Self-adaptive design method and system for puzzle solving game based on ecological balance

Abstract

The invention discloses a puzzle solution game adaptive design method and system based on ecological balance, and the method specifically comprises the steps: receiving ecological scene data selected by a player, carrying out the evaluation processing of the capability level of the player through combining with a player capability evaluation model, and generating an evaluation result; dynamically generating hierarchical ecological problems according to an evaluation result, and synchronously activating Agent behavior tree loaders of corresponding hierarchies to form an Agent modeling system; simulating a biological interaction network through an Agent modeling system, inputting real-time operation data of the biological interaction network into an LSTM double-time-sequence prediction engine for prediction processing, and generating an ecological evolution graph; and generating a dynamic visual interface according to the ecological evolution map, and triggering a difficulty regulator when detecting that the ecological imbalance index exceeds a difficulty threshold value of the current player. Complex characteristics of an ecological system are effectively simulated, more challenging and interesting puzzle solving game experience is provided for players, and meanwhile understanding and protection awareness of the players to the ecological system can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of game development, and particularly relates to a puzzle game adaptive design method and system based on ecological balance. BACKGROUND

[0002] The current market puzzle games present a diversified development trend, covering a variety of forms such as logical reasoning, physical mechanism, and graphic splicing. Such games challenge the intelligence and observation of players by means of carefully set various puzzles and levels, not only have high entertainment, but also play a certain educational role, and are deeply loved by players.

[0003] However, the existing puzzle games have significant limitations. Most of the puzzle games focus on a single puzzle element, and lack simulation and deep thinking of a wider and more complex system. Taking the ecological system as an example, ecological balance is a very complex and dynamic system, which involves complex interactions between multiple species and between species and environmental factors. However, there are few puzzle games on the market that can effectively simulate these key characteristics of the ecological system.

[0004] In the ecological system, the construction and maintenance of the biological chain, the rational allocation and recycling of resources, etc. are key elements to maintain balance. The existing puzzle games fail to fully demonstrate the internal relationship and dynamic changes between these elements, making it difficult for players to understand the overall picture and deep running mechanism of the ecological system during the game process. This not only limits the further improvement of the interest and challenge of the game, but also makes it difficult for players to deeply understand the importance of the ecological system during the game process, thereby making it difficult to effectively improve the players' awareness of protecting the ecological system.

[0005] With the rapid development of artificial intelligence (AI) technology, its application in the game field is increasingly widespread and in-depth. However, in the application of AI in the ecological system balance and puzzle design of the game, there are still many technical problems to be solved. For example, how to dynamically generate adaptive ecological problems according to the player's ability level to provide personalized game experience; how to accurately simulate the biological interaction network through AI modeling and realize accurate prediction of ecological evolution; and how to use AI technology to monitor the ecological imbalance in real time and intelligently adjust the game difficulty to ensure that the challenge and interest of the game always remain within a reasonable range. SUMMARY

[0006] The present application aims to provide a puzzle game adaptive design method and system based on ecological balance, which effectively simulates the complex characteristics of the ecological system, provides players with a more challenging and interesting puzzle game experience, and helps to improve players' understanding and protection awareness of the ecological system, to solve at least one of the above technical problems.

[0007] In a first aspect, the present application provides an ecological balance-based puzzle game adaptive design method, which specifically comprises: receiving ecological scene data selected by a player, evaluating the player's ability level in combination with a player ability evaluation model, and generating an evaluation result; dynamically generating hierarchical ecological problems according to the evaluation result, synchronously activating an Agent behavior tree loader of a corresponding level to form an Agent modeling system; simulating a biological interaction network through the Agent modeling system, and inputting real-time operation data of the biological interaction network into an LSTM double-time sequence prediction engine for prediction processing to generate an ecological evolution map; generating a dynamic visual interface according to the ecological evolution map, and triggering a difficulty adjuster when detecting that an ecological imbalance index exceeds a current player difficulty threshold; updating ability evaluation parameters according to the topological integrity of player operation chain data, and optimizing the player ability evaluation model in combination with a reinforcement learning algorithm.

[0008] In a second aspect, the present application provides an ecological balance-based puzzle game adaptive design system, which specifically comprises: an ability evaluation module for receiving ecological scene data selected by a player, evaluating the player's ability level in combination with a player ability evaluation model, and generating an evaluation result; a behavior modeling module for dynamically generating hierarchical ecological problems according to the evaluation result, synchronously activating an Agent behavior tree loader of a corresponding level to form an Agent modeling system; an ecological evolution module for simulating a biological interaction network through the Agent modeling system, and inputting real-time operation data of the biological interaction network into an LSTM double-time sequence prediction engine for prediction processing to generate an ecological evolution map; a difficulty adjustment module for generating a dynamic visual interface according to the ecological evolution map, and triggering a difficulty adjuster when detecting that an ecological imbalance index exceeds a current player difficulty threshold; a parameter optimization module for updating ability evaluation parameters according to the topological integrity of player operation chain data, and optimizing the player ability evaluation model in combination with a reinforcement learning algorithm.

[0009] In a third aspect, the present application provides a computer device, which comprises a memory, a processor, and a computer program stored in the memory, and when the computer program is executed on the processor, the ecological balance-based puzzle game adaptive design method according to any one of the above methods is implemented.

[0010] In a fourth aspect, the present application provides a computer readable storage medium having stored thereon a computer program, which, when executed by a processor, implements the ecological balance-based puzzle game adaptive design method according to any one of the above methods.

[0011] Compared with the prior art, the present application has at least one of the following technical effects: 1. The present application effectively simulates the complex characteristics of an ecological system, providing players with a more challenging and interesting puzzle game experience, while helping to improve players' understanding and protection awareness of the ecological system.

[0012] 2. The present application generates a player ability vector by collecting historical operation sequences and calculating multiple indicators, providing a data basis for accurately assessing the player's ability level, making the assessment more scientific and comprehensive.

[0013] 3. The present application multi-dimensionally deconstructs, classifies, analyzes and predicts the player ability vector, taking into account the partiality and potential ability, to obtain a more accurate and reasonable player ability rating.

[0014] 4. The present application generates hierarchical ecological problems and determines the depth of the behavior tree according to the evaluation results of each indicator, and constructs an Agent modeling system containing key parameters, achieving the adaptation of ecological problems and player ability.

