A Deep Learning-Based Optimized System and Method for View Data Storage and Switching

Abstract

The present invention discloses an optimized system and method for view data storage and switching based on deep learning, which relates to the field of data storage and switching optimization. The present invention screens the views of the currently to-be-selected replacement reference view, and then replaces or fragments the current reference view again according to the usage effect after replacement, further ensuring the usage effect of the current reference view; by comparing the current usage effect parameters with the usage effect parameters before replacement, it can be determined whether the current reference view can improve the efficiency of switching and view updating; the difference data between the current usage effect parameters and the usage effect parameters before replacement is compared with the first effect difference threshold and the second effect difference threshold, and different comparison results are processed in different ways, improving the diversity of processing schemes, which is conducive to more precisely solving the problem of improving the efficiency of switching and view updating of the current reference view.

Description

Technical Field

[0001] The present invention belongs to the field of data storage and switching optimization. Specifically, it particularly relates to a view data storage and switching optimization system and method based on deep learning. Background Art

[0002] Three-dimensional visualization scenarios often involve the optimization of large amounts of data models. It is very common for the number of rendered triangular patches to be in the millions or tens of millions. For such large-data scenarios, if one wants to achieve the storage and switching of views through simple data storage, a large amount of memory is required to support it. Moreover, when switching views, due to the large amount of data and the large number of object states to be synchronized, there will be problems such as slow view switching speed, or even program freezing and browser crashes, bringing a very bad experience to users. Summary of the Invention

[0003] In view of the problems in the related art, the present invention proposes a view data storage and switching optimization system and method based on deep learning to overcome the above-mentioned technical problems existing in the existing related technologies.

[0004] To solve the above technical problems, the present invention is realized through the following technical solutions:

[0005] The present invention is a view data storage and switching optimization method based on deep learning, including the following steps:

[0006] S1. Collect the variation equations of the usage frequencies of multiple candidate scene views over time, the average object activity data at multiple time points, and the spatial coverage data to obtain the current usage frequency variation equation set, the current average object activity data set, and the current average spatial coverage data set;

[0007] S2. Collect the probability data of each candidate scene view being used as a reference view, the variation equations of the usage frequencies of each candidate scene view over time, the average object activity data at multiple time points, and the spatial coverage data, and construct a final view base probability mapping equation;

[0008] S3. Select the current reference view according to the final view base probability mapping equation, the current usage frequency variation equation set, the current average object activity data set, and the current average spatial coverage data set;

[0009] S4. Determine whether it is necessary to switch or fragment the current reference view by collecting the effect parameters after using the current reference view to obtain a determination result;

[0010] S5. Fragment the current reference view according to the determination result to obtain the final number of current view fragments;

[0011] This solution first screens the views of the current candidate replacement reference view based on historical data, ensuring the rationality of the selected current reference view based on historical data. After the selection, the current reference view is replaced or segmented again according to the usage effect after replacement, further ensuring the usage effect of the current reference view.

[0012] Preferably, S1 includes the following steps:

[0013] S11. Set the three-dimensional scene to be optimized and the corresponding several views to obtain a set of candidate scene views; then set several parameter types that affect the possibility of a view being used as a reference view to obtain a set of reference view selection influence parameter types , a 1. a 2. a 3 respectively represent the view usage frequency, object activity, and space coverage rate;

[0014] S12. Set the first current statistical period; in combination with the first current statistical period and the set of reference view selection influence parameter types, obtain the change equation of the usage frequency of each view in the set of candidate scene views over time, the average object activity data at multiple time points, and the space coverage rate data, to obtain the current usage frequency change equation set, the current average object activity data set, and the current average space coverage rate data set;

[0015] In this solution, by selecting several views of the three-dimensional scene to be optimized and setting several reference view selection influence parameters, it is used to subsequently select the most suitable view from the set of candidate scene views as the reference view, thereby reducing the data volume of the view increment when replacing the view subsequently, and thus improving the efficiency of view switching and the data processing volume of the server; for the view usage frequency, since the number of times the view is accessed at the most recent moment is more important, therefore, by collecting the change equation of the usage frequency of each candidate view over time, it is convenient to subsequently construct a more accurate view selection model.

[0016] Preferably, S2 includes the following steps:

[0017] S21. Set the historical statistical period; in combination with the set of candidate scene views and the set of reference view selection influence parameter types, collect the probability data of each candidate scene view being used as the reference view, the change equation of the usage frequency of each candidate scene view over time, the average object activity data at multiple time points, and the space coverage rate data during the historical statistical period, to obtain the historical candidate scene view as the base probability data set , the historical usage frequency change equation set , the historical average object activity data set and the historical average spatial coverage rate dataset ;

[0018] ; ;

[0019] ; ;

[0020] Among them, 、 、 、 respectively represent the base probability data of the historical candidate scenario views, the change equation of the historical usage frequency over time, the average object activity data at multiple historical time points, and the spatial coverage rate data of the i th candidate scenario view in the candidate scenario view set to be selected, represents the total number of candidate scenario views set;

[0021] S22. Construct a final view-based probability mapping equation and a final usage frequency weight change equation set by using the historical candidate scenario view-based probability data set, the historical usage frequency change equation set, the historical average object activity data set, and the historical average spatial coverage rate data set;

[0022] By constructing a final view-based probability mapping equation, a calculation tool for calculating the candidate probability of each candidate scenario view is provided for subsequent selection of the current reference view from the candidate scenario view set, and then the most suitable current reference view is selected; among them, by collecting the probability data of each candidate scenario view being used as the reference view, the change equation of the usage frequency of each candidate scenario view over time, the average object activity data at multiple time points, and the spatial coverage rate data during the historical statistical period, data support is provided for subsequent construction of the final view-based probability mapping equation.

