Renewable resource recycling management method and system based on big data

Abstract

The present invention discloses a renewable resource recycling management method and system based on big data, including data collection, data anomaly identification, establishment of market demand forecasting model, model parameter training and comprehensive management. The present invention relates to the field of renewable resource recycling technology, and specifically refers to a renewable resource recycling management method and system based on big data. The scheme improves the accuracy of anomaly identification and reduces the misjudgment rate by defining a gravity weight function, setting a neighborhood width and defining a local density to remove abnormal data; establishes a model by designing an excitation function, an initial adjustment layer, a memory control layer, an attention layer, a correction layer and a loss function to comprehensively capture the features in the data, and improves the accuracy, accuracy and stability of the prediction; by defining a dynamic matching factor and a training origin position, designing a parameter training strategy and a parameter training rule, the flexibility and robustness of the training process are enhanced, making the training process more efficient and comprehensive.

Description

Technical Field

[0001] The present invention relates to the field of renewable resource recycling technology, and specifically to a renewable resource recycling management method and system based on big data. Background Art

[0002] With the development of social economy and the continuous growth of resource demand, the recycling of renewable resources has become increasingly important. The recycling management method and system of renewable resources based on big data use big data technology and artificial intelligence technology to manage the recycling process of renewable resources, thereby improving the efficiency and benefits of renewable resource recycling. However, the data anomaly identification in traditional renewable resource recycling management methods has the problems of low accuracy and easy misjudgment; the traditional market demand forecasting method has the problems of low prediction accuracy, weak model adaptability, and difficulty in capturing dependencies in time series data; the traditional model parameter training method has the problems of low training efficiency, easy to fall into local optimal solutions, and difficulty in finding the global optimal parameter configuration. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a renewable resource recycling management method and system based on big data. In view of the problems that data anomaly identification in traditional renewable resource recycling management methods is not accurate and easy to misjudge, this solution removes abnormal data by defining a gravity weight function, setting a neighborhood width, defining a local density and calculating anomaly scores, thereby improving the accuracy of anomaly identification and reducing the misjudgment rate; in view of the problems that traditional market demand forecasting methods have low prediction accuracy, weak model adaptability and difficulty in capturing dependencies in time series data, this solution establishes a model by constructing a model, designing an excitation function, designing an initial adjustment layer, designing a memory control layer, designing an attention layer, designing a correction layer, designing an output layer and designing a loss function, thereby more comprehensively capturing the features and information in the data and improving the accuracy, accuracy and stability of the prediction; in view of the problems that traditional model parameter training methods have low training efficiency, are prone to falling into local optimal solutions and are difficult to find global optimal parameter configurations, this solution trains model parameters by initialization, defining dynamic matching factors, defining training origin positions, designing parameter training strategies and designing parameter training rules, thereby enhancing the flexibility and robustness of the training process and making the training process more efficient and comprehensive.

[0004] The technical solution adopted by the present invention is as follows: a renewable resource recycling management method based on big data, the method comprising the following steps:

[0005] Step S1: data collection;

[0006] Step S2: data anomaly identification;

[0007] Step S3: Establishing a market demand forecasting model;

[0008] Step S4: model parameter training;

[0009] Step S5: Comprehensive management.

[0010] Furthermore, in step S1, the data collection is to collect historical renewable resource transaction data, production data, population data, substitute development data and market demand level; the historical renewable resource transaction data includes historical renewable resource transaction volume, transaction price and transaction time series data; the production data refers to the output of products using renewable resources as raw materials; the population data refers to population size, population structure and population mobility; the substitute development data refers to the development trend of substitute products of renewable resources, including three development trend levels of accelerated development, stable development and decelerated development; the market demand level includes strong demand, medium demand and low demand.

[0011] Furthermore, in step S2, the data anomaly identification specifically includes the following steps:

[0012] Step S21: define the gravity weight function, which is expressed as follows:

[0013] ;

[0014] in, represents the Euclidean distance between the data for calculating the gravity weight, represents the gravitational weight function, e represents the natural constant, represents the weight adjustment coefficient;

[0015] Step S22: Set the neighborhood width, calculate the Euclidean distance between all data points, select the two data points with the farthest distance, set the Euclidean distance between them as the domain width of the data, and set the neighborhood width of the data, which is expressed as follows:

[0016] ;

[0017] in, represents the neighborhood width of the data, Indicates the domain width of the data, Indicates the total number of data;