[0015] 5. The present application generates time series data by simulating biological interaction networks, uses an LSTM double-time series prediction engine for short-term and long-term prediction, and comprehensively presents the ecological evolution situation.

[0016] 6. The present application determines the long-term prediction path according to the short-term prediction results, predicts the reproduction chain or extinction chain for different key species change rates, and generates an intuitive and information-rich ecological evolution map.

[0017] 7. The present application monitors the ecological evolution map in real time to calculate the ecological imbalance index, compares it with the player difficulty threshold, and triggers difficulty adjustment, ensuring that the game difficulty always adapts to the player's level.

[0018] 8. The present application analyzes the player operation chain data topology to calculate multiple indicators, dynamically adjusts the ability parameters and optimizes the model by combining the reinforcement learning algorithm, making the player ability assessment more in line with the actual performance.

[0019] 9. The present application generates operation suggestions through a reinforcement learning strategy network, analyzes the actual utility through a value network and feeds back, and updates the weights through gradient backpropagation, achieving precise optimization of the player ability assessment model weights. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0021] Figure 1 is a flowchart of a method for adaptive design of a mystery game based on ecological balance provided by an embodiment of the present application; Figure 2 is a structural diagram of a system for adaptive design of a mystery game based on ecological balance provided by an embodiment of the present application; Figure 3 is a structural diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0022] In the following description, specific details are set forth in order to provide a thorough understanding of embodiments of the present application. However, persons of ordinary skill in the art will readily recognize that embodiments of the present application can be practiced without these specific details, in other instances, well-known structures, devices, circuits, and processes have not been described in detail in order to avoid obscuring embodiments of the present application.

[0023] It should be understood that, when used in the specification and the appended claims of the present application, the term "comprising" indicates the presence of the described features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0024] It should also be understood that the term "and / or" as used in the specification and the appended claims of the present application means any combination of one or more of the associated listed items and all possible combinations thereof, and includes these combinations.

[0025] As used in the specification and the appended claims of the present application, the term "if" can be interpreted as "when" or "upon" or "in response to a determination" or "in response to detecting" depending on the context. Similarly, the phrase "if it is determined" or "if [a described condition or event] is detected" can be interpreted as meaning "upon determining" or "in response to determining" or "upon detecting [a described condition or event]" or "in response to detecting [a described condition or event]" depending on the context.

[0026] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0027] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0028] In this application embodiment, the entity executing the process includes a terminal device. This terminal device includes, but is not limited to, devices capable of executing the methods disclosed in this application, such as servers, computers, smartphones, and tablets. Figure 1 A flowchart illustrating an adaptive design method for puzzle games based on ecological balance, as disclosed in an embodiment of the present invention, is shown below in detail: S101 receives the ecological scene data selected by the player, combines it with the player ability assessment model to evaluate the player's ability level, and generates the assessment result.

[0029] In this embodiment, after the game starts, the system presents players with multiple ecological scene options of different types and difficulties. These ecological scenes cover various types, including forest ecosystems, marine ecosystems, and grassland ecosystems, each with unique biological species, environmental factors, and ecological balance characteristics. For example, a forest ecosystem scene includes various organisms such as trees, animals, and microorganisms, as well as environmental factors such as light, water, and soil, all of which have complex interrelationships. Players select the ecological scene of interest from these options using interactive buttons on the game interface. The system records the ecological scene information selected by the player in real time and uses it as one of the input data for subsequent evaluation.

[0030] If this is the player's first time entering the game, this step is skipped; if the player has a playthrough history, the system will retrieve the player's historical game data from the database. This data includes the types of ecological scenarios the player chose in past games, the time taken to complete puzzle tasks, the accuracy rate, the number of attempts, and the operational strategies. By analyzing this historical data, a preliminary understanding of the player's game performance and ability preferences in different ecological scenarios can be obtained.

[0031] The player ability assessment model is a comprehensive evaluation system that considers multiple factors and is built upon a wealth of game data and expert experience. The model includes multiple assessment dimensions, each corresponding to different ability indicators, such as ecological knowledge, logical reasoning, observation skills, and adaptability.

[0032] The system inputs data from the player's chosen ecological scenario and collected historical game data into a player ability assessment model. Based on the characteristics of the ecological scenario and the player's past performance, the model scores the player's abilities across various assessment dimensions. For example, if a player chooses a marine ecosystem scenario and demonstrates some knowledge of marine life in past games, but performs only moderately well in logical reasoning to solve marine ecological problems, the model will assign a higher score to the ecological knowledge dimension and a relatively lower score to the logical reasoning dimension. Then, the scores are weighted and summed according to the weights of each dimension to obtain the player's comprehensive ability assessment result for that ecological scenario. The system generates a detailed assessment report based on the results, including the name of the ecological scenario chosen by the player, the scores on each assessment dimension, the comprehensive ability assessment score, and corresponding ability descriptions and suggestions.

[0033] In this embodiment, the ability level of players in the selected ecological scenario can be accurately and comprehensively evaluated. By combining ecological scenario data and players' historical game data, the evaluation results are more targeted and personalized.

[0034] S102, based on the evaluation results, dynamically generate hierarchical ecological problems, and simultaneously activate the corresponding level of Agent behavior tree loader to form an Agent modeling system.

[0035] In this embodiment, the evaluation results include player scores across multiple dimensions, including ecological knowledge, logical reasoning, observation skills, and adaptability. Based on the overall difficulty gradient of the game design and the players' ability range, a tiered standard is established. Player abilities can be divided into three levels: beginner, intermediate, and advanced, each corresponding to a different score range. For example, the beginner level corresponds to 0-40 points, the intermediate level to 41-70 points, and the advanced level to 71-100 points. Simultaneously, a corresponding range of ecological problem difficulty and the complexity of the Agent behavior tree are determined for each level. Beginner-level ecological problems are relatively simple, mainly involving basic ecological concepts and simple biological relationships; intermediate-level ecological problems are of moderate difficulty, requiring players to use some logical reasoning and observation skills to solve; advanced-level ecological problems are more complex, involving the dynamic changes of the ecosystem and the interactions between multiple elements.

[0036] For beginner players, choose simple question types focusing on basic ecological knowledge, such as multiple-choice, true / false, and simple fill-in-the-blank questions. For example, present a multiple-choice question: "Which of the following organisms is a producer in a forest ecosystem? A. Tiger B. Tree C. Mushroom." This simple question helps players solidify their basic ecological knowledge. Adjust the number and frequency of questions according to the beginner players' game pace and attention span. In the initial stages of the game, a simple question can appear every 3-5 minutes; as the game progresses, appropriately increase the number and difficulty of the questions.