[0023] Preferably, the S22 includes the following steps:

[0024] S221. Cooperate with the historical candidate scenario view-based probability data set, the historical usage frequency change equation set, the historical average object activity data set, and the historical average spatial coverage rate data set to construct an initial view-based probability mapping equation; as follows,

[0025] ;

[0026] Among them, is the dependent variable of the initial view-based probability mapping equation, representing the probability data of using the candidate scenario view as the reference view

[0027] ; Represents the mapping relationship of the initial view as the base probability mapping equation; and and and are all independent variables of the equation; respectively represent the equation of the change of the weight value corresponding to the historical usage frequency over time, the equation of the change of the historical usage frequency over time, the historical average object activity data, and the historical average space coverage data;

[0028] S222. Randomly generate corresponding usage frequency weight change equations for each historical usage frequency change equation in the historical usage frequency change equation set to obtain the usage frequency weight change equation set , represents the usage frequency weight change equation of the i th candidate scenario view in the candidate scenario view set to be selected;

[0029] Substitute each data in the historical usage frequency change equation set, the historical average object activity data set, the usage frequency weight change equation set, and the historical average space coverage data set into the initial view as the base probability mapping equation for mapping to obtain the historical candidate scenario view as the base probability initial mapping data set , b i represents the data obtained by substituting the i th data in the historical usage frequency change equation set, the historical average object activity data set, the usage frequency weight change equation set, and the historical average space coverage data set into the initial view as the base probability mapping equation for mapping;

[0030] S223. Set the mapping error threshold; calculate the error data between the historical candidate scenario view as the base probability initial mapping data set and the historical candidate scenario view as the base probability data set to obtain the initial mapping error data ; The calculation formula is as follows,

[0031] ;

[0032] When the initial mapping error data is less than the mapping error threshold, use the initial view as the base probability mapping equation as the final view as the base probability mapping equation; otherwise, adjust the initial view as the base probability mapping equation and each usage frequency weight change equation in the usage frequency weight change equation set until the initial mapping error data is less than the mapping error threshold;

[0033] By setting each independent variable of the initial view as the base probability mapping equation to the historical usage frequency change equation, the historical average object activity data, the historical average space coverage data, and the usage frequency weight change equation, and setting the dependent variable to the candidate scenario view as the base probability data, a mapping from the historical usage frequency change equation, the historical average object activity data, the historical average space coverage data, and the usage frequency weight change equation to the candidate scenario view as the base probability data is achieved; by substituting the collected relevant history into the constructed initial view as the base probability mapping equation, through preliminary mapping, the mapping accuracy of the initial view as the base probability mapping equation is detected; and then it is determined whether it is necessary to adjust the initial view as the base probability mapping equation.

[0034] Preferably, S3 includes the following steps:

[0035] S31. Set the current original reference view; substitute each data in the current usage frequency change equation set, the current average object activity data set, the current average space coverage data set, and the final usage frequency weight change equation set into the final view as the base probability mapping equation for mapping to obtain the current candidate scenario view as the base probability data set;

[0036] S32. When the candidate scenario view corresponding to the largest base probability data in the current candidate scenario view as the base probability data set is the same as the current original reference view, the current reference view is not replaced; otherwise, the candidate scenario view corresponding to the largest base probability data in the current candidate scenario view as the base probability data set is used as the current reference view, both are recorded as the current reference view;

[0037] Since the setting of the view usage weight is generally the same in other identical situations (such as the time difference between the call time and the time when it is set as the reference view), therefore, when performing probability data analysis on the current reference view from the candidate scenario view set, the final usage frequency weight change equation set is still used as the weight value corresponding to each current usage frequency change equation; by selecting the candidate scenario view corresponding to the largest base probability data as the current reference view, it follows the perspective of historical selection habits, and the most common reference view in historical selection habits means that the size of the view increment data when it is replaced is relatively reasonable, thus ensuring that after the current reference view is set, the resulting view increment is relatively reasonable, and to a certain extent, ensuring the efficiency of view switching.

[0038] Preferably, S4 includes the following steps:

[0039] S41. Set the second current statistical period; set a number of time points within the second current statistical period to obtain the current time point set; set a number of types of effect parameters that can reflect the effects after using the reference view to obtain the reference view effect parameter type set;

[0040] S42. In coordination with the reference view effect parameter type set and the current time point set, collect the effect parameters after using the current reference view to obtain the current usage effect parameter matrix ; as follows,

[0041] ;

[0042] where, represents the effect parameter of the i th type obtained by collecting at the j th current time point after using the current reference view, , respectively represent the total number of set effect parameter types and current time points; then collect the average effect parameters of a number of time points before using the current reference view to obtain the historical average effect parameter set , represents the mean value of the effect parameters of the j th type collected at a number of time points before using the current reference view; calculate the difference data between the historical average effect parameter set and each row of data in the current usage effect parameter matrix to obtain the current effect difference data set , represents the effect difference data between the historical average effect parameter set and the i th row of data in the current usage effect parameter matrix; the calculation formula is as follows,

[0043] ;

[0044] S43. Set the first effect difference threshold and the second effect difference threshold; when there is current effect difference data in the current effect difference data set that is less than the second effect difference threshold, switch the current reference view back to the current original reference view; when there is current effect difference data in the current effect difference data set that is greater than or equal to the second effect difference threshold and less than the first effect difference threshold, enter S5; otherwise, enter S44;

[0045] S44. Predict the usage effect parameters at future time points based on the current usage effect parameter matrix and using a BP neural network model to obtain a future usage effect parameter matrix; then calculate the difference data between each row of data in the historical average effect parameter set and the future usage effect parameter matrix to obtain a future effect difference data set; when there is future effect difference data in the future effect difference data set that is less than the second effect difference threshold, proceed to S5; otherwise, do nothing.