[0018] Step S23: define local density, take the data point as the center, set the area within the neighborhood width of the data point as the neighborhood of the data point, and calculate the local density, which is expressed as follows:

[0019] ;

[0020] Where m represents the index of the data point, represents the local density of the mth data point, represents the set of all data indexes within the neighborhood of the mth data point, Represents the index of the data in the neighborhood of the mth data point, Indicates that it belongs to the symbol, represents the mth data point and the The Euclidean distance between data points, div represents the dimension of the data;

[0021] Step S24: Calculate the anomaly score, and set the inverse of the local density of the data as the anomaly score, which is expressed as follows:

[0022] ;

[0023] in, Represents the abnormal score of the mth data;

[0024] Step S25: Remove abnormal data, calculate the abnormal scores of all data points, set the 1% of data points with the highest abnormal scores as abnormal data, and remove the abnormal data.

[0025] Furthermore, in step S3, the establishment of the market demand forecasting model specifically includes the following steps:

[0026] Step S31: construct a model, the market demand forecasting model includes an initial adjustment layer, a memory control layer, an attention layer, a correction layer and an output layer; at the same time, the market demand level is set as the label data of the model;

[0027] Step S32: Design an activation function, which is expressed as follows:

[0028] ;

[0029] in, represents the enhanced control activation function, represents the input of the reinforcement control activation function, represents the mean fitness, It means square root. represents the incentive weight, represents the linear rectification function, represents the excitation threshold;

[0030] Step S33: Design the initial adjustment layer, as shown below:

[0031] ;

[0032] in, represents the output value of the initial adjustment layer, p represents the layer index of the initial adjustment layer, and Represent the weight and bias of the initial adjustment layer of the pth layer, Represents the input value of the initial adjustment layer of the pth layer;

[0033] Step S34: Design a memory control layer, which is composed of an input control unit, an abandonment control unit, an update control unit, and an output control unit, and specifically includes the following contents:

[0034] Step S341: Design an input control unit, as shown below:

[0035] ;

[0036] Among them, t represents the index of the time point, represents the output value of the input control unit at time t, represents the sigmoid function, and denote the weight and bias of the input control unit, respectively. represents the control state of the memory control layer at time t-1, represents the input value of the input control unit at time t;

[0037] Step S342: Design an abandoned control unit, as shown below:

[0038] ;

[0039] in, represents the output value of the abandoned control unit at time t, and denote the weight and bias of the abandoned control unit, respectively. represents the input value of the abandoned control unit at time t, Indicates taking the L2 norm;

[0040] Step S343: Design and update the control unit, as shown below:

[0041] ;

[0042] in, represents the output value of the updated control unit at time t, represents the hyperbolic tangent activation function, and Represent the updated control weight and updated control bias respectively, represents the input value of the updated control unit at time t, represents element-wise multiplication operation, represents the control state of the memory control layer at time t-2, represents the input value of the updated control unit at time t-1;

[0043] Step S344: Design an output control unit, as shown below:

[0044] ;

[0045] in, represents the output value of the output control unit at time t, and denote the weight and bias of the output control unit, respectively. Represents the input value of the output control unit at time t;

[0046] Step S345: Update the memory control layer control state, as shown below:

[0047] ;

[0048] in, represents the control state of the memory control layer at time t, represents the input value of the output control unit at time t-1;

[0049] Step S35: Design the attention layer, calculate the attention weight of each time point, and assign the attention weight to each time point by multiplication, as shown below:

[0050] ;

[0051] in, represents the attention weight at time t, represents the normalized exponential function, U represents the basic attention weight, and Represent the weight and bias of the attention layer respectively;

[0052] Step S36: Design the correction layer, as shown below:

[0053] ;

[0054] in, represents the output of the correction layer, and Represent the weight and bias of the second layer of the correction layer, respectively. and Represent the weight and bias of the first layer of the correction layer, respectively. Represents the base input value of the first layer of the correction layer;

[0055] Step S37: Design the output layer, which is expressed as follows:

[0056] ;

[0057] in, Represents the output result of the model output layer, , and Represent the weight, input value and bias of the output layer respectively;

[0058] Step S38: Design a loss function, which is expressed as follows:

[0059] ;

[0060] in, represents the loss value of the model, j represents the index of the data for training the model, W represents the total number of data for training the model, represents the true value of the jth data, represents the model prediction result of the jth data, represents the scaling factor, Indicates taking the absolute value, and represents the loss weight, represents logarithmic operation, represents the predicted distribution of the data, represents the true distribution of the data, It represents the KL divergence between the predicted distribution and the true distribution of the data.