[0037] For intermediate-level players, in addition to multiple-choice and fill-in-the-blank questions, some question types requiring logical reasoning and observation skills will be added, such as case analysis questions and ranking questions. For example, a case study of an abnormal change in the population of a certain organism in a marine ecosystem might be presented, asking players to analyze possible causes and propose solutions; or a series of organisms might be given, and players might be asked to rank them according to their roles in the ecosystem. The difficulty of the questions will be dynamically adjusted based on the player's performance in the game. If a player answers several intermediate-level questions correctly in a row, the complexity of the questions can be appropriately increased; if a player answers many questions incorrectly, the difficulty of the questions will be reduced, ensuring that players can gradually improve their abilities through challenges.

[0038] For advanced players, complex ecological problems are designed that require a combination of ecological knowledge, logical reasoning, and creativity to solve. For example, given a virtual ecosystem scenario with various biological and environmental factors, players are required to design an experiment to study the impact of a particular ecological factor on the ecosystem and predict the experimental results. Dynamic problems are generated in conjunction with the real-time ecological evolution within the game. For instance, after simulating the impact of a natural disaster (such as a flood) on the ecosystem, related questions are immediately generated, allowing players to analyze the disaster's impact on the food chain, resource allocation, and other aspects, and to develop corresponding strategies.

[0039] An agent behavior tree is a tree-structured model used to describe the behavior of agents (creatures or environmental elements in a game). It defines the various behaviors that an agent can take under different circumstances and the logical relationships between these behaviors. For example, in a forest ecosystem, a wolf's agent behavior tree could include behavior nodes such as finding prey, chasing prey, hunting prey, and resting, as well as the triggering conditions and execution order between these nodes.

[0040] When a player is classified as a beginner, the system simultaneously activates the beginner-level Agent behavior tree loader. The beginner-level behavior tree is relatively simple, mainly containing basic biological behaviors such as movement, foraging, and simple interactions. For example, in a marine ecosystem scenario, a small fish's Agent behavior tree might only have two main behavioral nodes: random swimming and searching for plankton to feed.

[0041] For intermediate-level players, activate the intermediate-level Agent behavior tree loader. The intermediate-level behavior tree adds more complex interactions and decision-making behaviors to the basic behaviors. For example, in a grassland ecosystem, when a lion chases an antelope, it adjusts its pursuit strategy based on the antelope's speed and direction, as well as surrounding environmental factors (such as obstacles). These decision-making behaviors are reflected in the intermediate-level behavior tree.

[0042] When a player reaches an advanced level, the advanced-level Agent behavior tree loader is activated. Advanced-level behavior trees are highly complex and dynamic, capable of simulating intricate interactions between organisms and the dynamic changes within an ecosystem. For example, in an ecosystem simulating global climate change, the Agent behavior trees of different organisms will dynamically adjust according to changes in environmental factors such as temperature and precipitation, and organisms will exhibit more complex behavioral patterns due to competition and cooperation for resources.

[0043] By activating the corresponding agent behavior tree loader, the system loads the appropriate behavior tree for each organism and environmental element in the ecological scenario, thus forming a complete agent modeling system. Within this system, each agent autonomously behaves and interacts according to its behavior tree, simulating the dynamic changes of a real ecosystem. For example, in a forest ecosystem, trees grow according to environmental factors such as light and water, animals forage and avoid predators based on food resources and predator availability, and organisms influence each other through predation, symbiosis, and other relationships to jointly maintain the balance of the ecosystem.

[0044] In this embodiment, hierarchical ecological problems can be dynamically generated based on the player's ability assessment results, and a corresponding agent modeling system can be built simultaneously. This allows the game content to better adapt to the player's ability level and provide the player with a personalized gaming experience.

[0045] S103 simulates biological interaction networks through an agent modeling system, and inputs the real-time operational data of the biological interaction networks into an LSTM dual-time-series prediction engine for prediction processing to generate an ecological evolution map.

[0046] In this embodiment, a series of attributes are assigned to each biological entity (Agent), reflecting the characteristics of the organism within the ecosystem. Taking the rabbit Agent as an example, its attributes may include age, health status, hunger level, reproductive capacity, and activity range. The age attribute records the rabbit's growth stage; rabbits of different ages may differ in behavior and survival ability. Health status affects the rabbit's mobility and resistance to disease. Hunger level determines the frequency and intensity with which the rabbit searches for food. Reproductive capacity relates to the reproduction of the rabbit population. Activity range limits the rabbit's activity area within the ecosystem.

[0047] Based on the ecological habits and behavioral characteristics of organisms, corresponding behavioral rules are defined for each Agent. The behavioral rules for the rabbit Agent can include foraging, avoiding predators, and reproduction. Foraging behavior rules are triggered by the distribution of food resources and the rabbit's hunger level. When a rabbit feels hungry and there are edible plants nearby, it will move towards the food and eat. Avoiding predators behavior rules are determined by the location and distance of predators (such as foxes) in the surrounding environment. When a fox approaches within a certain distance, the rabbit will quickly run away. Reproduction behavior rules are determined based on the rabbit's age, health condition, and reproductive capacity. When certain conditions are met, the rabbit will reproduce, producing new rabbit Agents.

[0048] In an ecosystem, predator-prey relationships exist between different organisms. For example, in a forest ecosystem, a fox preys on a rabbit. In agent-based modeling systems, this interaction is achieved by defining predator-prey relationships. When a fox agent and a rabbit agent encounter each other within a certain range, a predation behavior is determined based on the fox's hunting ability and the rabbit's evasion ability. If the fox successfully preys on the rabbit, the rabbit's status changes to "prey," and it is removed from the ecosystem, while the fox's hunger is alleviated to some extent.

[0049] Some organisms have symbiotic relationships, such as bees and flowers. Bees help flowers spread pollen and promote reproduction while collecting nectar. In the modeling system, when a bee agent approaches a flower agent, a symbiotic interaction is triggered. After the bee collects nectar, the flower's reproduction probability increases, and the bee obtains food resources and replenishes its energy.

