects the cooling load of the air conditioning system can be used as a simulation parameter. This application does not impose any specific limitations on this.

[0051] Step S102: Based on the first quantity corresponding to the first weight, select the first quantity of second collected data from the second collected dataset; and based on the second quantity corresponding to the second weight, select the second quantity of simulation data from the simulation dataset.

[0052] In practice, after determining the first weight corresponding to the second data acquisition dataset and the second weight corresponding to the simulation dataset, a first number of second data acquisition datasets are selected based on the first weight, and a second number of simulation datasets are selected based on the second weight.

[0053] Step S103: Based on the first collected data, the selected second collected data, and the selected simulation data, train the previous round of cooling load prediction model to obtain the current round of cooling load prediction model; wherein, the first round of cooling load prediction model was obtained by training based on the simulation dataset.

[0054] After selecting the second set of collected data and simulation data, the selected second set of collected data and simulation data are combined with the first set of collected data to participate in the training process of the cooling load prediction model.

[0055] In addition, when no data is collected, a first-round cooling load prediction model is trained based on the simulation dataset. In this way, a corresponding cooling load prediction model is available for the initial use of the air conditioning system.

[0056] See Figure 2 As shown, taking the current time corresponding to the Nth target time period as an example, in the initial stage of the use of the air conditioning system (before time t0), the simulation dataset is used to train the first round (1st round) of the cooling load prediction model; After the first target period, the first data collection 1 is generated, but there is no historical data collection (second data collection) at this time. Therefore, after determining the second weight, we only need to select the simulation data (denoted as simulation data 1) and train the cooling load prediction model 1 based on the first data collection 1 + simulation data 1 to obtain the second round of cooling load prediction model. After the second target time period, the first data collection 2 is generated. At this time, the second data collection is already available. The second data collection (denoted as the second data collection 2) is selected based on the first weight, and the simulation data (denoted as the simulation data 2) is selected based on the second weight. The cooling load prediction model 2 is trained based on the first data collection 2 + the second data collection 2 + the simulation data 2 to obtain the cooling load prediction model for the third round. After the Nth target time period, the first data N is generated. The second data (denoted as the second data N) is selected based on the first weight, and the simulation data (denoted as the simulation data N) is selected based on the second weight. The cooling load prediction model N-1 is trained based on the first data N + the second data N + the simulation data N to obtain the cooling load prediction model for the N+1th round.

[0057] It is understandable that in the above process, the first weight is different in each round, and the second weight is also different.

[0058] The above scheme, when no data is collected, first trains a cooling load prediction model based on a simulation dataset to obtain the initial cooling load prediction model. In this way, there is a corresponding cooling load prediction model to predict the cooling load in the early stage of the air conditioning system's use. Then, after data is collected, considering the insufficient amount of collected data, the first weight (the weight of historical collected data) and the second weight (the weight of simulation data) corresponding to the second collected dataset are obtained based on the total number of accumulated collected data. The second collected data is selected based on the first weight, and the simulation data is selected based on the second weight. The proportion of simulation data and second collected data used in model training is gradually reduced, thereby achieving a smooth filtering of model training from simulation data dominance to collected data dominance, so that the cooling load predicted by the cooling load prediction model is closer and closer to the current actual usage of the air conditioning system.

[0059] In some alternative implementations, the first weight described above can be obtained in, but is not limited to, the following ways, see [reference]. Figure 3 As shown, it includes the following steps: Step S301: Based on the total number of accumulated collected data and the first attenuation coefficient, obtain the first attenuation parameter; wherein, the first attenuation parameter is inversely proportional to the total number of accumulated collected data; Step S302: The product of the first preset ratio and the first attenuation parameter is determined as the first weight.

[0060] In this embodiment, a first attenuation coefficient is set, which represents the attenuation rate of the first weight as the total number of accumulated collected data increases; the larger the first attenuation coefficient, the greater the attenuation rate. A first preset ratio is also set, which is the initial weight (sampling ratio) of historical data.

