Building complex demand load optimization method and system based on multimodal data fusion

Abstract

The present application relates to the field of electric power technology, and in particular to a method and system for optimizing the demand load of a building complex based on multimodal data fusion. The method includes: cleaning the historical data set of the building complex to obtain a cleaned historical data set; obtaining the reference load curve of each building in different time periods based on the cleaned historical data set; training the target neural network model using the historical data set of the building complex to obtain a trained target neural network model; using the trained target neural network model to predict the load data of the target building in the target time period to obtain the predicted load curve of the target building in the target time period; obtaining the target load curve of the target building in the target time period based on the predicted load curve and the corresponding reference load curve. The present invention can improve the accuracy of the model's prediction of the demand load of the building complex, and then adjust and optimize the demand load of the building complex based on the corresponding prediction results.

Description

Technical Field

[0001] The present invention relates to the field of electric power, and in particular to a method and system for optimizing building load demand based on multi-modal data fusion. Background Art

[0002] The building complex includes at least 2 buildings, and the building complex demand load refers to the total load required by all the buildings included in the building complex. The prior art discloses a method for predicting the load of a building based on load influencing factors. For example, the Chinese patent application with publication number CN114037182A discloses a building load prediction model training method, device and non-volatile storage medium. The patent application uses cleaned historical data as training samples, and the cleaning process includes cleaning out the sudden increase and decrease of loads; however, these sudden increase and decrease of loads may correspond to the real loads under extreme conditions (such as high temperature red warning weather). Cleaning out these sudden increase and decrease of loads is not conducive to the training of the building load prediction model, so that the building load prediction model does not have the function of predicting the load under extreme conditions, and the load under these extreme conditions is often the main load optimization object. How to improve the accuracy of the model's prediction of the building complex demand load, and then adjust and optimize the building complex demand load based on the prediction results of the building complex demand load, is an urgent problem to be solved. Summary of the invention

[0003] The purpose of the present invention is to provide a building complex demand load tuning method and system based on multimodal data fusion, so as to improve the accuracy of the model's prediction of the building complex demand load, and then adjust and optimize the building complex demand load based on the prediction results of the building complex demand load.

[0004] According to a first aspect of the present invention, a method for optimizing building load demand based on multimodal data fusion is provided, the method comprising the following steps:

[0005] S100, obtaining a historical data set of a building complex; the building complex includes at least two buildings, and the historical data set of each building includes load data and load influencing factor data of the corresponding building in a historical time period.

[0006] S200, cleaning the historical data set of the building complex to obtain a cleaned historical data set; the cleaning process includes deleting the load data of sudden increase and decrease in the historical data set.

[0007] S300, obtaining reference load curves of each building in different time periods according to the cleaned historical data set.

[0008] S400, training the target neural network model using the historical data set of the building complex to obtain a trained target neural network model.

[0009] S500, using the trained target neural network model to predict the load data of the target building in the target time period, and obtain the predicted load curve of the target building in the target time period. 1 ; The target building is any building in the building complex.

[0010] S600, obtaining a reference load curve of the target building in a time period matching the target time period 2 .

[0011] S700, according to cur 1 and cur 2 Obtain the target load curve of the target building in the target time period.

[0012] According to a second aspect of the present invention, a building complex demand load tuning system based on multimodal data fusion is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned building complex demand load tuning method based on multimodal data fusion when executing the computer program.

[0013] Compared with the prior art, the present invention has at least the following beneficial effects:

[0014] The present invention obtains a trained neural network model based on the historical data set of the building group, and the trained neural network model has the function of inferring the corresponding load data according to the input load influencing factor data; since the historical data set of the building group based on the present invention is a data set after the processing of the load data with sudden increase and sudden decrease is not deleted, it is a relatively real data set, therefore, the load data of the target building in the target time period predicted based on the trained neural network model is also relatively real load data, in which there may also be sudden increase or sudden decrease load data; for the load curve of the target building in the target time period obtained based on the predicted load data, the present invention corrects it according to the reference load curve of the target building in the corresponding same time period (obtained based on the data set after the processing of the load data with sudden increase and sudden decrease is deleted, which can meet the low-carbon operation requirements of the building), and uses the corrected target load curve as the load adjustment reference load curve of the target building in the target time period to optimize the power demand of the target building in the target time period. The target load curve determined by the present invention is obtained by adjusting the reference load curve of the same period in the historical time period on the basis of the predicted load curve, so that the target load curve can take into account the actual power demand and low-carbon operation requirements of the target building, and based on the target load curve, the optimization of the demand load data of the target building can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 A flow chart of a building complex demand load optimization method based on multimodal data fusion provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0018] Embodiment 1:

[0019] According to this embodiment, Figure 1 As shown, a building complex demand load tuning method based on multimodal data fusion is provided, and the method comprises the following steps:

[0020] S100, obtaining a historical data set of a building complex; the building complex includes at least two buildings, and the historical data set of each building includes load data and load influencing factor data of the corresponding building in a historical time period.