[0050] Organisms also compete for resources such as food and habitat. For example, in grassland ecosystems, various herbivores (such as antelopes and zebras) may compete for limited forage resources. In modeling systems, when different herbivorous agents simultaneously approach a forage area, the forage resources are allocated based on their competitive ability and hunger level. Animals with stronger competitive abilities may obtain more forage, while animals with weaker competitive abilities may obtain less or no forage, thus affecting their survival and growth.

[0051] As the game progresses, interactions between the creature agents occur continuously. The system updates the status and attributes of each agent in real time to reflect the results of these interactions. For example, after a series of predation, symbiosis, and competition interactions, the rabbit population may decrease, foxes may improve their health due to abundant food, and flowers may reproduce more abundantly due to pollination by bees. These changes are recorded in each agent's attributes in real time and affect subsequent interactions.

[0052] Real-time operational data of biological interaction networks is collected, including the location information, state change information, and interaction behavior information of biological agents. Location information records the specific location of an organism in the ecosystem, such as the coordinates of a rabbit in a forest; state change information reflects changes in the attributes of an organism, such as a rabbit's hunger level changing from "hunger" to "full"; interaction behavior information records specific interaction events between organisms, such as a fox preying on a rabbit or a bee collecting nectar.

[0053] Based on the game's pace and the dynamic changes in the ecosystem, an appropriate data collection frequency should be set. During periods of slow ecosystem change, the data collection frequency can be reduced to lessen the system load; conversely, during periods of significant ecosystem change (such as natural disasters or invasive species) or frequent biological interactions, the data collection frequency should be increased to ensure accurate capture of ecosystem dynamics. The collected real-time operational data should be stored in a dedicated data storage module.

[0054] LSTM (Long Short-Term Memory) is a special type of recurrent neural network capable of handling long-term dependencies in sequential data. Employing an LSTM dual-time-series prediction engine, it can simultaneously consider historical and real-time data from biological interaction networks, leading to more accurate predictions of ecosystem evolution trends. Historical data reflects the long-term development patterns of ecosystems, while real-time data reflects immediate changes. Combining these two types of data improves the accuracy and reliability of predictions.

[0055] Before performing predictive processing, the collected real-time operational data of the biological interaction network is preprocessed. Preprocessing steps include data cleaning and data normalization. Data cleaning removes noise and outliers to ensure data quality; data normalization unifies different types of data to the same scale range, facilitating processing by the LSTM model. For example, the location coordinates of organisms are normalized to the 0-1 range, and state change information is converted into numerical form for normalization.

[0056] The LSTM dual-time-series prediction engine was trained using historical biological interaction network data. During training, the historical data was divided into training and validation sets. The model parameters (such as the number of hidden layer nodes and the learning rate) were continuously adjusted to minimize the prediction error on the training set, while the model was validated on the validation set to ensure good generalization ability.

[0057] The preprocessed real-time operational data is input into a trained LSTM dual-time-series prediction engine for ecological evolution prediction. Based on the input data and patterns in historical data, the prediction engine outputs the evolution trend of the ecosystem over a future period, including changes in biological populations and adjustments in ecological relationships.

[0058] Based on the prediction results of the LSTM dual-time-series prediction engine, the content of the ecological evolution map is planned. The ecological evolution map should include aspects such as the distribution of biological populations in the ecosystem, changes in relationships between organisms, and the evolution of the ecological environment. For example, different colors and shapes can be used to represent different biological populations in the map, arrows can be used to represent predation, symbiosis, and other relationships between organisms, and changes in background color can be used to represent changes in the ecological environment (such as a forest changing from green to yellow to indicate forest degradation).

[0059] Choose an appropriate visualization method to present the ecological evolution map to players. Two-dimensional or three-dimensional graphical interfaces can be used, combined with animation effects, to make the ecological evolution process more intuitive and vivid. For example, in a two-dimensional graphical interface, the location and quantity of biological populations will dynamically change over time, and the arrows representing the relationships between organisms will adjust accordingly; in a three-dimensional graphical interface, players can observe the evolution of the ecosystem from different angles, enhancing immersion.

[0060] S104 generates a dynamic visualization interface based on the ecological evolution map. When the ecological imbalance index is detected to exceed the current player difficulty threshold, the difficulty adjuster is triggered.

[0061] In this embodiment, a dynamic visualization interface layout is designed based on the characteristics of the ecosystem and the player's observation habits. The interface is divided into different areas to display different elements of the ecosystem. For example, a main area can be set up to display the distribution and dynamic changes of biological populations, using different icons or colors to represent different species, and animation effects to show the movement, reproduction, predation, and other behaviors of the organisms; another area can be set up to display changes in environmental factors, using dashboards or charts to intuitively present real-time changes in data such as temperature, humidity, and light; and a third area can be set up to display the relationship network between organisms, using lines to connect different organisms, with the thickness of the lines indicating the strength of the relationship.

[0062] Based on data from the ecological evolution map, the various elements in the interface are dynamically rendered. For biological populations, the size or number of icons is adjusted according to changes in their population size. For example, if the population of a certain organism increases, the number of its icons increases or its size increases accordingly; if the population decreases, the number of icons decreases or its size decreases. For environmental factors, the display status of the dashboard or charts changes according to changes in their values. For example, when the temperature rises, the pointer in the dashboard moves towards higher temperatures, and the curve in the chart rises. For the relationship network between organisms, the thickness and color of the lines are adjusted according to changes in the strength of the relationship. The stronger the relationship, the thicker the line and the darker the color; the weaker the relationship, the thinner the line and the lighter the color.

[0063] Add interactive features to the dynamic visualization interface, allowing players to interact with the interface and gain a deeper understanding of the ecosystem. For example, players can click on a creature icon, and a detailed information window will pop up, displaying the creature's name, attributes, ecological habits, etc. Players can also drag elements in the interface to observe the interactions between different elements. For instance, dragging one creature icon near another will display possible interactions between them, such as predation and symbiosis.

[0064] Select key indicators that reflect the imbalance of the ecosystem, such as the fluctuation range of biological populations, changes in biodiversity, and the degree of abnormality of environmental factors. Determine the corresponding weights based on the importance of each indicator to the ecosystem balance. For example, changes in biodiversity have a greater impact on ecosystem stability and can be assigned a higher weight; while short-term fluctuations in the population size of a single biological species have a relatively smaller impact on the ecosystem and can be assigned a lower weight.