[0061] During implementation, as the amount of collected data increases, a first attenuation parameter is obtained that gradually decreases based on the total number of collected data and the first attenuation coefficient; then, a first weight is determined based on the product of the first preset ratio and the first attenuation parameter, so that the first weight also decreases as the total number of collected data increases.

[0062] For example, the first attenuation parameter A1 = e(-β1) V0); First weight W1=α1 A1=α1 e(-β1 V0); where β1 is the first attenuation coefficient (β1>0), V0 is the total number of accumulated data, and α1 is the first preset ratio (0<α<1).

[0063] This application does not specify the specific data for the above parameters. For example, β1 can be any value between 0.0001 and 0.001, and α1 can be any value between 0.25 and 0.35.

[0064] The above scheme sets a first attenuation coefficient and a first preset ratio, and obtains a first attenuation parameter based on the total number of accumulated collected data and the first attenuation coefficient, so that the first attenuation parameter decreases as the total number of accumulated collected data increases; then, based on the product of the first preset ratio and the first attenuation parameter, a first weight is determined, so that the first weight also decreases as the total number of accumulated collected data increases.

[0065] In some alternative implementations, the second weight described above can be obtained in, but is not limited to, the following ways, see [reference]. Figure 4 As shown, it includes the following steps: Step S401: Based on the total number of accumulated collected data and the second attenuation coefficient, obtain the second attenuation parameter; wherein the second attenuation parameter is inversely proportional to the total number of accumulated collected data; Step S402: The product of the second preset ratio and the second attenuation parameter is determined as the second weight.

[0066] In this embodiment, a second attenuation coefficient is set, which characterizes the attenuation rate of the second weight as the total number of accumulated collected data increases; the larger the second attenuation coefficient, the greater the attenuation rate. A second preset ratio is also set, which is the initial weight (sampling ratio) of the second collected data.

[0067] During implementation, as the amount of collected data increases, a gradually decreasing second attenuation parameter is obtained based on the total number of collected data and the second attenuation coefficient; then, based on the product of the second preset ratio and the second attenuation parameter, a second weight is determined, so that the second weight also decreases as the total number of collected data increases.

[0068] For example, the second attenuation parameter A2 = e(-β2) V0); Second weight W2=α2 A2=α2 e(-β2 V0); where β2 is the second attenuation coefficient (β2>0), V0 is the total number of accumulated data, and α2 is the second preset ratio (0<α<1).

[0069] This application does not specify the specific data for the above parameters. For example, β2 can be any value between 0.001 and 0.003, and α2 can be any value between 0.65 and 0.75.

[0070] The above scheme sets a second attenuation coefficient and a second preset ratio, and obtains a second attenuation parameter based on the total number of accumulated collected data and the second attenuation coefficient, so that the second attenuation parameter decreases as the total number of accumulated collected data increases; then, based on the product of the second preset ratio and the second attenuation parameter, a second weight is determined, so that the second weight also decreases as the total number of accumulated collected data increases.

[0071] In some alternative implementations, the first attenuation coefficient is smaller than the second attenuation coefficient. That is, as the amount of collected data increases, the attenuation rate of the first weight is smaller than that of the second weight. This effectively reduces the number of samples in the simulation data when there is a large amount of collected data, and ensures that the number of samples in the historical data does not drop sharply.

[0072] In some optional implementations, the sum of the first preset ratio and the second preset ratio is 1, so that too much or too little simulation data and historical data are sampled.

[0073] In some alternative implementations, the first quantity and the second quantity can be obtained in, but are not limited to, the following ways, see [reference]. Figure 5 As shown, it includes the following steps: Step S501: Based on the first weight and the second weight, obtain the third weight corresponding to the first collected data.