[0061] Furthermore, in step S4, the model parameter training specifically includes the following steps:

[0062] Step S41: Initialization, determine the training parameters of the model, including the adaptive mean, the excitation threshold, the weight and bias parameters of the model, set the opposite of the loss value of the model as the parameter performance value, create a parameter training space, and generate Initial parameter training points, calculate the average position of all initial parameter training points in the parameter training space, calculate the parameter performance values ​​corresponding to all parameter training points, and select the training points with the highest and lowest parameter performance values;

[0063] Step S42: Define a dynamic matching factor, which is expressed as follows:

[0064] ;

[0065] Among them, k represents the number of parameter training times, represents the dynamic matching factor during the k-th parameter training, represents the average position of the initial parameter training points, Indicates the parameter position with the highest current parameter performance. Indicates the parameter position with the lowest current parameter performance. It means taking the maximum value of the absolute value of the difference between the coordinates of all dimensions of the two points. The index representing the number of times the parameter is trained. Indicates the parameter position during the fqth parameter training, Indicates the Euclidean distance. Indicates the maximum number of parameter training times;

[0066] Step S43: define the training origin position, as shown below:

[0067] ;

[0068] in, represents the origin position generated during the k-th parameter training, Represents a random number between 0 and 1. represents the upper bound of the parameter training space, represents the lower bound of the parameter training space, Represents a random number that follows a standard normal distribution;

[0069] Step S44: Design parameter training strategy, as shown below:

[0070] ;

[0071] in, represents the parameter position during the kth parameter training, represents the parameter position during the k-1th parameter training, Represents a random number that follows a standard normal distribution;

[0072] Step S45: Design parameter training rules, set the maximum number of parameter training times and parameter performance threshold, and perform parameter training. Each time parameter training is performed, the parameter performance values ​​of the parameter points obtained through training are calculated, and the parameter positions with the highest parameter performance and the parameter positions with the lowest parameter performance are updated. If during a training process, the performance values ​​of more than 65% of the parameter points are greater than the parameter performance threshold, the training is terminated, and the parameter with the highest parameter performance at this time is set as the model parameter; if the maximum number of parameter training times is reached, perform parameter training again; otherwise, the number of training times is increased by one, and training continues.

[0073] Furthermore, in step S5, the comprehensive management is to predict the demand for renewable resources using a market demand forecasting model, and manage the recycling of renewable resources according to the market demand. When the market demand level is strong demand, the recycling of renewable resources is increased; when the market demand level is medium demand, the recycling of renewable resources is maintained; when the market demand level is low demand, the recycling of renewable resources is reduced.

[0074] The renewable resource recycling management system based on big data provided by the present invention includes a data acquisition module, a data anomaly identification module, a market demand forecasting model establishment module, a model parameter training module and a comprehensive management module;

[0075] The data collection module collects historical renewable resource transaction data, production data, population data, substitute development data and market demand level, and sends the data to the data anomaly identification module;

[0076] The data anomaly identification module receives the data sent by the data acquisition module, removes the abnormal data, and sends the data to the market demand forecasting model establishment module;

[0077] The module for establishing a market demand forecasting model receives data sent by the data anomaly identification module, establishes a market demand forecasting model, and sends the data to the model parameter training module and the comprehensive management module;

[0078] The model parameter training module receives the data sent by the market demand forecasting model establishment module, trains the parameters of the model, and sends the data to the comprehensive management module;

[0079] The comprehensive management module receives data sent by the market demand forecasting model building module and the model parameter training module, uses the market demand forecasting model to forecast the demand for renewable resources, and manages the recycling of renewable resources according to the market demand.

[0080] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0081] (1) In order to solve the problem of low accuracy and easy misjudgment in data anomaly identification in traditional renewable resource recycling management methods, this scheme removes abnormal data by defining a gravity weight function, setting the neighborhood width, defining the local density and calculating the anomaly score, thereby improving the accuracy of anomaly identification and reducing the misjudgment rate.

[0082] (2) In view of the problems of low prediction accuracy, weak model adaptability and difficulty in capturing dependencies in time series data in traditional market demand forecasting methods, this solution builds a model by constructing a model, designing an incentive function, designing an initial adjustment layer, designing a memory control layer, designing an attention layer, designing a correction layer, designing an output layer and designing a loss function, so as to more comprehensively capture the features and information in the data and improve the prediction accuracy, precision and stability.