[0021] In this embodiment, the historical data set of each building includes the load data and load influencing factor data of the corresponding building at each sampling moment in the historical time period; the interval between adjacent sampling moments is an empirical value, and optionally, the length of the historical time period is 1 year, and the interval between adjacent sampling moments is 1 hour or 0.5 hour.

[0022] In this embodiment, the load influencing factors are preset; as an optional specific implementation method, the load influencing factors include the attributes of the building and the external environment, etc. The attributes of the building include the type of use of the building (residential building, office building, commercial building or industrial building, etc.) and the area of ​​the building, etc., and the external environment includes external humidity and external temperature, etc.

[0023] S200, cleaning the historical data set of the building complex to obtain a cleaned historical data set; the cleaning process includes deleting the load data of sudden increase and decrease in the historical data set.

[0024] As a specific implementation, S200 also includes a judgment process for the sudden increase and decrease of load data in the historical data set, and the judgment process includes: if the difference between the load data at a certain sampling moment and the load data at the previous sampling moment is greater than a preset difference value threshold, then the load data at the sampling moment is judged as a sudden increase of load data; if the inverse of the difference between the load data at a certain sampling moment and the load data at the previous sampling moment is greater than the preset difference value threshold, then the load data at the sampling moment is judged as a sudden decrease of load data.

[0025] Optionally, the difference value threshold is an empirical value.

[0026] S300, obtaining reference load curves of each building in different time periods according to the cleaned historical data set.

[0027] In this embodiment, the time periods corresponding to the reference load curves are equal or unequal in length. As an optional specific implementation, S300 includes:

[0028] S301, obtaining cleaned load data corresponding to a target building in the cleaned historical data set.

[0029] S302, obtaining a load curve set corresponding to the target building according to the cleaned load data corresponding to the target building; the time periods corresponding to any load curve corresponding to the target building are equal in length.

[0030] In this embodiment, the horizontal axis of the load curve is time, and the vertical axis is load. As a specific implementation, the duration of the time period corresponding to any load curve is 24 hours.

[0031] S303: Divide the load curve set corresponding to the target building into several types of load curves according to several preset time points.

[0032] As an optional specific implementation, the interval between adjacent preset time points is 30 days. Therefore, the load curve set is divided into 12 categories, and different categories of load curves correspond to load curves of different months.

[0033] As an optional specific implementation, the interval between adjacent preset time points is 7 days. Therefore, the load curve set is divided into 52 categories, and different types of load curves correspond to load curves of different weeks.

[0034] S304: Determine the central load curve of each type of load curve as a reference load curve for a time period corresponding to the corresponding type.

[0035] In this embodiment, the central load curve of any type of load curve is the average of the load curves in this type, that is, the load data at the i-th position (i.e., the i-th point of the curve) of the central load curve of this type of load curve is the average load data at the i-th position of the load curve in this type.

[0036] Based on S301-S304, the load curve can be quickly classified, which is conducive to quickly obtaining the reference load curve; and the time periods corresponding to the load curves in each category are continuous, and the probability that the distance between the load curves corresponding to the continuous time periods is small is large, which can basically meet the condition that the distance between the load curves in the same category is small.

[0037] As a preferred specific implementation, S300 includes:

[0038] S310, obtaining cleaned load data corresponding to the target building in the cleaned historical data set.

[0039] S320, obtaining a load curve set corresponding to the target building according to the cleaned load data corresponding to the target building; the time periods corresponding to any load curve corresponding to the target building are equal in length.

[0040] S330, clustering the load curves corresponding to the target building according to the preset clustering conditions to obtain a plurality of clusters of load curves; the preset clustering conditions include: if the time period corresponding to a load curve is not continuous with the time period corresponding to any load curve in a cluster, then the load curve is not divided into the cluster.