[0074] The first and second weights are both sampling ratios, but the total number of samples is unknown. However, the number of the first data points is known. Based on the first and second weights, the third weight corresponding to the first data points can be obtained, which is the sampling ratio of the first data points. Thus, based on the number of the first data points and the third weight, the total number of samples can be deduced.

[0075] For example, based on W1+W2+W3=1, we get W3=1-W1-W2; where W1 is the first weight, W2 is the second weight, and W3 is the third weight.

[0076] Step S502: Obtain the total number of samples based on the number of the first collected data and the third weight.

[0077] In practice, after obtaining the third weight, the total number of samples can be deduced from the number of first collected data and the third weight.

[0078] For example, the total number of samples D0 = D3 / W3; where D3 is the number of the first data collections and W3 is the third weight.

[0079] Step S503: The product of the total number of samples and the first weight is determined as the first quantity, and the product of the total number of samples and the second weight is determined as the second quantity.

[0080] In this embodiment, the first weight is the sampling ratio of the second collected data. Therefore, the product of the total number of samples and the first weight is the number of samples of the second collected data (the first number). The second weight is the sampling ratio of the simulation data. Therefore, the product of the total number of samples and the second weight is the number of samples of the simulation data (the second number).

[0081] The formula is expressed as: First quantity D1 = D0 W1; where D0 is the total number of samples and W1 is the first weight; The second quantity D2=D0 W2; where D0 is the total number of samples and W2 is the second weight.

[0082] The above scheme obtains a third weight corresponding to the first collected data based on the first weight and the second weight, and obtains the total number of samples based on the number of the first collected data and the third weight. In this way, with the number of the first collected data remaining unchanged, the total number of samples will dynamically change with the total number of accumulated collected data. After obtaining the total number of samples, a more suitable number of samples for the second collected data is obtained based on the product of the total number of samples and the first weight, and a more suitable number of samples for the simulation data is obtained based on the product of the total number of samples and the second weight.

[0083] In some alternative implementations, the second data collection can be selected in, but is not limited to, the following ways, see [reference]. Figure 6 As shown, it includes the following steps: Step S601: Obtain the acquisition time of each second acquisition data in the second acquisition dataset; Step S602: Select a first number of second collection data from the second collection dataset whose collection time is adjacent to the target time period.

[0084] In practice, the more similar the selected second collection data is to the first collection data, the better the training effect. For scenarios such as air conditioning system cooling load prediction, the collection time of the data is often closer and the data distribution is more similar. Based on this, the embodiments of this application select the second collection data whose collection time is adjacent to the target time period.

[0085] The above scheme selects a second data point whose collection time is adjacent to the target time period, so that the collection time of the selected second data point is close to that of the first data point. In this way, the data distribution of the selected second data point is closer to that of the first data point.

[0086] In some alternative implementations, simulation data can be selected in, but is not limited to, the following ways, see [reference]. Figure 7 As shown, it includes the following steps: Step S701: Obtain the target environment parameters corresponding to the first collected data; Step S702: Based on the target environment parameters, select the second number of simulation data from the simulation dataset.

[0087] In practice, the more similar the selected simulation data is to the first collected data, the better the training effect. For scenarios such as air conditioning system cooling load prediction, the more similar the environmental parameters of the simulation data and the collected data are, the more similar the data distribution is. Based on this, the embodiments of this application select simulation data according to the target environmental parameters of the first collected data.

[0088] For example, taking environmental parameters including temperature and humidity as an example, each second number of simulation data is taken as a simulation segment, and the temperature similarity and humidity similarity between each simulation segment and the first collected data are calculated; based on the temperature similarity and humidity similarity, the target similarity is obtained by weighting; and the simulation segment with the highest target similarity is selected.

[0089] The process of selecting simulation data described above is merely an illustrative example. In practice, other parameters may be selected as target environment parameters, and this application does not impose any specific limitations on this.

[0090] The above scheme selects simulation data based on the target environmental parameters of the first collected data, making the selected simulation data closer to the environmental parameters of the first collected data. This makes the selected simulation data more similar to the data distribution of the first collected data.