[0065] The ecological imbalance index is calculated based on the selected indicators and determined weights. A weighted average method can be used, where the values ​​of each indicator are multiplied by their respective weights and then summed to obtain the ecological imbalance index. For example, let the indicator of population fluctuation amplitude be A, with a weight of w_1; the indicator of biodiversity change be B, with a weight of w_2; and the indicator of the degree of environmental factor anomaly be C, with a weight of w_3. Then the ecological imbalance index I = w_1*A + w_2*B + w_3*C.

[0066] Establish a real-time update mechanism for the ecological imbalance index. As the ecosystem dynamically changes, update the values ​​of each indicator in a timely manner and recalculate the ecological imbalance index. For example, at regular intervals (e.g., every minute), obtain the latest data from the ecological evolution map, update the indicator values, and then recalculate the ecological imbalance index to ensure that the index accurately reflects the current imbalance state of the ecosystem.

[0067] Based on player ability assessments, a personalized difficulty threshold for the ecological imbalance index is set for each player. Players with higher abilities can withstand a higher ecological imbalance index, thus receiving a higher difficulty threshold; players with lower abilities adapt less to ecological imbalance, receiving a lower difficulty threshold. The difficulty threshold is dynamically adjusted as players' abilities improve or decline during gameplay. If a player successfully completes a series of high-difficulty puzzles over a period of time, it indicates an improvement in their ability, and the difficulty threshold can be appropriately increased; conversely, if a player fails multiple times, demonstrating difficulty adapting to the current level, the difficulty threshold is decreased.

[0068] The system monitors the ecological imbalance index and player difficulty threshold in real time. When the ecological imbalance index exceeds the current player's difficulty threshold, a difficulty regulator is triggered. Based on the specific circumstances of the ecological imbalance and the player's skill level, corresponding difficulty adjustment strategies are formulated. Difficulty adjustment can be implemented in several ways, such as adjusting the complexity of ecological problems, altering the rules of biological interaction, and increasing or decreasing environmental disturbance factors. For example, if ecological imbalance is caused by the overgrowth of a certain species' population, the system can increase the complexity of the control problem targeting that species; or it can change the rules of interaction between that species and other species, making it more susceptible to the constraints of other organisms.

[0069] Implement difficulty adjustments according to the established strategy, and promptly provide feedback to players. During difficulty adjustments, a dynamic visual interface can be used to display the specific content and reasons for the adjustments, allowing players to understand the basis for the change in game difficulty. For example, a pop-up window can inform players that "due to an abnormal increase in the population of a certain species, leading to an imbalance in the ecosystem, control issues related to this species have been added to increase the game difficulty." Simultaneously, observe player reactions to the difficulty adjustments and further optimize the strategy based on player feedback.

[0070] S105 updates the ability assessment parameters based on the topological integrity of the player's operation chain data, and optimizes the player ability assessment model by combining reinforcement learning algorithms.

[0071] In this embodiment, every player action is recorded in real time during game execution, including the type of action (such as selecting a creature, adjusting environmental parameters, performing interactive actions, etc.), the object of the action (specifically, which creature or environmental factor), the time of the action, and the order of the actions. This information is recorded sequentially in chronological order to form the player's action chain data. For example, if a player first selects a plant, then adjusts the light intensity in the area where the plant is located, and then interacts with the plant using an insect, all these actions and their related information will be fully recorded.

[0072] The player's action chain data is transformed into a topological structure, with action types as nodes and the order and logical relationships between actions as edges, constructing a directed graph. For example, if a player first performs action A, and then performs action B based on the result of action A, then there exists an edge in the topological structure pointing from node A to node B. In this way, the logical connections and flow between player actions can be clearly displayed.

[0073] A series of metrics are defined to measure the topological integrity of the operation chain, including operational coherence, logical rationality, and goal orientation. Based on these metrics, the topological integrity of the operation chain is comprehensively evaluated. A weighted average method can be used, assigning a certain weight to each metric and summing the evaluation results to obtain a comprehensive evaluation score. The weights can be determined based on the importance of each metric in reflecting the player's ability; for example, logical rationality is more important in reflecting the player's puzzle-solving ability and can be given a higher weight.

[0074] Based on the comprehensive evaluation results of the operation chain's topological integrity, corresponding rules for adjusting ability evaluation parameters are formulated. For example, if the comprehensive evaluation score for the operation chain's topological integrity is high, it indicates that the player demonstrates strong logic, coherence, and goal orientation during puzzle-solving; therefore, the evaluation parameters for the player's logical reasoning and problem-solving abilities can be appropriately increased. Conversely, if the evaluation score is low, the corresponding evaluation parameters should be decreased. Simultaneously, considering the specific ecological knowledge and operation types involved in the operation chain, the evaluation parameters for the player's mastery of ecological knowledge are adjusted. For example, if the player correctly applies knowledge of food chains in an ecosystem during the operation, the evaluation parameters for the food chain-related parts of their ecological knowledge mastery can be increased.

[0075] According to the established parameter adjustment rules, the player's ability assessment parameters are updated in real time. Each time a player completes a puzzle or a level, the parameters are adjusted based on the latest operation chain topology integrity assessment results to ensure that the ability assessment parameters can reflect the player's current ability level in a timely and accurate manner.

[0076] This study uses an ecological balance-based puzzle game as a reinforcement learning environment. Player actions within the game are considered the agent's actions, and game outcomes (such as successful puzzle solving, time taken, and score) serve as reward signals. For example, a positive reward is given for successful puzzle solving, while a negative reward is given for failed puzzle solving or excessive time spent. Simultaneously, player ability evaluation parameters are used as state information to describe the agent's state within the current environment.

[0077] Choose a suitable reinforcement learning algorithm, such as Q-learning or deep reinforcement learning (e.g., Deep Q-Network, DQN). Use the player's ability assessment parameters as input states, the player's actions as actions, and the game result as the reward to train the reinforcement learning algorithm. During training, the algorithm continuously tries different actions and adjusts its policy based on the reward signals obtained to maximize long-term rewards. For example, the Q-learning algorithm updates the Q-table to record the expected rewards obtained by taking different actions in different states. As training progresses, the Q-table gradually converges, and the algorithm can find the optimal action policy for a given state.