[0041] When clustering the load curves and concentrated load curves corresponding to the target building, each load curve is taken as an object to be clustered, and the sum of the load data difference values ​​corresponding to all positions on the two load curves is determined as the distance between the two load curves; the load data difference value is the absolute value of the difference between the load data. As a specific implementation method, a density clustering method is used to cluster the load curves and concentrated load curves corresponding to the target building. Optionally, the density clustering method is a DBSCAN clustering method. The DBSCAN clustering method is first used to cluster the load curves and concentrated load curves corresponding to the target building into several clusters, and then each cluster is judged. If the time periods corresponding to the load curves included in the cluster are all continuous, the cluster is not processed; if the time periods corresponding to the load curves included in the cluster are not continuous, the cluster is divided so that the time periods corresponding to the load curves in any cluster after division are all continuous and the number of clusters obtained after division is minimized. For example, a cluster includes 5 load curves, and the time periods corresponding to these 5 load curves are January 1, 2023, January 2, 2023, January 13, 2023, January 14, 2023, and February 2, 2023. This cluster does not meet the condition that the time periods corresponding to the load curves included in the cluster are continuous, among which the time periods corresponding to the load curves on January 1, 2023 and January 2, 2023 are continuous, and the time periods corresponding to the load curves on January 13, 2023 and January 14, 2023 are continuous. If the time periods corresponding to the load curves of the cluster are continuous, and the time period corresponding to the load curve of February 2, 2023 is not continuous with the time periods corresponding to any other load curves in the cluster, then the number of clusters obtained after the cluster is divided is 3, the first cluster includes the load curves corresponding to the time periods of January 1, 2023 and January 2, 2023, the second cluster includes the load curves corresponding to the time periods of January 13, 2023 and January 14, 2023, and the third cluster includes the load curve corresponding to the time period of February 2, 2023. For another example, a cluster includes 4 load curves, and the time periods corresponding to these 4 load curves are January 3, 2023, January 4, 2023, January 5, 2023, and January 6, 2023, respectively. If the cluster meets the condition that the time periods corresponding to the load curves included in the cluster are continuous, then the cluster will not be processed.

[0042] Therefore, the distance between the load curves in the cluster obtained based on S330 is not only small, but also the time periods corresponding to the load curves in the cluster obtained based on S330 are continuous.

[0043] S340: Obtain the variance of the central load curve of the plurality of clusters of load curves.

[0044] In this embodiment, the central load curve of any cluster load curve is the average of the load curves in the cluster, that is, the load data at the jth position of the central load curve of the cluster load curve is the average load data at the jth position of the load curve in the cluster.

[0045] In this embodiment, the sum of the load data difference values ​​corresponding to all positions on the two central load curves is determined as the distance between the two central load curves. Those skilled in the art know that any variance calculation method in the prior art falls within the protection scope of the present invention.

[0046] S350: Integrate the plurality of clusters of load curves according to the variance to obtain a plurality of integrated groups of load curves.

[0047] In this embodiment, the variance of the central load curve of the plurality of clusters of load curves can represent the difference between the load curves included in different clusters of the plurality of clusters. As a preferred specific implementation, S350 includes:

[0048] S351: If the variance is greater than or equal to a preset variance threshold, proceed to S352.

[0049] In this embodiment, the preset variance threshold is an empirical value. If the variance is greater than or equal to the preset variance threshold, it is determined that the differences between the load curves included in different clusters among the plurality of clusters are relatively large.

[0050] S352, performing a first integration process on the plurality of clusters of load curves to obtain a plurality of groups of load curves after the first integration process; the first integration process includes: judging whether the distance between the central load curves of any two clusters in a corresponding time period is less than a preset distance threshold, if the distance between the central load curves of two clusters in a corresponding time period is less than the preset distance threshold, merging the two clusters into one cluster; repeating the judgment until the distance between the central load curves of any two clusters in a corresponding time period is greater than or equal to the preset distance threshold, and determining each cluster as each group.

[0051] Optionally, the preset distance threshold is an empirical value.

[0052] In this embodiment, if two clusters with corresponding continuous time periods are merged into one cluster, it is further determined whether the merged cluster needs to be further merged with other clusters until the distance between the central load curves of any two clusters with corresponding continuous time periods is greater than or equal to the preset distance threshold. Thus, among the multiple groups of load curves after the first integration process, the distance between the load curves in the same group of load curves is small, the similarity is high, and the distance between the load curves in different groups of load curves is large, and the similarity is low.