[0091] Figure 8 A flowchart illustrating the second model training method provided in this application embodiment is shown below. Figure 8 As shown, it includes the following steps: Step S801: After obtaining the first collection data within the target time period, determine the first weight corresponding to the second collection dataset and the second weight corresponding to the simulation dataset based on the total number of accumulated collection data; wherein, the second collection dataset includes multiple second collection data outside the target time period.

[0092] Step S802: Based on the first quantity corresponding to the first weight, select the first quantity of second collected data from the second collected dataset; and based on the second quantity corresponding to the second weight, select the second quantity of simulation data from the simulation dataset.

[0093] The specific implementation methods of steps S801 to S802 can be found in other embodiments, and will not be repeated here.

[0094] Step S803: Based on the first collected data, the selected second collected data, and the selected simulation data, the cooling load prediction model of the previous round is trained iteratively multiple times to obtain the cooling load prediction model of the current round; wherein, each iteration includes: Step S8031: Select data sequences from each training dataset and obtain the target cold load corresponding to each selected data sequence; wherein, the data sequence is a preset number of consecutive data in the corresponding training dataset, and the target cold load is the cold load of the data sequence at the next moment in the corresponding training dataset; the training dataset is the first collected data, the selected second collected data, or the selected simulation data.

[0095] In practice, the cold load at the next moment of the sliding window is predicted based on a continuous, preset amount of data as a data sequence. Based on this, when training the model, it is also necessary to obtain multiple data sequences and the target cold load corresponding to each data sequence based on the first collected data, the selected second collected data, and the selected simulation data.

[0096] For example, a preset number of consecutive data points can be obtained by using a sliding window.

[0097] See Figure 9 As shown, the target time period includes n first collected data points: a1, a2, a3, ..., an. Taking a preset quantity of 3 as an example (the length of the sliding window is 3 time units), a1, a2, and a3 are data sequence 1, and the corresponding target cooling load is the actual cooling load in a4; a2, a3, and a4 are data sequence 2, and the corresponding target cooling load is the actual cooling load in a5; ... a(n-3), a(n-2), and a(n-1) are data sequence n-3, and the corresponding target cooling load is the actual cooling load in an.

[0098] The same approach can be used for the selected second collection data and the selected simulation data. Multiple data sequences and their corresponding target cooling loads can be obtained through a sliding window. Examples will not be given here.

[0099] In some alternative implementations, each training dataset can be preprocessed before selecting data sequences for each training dataset individually. For example: For missing data, fill in the missing data based on adjacent data; Standardize all sub-data in all training datasets, scaling them to a single range to eliminate the influence of unit dimensions.

[0100] In implementation, all data sequences can be divided into training and testing sets. Parameter tuning is performed using the training set, and the tuned model is tested using the testing set. This application does not specify a particular ratio for the training and testing sets; for example, 80% of the data sequences can be used as the training set and 20% as the testing set.

[0101] Step S8032: Based on the selected data sequence, obtain the predicted cooling load.

[0102] In practice, the previous round of cooling load prediction model is input for each data sequence, and the corresponding predicted cooling load is obtained by using the cooling load prediction model.

[0103] Step S8033: Based on the difference between the predicted cooling load and the corresponding target cooling load, obtain the target loss.

[0104] The higher the certainty of the cooling load forecasting model, the closer the forecasted cooling load is to the corresponding target cooling load; Based on this, the embodiments of this application obtain the target loss based on the difference between the predicted cooling load and the corresponding target cooling load.

[0105] For example, the root mean square error (RMSE) or mean absolute percentage error (MAPE) between the predicted cooling load and the corresponding target cooling load can be used as the target loss. Alternatively, the target loss can be obtained by analyzing the difference between the predicted cooling load and the corresponding target cooling load using a target loss function (such as the Huber loss function).