[0083] (3) In view of the problems of low training efficiency, easy to fall into local optimal solutions, and difficulty in finding the global optimal parameter configuration in traditional model parameter training methods, this scheme trains model parameters by initialization, definition of dynamic matching factors, definition of training origin locations, design of parameter training strategies, and design of parameter training rules, thereby enhancing the flexibility and robustness of the training process and making the training process more efficient and comprehensive. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 A schematic diagram of the process of the renewable resource recycling management method based on big data provided by the present invention;

[0085] Figure 2 A schematic diagram of a renewable resource recycling management system based on big data provided by the present invention;

[0086] Figure 3 is a schematic flow chart of step S2;

[0087] Figure 4 is a schematic flow chart of step S3;

[0088] Figure 5 It is a schematic diagram of the process of step S4.

[0089] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0090] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0091] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0092] Example 1, see Figure 1 The present invention provides a renewable resource recycling management method based on big data, which comprises the following steps:

[0093] Step S1: data collection, wherein the data collection is to collect historical renewable resource transaction data, production data, population data, substitute development data and market demand level;

[0094] Step S2: Data anomaly identification, removing abnormal data by defining the gravity weight function, setting the neighborhood width, defining the local density and calculating the anomaly score;

[0095] Step S3: Establishing a market demand forecasting model, by constructing a model, designing an incentive function, designing an initial adjustment layer, designing a memory control layer, designing an attention layer, designing a correction layer, designing an output layer and designing a loss function to establish a market demand forecasting model;

[0096] Step S4: model parameter training, training the model parameters by initializing, defining dynamic matching factors, defining training origin positions, designing parameter training strategies, and designing parameter training rules;

[0097] Step S5: Comprehensive management, using the market demand forecasting model to forecast the demand for renewable resources, and managing the recycling of renewable resources according to the market demand.

[0098] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the data collection is to collect historical renewable resource transaction data, production data, population data, substitute development data and market demand level; the historical renewable resource transaction data includes historical renewable resource transaction volume, transaction price and transaction time series data; the production data refers to the output of products using renewable resources as raw materials; the population data refers to population size, population structure and population mobility; the substitute development data refers to the development trend of substitute products of renewable resources, including three development trend levels of accelerated development, stable development and decelerated development; the market demand level includes strong demand, medium demand and low demand.

[0099] Example 3, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S2, the data anomaly identification specifically includes the following steps:

[0100] Step S21: define the gravity weight function, which is expressed as follows:

[0101] ;

[0102] in, represents the Euclidean distance between the data for calculating the gravity weight, represents the gravitational weight function, e represents the natural constant, represents the weight adjustment coefficient;

[0103] Step S22: Set the neighborhood width, calculate the Euclidean distance between all data points, select the two data points with the farthest distance, set the Euclidean distance between them as the domain width of the data, and set the neighborhood width of the data, which is expressed as follows:

[0104] ;

[0105] in, represents the neighborhood width of the data, Indicates the domain width of the data, Indicates the total number of data;

[0106] Step S23: define local density, take the data point as the center, set the area within the neighborhood width of the data point as the neighborhood of the data point, and calculate the local density, which is expressed as follows:

[0107] ;

[0108] Where m represents the index of the data point, represents the local density of the mth data point, represents the set of all data indexes within the neighborhood of the mth data point, Represents the index of the data in the neighborhood of the mth data point, Indicates that it belongs to the symbol, represents the mth data point and the The Euclidean distance between data points, div represents the dimension of the data;

[0109] Step S24: Calculate the anomaly score, and set the inverse of the local density of the data as the anomaly score, which is expressed as follows:

[0110] ;

[0111] in, Represents the abnormal score of the mth data;

[0112] Step S25: Remove abnormal data, calculate the abnormal scores of all data points, set the 1% of data points with the highest abnormal scores as abnormal data, and remove the abnormal data.

[0113] By performing the above operations, in order to address the problem of low accuracy and easy misjudgment of data anomaly identification in traditional renewable resource recycling management methods, this solution removes abnormal data by defining a gravity weight function, setting the neighborhood width, defining the local density and calculating the anomaly score, thereby improving the accuracy of anomaly identification and reducing the misjudgment rate.