[0046] After selecting the view as the current reference view, it is necessary to detect and determine the usage effect brought by this current reference view in order to determine whether changing to the current reference view can improve the efficiency of switching and view updating; furthermore, it provides a basis for the next step of processing; by setting the reference view effect parameter type set, it provides a basis for collecting data for determining the usage effect of the current reference view in the future; then by comparing the current usage effect parameters with the usage effect parameters before replacement, it is possible to determine whether the current reference view can improve the efficiency of switching and view updating; among them, by setting the first effect difference threshold and the second effect difference threshold, the difference data between the current usage effect parameters and the usage effect parameters before replacement is compared with the first effect difference threshold and the second effect difference threshold, and different comparison results are processed in different ways, improving the diversity of the processing scheme, dealing with different situations specifically, which is conducive to more precisely solving the problem of whether the current reference view can improve the efficiency of switching and view updating; through prediction, the possible efficiency problems in the future can be discovered and solved in a timely manner.

[0047] Among them, when the current effect difference data is too small, it is solved by directly switching back to the current original reference view. Direct switching is more efficient, thus ensuring the normal progress of the 3D view change operation.

[0048] Preferably, S5 includes the following steps:

[0049] S51. Set the initial number of current view slices; use the initial number of current view slices to slice and store the current reference view data; after slicing and storing the current reference view data, collect the current effect parameters according to the reference view effect parameter type set to obtain a current sliced effect parameter set; calculate the difference data between the current sliced effect parameter set and the historical average effect parameter set to obtain the current sliced difference data.

[0050] S52. When the difference data after the current fragmentation is greater than or equal to the first effect difference threshold, use the initial current view fragmentation number as the final current view fragmentation number; otherwise, adjust the initial current view fragmentation number until the difference data after the current fragmentation is greater than or equal to the first effect difference threshold.

[0051] By adopting multi-reference sharding storage when the increment of a single reference view is too large, the data can be dispersed to multiple nodes or tables, reducing the data volume of a single node or table, thereby improving the query speed; if a certain shard fails, it will only affect the data of that shard, rather than the data of other shards, thus improving the fault tolerance of the system; different shards can be allocated different hardware resources according to actual needs, such as memory, CPU, etc., improving the utilization rate of hardware resources; since the data is distributed on multiple shards, the data volume of a single shard is relatively small, and the difficulty of maintenance and management is also relatively low, thereby reducing the maintenance cost of the system.

[0052] Preferably, the adjustment of the initial current view fragmentation number in S52 includes the following steps:

[0053] S521. Set the value range of the initial current view fragmentation number to obtain the current fragmentation number value range

[0054] , . respectively represent the lower limit and the upper limit of the value of the initial current view fragmentation number; construct a spotted hyena population for view fragmentation number adjustment; set the maximum number of iterations of the spotted hyena population for view fragmentation number adjustment to and the current number of iterations to , which are respectively recorded as the maximum number of iterations for shard adjustment and the current number of iterations for shard adjustment; the search space dimension of the spotted hyena population for view fragmentation number adjustment is one-dimensional.

[0055] S522. Set the initial position of each spotted hyena in the spotted hyena population for view fragmentation number adjustment according to the current fragmentation number value range to obtain the second initial position set , e i represents the initial position of the i th spotted hyena in the spotted hyena population for view fragmentation number adjustment, represents the scale of the spotted hyena population for view fragmentation number adjustment; e i The calculation formula of

[0056] is as follows,

[0057] In the formula, represents fore i A random number between 0 and 1 generated

[0058] S523. Construct a fitness function for the spotted hyena population that adjusts the number of view shards ; as follows,

[0059] ;

[0060] In the formula, represents using the current number of view shards obtained in each iteration process for the sharding operation of the current reference view and calculating the difference data between the effect parameters corresponding to the sharded current reference view and the historical average effect parameter set;

[0061] S524. Start the iteration. Before the iteration, set the current iteration number of the sharding adjustment to 1; in the first round of the iteration process, use the fitness function of the spotted hyena population that adjusts the number of view shards Calculate the fitness values of the initial positions of each spotted hyena in the second initial position set to obtain a third fitness value set; take the maximum fitness value in the third fitness value set and the corresponding initial position of the spotted hyena as the third global best fitness and the third global best position respectively; update the initial positions of each spotted hyena in the second initial position set according to the third global best fitness and the third global best position; after the update is completed, increment the current iteration number of the sharding adjustment by 1 and enter the next round of iteration;

[0062] In each subsequent round of the iteration process, use the fitness function of the spotted hyena population that adjusts the number of view shards Calculate the fitness values of the positions of each spotted hyena in the spotted hyena population that adjusts the number of view shards updated in the previous round of the iteration process to obtain a fourth fitness value set; take the maximum fitness value in the fourth fitness value set and the corresponding position of the spotted hyena as the fourth global best fitness and the fourth global best position respectively; update the positions of each spotted hyena in the spotted hyena population that adjusts the number of view shards updated in the previous round of the iteration process according to the fourth global best fitness and the fourth global best position; after the update is completed, increment the current iteration number of the sharding adjustment by 1 and enter the next round of iteration;

[0063] S525. When stop the iteration to obtain the second final global best fitness and the second final global best position; otherwise, continue the iteration until until; use the second final global best fitness as the optimized current sharded differential data; when the optimized current sharded differential data is greater than or equal to the first effect difference threshold, use the second final global best position to perform sharding processing and storage on the current reference view data to obtain the final current view shard count; otherwise, return to S524 to continue the iteration until the optimized current sharded differential data is greater than or equal to the first effect difference threshold;

[0064] By using the Spotted Hyena Optimization Algorithm to perform multiple iterative optimizations on the initial current view shard count, and using the differential data between the effect parameters after using the current reference view and the initial current view shard count as the fitness function; therefore, as the iteration progresses, the differential data between the effect parameters after using the current reference view and the initial current view shard count becomes larger and larger, that is, the effect after using the current reference view is getting better and better.

