시계열 모형을 이용한 범죄예측 사례연구

주일엽 한국경호경비학회 2012 시큐리티연구 Vol.- No.30

본 연구는 살인, 강도, 강간, 절도, 폭력 등 주요 범죄를 예측할 수 있는 시계열 모형을 도출하고 이를 이용한 주요 범죄의 발생 전망을 파악하여 범죄 발생에 대한 과학적인 치안 정책 수립에 기여하는데 그 목적이 있다. 이와 같은 목적을 달성하기 위하여 2002년부터 2010년까지의 살인, 강도, 강간, 절도, 폭력 등 주요범죄에 대한 월별 발생건수를 IBM PASW(SPSS) 19.0을 사용하여 주요 범죄의 시계열 예측모형을 규명하기 위한 시계열 모형생성(C), 주요 범죄의 시계열 예측모형에 대한 정확도 규명을 위한 시계열 모형생성(C) 및 시계열 순차도표(N)를 실시하였다. 이와 같은 연구목적과 연구방법을 통하여 도출한 연구결과는 다음과 같다. 첫째, 살인, 강도, 강간, 절도, 폭력 등 주요 범죄에 대한 시계열 예측모형은 각각 단순계 절, Winters 승법, ARIMA(0,1,1)(0,1,1), ARIMA(1,1,0)(0,1,1), 단순계절로 나타났다. 둘째, 살인, 강도, 강간, 절도, 폭력 등 주요 범죄에 대하여 시계열 예측모형을 이용한 주요 범죄에 대한 단기적 발생 전망이 가능한 것으로 나타났다. 이러한 연구결과를 토대로 범죄 발생에 대한 지속적인 시계열 예측모형 제시, 분기별, 연도별 범죄 발생건수를 기초로 하는 중 ․ 장기 시계열 예측모형에 대한 관심이 요구된다. The purpose of this study is to contribute to establishing the scientific policing policies through deriving the time series models that can forecast the occurrence of major crimes such as murder, robbery, burglary, rape, violence and identifying the occurrence of major crimes using the models. In order to achieve this purpose, there were performed the statistical methods such as Generation of Time Series Model(C) for identifying the forecasting models of time series, Generation of Time Series Model(C) and Sequential Chart of Time Series(N) for identifying the accuracy of the forecasting models of time series on the monthly incidence of major crimes from 2002 to 2010 using IBM PASW(SPSS) 19.0. The following is the result of the study. First, murder, robbery, rape, theft and violence crime's forecasting models of time series are Simple Season, Winters Multiplicative, ARIMA(0,1,1)(0,1,1), ARIMA(1,1,0 )(0,1,1) and Simple Season. Second, it is possible to forecast the short-term's occurrence of major crimes such as murder, robbery, burglary, rape, violence using the forecasting models of time series. Based on the result of this study, we have to suggest various forecasting models of time series continuously, and have to concern the long-term forecasting models of time series which is based on the quarterly, yearly incidence of major crimes.