[0053] S353: Determine the plurality of groups of load curves that have undergone the first integration process as the plurality of groups of integrated load curves.

[0054] Based on S352-S353, this embodiment realizes the grouping of load curves corresponding to the target building when the variance of the central load curves of several clusters obtained by clustering the concentrated load curves is large. After grouping, the similarity of the load curves in the same group is high, and the similarity of the load curves in different groups is low.

[0055] As a preferred specific implementation, S351 further includes: if the variance is less than a preset variance threshold, entering S354.

[0056] In this embodiment, if the variance is smaller than a preset variance threshold, it is determined that the differences between the load curves included in different clusters among the plurality of clusters are small.

[0057] S354, performing a second integration process on the plurality of clusters of load curves to obtain a plurality of groups of load curves after the second integration process; the second integration process includes: making judgments according to the order of the corresponding time periods, and if the number of load curves included in a cluster is less than a preset number threshold, merging the cluster and the associated cluster of the cluster into one cluster; repeating the judgments until the number of load curves included in any cluster is greater than or equal to the preset number threshold, and determining each cluster as a group; the associated cluster of the cluster is a cluster whose time period corresponding to the cluster is continuous and whose central load curve is at a smaller distance from the central load curve of the cluster.

[0058] Optionally, the preset quantity threshold is an empirical value.

[0059] In this embodiment, if the merged cluster still does not meet the condition that the number of load curves in the cluster is greater than or equal to the preset number threshold, the merged cluster and its associated clusters will continue to be merged until the number of load curves included in any cluster is greater than or equal to the preset number threshold. Thus, the load curves of the several clusters can be quickly grouped based on the number of load curves in the cluster, and the number of load curves included in each group after grouping is greater than or equal to the preset number threshold. The preset number threshold determines the number of groups. When the preset number threshold is larger, the number of groups obtained in the end is smaller; when the preset number threshold is smaller, the number of groups obtained in the end is larger; different numbers of groups can be obtained by adjusting the preset number threshold, which is more flexible.

[0060] S355, determining the plurality of groups of load curves after the second integration process as the plurality of groups of integrated load curves.

[0061] Based on S354-S355, this embodiment realizes the rapid grouping of load curves corresponding to the target building when the variance of the central load curves of several clusters obtained by clustering the concentrated load curves is small. The distance between the load curves in the same group is small and the similarity is high. The number of groups can be adjusted by adjusting the preset number threshold, which has high flexibility.

[0062] S360: Determine the central load curve of each group of load curves as a reference load curve for a time period corresponding to the corresponding group.

[0063] Based on S310-S360, this embodiment obtains reference load curves of different time periods, and the distance between any reference load curve and any load curve in the corresponding group is small, and the similarity is high. For example, a group of load curves includes 7 load curves, and the time periods corresponding to these 7 load curves are June 10, 2023, June 11, 2023, June 12, 2023, June 13, 2023, June 14, 2023, June 15, 2023, and June 16, 2023, respectively. Then the time period corresponding to the group is June 10, 2023-June 16, 2023; the reference load curve and the corresponding time length of any load curve in the group are equal, for example, both are 24 hours.

[0064] S400, training the target neural network model using the historical data set of the building complex to obtain a trained target neural network model.

[0065] Those skilled in the art know that any training method in the prior art falls within the protection scope of the present invention. As a specific implementation, S400 includes:

[0066] S410, obtaining a load influencing factor data sequence of each building in the building complex at each sampling moment; the load influencing factor data sequence is a sequence composed of load influencing factor data.

[0067] S420, determine the load influencing factor data sequence of each building in the building complex at each sampling moment as a training sample, determine the load data of each building in the building complex at each sampling moment as the label of the corresponding training sample, and train the target neural network model.

[0068] This embodiment trains the target neural network model based on the load influencing factor data sequence and the corresponding load data of all buildings in the building complex at all sampling moments to obtain a trained target neural network model. The trained target neural network model has the function of acquiring the corresponding load data according to the input load influencing factor data sequence.

[0069] S500, using the trained target neural network model to predict the load data of the target building in the target time period, and obtain the predicted load curve of the target building in the target time period. 1 ; The target building is any building in the building complex.

[0070] As a specific implementation, S500 includes:

[0071] S510, obtaining load influencing factor data of a target building in a target time period.

[0072] S520, inputting the load influencing factor data of the target building in the target time period into the trained target neural network model.