[0065] A view data storage and switching optimization system based on deep learning, including a base possibility influence parameter setting module, a current reference view selection influence parameter acquisition module, a historical reference view selection influence parameter acquisition module, a view base probability mapping equation construction module, a current reference view setting module, an effect parameter determination module, and a current reference view sharding module.

[0066] The present invention has the following beneficial effects:

[0067] 1. In the present invention, by screening the current view to be selected as the replacement reference view based on historical data, the rationality of the selected current reference view is ensured based on historical data; after the selection is completed, the current reference view is replaced or sharded again according to the usage effect after replacement, further ensuring the usage effect of the current reference view.

[0068] 2. In the present invention, by comparing the current usage effect parameters with the usage effect parameters before replacement, it can be determined whether the current reference view can improve the switching and view update efficiency; among them, by setting the first effect difference threshold and the second effect difference threshold, the differential data between the current usage effect parameters and the usage effect parameters before replacement is compared with the first effect difference threshold and the second effect difference threshold, and different processing methods are used for different comparison results, improving the diversity of the processing scheme, and performing targeted processing for different situations, which is beneficial to more accurately solve the problem of improving the switching and view update efficiency of the current reference view; through prediction, potential efficiency problems that may occur in the future can be discovered and solved in a timely manner.

[0069] 3. In the present invention, the initial view is used as the base probability mapping equation by adopting the spotted hyena optimization algorithm, and multiple constant coefficients of each usage frequency weight change equation in the usage frequency weight change equation set are iteratively adjusted multiple times, and the mapping accuracy of the base probability mapping equation with the initial view is used as the fitness function; therefore, as the iteration progresses, the mapping accuracy of the base probability mapping equation with the initial view becomes higher and higher, and finally meets the mapping requirements.

[0070] Of course, it is not necessary for any product implementing the present invention to simultaneously achieve all the above-mentioned advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0072] Figure 1 It is a schematic flow chart of a method for optimizing view data storage and switching based on deep learning according to the present invention;

[0073] Figure 2 It is a schematic flow chart of constructing the base probability mapping equation for the final view according to the present invention;

[0074] Figure 3 It is a schematic flow chart of switching or slicing the current reference view according to the present invention;

[0075] Figure 4 It is a schematic module diagram of a system for optimizing view data storage and switching based on deep learning according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0076] The technical solutions in the embodiments of the invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the invention. Obviously, the described embodiments are only a part of the embodiments of the invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the invention belong to the scope of protection of the invention.

[0077] Embodiment 1

[0078] Please refer to Figures 1-3 , this embodiment is a method for optimizing view data storage and switching based on deep learning, including the following steps:

[0079] S1. Collect the equations of the usage frequencies of multiple candidate scene views over time, the average object activity data at multiple time points, and the spatial coverage data, to obtain the current set of usage frequency change equations, the current set of average object activity data sets, and the current set of average spatial coverage data sets;

[0080] S1 includes the following steps:

[0081] S11. Set the three-dimensional scene to be optimized and the corresponding several views to obtain a set of candidate scene views; then set several types of parameters that affect the possibility of a view being used as a reference view to obtain a set of reference view selection influence parameter types , a 1. a 2. a 3 respectively represent the view usage frequency, object activity, and spatial coverage;

[0082] S12. Set the first current statistical period; in combination with the first current statistical period and the set of reference view selection influence parameter types, obtain the equations of the usage frequencies of each view in the set of candidate scene views over time, the average object activity data at multiple time points, and the spatial coverage data, to obtain the current set of usage frequency change equations, the current set of average object activity data sets, and the current set of average spatial coverage data sets; for the object activity data, it is necessary to determine which objects are frequently modified in the view, such as objects with many changes in attributes such as visibility and position; the more active objects there are, the more effective the view may be as a reference view because there will be less incremental data; for the spatial coverage, it is necessary to consider the spatial range covered by the camera view of the view and the three-dimensional bounding box coverage rate of the objects in the view; views with a high coverage rate may be more suitable as a reference because they can reduce the amount of data stored incrementally;

[0083] S2. Collect the probability data of each candidate scene view being used as a reference view, the equations of the usage frequencies of each candidate scene view over time, the average object activity data at multiple time points, and the spatial coverage data, and construct a final view-based probability mapping equation;

[0084] S2 includes the following steps:

[0085] S21. Set the historical statistical period; in combination with the set of candidate scene views and the set of reference view selection influence parameter types, collect the probability data of each candidate scene view being used as a reference view, the equations of the usage frequencies of each candidate scene view over time, the average object activity data at multiple time points, and the spatial coverage data within the historical statistical period, to obtain the historical candidate scene view-based probability data set , the historical set of usage frequency change equations , historical average object activity dataset and historical average space coverage dataset ;

[0086] ; ;

[0087] ; ;

[0088] Among them, , , , respectively represent the base probability data of the historical candidate scene views, the change equation of the historical usage frequency over time, the average object activity data at multiple historical time points, and the space coverage data of the i th candidate scene view in the candidate scene view set to be selected, represents the total number of candidate scene views to be set;

[0089] S22. Construct a final view base probability mapping equation and a final usage frequency weight change equation set using the historical candidate scene view base probability data set, the historical usage frequency change equation set, the historical average object activity data set, and the historical average space coverage data set;

[0090] The S22 includes the following steps:

[0091] S221. Cooperate with the historical candidate scene view base probability data set, the historical usage frequency change equation set, the historical average object activity data set, and the historical average space coverage data set to construct an initial view base probability mapping equation; as follows,

[0092] ;

[0093] Among them, is the dependent variable of the initial view base probability mapping equation, representing the probability data of taking the candidate scene view as the reference view