[0114] Example 4, see Figure 1 and Figure 4This embodiment is based on the above embodiment. In step S3, the establishment of the market demand forecasting model specifically includes the following steps:

[0115] Step S31: construct a model, the market demand forecasting model includes an initial adjustment layer, a memory control layer, an attention layer, a correction layer and an output layer; at the same time, the market demand level is set as the label data of the model;

[0116] Step S32: Design an activation function, which is expressed as follows:

[0117] ;

[0118] in, represents the enhanced control activation function, represents the input of the reinforcement control activation function, represents the mean fitness, It means square root. represents the incentive weight, represents the linear rectification function, represents the excitation threshold;

[0119] Step S33: Design the initial adjustment layer, as shown below:

[0120] ;

[0121] in, represents the output value of the initial adjustment layer, p represents the layer index of the initial adjustment layer, and Represent the weight and bias of the initial adjustment layer of the pth layer, Represents the input value of the initial adjustment layer of the pth layer;

[0122] Step S34: Design a memory control layer, which is composed of an input control unit, an abandonment control unit, an update control unit, and an output control unit, and specifically includes the following contents:

[0123] Step S341: Design an input control unit, as shown below:

[0124] ;

[0125] Among them, t represents the index of the time point, represents the output value of the input control unit at time t, represents the sigmoid function, and denote the weight and bias of the input control unit, respectively. represents the control state of the memory control layer at time t-1, represents the input value of the input control unit at time t;

[0126] Step S342: Design an abandoned control unit, as shown below:

[0127] ;

[0128] in, represents the output value of the abandoned control unit at time t, and denote the weight and bias of the abandoned control unit, respectively. represents the input value of the abandoned control unit at time t, Indicates taking the L2 norm;

[0129] Step S343: Design and update the control unit, as shown below:

[0130] ;

[0131] in, represents the output value of the updated control unit at time t, represents the hyperbolic tangent activation function, and Represent the updated control weight and updated control bias respectively, represents the input value of the updated control unit at time t, represents element-wise multiplication operation, represents the control state of the memory control layer at time t-2, represents the input value of the updated control unit at time t-1;

[0132] Step S344: Design an output control unit, as shown below:

[0133] ;

[0134] in, represents the output value of the output control unit at time t, and denote the weight and bias of the output control unit, respectively. Represents the input value of the output control unit at time t;

[0135] Step S345: Update the memory control layer control state, as shown below:

[0136] ;

[0137] in, represents the control state of the memory control layer at time t, represents the input value of the output control unit at time t-1;

[0138] Step S35: Design the attention layer, calculate the attention weight of each time point, and assign the attention weight to each time point by multiplication, as shown below:

[0139] ;

[0140] in, represents the attention weight at time t, represents the normalized exponential function, U represents the basic attention weight, and Represent the weight and bias of the attention layer respectively;

[0141] Step S36: Design the correction layer, as shown below:

[0142] ;

[0143] in, represents the output of the correction layer, and Represent the weight and bias of the second layer of the correction layer, respectively. and Represent the weight and bias of the first layer of the correction layer, respectively. Represents the base input value of the first layer of the correction layer;

[0144] Step S37: Design the output layer, which is expressed as follows:

[0145] ;

[0146] in, Represents the output result of the model output layer, , and Represent the weight, input value and bias of the output layer respectively;

[0147] Step S38: Design a loss function, which is expressed as follows:

[0148] ;

[0149] in, represents the loss value of the model, j represents the index of the data for training the model, W represents the total number of data for training the model, represents the true value of the jth data, represents the model prediction result of the jth data, represents the scaling factor, Indicates taking the absolute value, and represents the loss weight, represents logarithmic operation, represents the predicted distribution of the data, represents the true distribution of the data, It represents the KL divergence between the predicted distribution and the true distribution of the data.

[0150] By performing the above operations, in order to address the problems of low prediction accuracy, weak model adaptability and difficulty in capturing dependencies in time series data in traditional market demand forecasting methods, this solution builds a model by constructing a model, designing an excitation function, designing an initial adjustment layer, designing a memory control layer, designing an attention layer, designing a correction layer, designing an output layer and designing a loss function, so as to more comprehensively capture the features and information in the data and improve the accuracy, precision and stability of the prediction.