[0073] In this embodiment, based on the load influencing factor data sequence of the target building at each moment in the target time period, the load data of the target building at the corresponding moment in the target time period can be obtained; based on the load data of all corresponding moments in the target time period, the predicted load curve of the target building in the target time period can be drawn. For example, the target time period is 24 hours, and the interval between any two adjacent moments in the target time period is 1 hour.

[0074] S600, obtaining a reference load curve of the target building in a time period matching the target time period 2 .

[0075] Optionally, the time period that matches the target time period refers to a time period in a different year but with the same month and date. For example, if the target time period is December 1, 2024, and the time period that matches the target time period is December 1, 2023, the reference load curve of the target building in the time period that matches the target time period is the reference load curve of the corresponding group including December 1, 2023.

[0076] S700, according to cur 1 and cur 2 Obtain the target load curve of the target building in the target time period.

[0077] In this embodiment, according to cur 2 Cur 1 As a preferred embodiment, S700 includes:

[0078] S710, get cur 1 and cur 2 The difference value of load data at the same position in .

[0079] S720: If the load data difference value at a certain location is greater than or equal to a preset difference value threshold, then 1The load data at that position in is replaced with the corresponding specified data; when cur 1 The load data at this position is greater than cur 2 When the load data at this position is in the 2 The sum of the load data at this location and the preset difference value threshold; when cur 1 The load data at this position is less than cur 2 When the load data at this position is in the 2 The difference between the load data at the position and a preset difference value threshold; the preset difference value threshold is greater than 0.

[0080] S730, the replaced cur 1 Determine the target load curve for the target building in the target time period.

[0081] In this embodiment, the target load curve of the target building in the target time period is output to relevant personnel of the building complex, so that the relevant personnel can adjust and optimize the demand load of the target building in the target time period according to the target load curve of the target building in the target time period.

[0082] This embodiment obtains a trained neural network model based on the historical data set of the building complex, and the trained neural network model has the function of inferring the corresponding load data according to the input load influencing factor data; since the historical data set of the building complex based on this embodiment is a data set that has not been processed by deleting the sudden increase and sudden decrease of load data, it is a relatively real data set, therefore, the load data of the target building in the target time period predicted based on the trained neural network model is also a relatively real load data, in which there may also be sudden increase or sudden decrease of load data; for the load curve of the target building in the target time period obtained based on the predicted load data, this embodiment corrects it according to the reference load curve of the target building in the corresponding same time period (obtained based on the data set processed by deleting the sudden increase and sudden decrease of load data, which can meet the low-carbon operation requirements of the building), and uses the corrected target load curve as the load adjustment reference load curve of the target building in the target time period to optimize the electricity demand of the target building in the target time period. The target load curve determined in this embodiment is obtained by adjusting the reference load curve of the same period in the historical time period on the basis of the predicted load curve, so that the target load curve can take into account the actual electricity demand and low-carbon operation requirements of the target building, and based on the target load curve, the optimization of the demand load data of the target building can be achieved.

[0083] Embodiment 2:

[0084] According to this embodiment, a building complex demand load tuning system based on multimodal data fusion is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the building complex demand load tuning method based on multimodal data fusion in Embodiment 1 is implemented.

[0085] Although some specific embodiments of the present invention have been described in detail by way of example, it will be appreciated by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present invention. It will also be appreciated by those skilled in the art that various modifications may be made to the embodiments without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.

Claims

1. A building complex demand load optimization method based on multimodal data fusion, characterized in that: The method comprises the following steps: S100, obtaining a historical data set of a building complex; the building complex includes at least two buildings, and the historical data set of each building includes load data and load influencing factor data of the corresponding building in a historical time period; S200, cleaning the historical data set of the building complex to obtain a cleaned historical data set; the cleaning process includes deleting the load data of sudden increase and decrease in the historical data set; S300, obtaining a reference load curve of each building in different time periods according to the cleaned historical data set; S400, training the target neural network model using the historical data set of the building complex to obtain a trained target neural network model; S500, using the trained target neural network model to predict the load data of the target building in the target time period, to obtain a predicted load curve cur1 of the target building in the target time period; the target building is any building in the building complex; S600, obtaining a reference load curve cur2 of the target building in a time period matching the target time period; S700, obtaining a target load curve of a target building in a target time period according to cur1 and cur2; S300 includes: S310, obtaining cleaned load data corresponding to the target building in the cleaned historical data set; S320, obtaining a load curve set corresponding to the target building according to the cleaned load data corresponding to the target building; the time periods corresponding to any load curve corresponding to the target building are equal in length; S330, clustering the load curves corresponding to the target building according to a preset clustering condition to obtain a plurality of clusters of load curves; the preset clustering condition includes: if the time period corresponding to a load curve is not continuous with the time period corresponding to any load curve in a cluster, then the load curve is not classified into the cluster; S340, obtaining the variance of the central load curve of the plurality of clusters of load curves; S350, integrating the plurality of clusters of load curves according to the variance to obtain a plurality of integrated groups of load curves; S360: Determine the central load curve of each group of load curves as a reference load curve for a time period corresponding to the corresponding group.