[0094] ; represents the mapping relationship of the initial view base probability mapping equation; such as a direct proportional relationship, an inverse proportional relationship, a relationship proportional to the product of the independent variables, or an inverse proportional relationship, etc. between the independent variables of the initial view base probability mapping equation; , , , are all independent variables of the equation; respectively represent the change equation of the weight value corresponding to the historical usage frequency over time, the change equation of the historical usage frequency over time, the historical average object activity data, and the historical average space coverage data;

[0095] S222. Randomly generate a corresponding usage frequency weight change equation for each historical usage frequency change equation in the historical usage frequency change equation set to obtain a usage frequency weight change equation set , denote the usage frequency weight change equation of the i th candidate scenario view in the candidate scenario view set to be selected;

[0096] Substitute the data of the historical usage frequency change equation set, the historical average object activity dataset, the usage frequency weight change equation set, and the historical average space coverage dataset into the initial view-based probability mapping equation respectively for mapping to obtain a historical candidate scenario view-based probability initial mapping dataset , b i denote the data obtained by substituting the i th data of the historical usage frequency change equation set, the historical average object activity dataset, the usage frequency weight change equation set, and the historical average space coverage dataset into the initial view-based probability mapping equation for mapping;

[0097] S223. Set a mapping error threshold; calculate the error data between the historical candidate scenario view-based probability initial mapping dataset and the historical candidate scenario view-based probability dataset to obtain initial mapping error data ; The calculation formula is as follows,

[0098] ;

[0099] When the initial mapping error data is less than the mapping error threshold, use the initial view-based probability mapping equation as the final view-based probability mapping equation; otherwise, adjust the initial view-based probability mapping equation and each usage frequency weight change equation in the usage frequency weight change equation set until the initial mapping error data is less than the mapping error threshold;

[0100] The adjustment of the initial view-based probability mapping equation and each usage frequency weight change equation in the usage frequency weight change equation set in S223 includes the following steps:

[0101] S2231. Set several constant coefficient value ranges for the initial view-based probability mapping equation and each usage frequency weight change equation in the usage frequency weight change equation set to obtain a first constant coefficient value range set and a second constant coefficient value range matrix ; Respectively as follows,

[0102] ;

[0103] ;

[0104] Among them, and respectively represent the lower limit and the upper limit of the value of the i th constant coefficient of the base probability mapping equation of the initial view; and respectively represent the lower limit and the upper limit of the value of the j th constant coefficient of the i th usage frequency weight change equation in the set of usage frequency weight change equations, and respectively represent the total number of constant coefficients of the base probability mapping equation of the initial view and each usage frequency weight change equation;

[0105] Construct a spotted hyena population for adjusting the base probability-frequency weight mapping constant coefficients; set the maximum number of iterations of the spotted hyena population for adjusting the base probability-frequency weight mapping constant coefficients to be c 1 and the current number of iterations to be c 2, which are respectively denoted as the maximum number of iterations for coefficient adjustment and the current number of iterations for coefficient adjustment; the number of search space dimensions of the spotted hyena population for adjusting the base probability-frequency weight mapping constant coefficients is the same as that of ;

[0106] S2232. According to the first constant coefficient value interval set and the second constant coefficient value interval matrix, set the initial positions of each spotted hyena in the spotted hyena population for adjusting the base probability-frequency weight mapping constant coefficients to obtain the first initial position matrix set , represents the initial position matrix of the k th spotted hyena in the spotted hyena population for adjusting the base probability-frequency weight mapping constant coefficients, represents the size of the spotted hyena population for adjusting the base probability-frequency weight mapping constant coefficients; As follows,

[0107] ;

[0108] Among them, and respectively represent the position components of the initial position of the k th spotted hyena in the spotted hyena population for adjusting the base probability-frequency weight mapping constant coefficients on the i th constant coefficient dimension of the base probability mapping equation of the initial view and the j th usage frequency weight change equation in the set of usage frequency weight change equations on the i th constant coefficient dimension; the calculation formulas are as follows respectively,

[0109] ;

[0110] ;

[0111] In the formula, rand k1i , rand k2ji respectively represent random numbers between 0 and 1 generated for , ;

[0112] S2233. Set the fitness function of the spotted hyena population by adjusting the base probability frequency weight mapping constant coefficient ; as follows,

[0113] ;

[0114] wherein, d Set the constant coefficient of the base probability mapping equation for a set of initial views obtained in each iteration process and a set of constant coefficients of each usage frequency weight change equation in the usage frequency weight change equation set, and substitute them into the base probability mapping equation of the initial view and the corresponding usage frequency weight change equation respectively. Then, combine the historical usage frequency change equation set, the historical average object activity dataset, and the historical average space coverage dataset and input them into the base probability mapping equation of the initial view to map the error between the data and the actual data;

[0115] S2234. Start the iteration. Before the iteration, set the current iteration number of the coefficient adjustment to 1; in the first iteration process, use the fitness function of the spotted hyena population adjusted by the base probability frequency weight mapping constant coefficient Calculate the fitness value of the initial position matrix of each spotted hyena in the first initial position matrix set to obtain the first fitness value set; take the maximum fitness in the first fitness value set and the initial position matrix of the corresponding spotted hyena as the first global best fitness and the first global best position respectively; update the initial position matrix of each spotted hyena in the first initial position matrix set according to the first global best fitness and the first global best position; after the update is completed, add 1 to the current iteration number of the coefficient adjustment and enter the next iteration;