[0151] Example 5, see Figure 1 and Figure 5 This embodiment is based on the above embodiment. In step S4, the model parameter training specifically includes the following steps:

[0152] Step S41: Initialization, determine the training parameters of the model, including the adaptive mean, the excitation threshold, the weight and bias parameters of the model, set the opposite of the loss value of the model as the parameter performance value, create a parameter training space, and generate Initial parameter training points, calculate the average position of all initial parameter training points in the parameter training space, calculate the parameter performance values ​​corresponding to all parameter training points, and select the training points with the highest and lowest parameter performance values;

[0153] Step S42: Define a dynamic matching factor, which is expressed as follows:

[0154] ;

[0155] Among them, k represents the number of parameter training times, represents the dynamic matching factor during the k-th parameter training, represents the average position of the initial parameter training points, Indicates the parameter position with the highest current parameter performance. Indicates the parameter position with the lowest current parameter performance. It means taking the maximum value of the absolute value of the difference between the coordinates of all dimensions of the two points. The index representing the number of times the parameter is trained. Indicates the parameter position during the fqth parameter training, Indicates the Euclidean distance. Indicates the maximum number of parameter training times;

[0156] Step S43: define the training origin position, as shown below:

[0157] ;

[0158] in, represents the origin position generated during the k-th parameter training, Represents a random number between 0 and 1. represents the upper bound of the parameter training space, represents the lower bound of the parameter training space, Represents a random number that follows a standard normal distribution;

[0159] Step S44: Design parameter training strategy, as shown below:

[0160] ;

[0161] in, represents the parameter position during the kth parameter training, represents the parameter position during the k-1th parameter training, Represents a random number that follows a standard normal distribution;

[0162] Step S45: Design parameter training rules, set the maximum number of parameter training times and parameter performance threshold, and perform parameter training. Each time parameter training is performed, the parameter performance values ​​of the parameter points obtained through training are calculated, and the parameter positions with the highest parameter performance and the parameter positions with the lowest parameter performance are updated. If during a training process, the performance values ​​of more than 65% of the parameter points are greater than the parameter performance threshold, the training is terminated, and the parameter with the highest parameter performance at this time is set as the model parameter; if the maximum number of parameter training times is reached, perform parameter training again; otherwise, the number of training times is increased by one, and training continues.

[0163] By performing the above operations, in order to address the problems of low training efficiency, easy to fall into local optimal solutions, and difficulty in finding the global optimal parameter configuration in traditional model parameter training methods, this solution trains model parameters by initialization, definition of dynamic matching factors, definition of training origin locations, design of parameter training strategies, and design of parameter training rules, thereby enhancing the flexibility and robustness of the training process and making the training process more efficient and comprehensive.

[0164] Example 6, see Figure 1This embodiment is based on the above embodiment. In step S5, the comprehensive management is to use the market demand forecasting model to predict the demand for renewable resources, and manage the recycling of renewable resources according to the market demand. When the market demand level is strong demand, the recycling of renewable resources is increased; when the market demand level is medium demand, the recycling of renewable resources is maintained; when the market demand level is low demand, the recycling of renewable resources is reduced.

[0165] Embodiment 7, see Figure 2 , this embodiment is based on the above embodiment, the renewable resource recycling management system based on big data provided by the present invention includes a data acquisition module, a data anomaly identification module, a market demand prediction model establishment module, a model parameter training module and a comprehensive management module;

[0166] The data collection module collects historical renewable resource transaction data, production data, population data, substitute development data and market demand level, and sends the data to the data anomaly identification module;

[0167] The data anomaly identification module receives the data sent by the data acquisition module, removes the abnormal data, and sends the data to the market demand forecasting model establishment module;

[0168] The module for establishing a market demand forecasting model receives data sent by the data anomaly identification module, establishes a market demand forecasting model, and sends the data to the model parameter training module and the comprehensive management module;

[0169] The model parameter training module receives the data sent by the market demand forecasting model establishment module, trains the parameters of the model, and sends the data to the comprehensive management module;

[0170] The comprehensive management module receives data sent by the market demand forecasting model building module and the model parameter training module, uses the market demand forecasting model to forecast the demand for renewable resources, and manages the recycling of renewable resources according to the market demand.