Temporal Fusion Transformer를 이용한 대형마트 판매량의 다단계 시계열 수요예측

안세희,정재윤 한국전자거래학회 2023 한국전자거래학회지 Vol.28 No.3

Demand forecast is used as basic data for business and operation planning in all industries. In this paper, the Temporal Fusion Transformer (TFT) architecture was applied to the data of the M5 Competition, a famous forecasting competition, and the accuracy of the TFT-based forecasting method was compared with that of the DRFAM method, that had won the competition. The performance was evaluated for the sales data of CA_1 store in the Walmart dataset of the M5 Competition. The TFT models were trained with two data pools at the store level and category level, respectively, and the final forecast was calculated by arithmetically averaging the prediction results of the two models. As a result, the TFT-based method obtained better forecasts than the DRFAM method, which trained six LightGBM models with direct forecasting and recursive forecasting for three levels of data pools and predicted with the arithmetic average of the six trained models. It was found that the TFT-based method had sufficiently learned the relationship between variables and sales volumes in the time-series using the self-attention structure of TFT. While the direct and recursive forecasting models of the DRFAM method require 28 repeated calls for 28 days of forecasting, the TFT-based method can obtain 28 time-series forecasts with a single model call because of its multi-output structure. The proposed TFT-based forecasting method is expected to be applicable to various fields by providing faster and more accurate time-series forecasts. 수요예측은 모든 산업에서 사업 기획 및 운영 계획의 중요한 기초 자료로 사용된다. 본 논문에서는 수요예측 경진대회인 M5 Competition 데이터를 대상으로 Temporal Fusion Transformer(TFT) 모형을 적용하였고, 이 대회에서 우승한 DRFAM 기법과 정확도를 비교하였다. M5 Competition의 Walmart 데이터셋 중 CA_1 매장의 판매량 데이터를 대상으로 성능을 평가하였으며, 매장(store) 수준과 카테고리(category) 수준의 데이터풀(data pool)로 각각 TFT 모형을 학습한 후 예측값을 산술평균하는 방식을 사용하였다. 그 결과, 세 가지 수준의 데이터풀에 대해 직접적 예측모형(direct forecasting)과 재귀적 예측모형(recursive forecasting)으로 총 6개의 LightGBM 모형을 학습하여 산술평균으로 예측하는 DRFAM 기법보다 평균적으로 개선된 예측 정확도를 달성하였다. 이를 통해 TFT 모형이 자기-어텐션 구조를 사용하여 시계열에서 변수와 판매량 간의 관계를 충분히 학습하였음을 알 수 있었다. DRFAM 기법의 직접적 예측모형과 재귀적 예측모형이 28일 간의 예측을 위하여 28회 반복호출을 해야 하지만, TFT 모형은 다중 출력 구조이기 때문에 한번 모형 호출로 28개의 시계열 예측이 가능하다. 본 논문에서 제안한 TFT 기반의 예측모형은 보다 빠르고 정확한 시계열 예측을 제공하여 다양한 분야에 확대 적용할 수 있을 것으로 기대한다.

Design for Forecasting System of Tobacco Sales Based on Time Series

Zhenyu Yang 보안공학연구지원센터 2016 International Journal of u- and e- Service, Scienc Vol.9 No.8

The sales forecast in the tobacco sales management plays a great role in the production and operation activities of an enterprise as a bridge between the production activity and the economic benefit. Forecasting system based on time series is put forward in accordance with the characteristics of tobacco sales by knowing about market supply and demand as well as development tendency at home and abroad through the sales forecast, with the help of studies on the sales forecasting technology. Time series is a sequence of random variables in chronological order and the time series analysis technique includes the moving average forecasting, exponential smoothing and regression analysis. According to the characteristics of tobacco sales data, the moving average forecasting model, exponential smoothing model and regression analysis model are established. Through programming, a forecasting system of tobacco sales is set up, which is used in combination with examples to simulate the forecasting model, and then model tests are conducted on the simulation results to get the appropriate model, which is used for sales forecast. The establishment of the system has an important influence on the management of tobacco companies. According to the forecast of tobacco sales, they will know the sales of the next period, so that they can take effective sales management strategies to improve the accuracy and efficiency of the decisions on the company's production and operation.

Temporal Fusion Transformers와 심층 학습 방법을사용한 다층 수평 시계열 데이터 분석

김인경,김대희,이재구 한국정보처리학회 2022 정보처리학회논문지. 소프트웨어 및 데이터 공학 Vol.11 No.2

Given that time series are used in various fields, such as finance, IoT, and manufacturing, data analytical methods for accurate time-seriesforecasting can serve to increase operational efficiency. Among time-series analysis methods, multi-horizon forecasting provides a betterunderstanding of data because it can extract meaningful statistics and other characteristics of the entire time-series. Furthermore, time-seriesdata with exogenous information can be accurately predicted by using multi-horizon forecasting methods. However, traditional deeplearning-based models for time-series do not account for the heterogeneity of inputs. We proposed an improved time-series predicting method,called the temporal fusion transformer method, which combines multi-horizon forecasting with interpretable insights into temporal dynamics. Various real-world data such as stock prices, fine dust concentrates and electricity consumption were considered in experiments. Experimentalresults showed that our temporal fusion transformer method has better time-series forecasting performance than existing models 시계열 데이터는 주식, IoT, 공장 자동화와 같은 다양한 실생활에서 수집되고 활용되고 있으며, 정확한 시계열 예측은 해당 분야에서 운영 효율성을 높일 수 있어서 전통적으로 중요한 연구 주제이다. 전반적인 시계열 데이터의 향상된 특징을 추출할 수 있는 대표적인 시계열 데이터 분석방법인 다층 수평 예측은 최근 부가적 정보를 포함하는 시계열 데이터에 내재한 이질성(heterogeneity)까지 포괄적으로 분석에 활용하여 향상된시계열 예측한다. 하지만 대부분의 심층 학습 기반 시계열 분석 모델들은 시계열 데이터의 이질성을 반영하지 못했다. 따라서 우리는 잘 알려진temporal fusion transformers 방법을 사용하여 실생활과 밀접한 실제 데이터를 이질성을 고려한 다층 수평 예측에 적용하였다. 결과적으로 주식,미세먼지, 전기 소비량과 같은 실생활 시계열 데이터에 적용한 방법이 기존 예측 모델보다 향상된 정확도를 가짐을 확인할 수 있었다.