[0116] In each subsequent iteration process, use the fitness function of the spotted hyena population adjusted by the base probability frequency weight mapping constant coefficient Calculate the fitness values of the position matrices of each spotted hyena in the spotted hyena population by adjusting the base probability frequency weight mapping constant coefficient obtained in the previous iteration process to obtain a second fitness value set; use the maximum fitness value in the second fitness value set and the corresponding position matrix of the spotted hyena as the second global best fitness and the second global best position respectively; update the position matrices of each spotted hyena in the spotted hyena population with the base probability frequency weight mapping constant coefficient adjusted in the previous iteration process according to the second global best fitness and the second global best position; after the update is completed, increment the current iteration count of the coefficient by 1 and enter the next iteration;

[0117] S2235. When is satisfied, stop the iteration to obtain the first final global best position and the first final global best fitness; otherwise, return to S2234 to continue the iteration until is satisfied; use the first final global best fitness as the optimized mapping error data; when the optimized mapping error data is less than the mapping error threshold, substitute each position component of the first final global best position into the initial view-based probability mapping equation and each usage frequency weight change equation in the usage frequency weight change equation set to obtain the final view-based probability mapping equation and the final usage frequency weight change equation set;

[0118] The spotted hyena optimization algorithm can quickly converge to the local optimal solution in the local search stage, improving the search efficiency of the algorithm; it can be extended to higher dimensions and shows good performance in dealing with high-dimensional optimization problems; it can converge to the optimal solution in a short time, improving the efficiency of the algorithm; it has fewer parameters, is easy to implement and adjust, reducing the complexity of the algorithm; it has strong robustness to the selection of the initial population and the change of parameters and can find the optimal solution under different initial conditions; based on the above advantages, the spotted hyena optimization algorithm is used in this solution to iteratively adjust multiple constant coefficients of the initial view-based probability mapping equation and each usage frequency weight change equation in the usage frequency weight change equation set, and use the mapping accuracy of the initial view-based probability mapping equation as the fitness function; therefore, as the iteration progresses, the mapping accuracy of the initial view-based probability mapping equation becomes higher and higher, finally meeting the mapping requirements;

[0119] S3. Select the current reference view according to the final view-based probability mapping equation, the current usage frequency change equation set, the current average object activity dataset, and the current average space coverage dataset;

[0120] S3 includes the following steps:

[0121] S31, setting the current original reference view; substituting each data in the current usage frequency change equation set, the current average object activity data set, the current average space coverage rate data set, and the final usage frequency weight change equation set into the final view-based probability mapping equation for mapping, to obtain the current selected scene view-based probability data set;

[0122] S32, when the scene view to be selected corresponding to the largest base probability data in the current scene view to be selected base probability data set is the same as the current original reference view, the current reference view is not replaced; otherwise, the scene view to be selected corresponding to the largest base probability data in the current scene view to be selected base probability data set is used as the current reference view, and both are recorded as the current reference view;

[0123] S4, determining whether it is necessary to switch or slice the current reference view by collecting effect parameters after using the current reference view, and obtaining a determination result;

[0124] The S4 comprises the following steps:

[0125] S41, setting a second current statistical period; setting a number of time points in the second current statistical period to obtain a current time point set; setting a number of types of effect parameters that can reflect the effect after using the reference view to obtain a reference view effect parameter type set; the reference view effect parameter type set includes memory usage, switching delay, and incremental data volume, etc.;

[0126] S42: Collect effect parameters after using the current reference view in conjunction with the reference view effect parameter type set and the current time point set to obtain a currently used effect parameter matrix. ;as follows,

[0127] ;

[0128] in, Indicated in i The first j types of effect parameters after using the current reference view, , respectively represent the set effect parameter type and the total number of the current time point; then collect the average effect parameters of several time points before the current reference view is used to obtain the historical average effect parameter set , Indicates the first time point before the current reference view is used for acquisition j Calculate the difference between the historical average effect parameter set and each row of data in the currently used effect parameter matrix to obtain the current effect difference data set , represents the effect difference data between the historical average effect parameter set and the data in the i th row of the current usage effect parameter matrix; the calculation formula is as follows,

[0129] ;

[0130] S43. Set the first effect difference threshold and the second effect difference threshold; when there is current effect difference data in the current effect difference data set that is less than the second effect difference threshold, switch the current reference view back to the current original reference view again; when there is current effect difference data in the current effect difference data set that is greater than or equal to the second effect difference threshold and less than the first effect difference threshold, go to S5; otherwise, go to S44;

[0131] S44. Predict the usage effect parameters at future time points based on the current usage effect parameter matrix and using a BP neural network model to obtain a future usage effect parameter matrix; then calculate the difference data between the historical average effect parameter set and the data in each row of the future usage effect parameter matrix to obtain a future effect difference data set; when there is future effect difference data in the future effect difference data set that is less than the second effect difference threshold, go to S5; otherwise, do not process;

[0132] S5. Perform slicing processing on the current reference view according to the determination result to obtain the final number of current view slices;

[0133] The S5 includes the following steps:

[0134] S51. Set the initial number of current view slices; perform slicing processing and storage on the current reference view data using the initial number of current view slices; after completing the slicing processing and storage of the current reference view data, collect the current effect parameters according to the reference view effect parameter type set to obtain a current sliced effect parameter set; calculate the difference data between the current sliced effect parameter set and the historical average effect parameter set to obtain the current sliced difference data;

[0135] S52. When the current sliced difference data is greater than or equal to the first effect difference threshold, use the initial number of current view slices as the final number of current view slices; otherwise, adjust the initial number of current view slices until the current sliced difference data is greater than or equal to the first effect difference threshold;

[0136] The adjustment of the initial number of current view slices in S52 includes the following steps:

[0137] S521. Set the value range of the initial current view shard count to obtain the current shard count value range

[0138] , 、 respectively represent the lower limit and the upper limit of the value of the initial current view shard count; construct a spotted hyena population for adjusting the view shard count; set the maximum number of iterations of the spotted hyena population for adjusting the view shard count to and the current iteration count to , which are respectively denoted as the maximum iteration count for shard adjustment and the current iteration count for shard adjustment; the search space dimension of the spotted hyena population for adjusting the view shard count is one-dimensional;