[0106] Step S8034: Adjust the parameters of the previous round of cold load prediction model based on the target loss.

[0107] Since a smaller target loss indicates a more accurate set of parameters in the current cooling load prediction model, this application embodiment adjusts the parameters of the previous cooling load prediction model based on the target loss.

[0108] This application does not limit the specific implementation method for parameter tuning; an optimizer can be used. The optimizer can be any optimizer capable of implementing the model parameters, such as the Adaptive Moments Estimator (Adam) optimizer with an initial learning rate of 0.001. During parameter tuning, if the target loss is found to increase, the system can revert to the previous model parameters.

[0109] The embodiments of this application do not specifically limit the iteration stopping condition. For example, a preset number of iterations (such as 200 times) can be set, and the iteration will stop when the current number of iterations equals the preset number of iterations.

[0110] This application does not specifically limit the architecture of the above-mentioned cooling load prediction model in its embodiments; please refer to [reference needed]. Figure 10 As shown, it includes an input layer, a first LSTM layer (LSTM 1), a Dropout layer, an attention layer, a second LSTM layer (LSTM 2), a fully connected layer, and an output layer.

[0111] In implementation, m is generated based on a data sequence. A tensor of n is input to the input layer, where m is a preset number (the total number of data points contained in a data sequence), and n is the total number of sub-data points contained in a data sequence. This tensor represents the n-dimensional feature data of m consecutive time steps.

[0112] LSTM 1 contains multiple (e.g., 64) neurons, returns the output of each time step (i.e., returns the hidden states of m time steps), not just the output of the last time step, to perform preliminary feature learning and provide rich sequence information for subsequent attention layers.

[0113] Dropout layers, with a dropout rate (e.g., 0.2), randomly "drop out" (temporarily ignore) neurons in a layer at that dropout rate during the training phase. For models containing complex LSTM layers with numerous parameters, this allows them to learn more robust and generalized features, thus avoiding over-reliance on individual neurons or feature combinations.

[0114] The attention layer calculates a set of weights (the attention weights can be calculated by a fully connected layer and a softmax function), and then sums the hidden states of the LSTM1 output at m time steps based on these weights, finally outputting a context vector. This context vector condenses the most critical information in the sequence.

[0115] LSTM 2 contains multiple (e.g., 32) neurons, further abstracts the attention-weighted context vector, and only returns the output of the last time step.

[0116] The fully connected layer contains multiple (e.g., 32) neurons, which combine and transform the output of LSTM 2 and map it to the final output space.

[0117] The output layer contains one neuron and uses a linear activation function to transform the output of the fully connected layer into a prediction result (predicting cold load).

[0118] In combination with the above Figure 2 As shown, the embodiments of this application are incremental learning. Therefore, the parameters of LSTM1, the attention layer, LSTM2 and the fully connected layer can be tuned only when training the cold load prediction model in the first round. In the subsequent model training process, LSTM1 can be frozen and only the attention layer, LSTM2 and the fully connected layer can be tuned.

[0119] The above Figure 10 This is merely an illustrative example of a cooling load forecasting model; in practice, other architectures of cooling load forecasting models may also be used.

[0120] Figure 11 This is a flowchart illustrating a cooling load prediction method provided in an embodiment of this application, as shown below. Figure 11 As shown, it includes the following steps: Step S1101: Obtain the data sequence corresponding to the target time.

[0121] Step S1102: Input the data sequence into the current round of cooling load prediction model, and obtain the predicted cooling load at the target time through the current round of cooling load prediction model.

[0122] The implementation method of the data sequence and the training method of the cold load prediction model in this round can be referred to the above embodiments, and will not be repeated here.