다변량 시계열 예측 개선을 위한 입력 시계열의 약정상화

Ranjai Baidya,Sang-Woong Lee 한국차세대컴퓨팅학회 2022 한국차세대컴퓨팅학회 학술대회 Vol.2022 No.05

Time series forecasting is relevant in many real-world applications. However, most real-world time series data are non-stationary, which means their statistical properties like mean, and variance varies with time. This property of time series is not considered by most modern deep learning forecasting models, causing the distribution of the training and test sets to be different. Eventually, the accuracy of the forecasting model is significantly affected by the distribution shift. To tackle this problem, we suggest a simple solution called 'Pseudo-Stationarizer.’ This block can be used seamlessly alongside pre-existing forecasting models to obtain better forecasts. ‘Pseudo-Stationarizer’ performs differencing on the original time series to make the data weakly stationary and helps in minimizing the distribution shift. Via thorough experimentation, we prove that the usage of the proposed block aids the forecasting models in getting significant improvements in their performance by diminishing the distribution shift and making the time series weakly stationary.

시계열 예측을 위한 스타일 기반 트랜스포머

김동건 ( Dong-keon Kim ),김광수 ( Kwangsu Kim ) 한국정보처리학회 2021 정보처리학회논문지. 소프트웨어 및 데이터 공학 Vol.10 No.12

시계열 예측은 과거 시점의 정보를 토대로 미래 시점의 정보를 예측하는 것을 말한다. 향후 시점의 정보를 정확하게 예측하는 것은 다양한 분야 전략 수립, 정책 결정 등을 위해 활용되기 때문에 매우 중요하다. 최근에는 트랜스포머 모델이 시계열 예측 모델로서 주로 연구되고 있다. 그러나 기존의 트랜스포머의 모델은 예측 순차를 출력할 때 출력 결과를 다시 입력하는 자가회귀 구조로 되어 있다는 한계점이 있다. 이 한계점은 멀리 떨어진 시점을 예측할 때 정확도가 떨어진다는 문제점을 초래한다. 본 논문에서는 이러한 문제점을 개선하고 더 정확한 시계열 예측을 위해 스타일 변환 기법에 착안한 순차 디코딩 모델을 제안한다. 제안하는 모델은 트랜스포머-인코더에서 과거 정보의 특성을 추출하고, 이를 스타일-기반디코더에 반영하여 예측 시계열을 생성하는 구조로 되어 있다. 이 구조는 자가회귀 방식의 기존의 트랜스포머의 디코더 구조와 다르게, 예측 순차를 한꺼번에 출력하기 때문에 더 먼 시점의 정보를 좀 더 정확히 예측할 수 있다는 장점이 있다. 서로 다른 데이터 특성을 가지는 다양한 시계열 데이터셋으로 예측 실험을 진행한 결과, 본 논문에서 제시한 모델이 기존의 다른 시계열 예측 모델보다 예측 정확도가 우수하다는 것을 보인다. Time series forecasting refers to predicting future time information based on past time information. Accurately predicting future information is crucial because it is used for establishing strategies or making policy decisions in various fields. Recently, a transformer model has been mainly studied for a time series prediction model. However, the existing transformer model has a limitation in that it has an auto-regressive structure in which the output result is input again when the prediction sequence is output. This limitation causes a problem in that accuracy is lowered when predicting a distant time point. This paper proposes a sequential decoding model focusing on the style transformation technique to handle these problems and make more precise time series forecasting. The proposed model has a structure in which the contents of past data are extracted from the transformer-encoder and reflected in the style-based decoder to generate the predictive sequence. Unlike the decoder structure of the conventional auto-regressive transformer, this structure has the advantage of being able to more accurately predict information from a distant view because the prediction sequence is output all at once. As a result of conducting a prediction experiment with various time series datasets with different data characteristics, it was shown that the model presented in this paper has better prediction accuracy than other existing time series prediction models.