[0078] By combining a trained reinforcement learning algorithm with a player ability assessment model, the optimal action strategy output by the reinforcement learning algorithm is used to guide the optimization of the ability assessment model. For example, if the reinforcement learning algorithm discovers that taking a specific action in a certain state can yield a higher reward, it indicates that this action reflects a high level of player ability. Based on this information, the evaluation parameters and weights for relevant ability dimensions in the ability assessment model can be further adjusted, making the model more accurate in assessing the player's ability in similar states. Simultaneously, new action chain data of players in the game is continuously collected, and the above steps are repeated to iteratively optimize the ability assessment model, improving its accuracy and adaptability.

[0079] In some embodiments, step S101 above, which involves receiving the ecological scene data selected by the player, combining it with a player ability assessment model to assess the player's ability level, and generating an assessment result, specifically includes: Receive ecological scene data selected by the player and collect historical operation sequences in real time through the operation chain capture device; Based on historical operation sequences, calculate strategy complexity, operation accuracy, and stability metrics. The strategy complexity metric, operational accuracy metric, and stability metric are input into the vector mapping engine to generate a normalized player ability vector. Player ability vectors are input into a pre-defined player ability assessment model for processing, resulting in an assessment result.

[0080] In this embodiment, a dedicated ecological scene selection area is set up in the game interface to display different ecological scenes in an intuitive and vivid way. For example, high-definition images, animated demonstrations, or virtual reality (VR) / augmented reality (AR) technologies are used to present diverse ecosystem scenes such as forests, grasslands, oceans, and deserts. Each scene is accompanied by a detailed text description explaining its characteristics, ecological elements, and potential puzzle challenges, helping players understand and make choices. After the player makes a selection in the scene selection interface, the game system receives the selected ecological scene data through a specific data interface. This data includes the scene's unique identifier, the main biological species in the scene, environmental parameters (such as temperature, humidity, light intensity, etc.), and the initial ecological balance state. The received data is stored in the game server's database for subsequent retrieval and analysis during gameplay.

[0081] An action chain capture device is integrated into the game system. This device, which can be a software module, is responsible for monitoring and recording every action the player takes in the game in real time. Actions include, but are not limited to, selecting creatures, adjusting environmental parameters, performing creature interactions (such as predation and symbiosis), and using items. The action chain capture device must ensure that it can accurately record information such as the timestamp of the action, the type of action, the object of the action, and the specific parameters of the action. During game execution, the action chain capture device collects the player's action information in real time according to the time sequence and combines this information into a historical action sequence. For example, if the player first selects a certain plant in the game, then moves it to another location, and then adjusts the light intensity at that location, these actions will be recorded sequentially to form an action sequence.

[0082] This involves statistically analyzing the number and variety of different types of operations in historical operation sequences. For example, if a player uses multiple different types of operations during puzzle-solving, such as selecting different types of creatures or adjusting various environmental parameters, it indicates a diverse operational strategy. This operational diversity can be measured by calculating the Shannon entropy of operation types; the higher the Shannon entropy, the richer the operation types and the higher the strategy complexity.

[0083] Analyze the sequence and logical relationships between actions in historical action sequences. For example, does the player follow reasonable ecological principles, such as first adjusting environmental factors to influence the growth of organisms, and then performing interactions between organisms? This can be achieved by establishing a rule base for action sequences, comparing the player's action sequences with reasonable sequences in the rule base, and calculating the proportion of actions that conform to the rules. A higher proportion indicates stronger logical consistency in the action sequence and a correspondingly higher strategy complexity.

[0084] Based on the puzzle objectives within the ecological scenario, assess the contribution of each player's action to achieving those objectives. For example, if the puzzle objective is to restore ecosystem balance, evaluate whether the player's actions effectively adjusted biodiversity levels or improved environmental conditions. The degree of objective achievement can be measured by calculating the changes in key ecosystem indicators (such as biodiversity indices and ecological balance indices) before and after the actions; a larger change indicates higher accuracy in the actions.

[0085] Record player errors during gameplay, such as selecting the wrong creature or making excessively large deviations in environmental parameter adjustments. Analyze the frequency and severity of these errors; lower frequency and severity indicate higher skill level. Skill level can be assessed by setting an error threshold, classifying actions exceeding the threshold as erroneous, and calculating the proportion of erroneous actions to total actions.

[0086] The system analyzes player action frequency across different time periods to identify fluctuations. For example, it examines whether players maintain a relatively stable action rhythm during puzzle solving, or whether their actions are too frequent or too infrequent. The stability of action frequency can be measured by calculating its standard deviation; a smaller standard deviation indicates a more stable action frequency.

[0087] For the same type of operation and the same object, observe whether the results of multiple player operations are consistent. For example, after a player adjusts a certain environmental parameter multiple times, observe whether the reactions of related organisms in the ecosystem are similar. If the consistency of the operation results is high, it indicates that the player has a good grasp of the operation and strong stability. The consistency of operation results can be assessed by calculating the standard deviation of the results under the same operation.

[0088] Design a vector mapping engine, a software module responsible for mapping and transforming calculated strategy complexity, operational accuracy, and stability metrics to generate a normalized player ability vector. The vector mapping engine can employ pre-defined mapping rules and algorithms to ensure that different metrics can be represented and compared on the same scale. To eliminate differences in dimensions and numerical ranges between different metrics, the strategy complexity, operational accuracy, and stability metrics are normalized. The normalized strategy complexity, operational accuracy, and stability metrics are then combined in a specific order into a three-dimensional vector, which serves as the player ability vector.

[0089] A player ability assessment model is pre-defined. This model can be trained using machine learning algorithms, such as decision tree models or neural network models. During model construction, a large amount of historical player action data and corresponding actual ability assessment results are collected as training samples to train and optimize the model, enabling it to accurately predict a player's ability level based on their ability vectors. The generated player ability vectors are input into the pre-defined player ability assessment model. The model processes and analyzes the input vector information using pre-trained parameters and algorithms. For example, if it's a decision tree model, it will classify and judge the vectors layer by layer according to the rules of decision trees; if it's a neural network model, it will perform complex calculations and transformations on the vectors through connections between neurons and activation functions. After processing by the model, a player ability assessment result is generated. The assessment result can be a specific ability score or an ability level (such as beginner, intermediate, advanced, etc.).

[0090] In this embodiment, by collecting ecological scene data and historical operation sequences selected by the player, and calculating indicators in multiple dimensions, it is possible to more accurately reflect the player's strategy formulation ability, operation skills, and stability.