[0139] S522. Set the initial position of each spotted hyena in the spotted hyena population for adjusting the view shard count according to the current shard count value range to obtain the second initial position set , e i represents the initial position of the i th spotted hyena in the spotted hyena population for adjusting the view shard count, represents the size of the spotted hyena population for adjusting the view shard count; e i The calculation formula of is as follows,

[0140] ;

[0141] In the formula, represents a random number between 0 and 1 generated for e i ;

[0142] S523. Construct the fitness function of the spotted hyena population for adjusting the view shard count ; as follows,

[0143] ;

[0144] In the formula, represents calculating the difference data between the effect parameter corresponding to the current benchmark view after sharding and the historical average effect parameter set when using the current view shard count obtained in each iteration for sharding the current benchmark view;

[0145] S524. Start iteration. Before iteration, set the current iteration count for shard adjustment to 1; in the first round of iteration, use the fitness function of the spotted hyena population for adjusting the view shard count Calculate the fitness value of each spotted hyena's initial position in the second initial position set to obtain a third fitness value set; use the maximum fitness value in the third fitness value set and the corresponding initial position of the spotted hyena as the third global optimal fitness and the third global optimal position respectively; update the initial position of each spotted hyena in the second initial position set according to the third global optimal fitness and the third global optimal position; after the update is completed, adjust the current iteration number of the slice by 1 and enter the next round of iteration;

[0146] In each other iteration, the number of view fragments is used to adjust the fitness function of the spotted hyena population. Calculate the fitness value of the position of each spotted hyena in the spotted hyena population adjusted by the number of view slices updated in the previous iteration process to obtain a fourth fitness value set; use the maximum fitness value in the fourth fitness value set and the corresponding position of the spotted hyena as the fourth global optimal fitness and the fourth global optimal position respectively; update the position of each spotted hyena in the spotted hyena population adjusted by the number of view slices updated in the previous iteration process according to the fourth global optimal fitness and the fourth global optimal position; after the update is completed, add 1 to the current iteration number of the slice adjustment and enter the next iteration;

[0147] S525, when When , stop the iteration and get the second final global best fitness and the second final global best position; otherwise, continue to iterate until until the time; use the second final global best fitness as the difference data of the optimized current slice; when the difference data of the optimized current slice is greater than or equal to the first effect difference threshold, use the second final global best position to slice and store the current reference view data to obtain the final current view slice number; otherwise, return to S524 and continue to iterate until the difference data of the optimized current slice is greater than or equal to the first effect difference threshold.

[0148] Embodiment 2

[0149] See also Figure 4 This embodiment discloses a view data storage and switching optimization system based on deep learning, which can implement the method of the above embodiment, including a base possibility influence parameter setting module, a current reference view selection influence parameter acquisition module, a historical reference view selection influence parameter acquisition module, a view base probability mapping equation construction module, a current reference view setting module, an effect parameter determination module and a current reference view slicing module;

[0150] The base possibility influence parameter setting module sets several types of parameters that affect the possibility of a view being used as a reference view, and obtains a set of reference view selection influence parameter types;

[0151] The current reference view selection influence parameter acquisition module acquires the equations of the usage frequencies of multiple candidate scenario views over time, the average object activity data at multiple time points, and the spatial coverage data according to the set of reference view selection influence parameter types, and obtains a set of current usage frequency change equations, a set of current average object activity data, and a set of current average spatial coverage data;

[0152] The historical reference view selection influence parameter acquisition module acquires the probability data of each candidate scenario view being used as a reference view, the equations of the usage frequencies of each candidate scenario view over time, the average object activity data at multiple time points, and the spatial coverage data according to the set of reference view selection influence parameter types, and obtains a set of historical candidate scenario view base probabilities, a set of historical usage frequency change equations, a set of historical average object activity data, and a set of historical average spatial coverage data;

[0153] The view base probability mapping equation construction module constructs a final view base probability mapping equation using the set of historical candidate scenario view base probabilities, the set of historical usage frequency change equations, the set of historical average object activity data, and the set of historical average spatial coverage data;

[0154] The current reference view setting module selects the current reference view according to the final view base probability mapping equation, the set of current usage frequency change equations, the set of current average object activity data, and the set of current average spatial coverage data;

[0155] The effect parameter determination module determines whether it is necessary to switch or slice the current reference view by collecting the effect parameters after using the current reference view, and obtains a determination result;

[0156] The current reference view slicing module slices the current reference view according to the determination result, and obtains the final number of slices of the current view.

[0157] In the description of this specification, the descriptions referring to terms such as "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0158] The preferred embodiments of the invention disclosed above are only used to help illustrate the invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification to better explain the principle and practical application of the invention, so that those skilled in the art can well understand and utilize the invention.

Claims

1. A view data storage and switching optimization method based on deep learning, characterized in that: The following steps are involved: S1, collecting the usage frequency change equations of multiple scene views to be selected over time, the average object activity data and the space coverage data at multiple time points, and obtaining the current usage frequency change equation set, the current average object activity data set and the current average space coverage data set; S2, collecting probability data of each candidate scene view being used as a reference view, the time-varying equation of the usage frequency of each candidate scene view, the average object activity data at multiple time points, and the spatial coverage data, and constructing the final view-based probability mapping equation; S3, setting the current original reference view; Substituting each data in the current usage frequency change equation set, the current average object activity data set, the current average space coverage data set, and the final usage frequency weight change equation set into the final view-based probability mapping equation for mapping, to obtain the current scene view-based probability data set for selection; When the selected scene view corresponding to the largest base probability data in the current selected scene view base probability data set is the same as the current original reference view, the current reference view is not replaced; otherwise, the selected scene view corresponding to the largest base probability data in the current selected scene view base probability data set is used as the current reference view, and both are recorded as the current reference view; S4, determining whether it is necessary to switch or slice the current reference view by collecting effect parameters after using the current reference view, and obtaining a determination result; S5. Slice the current reference view according to the determination result to obtain a final number of slices of the current view.