Stock Forecasting Using Prophet vs. LSTM Model Applying Time-Series Prediction

Alshara, Mohammed Ali International Journal of Computer ScienceNetwork S 2022 International journal of computer science and netw Vol.22 No.2

Forecasting and time series modelling plays a vital role in the data analysis process. Time Series is widely used in analytics & data science. Forecasting stock prices is a popular and important topic in financial and academic studies. A stock market is an unregulated place for forecasting due to the absence of essential rules for estimating or predicting a stock price in the stock market. Therefore, predicting stock prices is a time-series problem and challenging. Machine learning has many methods and applications instrumental in implementing stock price forecasting, such as technical analysis, fundamental analysis, time series analysis, statistical analysis. This paper will discuss implementing the stock price, forecasting, and research using prophet and LSTM models. This process and task are very complex and involve uncertainty. Although the stock price never is predicted due to its ambiguous field, this paper aims to apply the concept of forecasting and data analysis to predict stocks.

Time-Series Forecasting Based on Multi-Layer Attention Architecture

Na Wang,Xianglian Zhao 한국인터넷정보학회 2024 KSII Transactions on Internet and Information Syst Vol.18 No.1

Time-series forecasting is extensively used in the actual world. Recent research has shown that Transformers with a self-attention mechanism at their core exhibit better performance when dealing with such problems. However, most of the existing Transformer models used for time series prediction use the traditional encoder-decoder architecture, which is complex and leads to low model processing efficiency, thus limiting the ability to mine deep time dependencies by increasing model depth. Secondly, the secondary computational complexity of the self-attention mechanism also increases computational overhead and reduces processing efficiency. To address these issues, the paper designs an efficient multi-layer attention-based time-series forecasting model. This model has the following characteristics: (i) It abandons the traditional encoder-decoder based Transformer architecture and constructs a time series prediction model based on multi-layer attention mechanism, improving the model's ability to mine deep time dependencies. (ii) A cross attention module based on cross attention mechanism was designed to enhance information exchange between historical and predictive sequences. (iii) Applying a recently proposed sparse attention mechanism to our model reduces computational overhead and improves processing efficiency. Experiments on multiple datasets have shown that our model can significantly increase the performance of current advanced Transformer methods in time series forecasting, including LogTrans, Reformer, and Informer.

The Prediction and Analysis of the Power Energy Time Series by Using the Elman Recurrent Neural Network

Chang-Yong Lee(이창용),Jinho Kim(김진호) 한국산업경영시스템학회 2018 한국산업경영시스템학회지 Vol.41 No.1

In this paper, we propose an Elman recurrent neural network to predict and analyze a time series of power energy consumption. To this end, we consider the volatility of the time series and apply the sample variance and the detrended fluctuation analyses to the volatilities. We demonstrate that there exists a correlation in the time series of the volatilities, which suggests that the power consumption time series contain a non-negligible amount of the non-linear correlation. Based on this finding, we adopt the Elman recurrent neural network as the model for the prediction of the power consumption. As the simplest form of the recurrent network, the Elman network is designed to learn sequential or time-varying pattern and could predict learned series of values. The Elman network has a layer of “context units” in addition to a standard feedforward network. By adjusting two parameters in the model and performing the cross validation, we demonstrated that the proposed model predicts the power con-sumption with the relative errors and the average errors in the range of 2%~5% and 3kWh~8kWh, respectively. To further confirm the experimental results, we performed two types of the cross validations designed for the time series data. We also support the validity of the model by analyzing the multi-step forecasting. We found that the prediction errors tend to be saturated although they increase as the prediction time step increases. The results of this study can be used to the energy management system in terms of the effective control of the cross usage of the electric and the gas energies.

적대적 훈련 기반의 시계열 데이터 증강 기법

신광훈,김도국 한국정보과학회 2023 정보과학회논문지 Vol.50 No.8

Recently, time series data are being generated in various industries with advancement of the Internet of Things (IoT). Accordingly, demands for time series forecasting in various industries are increasing. With acquisition of a large amount of time-series data, studies on traditional statistical method based time-series forecasting and deep learning-based forecasting methods have become active and the need for data augmentation techniques has emerged. In this paper, we proposed a novel data augmentation method for time series forecasting based on adversarial training. Unlike conventional adversarial training, the proposed method could fix the hyperparameter about the number of adversarial training iterations and utilize blockwise clipping of perturbations. We carried out various experiments to verify the performance of the proposed method. As a result, we were able to confirm that the proposed method had consistent performance improvement effect on various datasets. In addition, unlike conventional adversarial training, the necessity of blockwise clipping and the hyperparameter value fixing proposed in this paper were also verified through comparative experiments.