[0123] The above scheme, when no data is collected, first trains the initial cooling load prediction model based on a large amount of simulation data. This allows the model to predict the cooling load at the target time during the initial use of the air conditioning system. Furthermore, after data collection becomes available, considering the limited amount of available data, a first weight (for historical data) and a second weight (for simulation data) are determined based on the total number of collected data. Second collected data are selected based on the first weight, and simulation data is selected based on the second weight. This gradually reduces the proportion of simulation data and second collected data used in model training, achieving a smooth transition from simulation data dominance to collected data dominance. This makes the cooling load predicted by the model increasingly closer to the actual usage of the air conditioning system. Therefore, the cooling load prediction model trained in this way can typically predict the target cooling load with relatively high accuracy.

[0124] In some optional implementations, after obtaining the predicted cooling load at the target time, the method further includes: Obtain the error between the predicted cooling load and the actual cooling load within the target duration; When the error reaches the error threshold, the previous round's cooling load prediction model is selected to replace the current round's cooling load prediction model.

[0125] The above scheme indicates that when the error between the predicted cooling load and the actual cooling load within the target time period reaches the error threshold, it means that the current cooling load model is unable to predict the current cooling load. By replacing the current cooling load prediction model with the previous cooling load prediction model, that is, reverting to the previous cooling load prediction model to continue the cooling load prediction, even if the cooling load prediction model trained in a certain round is not accurate enough, there is still a backup cooling load prediction model from the previous round to continue the subsequent cooling load prediction.

[0126] like Figure 12 As shown in the figure, this application embodiment provides a model training device 1200, which includes: The weight determination module 1201 is used to determine the first weight corresponding to the second collection dataset and the second weight corresponding to the simulation dataset based on the total number of accumulated collection data after obtaining the first collection data within the target time period; wherein, the second collection dataset includes multiple second collection data outside the target time period; The data selection module 1202 is used to select the first quantity of second collected data from the second collected dataset based on the first quantity corresponding to the first weight; and to select the second quantity of simulation data from the simulation dataset based on the second quantity corresponding to the second weight. The training module 1203 is used to train the previous round of cooling load prediction model based on the first collected data, the selected second collected data, and the selected simulation data to obtain the current round of cooling load prediction model; wherein, the first round of cooling load prediction model is obtained by training based on the simulation dataset.

[0127] In some optional implementations, the weight determination module 1201 is specifically used for: A first attenuation parameter is obtained based on the total number of accumulated collected data and the first attenuation coefficient; wherein, the first attenuation parameter is inversely proportional to the total number of accumulated collected data; The product of the first preset ratio and the first attenuation parameter is determined as the first weight.

[0128] In some optional implementations, the weight determination module 1201 is specifically used for: A second attenuation parameter is obtained based on the total number of accumulated collected data and the second attenuation coefficient; wherein the second attenuation parameter is inversely proportional to the total number of accumulated collected data. The product of the second preset ratio and the second attenuation parameter is determined as the second weight.

[0129] In some optional implementations, the data selection module 1202 is further configured to obtain the first quantity and the second quantity in the following manner: Based on the first weight and the second weight, a third weight corresponding to the first collected data is obtained; The total number of samples is obtained based on the number of the first collected data and the third weight; The product of the total number of samples and the first weight is determined as the first quantity, and the product of the total number of samples and the second weight is determined as the second quantity.

[0130] In some optional implementations, the data selection module 1202 is specifically used for: From the second collected dataset, select a first number of second collected data points whose collection time is adjacent to the target time period.

[0131] In some optional implementations, the data selection module 1202 is specifically used for: Obtain the target environment parameters corresponding to the first collected data; Based on the target environment parameters, the second number of simulation data is selected from the simulation dataset.

[0132] In some optional implementations, the training module 1203 is specifically used for: Based on the first collected data, the selected second collected data, and the selected simulation data, the cooling load prediction model of the previous round is trained iteratively multiple times to obtain the cooling load prediction model of the current round; wherein each iteration includes: Data sequences are selected from each training dataset, and the target cold load corresponding to each selected data sequence is obtained; wherein, the data sequence is a preset number of consecutive data in the corresponding training dataset, and the target cold load is the cold load of the data sequence at the next moment in the corresponding training dataset; the training dataset is the first collected data, the selected second collected data, or the selected simulation data; Based on the selected data sequence, the predicted cooling load is obtained; The target loss is obtained based on the difference between the predicted cooling load and the corresponding target cooling load; The parameters of the previous round of cold load prediction model are adjusted based on the target loss.

