要研究地震预测的可能性,客观评估预测结果必不可少。鲜少发生的大地震和小地震的发生概率大相径庭,预测成功时的评估也大不相同。地震活动程度中的地域差也与评估相关。首先,需要确立适合各地域的基于以往经验的地震活动的基准预测。作为评估概率预测成绩的手段,信息增益比较合理。当新的预测模型被提出时,对比基准模型的预测,即可评估预测能力是否提高及其提高的程度。赤池信息量准则AIC及ABIC是仅仅使用现有的数据预先对提议模型将来预测准确度进行推测并给出评估分数是有用的。由于预测的演算方法和经验尚处于发展完善过程中,有时难以给出概率的数值预测。因此,大多数都属于警告型预测(二值预测)。本文将进一步就评估警告型地震预测的图示法和赌局计分法进行说明。为此,仍然必须设定地震的大小程度和活动度的基准概率(行情)的经验值。也就是在基于行情的公平的赌局中,比较警告型预测的成功或失败结果的得失分数的评估方法。作为经验基准概率,将古登堡-里克特关系(指数分布)当作地震大小的出现频度。时间和空间一样的地震发生模型(平稳泊松过程)可以考虑。然而,更为现实地将相应地震活动性空间不一致的泊松过程和地震的连锁过程等设定为基准模型,则现今的警告型预测要得到比基准预测更好的评估结果将变得更难。 To study the possibility of earthquake prediction, an objective assessment of the forecast results is essential. The rare occurrence of large earthquakes and small earthquakes probability of occurrence varies greatly, the assessment of the success of the prediction is also very different. The geographical difference in the degree of seismic activity is also related to the assessment. First, there is a need to establish a baseline prediction of seismic activity based on past experience that is appropriate for each region. As a means to assess the probability of predicting performance, the information gain is reasonable. When a new forecasting model is proposed, the prediction of the benchmarking model can be compared with the forecasting ability and the degree of its improvement. AIC and ABIC Guidelines on Akaike’s Information Volume It is useful to use the available data to predict in advance the accuracy of the proposed model’s predictions and to give an evaluation score using only the available data. As the prediction methods and experience are still in the process of development and perfection, it is sometimes difficult to predict the value of probability. Therefore, most of them belong to warning type prediction (binary prediction). This article will further illustrate the method of assessing the prediction of warning type earthquake prediction and gambling scores. For this reason, it is still necessary to set empirical values ​​of the benchmark probabilities (quotes) of the magnitude and the degree of activity of the earthquakes. That is, in the market-based fair betting method, a method of assessing the pros and cons of a successful or a failed result of a warning type prediction is compared. As empirical benchmark probabilities, the Gutenberg-Richter relationship (exponential distribution) is taken as the frequency of occurrence of the earthquake. Time and space-like earthquake occurrence models (stationary Poisson processes) can be considered. However, it is more realistic to set the Poisson process with inconsistencies in seismic activity and the interlocking process of earthquakes as the baseline model, and it will be harder for today’s warning predictions to get better assessment results than baseline predictions .