[0091] Furthermore, the step of inputting the player's ability vector into a preset player ability evaluation model for processing to form an evaluation result specifically includes: It receives the input of the player's ability vector and deconstructs it into a strategy complexity feature dimension, an operation accuracy feature dimension, and a stability feature dimension according to a preset weight ratio; The strategy complexity feature dimension, operation accuracy feature dimension, and stability feature dimension are input into the pre-built capability classification model for processing, and basic capability classification results are generated. Based on the basic ability classification results, a dimensional balance analysis and potential ability prediction are performed using a pre-built player ability assessment model, generating balance analysis results and potential ability prediction results. A subject imbalance index is generated based on the results of the balance analysis. When the subject imbalance index exceeds the preset subject imbalance index threshold, the ability correction requirement is marked. The growth coefficient is determined based on the potential ability prediction results. The ability correction needs and the growth coefficient are combined to calculate and output the player's ability rating, and the evaluation result is obtained.

[0092] In this embodiment, a capability classification model is pre-constructed, which can employ a decision tree algorithm. First, a large amount of historical capability data of players is collected, including specific values ​​for strategy complexity, operational precision, and stability, as well as the corresponding actual capability levels (e.g., beginner, intermediate, advanced). Then, this data is used to train the decision tree model, determining the node splitting rules and classification boundaries. For example, using the strategy complexity value as the first splitting node, when the strategy complexity is less than 0.3, the model enters the left subtree for further splitting based on operational precision; when the strategy complexity is greater than or equal to 0.3, the model enters the right subtree for splitting based on stability, ultimately classifying player capabilities into three basic categories: beginner, intermediate, and advanced.

[0093] The deconstructed strategy complexity, operational precision, and stability feature dimensions are input into the trained ability classification model. The model processes and analyzes the input data according to preset rules and algorithms, and classifies players into corresponding ability levels based on the numerical combinations of each feature dimension, generating basic ability classification results, such as beginner, intermediate, or advanced players.

[0094] Design a dimensional balance analysis method to calculate the degree of difference between the values ​​of three feature dimensions. For example, calculate the absolute value of the difference between strategy complexity and operational accuracy, strategy complexity and stability, and operational accuracy and stability, and then calculate the average of these absolute values. Compare this average value with a preset balance threshold. If the average value is less than the threshold, the player's ability dimensions are considered relatively balanced; if the average value is greater than or equal to the threshold, the player's ability is considered to be unbalanced. Simultaneously, record the specific dimensions of the imbalance, such as strategy complexity being much higher than operational accuracy and stability, or operational accuracy being much lower than the other two dimensions.

[0095] A method based on historical data and trend analysis is used to predict potential abilities. Player ability data is collected across multiple game levels or time periods, and the trends in ability changes are analyzed. For example, the improvement in a player's strategy complexity, operational precision, and stability over several consecutive levels is observed. If a player shows a continuous upward trend in a certain dimension, and the increase is significant, then the player is predicted to have high potential ability in that dimension. Simultaneously, by combining the results of basic ability classification and comprehensively considering the player's current ability level and potential development trend, a potential ability prediction result is generated, such as predicting the ability level the player may reach in the next few levels.

[0096] Based on the degree of difference in the numerical values ​​of the feature dimensions obtained from the dimensional balance analysis, a method for calculating the subject imbalance index is designed. For example, the subject imbalance index is obtained by dividing the difference between the maximum and minimum values ​​of the three feature dimensions by their average value. The larger the subject imbalance index, the more severe the player's subject imbalance.

[0097] A threshold for a subject-specific imbalance index is preset. When the index is detected to exceed this threshold, the assessment result indicates a need for ability correction. For example, if the threshold is set to 0.5, and the calculated index is 0.6, the player is marked as needing ability correction, and the assessment result details the specific dimensions and degree of the imbalance so that the game system can subsequently provide the player with targeted ability improvement suggestions or training content.

[0098] Based on the predicted potential abilities, a growth coefficient is determined for each player. If a player is predicted to have high potential abilities, a higher growth coefficient is assigned; if a player is predicted to have average potential abilities, a lower growth coefficient is assigned. For example, the growth coefficient is divided into three levels: 1.2 for high potential abilities, 1.0 for medium potential abilities, and 0.8 for low potential abilities.

[0099] The player's ability rating is calculated by combining the need for ability correction (which is quantified if applicable; for example, if the subject imbalance index exceeds a threshold, the index is converted into a negative factor affecting the ability rating at a certain ratio) with the growth coefficient. For example, based on the basic ability classification result, if the basic ability classification is intermediate player, the growth coefficient is 1.2, and there is no need for ability correction, the player's ability rating may be upgraded to intermediate-advanced player; if there is a need for ability correction, and the negative factor corresponding to the subject imbalance index is 0.1, and the growth coefficient is 0.8, the player's ability rating may be downgraded to beginner-intermediate player. The final generated player ability rating serves as the evaluation result and is fed back to the game system for dynamically adjusting game difficulty, providing personalized game content and suggestions, etc.

[0100] In some embodiments, step S102 above, which involves dynamically generating hierarchical ecological problems based on the evaluation results and simultaneously activating the corresponding level of the Agent behavior tree loader to form an Agent modeling system, specifically includes: The strategy complexity index in the evaluation results is analyzed, and a basic problem framework is selected from the ecological problem template library based on the strategy complexity index. The operational accuracy indicators in the evaluation results are analyzed, and dynamic perturbation parameters are injected into the basic problem framework based on the operational accuracy indicators to form a hierarchical ecological problem. Analyze the stability indicators in the evaluation results and determine the depth of the Agent behavior tree based on the stability indicators; Modular behavior units are configured based on the depth of the Agent behavior tree to generate a hierarchical behavior tree; By coupling hierarchical ecological problems with hierarchical behavior trees, an Agent modeling system is formed that includes behavior tree depth parameters, ecological tolerance thresholds, and early warning time windows.

[0101] In this embodiment, a rich ecological problem template library is constructed, storing various basic ecological problem frameworks of different types and difficulties. Each basic problem framework defines the basic scenario of the ecological problem, the ecological elements involved, and the initial state. For example, the template library may include a food chain balance problem framework in a forest ecosystem, a water resource allocation problem framework in a grassland ecosystem, etc. The basic problem frameworks are categorized and stored according to difficulty levels, with each difficulty level corresponding to a range of strategy complexity indicators. Based on the parsed strategy complexity indicator values, matching and selection are performed within the ecological problem template library. Different strategy complexity indicator ranges are set to correspond to different difficulty levels of basic problem frameworks. For example, when the strategy complexity indicator is between 0 and 0.3, a basic problem framework of beginner difficulty is selected; when the indicator is between 0.3 and 0.6, a basic problem framework of intermediate difficulty is selected; and when the indicator is between 0.6 and 1, a basic problem framework of advanced difficulty is selected. This method ensures that the selected basic problem framework matches the player's strategy-making ability.