2. The method for storing and optimizing view data based on deep learning according to claim 1, characterized in that: The S1 comprises the following steps: S11, setting a 3D scene to be optimized and a number of corresponding views to obtain a set of scene views to be selected; and then setting a number of parameter types that affect the possibility of a view being used as a reference view to obtain a set of parameter types that affect the selection of a reference view; S12. Set a first current statistical period; select an influencing parameter type set in conjunction with the first current statistical period and the baseline view, obtain an equation for how the usage frequency of each view in the selected scene view set changes over time, average object activity data at multiple time points, and spatial coverage data, and obtain a current usage frequency change equation set, a current average object activity data set, and a current average spatial coverage data set.

3. The method for storing and optimizing view data based on deep learning according to claim 2, characterized in that: The S2 comprises the following steps: S21, setting a historical statistical period; in conjunction with the candidate scene view set and the reference view selection influencing parameter type set, collecting probability data of each candidate scene view being used as a reference view, a usage frequency variation equation of each candidate scene view over time, average object activity data and space coverage data at multiple time points in the historical statistical period, and obtaining a historical candidate scene view as a base probability data set, a historical usage frequency variation equation set, a historical average object activity data set and a historical average space coverage data set; S22, using the historical candidate scene views as a base probability data set, the historical usage frequency change equation set, the historical average object activity data set and the historical average space coverage rate data set to construct a final view-based probability mapping equation and a final usage frequency weight change equation set.

4. The method for storing and optimizing view data based on deep learning according to claim 3, characterized in that: In S22, the final view-based probability mapping equation and the final set of frequency weight change equations are constructed using the spotted hyena optimization algorithm.

5. The method for storing and switching view data based on deep learning according to claim 4, characterized in that: The S4 comprises the following steps: S41, setting a second current statistical period; setting a number of time points in the second current statistical period to obtain a current time point set; setting a number of types of effect parameters that can reflect the effect after using the reference view to obtain a reference view effect parameter type set; S42, collecting effect parameters after using the current reference view in combination with the reference view effect parameter type set and the current time point set to obtain a currently used effect parameter matrix; then collecting average effect parameters at several time points before using the current reference view to obtain a historical average effect parameter set; calculating difference data between the historical average effect parameter set and each row of data in the currently used effect parameter matrix to obtain a current effect difference data set; S43, setting a first effect difference threshold and a second effect difference threshold; when the current effect difference data set contains current effect difference data less than the second effect difference threshold, switching the current reference view back to the current original reference view; when the current effect difference data set contains current effect difference data greater than or equal to the second effect difference threshold and less than the first effect difference threshold, entering S5; otherwise, entering S44; S44. Predict the usage effect parameters at future time points according to the current usage effect parameter matrix and adopt the BP neural network model to obtain the future usage effect parameter matrix; then calculate the difference data between the historical average effect parameter set and each row of data in the future usage effect parameter matrix to obtain a future effect difference data set; when the future effect difference data set has future effect difference data that is less than the second effect difference threshold, enter S5; otherwise, do not process.

6. The method for storing and optimizing view data based on deep learning according to claim 5, characterized in that: The S5 comprises the following steps: S51, setting an initial current view slice number; using the initial current view slice number to slice and store the current reference view data; after the current reference view data is sliced ​​and stored, collecting the current effect parameters according to the reference view effect parameter type set to obtain a current sliced ​​effect parameter set; calculating the difference data between the current sliced ​​effect parameter set and the historical average effect parameter set to obtain the current sliced ​​difference data; S52. When the difference data after the current slice is greater than or equal to the first effect difference threshold, the initial current view slice number is used as the final current view slice number; otherwise, the initial current view slice number is adjusted until the difference data after the current slice is greater than or equal to the first effect difference threshold.

7. The method for storing and optimizing view data based on deep learning according to claim 6, characterized in that: The step of adjusting the initial number of slices of the current view in S52 includes the following steps: S521, setting the value interval of the initial current view slice number to obtain the value interval of the current slice number; constructing the view slice number to adjust the spotted hyena population; setting the maximum number of iterations of the view slice number to adjust the spotted hyena population is And the current number of iterations is , respectively recorded as the maximum number of iterations for shard adjustment and the current number of iterations for shard adjustment; S522, adjusting the initial position of each spotted hyena in the spotted hyena population by setting the number of view slices according to the current slice number value interval, to obtain a second initial position set; S523, constructing a fitness function for adjusting the spotted hyena population by the number of view fragments; S524, start iteration, and before iteration, set the current iteration number of the slice adjustment to 1; in each round of iteration, use the view slice number to adjust the fitness function of the spotted hyena population to calculate the fitness value of the position of each spotted hyena in the view slice number adjustment spotted hyena population updated in the previous round of iteration, and update the position of each spotted hyena in the view slice number adjustment spotted hyena population updated in the previous round of iteration; S525, when When , stop the iteration and get the second final global best fitness and the second final global best position; otherwise, continue to iterate until until the time; use the second final global best fitness as the difference data of the optimized current slice; when the difference data of the optimized current slice is greater than or equal to the first effect difference threshold, use the second final global best position to slice and store the current reference view data to obtain the final current view slice number; otherwise, return to S524 and continue to iterate until the difference data of the optimized current slice is greater than or equal to the first effect difference threshold.

8. A system for implementing a view data storage and switching optimization method based on deep learning as described in any one of claims 1 to 7.

Images