[0171] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0172] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0173] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A renewable resource recycling management method based on big data, characterized in that: The method comprises the following steps: Step S1: data collection, wherein the data collection is to collect historical renewable resource transaction data, production data, population data, substitute development data and market demand level; Step S2: Data anomaly identification, removing abnormal data by defining the gravity weight function, setting the neighborhood width, defining the local density and calculating the anomaly score; Step S3: Establishing a market demand forecasting model, by constructing a model, designing an incentive function, designing an initial adjustment layer, designing a memory control layer, designing an attention layer, designing a correction layer, designing an output layer and designing a loss function to establish a market demand forecasting model; Step S4: model parameter training, training the model parameters by initializing, defining dynamic matching factors, defining training origin positions, designing parameter training strategies, and designing parameter training rules; Step S5: Comprehensive management, using the market demand forecasting model to forecast the demand for renewable resources, and managing the recycling of renewable resources according to the market demand; In step S1, the data collection is to collect historical renewable resource transaction data, production data, population data, substitute development data and market demand level; the historical renewable resource transaction data includes historical renewable resource transaction volume, transaction price and transaction time series data; the production data refers to the output of products using renewable resources as raw materials; the population data refers to the population size, population structure and population mobility; the substitute development data refers to the development trend of substitute products of renewable resources, including three development trend levels of accelerated development, stable development and decelerated development; the market demand level includes strong demand, medium demand and low demand; In step S2, the data anomaly identification specifically includes the following steps: Step S21: define the gravity weight function, which is expressed as follows: ； in, represents the Euclidean distance between the data for calculating the gravity weight, represents the gravitational weight function, e represents the natural constant, represents the weight adjustment coefficient; Step S22: Set the neighborhood width, calculate the Euclidean distance between all data points, select the two data points with the farthest distance, set the Euclidean distance between them as the domain width of the data, and set the neighborhood width of the data, which is expressed as follows: ； in, represents the neighborhood width of the data, Indicates the domain width of the data, Indicates the total number of data; Step S23: define the local density, take the data point as the center, set the area within the neighborhood width of the data point as the neighborhood of the data point, and calculate the local density, which is expressed as follows: ； Where m represents the index of the data point, represents the local density of the mth data point, represents the set of all data indexes within the neighborhood of the mth data point, Represents the index of the data in the neighborhood of the mth data point, Indicates that it belongs to the symbol, represents the mth data point and the The Euclidean distance between data points, div represents the dimension of the data; Step S24: Calculate the anomaly score, and set the inverse of the local density of the data as the anomaly score, which is expressed as follows: ； in, Represents the abnormal score of the mth data; Step S25: remove abnormal data, calculate the abnormal scores of all data points, set the data points with the highest abnormal scores as abnormal data, and remove the abnormal data; In step S5, the comprehensive management is to use the market demand forecasting model to predict the demand for renewable resources, and manage the recycling of renewable resources according to the market demand. When the market demand level is strong demand, the recycling of renewable resources is increased; when the market demand level is medium demand, the recycling of renewable resources is maintained; when the market demand level is low demand, the recycling of renewable resources is reduced.

2. The renewable resource recycling management method based on big data according to claim 1 is characterized in that: In step S3, the establishment of the market demand forecasting model specifically includes the following steps: Step S31: construct a model, the market demand forecasting model includes an initial adjustment layer, a memory control layer, an attention layer, a correction layer and an output layer; at the same time, the market demand level is set as the label data of the model; Step S32: Design an activation function, which is expressed as follows: ； in, represents the enhanced control activation function, represents the input of the reinforcement control activation function, represents the adaptive mean, e represents the natural constant, It means square root. represents the incentive weight, represents the linear rectification function, represents the excitation threshold; Step S33: Design the initial adjustment layer, as shown below: ； in, represents the output value of the initial adjustment layer, p represents the layer index of the initial adjustment layer, and Represent the weight and bias of the initial adjustment layer of the pth layer, Represents the input value of the initial adjustment layer of the pth layer; Step S34: Design a memory control layer, which is composed of an input control unit, an abandonment control unit, an update control unit, and an output control unit, and specifically includes the following contents: Step S341: Design an input control unit, as shown below: ； Among them, t represents the index of the time point, represents the output value of the input control unit at time t, represents the sigmoid function, and denote the weight and bias of the input control unit, respectively. represents the control state of the memory control layer at time t-1, represents the input value of the input control unit at time t; Step S342: Design an abandoned control unit, as shown below: ； in, represents the output value of the abandoned control unit at time t, and denote the weight and bias of the abandoned control unit, respectively. represents the input value of the abandoned control unit at time t, Indicates taking the L2 norm; Step S343: Design and update the control unit, as shown below: ； in, represents the output value of the updated control unit at time t, represents the hyperbolic tangent activation function, and Represent the updated control weight and updated control bias respectively, represents the input value of the updated control unit at time t, represents element-wise multiplication operation, represents the control state of the memory control layer at time t-2, represents the input value of the updated control unit at time t-1; Step S344: Design an output control unit, as shown below: ； in, represents the output value of the output control unit at time t, and denote the weight and bias of the output control unit, respectively. Represents the input value of the output control unit at time t; Step S345: Update the memory control layer control state, as shown below: ； in, represents the control state of the memory control layer at time t, represents the input value of the output control unit at time t-1; Step S35: Design the attention layer, calculate the attention weight of each time point, and assign the attention weight to each time point by multiplication, as shown below: ； in, represents the attention weight at time t, represents the normalized exponential function, U represents the basic attention weight, and Represent the weight and bias of the attention layer respectively; Step S36: Design the correction layer, as shown below: ； in, represents the output of the correction layer, and Represent the weight and bias of the second layer of the correction layer, respectively. and Represent the weight and bias of the first layer of the correction layer, respectively. Represents the base input value of the first layer of the correction layer; Step S37: Design the output layer, which is expressed as follows: ； in, Represents the output result of the model output layer, , and Represent the weight, input value and bias of the output layer respectively; Step S38: Design a loss function, which is expressed as follows: ； in, represents the loss value of the model, j represents the index of the data for training the model, W represents the total number of data for training the model, represents the true value of the jth data, represents the model prediction result of the jth data, represents the scaling factor, Indicates taking the absolute value, and represents the loss weight, represents logarithmic operation, represents the predicted distribution of the data, represents the true distribution of the data, It represents the KL divergence between the predicted distribution and the true distribution of the data.