[0133] Since this device is the same as the device in the model training method in the embodiments of this application, and the principle of the device in solving the problem is similar to that of the model training method, the implementation of this device can refer to the implementation of the model training method, and the repeated parts will not be described again.

[0134] like Figure 13 As shown in the figure, this application embodiment provides a cooling load prediction device 1300, which includes: The prediction module 1301 is used to input the data sequence corresponding to the target time into the current round of cold load prediction model after obtaining the data sequence, and obtain the predicted cold load at the target time through the current round of cold load prediction model; wherein, the current round of cold load prediction model is trained using any of the above-mentioned model training methods.

[0135] In some alternative implementations, a rollback module 1302 is also included for: Obtain the error between the predicted cooling load and the actual cooling load within the target duration; When the error reaches the error threshold, the previous round's cooling load prediction model is selected to replace the current round's cooling load prediction model.

[0136] Since this device is the same as the device in the cooling load forecasting method in the embodiments of this application, and the principle of the device in solving the problem is similar to that of the cooling load forecasting method, the implementation of this device can refer to the implementation of the cooling load forecasting method, and the repeated parts will not be described again.

[0137] Based on the same technical concept, this application also provides an electronic device 1400, such as... Figure 14As shown, it includes at least one processor 1401 and a memory 1402 connected to at least one processor. In this embodiment, the specific connection medium between the processor 1401 and the memory 1402 is not limited. Figure 14 Taking the connection between processor 1401 and memory 1402 via bus 1403 as an example. The bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, Figure 14 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0138] The processor 1401 is the control center of the electronic device, capable of connecting various parts of the device via various interfaces and lines. It performs data processing by running or executing instructions stored in the memory 1402 and accessing data stored in the memory 1402. Optionally, the processor 1401 may include one or more processing units. The processor 1401 may integrate an application processor and a modem processor. The application processor primarily handles the operating system, user interface, and applications, while the modem processor primarily handles issuing instructions. It is understood that the modem processor may not be integrated into the processor 1401. In some embodiments, the processor 1401 and the memory 1402 may be implemented on the same chip; in other embodiments, they may be implemented on separate chips.

[0139] Processor 1401 can be a general-purpose processor, such as a CPU, digital signal processor, application-specific integrated circuit (ASIC), field-programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of the model training method or cold load prediction method can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0140] Memory 1402, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. Memory 1402 may include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory, random access memory (RAM), static random access memory (SRAM), programmable read-only memory (PROM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), magnetic storage, magnetic disk, optical disk, etc. Memory 1402 can be any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited thereto. In the embodiments of this application, memory 1402 can also be a circuit or any other device capable of implementing storage functions for storing program instructions and / or data.

[0141] In this embodiment, the memory 1402 stores a computer program, which, when executed by the processor 1401, causes the processor 1401 to perform the steps of the above-described model training or the steps of the above-described cold load prediction method.