2. The building complex demand load optimization method based on multimodal data fusion according to claim 1 is characterized in that: S350 includes: S351, if the variance is greater than or equal to a preset variance threshold, proceed to S352; S352, performing a first integration process on the plurality of clusters of load curves to obtain a plurality of groups of load curves after the first integration process; the first integration process comprises: judging whether the distance between the central load curves of any two clusters in a corresponding time period is less than a preset distance threshold, if the distance between the central load curves of any two clusters in a corresponding time period is less than the preset distance threshold, merging the two clusters into one cluster; repeating the judgment until the distance between the central load curves of any two clusters in a corresponding time period is greater than or equal to the preset distance threshold, and determining each cluster as each group; S353: Determine the plurality of groups of load curves that have undergone the first integration process as the plurality of groups of integrated load curves.

3. The building complex demand load optimization method based on multimodal data fusion according to claim 2 is characterized in that: S351 also includes: if the variance is less than a preset variance threshold, entering S354; S354, performing a second integration process on the plurality of clusters of load curves to obtain a plurality of groups of load curves after the second integration process; the second integration process comprises: judging according to the order of the corresponding time periods, if the number of load curves included in a cluster is less than a preset number threshold, merging the cluster and the associated cluster of the cluster into one cluster; repeating the judgment until the number of load curves included in any cluster is greater than or equal to the preset number threshold, and determining each cluster as a group; the associated cluster of the cluster is a cluster whose time period corresponding to the cluster is continuous and whose central load curve is less distant from the central load curve of the cluster; S355, determining the plurality of groups of load curves after the second integration process as the plurality of groups of integrated load curves.

4. The building complex demand load optimization method based on multimodal data fusion according to claim 1 is characterized in that: S700 includes: S710, obtaining a load data difference value at the same position in cur1 and cur2; S720, if the load data difference value of a certain position is greater than or equal to a preset difference value threshold, the load data of the position in cur1 is replaced with the corresponding designated data; when the load data of the position in cur1 is greater than the load data of the position in cur2, the corresponding designated data is the sum of the load data of the position in cur2 and the preset difference value threshold; when the load data of the position in cur1 is less than the load data of the position in cur2, the corresponding designated data is the difference between the load data of the position in cur2 and the preset difference value threshold; the preset difference value threshold is greater than 0; S730, determining the replaced cur1 as the target load curve of the target building in the target time period.

5. The building complex demand load optimization method based on multimodal data fusion according to claim 1 is characterized in that: S500 includes: S510, obtaining load influencing factor data of a target building in a target time period; S520, inputting the load influencing factor data of the target building in the target time period into the trained target neural network model.

6. The building complex demand load optimization method based on multimodal data fusion according to claim 1 is characterized in that: S200 also includes a judgment process for the sudden increase and decrease of load data in the historical data set, and the judgment process includes: if the difference between the load data at a certain sampling moment and the load data at the previous sampling moment is greater than a preset difference value threshold, the load data at the sampling moment is judged as a sudden increase of load data; if the inverse of the difference between the load data at a certain sampling moment and the load data at the previous sampling moment is greater than the preset difference value threshold, the load data at the sampling moment is judged as a sudden decrease of load data.

7. The building complex demand load optimization method based on multimodal data fusion according to claim 1 is characterized in that: S400 includes: S410, obtaining a load influencing factor data sequence of each building in the building complex at each sampling moment; the load influencing factor data sequence is a sequence composed of load influencing factor data; S420, determine the load influencing factor data sequence of each building in the building complex at each sampling moment as a training sample, determine the load data of each building in the building complex at each sampling moment as the label of the corresponding training sample, and train the target neural network model.

8. A building complex demand load tuning system based on multimodal data fusion, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the building complex demand load optimization method based on multimodal data fusion as described in any one of claims 1 to 7 is implemented.

Images