3. The renewable resource recycling management method based on big data according to claim 1 is characterized in that: In step S4, the model parameter training specifically includes the following steps: Step S41: Initialization, determine the training parameters of the model, including the adaptive mean, the excitation threshold, the weight and bias parameters of the model, set the opposite of the loss value of the model as the parameter performance value, create a parameter training space, and generate Initial parameter training points, calculate the average position of all initial parameter training points in the parameter training space, calculate the parameter performance values ​​corresponding to all parameter training points, and select the training points with the highest and lowest parameter performance values; Step S42: Define a dynamic matching factor, which is expressed as follows: ； Among them, k represents the number of parameter training times, represents the dynamic matching factor during the k-th parameter training, represents the average position of the initial parameter training points, Indicates the parameter position with the highest current parameter performance. Indicates the parameter position with the lowest current parameter performance. It means taking the maximum value of the absolute value of the difference between the coordinates of all dimensions of the two points. The index representing the number of times the parameter is trained. Indicates the parameter position during the fqth parameter training, Indicates the Euclidean distance. Indicates the maximum number of parameter training times; Step S43: define the training origin position, as shown below: ； in, represents the origin position generated during the k-th parameter training, Represents a random number between 0 and 1. represents the upper bound of the parameter training space, represents the lower bound of the parameter training space, Represents a random number that follows a standard normal distribution; Step S44: Design parameter training strategy, as shown below: ； in, represents the parameter position during the kth parameter training, represents the parameter position during the k-1th parameter training, Represents a random number that follows a standard normal distribution; Step S45: Design parameter training rules, set the maximum number of parameter training times and parameter performance threshold, and perform parameter training. Each time parameter training is performed, the parameter performance values ​​of the parameter points obtained through training are calculated, and the parameter positions with the highest parameter performance and the parameter positions with the lowest parameter performance are updated. If during a training process, the performance values ​​of more than 65% of the parameter points are greater than the parameter performance threshold, the training is terminated, and the parameter with the highest parameter performance at this time is set as the model parameter; if the maximum number of parameter training times is reached, perform parameter training again; otherwise, the number of training times is increased by one, and training continues.

4. A renewable resource recycling management system based on big data, used to implement the renewable resource recycling management method based on big data as described in any one of claims 1 to 3, characterized in that: It includes data collection module, data anomaly identification module, market demand forecasting model building module, model parameter training module and comprehensive management module; The data collection module collects historical renewable resource transaction data, production data, population data, substitute development data and market demand level, and sends the data to the data anomaly identification module; The data anomaly identification module receives the data sent by the data acquisition module, removes the abnormal data, and sends the data to the market demand forecasting model establishment module; The module for establishing a market demand forecasting model receives data sent by the data anomaly identification module, establishes a market demand forecasting model, and sends the data to the model parameter training module and the comprehensive management module; The model parameter training module receives the data sent by the market demand forecasting model establishment module, trains the parameters of the model, and sends the data to the comprehensive management module; The comprehensive management module receives data sent by the market demand forecasting model building module and the model parameter training module, uses the market demand forecasting model to forecast the demand for renewable resources, and manages the recycling of renewable resources according to the market demand.

Images