[0142] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0143] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0144] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0145] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0146] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0147] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A model training method, characterized in that, Applied to electronic devices, the method includes: After obtaining the first set of collected data within the target time period, the first weight corresponding to the second set of collected data and the second weight corresponding to the simulation dataset are dynamically adjusted based on the total number of accumulated collected data. The second set of collected data includes multiple sets of second collected data outside the target time period. The first set of collected data consists of data actually collected by the air conditioning system during the target time period. The second set of collected data consists of data actually collected by the air conditioning system during historical times outside the target time period. Both the first weight and the second weight are inversely proportional to the total number of accumulated collected data; the first weight is the weight of historical collected data, and the second weight is the weight of simulation data. The simulation data includes at least building parameters, equipment parameters, meteorological parameters, and operational strategies. The building parameters include at least the thermal parameters of the building envelope and spatial attributes. The thermal parameters of the building envelope include at least wall materials, window and door glass types, and roof skylights. The spatial attributes include at least spatial division and lighting power density. The equipment parameters include at least cooling system parameters, terminal fan coil unit parameters, and combined air conditioning system parameters. The cooling system parameters include the equipment and performance attributes of chillers, chilled water pumps, cooling pumps, and cooling towers. The meteorological parameters include at least the project location and typical meteorological year data. The operational strategies include at least a timetable pattern, equipment control logic, and personnel activity status. Based on the first quantity corresponding to the first weight, select the first quantity of second collected data from the second collected dataset whose collection time is adjacent to the target time period; and, based on the second quantity corresponding to the second weight, obtain the target environmental parameters corresponding to the first collected data; based on the target environmental parameters, select the second quantity of simulation data from the simulation dataset, wherein the selected second quantity of simulation data is the simulation data with the highest target similarity to the first collected data among the second quantity of simulation data in the simulation dataset; the simulation data is the data corresponding to the air conditioning system generated by simulation software; Based on the first collected data, the selected second collected data, and the selected simulation data, the cooling load prediction model from the previous round is trained iteratively multiple times to obtain the cooling load prediction model for the current round; wherein each iteration includes: Data sequences are selected from each training dataset, and the target cold load corresponding to each selected data sequence is obtained; wherein, the data sequence is a preset number of consecutive data in the corresponding training dataset, and the target cold load is the cold load of the data sequence at the next moment in the corresponding training dataset; the training dataset is the first collected data, the selected second collected data, or the selected simulation data; Based on the selected data sequence, the predicted cooling load is obtained; The target loss is obtained based on the difference between the predicted cooling load and the corresponding target cooling load; The parameters of the previous round of cold load prediction model are tuned based on the target loss; wherein, the first round of cold load prediction model is obtained by training based on the simulation dataset; the first weights are different in each round, and the second weights are different in each round.

2. The method as described in claim 1, characterized in that, The first weight is obtained in the following way: A first attenuation parameter is obtained based on the total number of accumulated collected data and the first attenuation coefficient; wherein, the first attenuation parameter is inversely proportional to the total number of accumulated collected data; The product of the first preset ratio and the first attenuation parameter is determined as the first weight.

3. The method as described in claim 1, characterized in that, The second weight is obtained in the following manner: A second attenuation parameter is obtained based on the total number of accumulated collected data and the second attenuation coefficient; wherein the second attenuation parameter is inversely proportional to the total number of accumulated collected data. The product of the second preset ratio and the second attenuation parameter is determined as the second weight.

4. The method as described in claim 1, characterized in that, The first quantity and the second quantity are obtained in the following manner: Based on the first weight and the second weight, a third weight corresponding to the first collected data is obtained; The total number of samples is obtained based on the number of the first collected data and the third weight; The product of the total number of samples and the first weight is determined as the first quantity, and the product of the total number of samples and the second weight is determined as the second quantity.

5. A method for predicting cooling load, characterized in that, include: After obtaining the data sequence corresponding to the target time, the data sequence is input into the current round of cold load prediction model, and the predicted cold load at the target time is obtained through the current round of cold load prediction model; wherein, the current round of cold load prediction model is trained using the method described in any one of claims 1 to 4.

6. The method as described in claim 5, characterized in that, After obtaining the predicted cooling load at the target time, the process also includes: Obtain the error between the predicted cooling load and the actual cooling load within the target duration; When the error reaches the error threshold, the previous round's cooling load prediction model is selected to replace the current round's cooling load prediction model.

7. An electronic device, characterized in that, It includes at least one processor and at least one memory; wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the method as described in any one of claims 1 to 6.

Images