[0102] A set of dynamic perturbation parameters related to operational accuracy indicators is designed. These parameters are used to adjust the values ​​or states of ecological elements in the basic problem framework, increasing the complexity and uncertainty of the problem. For example, dynamic perturbation parameters may include the fluctuation range of biological populations and the rate of change of environmental parameters. Different operational accuracy indicator ranges correspond to different dynamic perturbation parameter values. Based on the extracted operational accuracy indicator values, the corresponding dynamic perturbation parameter values ​​are determined and injected into the selected basic problem framework. For example, if the operational accuracy indicator is low, a larger fluctuation range of biological populations and a faster rate of change of environmental parameters are injected to make the problem more challenging; if the operational accuracy indicator is high, a smaller fluctuation range and a slower rate of change are injected to make the problem moderately difficult. By injecting dynamic perturbation parameters, a hierarchical ecological problem is formed on the basis of the basic problem framework, with the difficulty and complexity of the problem matching the player's operational accuracy ability.

[0103] Establish association rules between the depth of the Agent behavior tree and stability metrics. The Agent behavior tree is a tree-like structure describing the agent's behavioral logic, and its depth represents the complexity and decision-making level of the behavior. Generally, a higher stability metric indicates more balanced player abilities and more stable operations, allowing the agent to employ more complex behavioral strategies; therefore, a larger behavior tree depth can be set. Conversely, a lower stability metric requires a smaller behavior tree depth. For example, a stability metric between 0 and 0.3 corresponds to a 3-level Agent behavior tree depth; between 0.3 and 0.6, a depth of 5 levels; and between 0.6 and 1, a depth of 7 levels. The depth of the Agent behavior tree is determined according to the extracted stability metric values ​​and the association rules.

[0104] Construct a modular behavior unit library, storing various basic agent behavior units such as environmental perception, decision-making, and action execution. Each behavior unit has clearly defined input and output interfaces for easy combination and invocation. Behavior units are categorized and stored according to their functional type, such as perception, decision-making, and action. Design a hierarchical behavior tree generation algorithm. This algorithm selects appropriate behavior units from the modular behavior unit library based on a determined agent behavior tree depth, combining and connecting them to form a hierarchical behavior tree structure. The algorithm constructs each node of the behavior tree sequentially from top to bottom and left to right, with each node representing a behavior unit or sub-behavior tree. For example, the root node of the behavior tree can be a decision-making behavior unit, and its child nodes can be multiple perception and action behavior units, thus constructing a hierarchical behavior tree. Run the hierarchical behavior tree generation algorithm, selecting suitable behavior units from the modular behavior unit library based on the determined agent behavior tree depth to generate a hierarchical agent behavior tree.

[0105] The design incorporates parameters for coupling hierarchical ecological problems and hierarchical behavior trees, including behavior tree depth parameters, ecological tolerance thresholds, and warning time windows. The behavior tree depth parameter ensures that the depth of the agent's behavior tree matches the complexity of the hierarchical ecological problems; the ecological tolerance threshold defines the permissible range of change for each element in the ecosystem, requiring the agent to take appropriate adjustments when an ecological element exceeds this range; and the warning time window sets the time interval for the agent to detect ecological problems and respond.

[0106] Design a coupled processing algorithm that integrates hierarchical ecological problems and hierarchical behavior trees. The algorithm takes relevant information about the ecological problem (such as the state of ecological elements and problem objectives) as input to the agent's behavior tree, and simultaneously feeds the execution results of the behavior tree back to the ecological problem model, achieving dynamic interaction and collaboration between the two. For example, the algorithm can adjust the agent's decision-making and execution processes in real time according to changes in the ecological problem, ensuring that the agent can effectively respond to the ecological problem.

[0107] A coupled processing algorithm is used to couple hierarchical ecological problems and hierarchical behavior trees, forming an Agent modeling system that includes behavior tree depth parameters, ecological tolerance thresholds, and early warning time windows. This system can dynamically generate suitable ecological problems and corresponding Agent behavior patterns based on player ability assessment results, providing players with a personalized and intelligent gaming experience.

[0108] In this embodiment, the Agent modeling system can dynamically adjust its behavioral strategies according to changes in ecological issues, enabling effective interaction and collaboration with players and providing them with a more realistic and intelligent gaming experience.

[0109] In some embodiments, step S103 above, which involves simulating a biological interaction network using an Agent modeling system and inputting the real-time operational data of the biological interaction network into an LSTM dual-time-series prediction engine for prediction processing to generate an ecological evolution map, specifically includes: Simulate biological interaction networks using an agent-based modeling system to generate real-time streams of species interaction events. Convert species interaction event streams into time-series structured data and encapsulate them into sliding time window data packets; The sliding time window data packets are input into the LSTM dual time series prediction engine to perform short-term ecological change prediction and obtain short-term prediction results. Based on the short-term forecast results, long-term ecosystem effect forecasts are triggered, generating long-term forecast results. The short-term and long-term forecast results are then merged to form an ecological evolution map.

[0110] In this embodiment, a corresponding Agent instance is created for each species of creature in the game. Each Agent instance contains basic attribute information about the creature, such as species name, survivability, reproductive capacity, and food preferences. Simultaneously, behavioral rules for the Agents are defined, including foraging behavior, reproductive behavior, and predator avoidance behavior. For example, for the Rabbit Agent, its foraging behavior is set to search for grassland resources, its reproductive behavior is set to occur when certain energy conditions are met, and its predator avoidance behavior is set to flee when a Fox Agent is detected approaching.

[0111] The agent modeling system is activated, allowing agent instances to interact within a virtual ecosystem according to predefined behavioral rules. During these interactions, real-time monitoring and recording of interspecies interaction events are conducted, such as predation (fox preying on rabbit), competition (two herbivores vying for grassland resources), and symbiotic relationships (mutualistic relationships between certain plants and insects). These interaction events are then arranged chronologically to form a stream of species interaction events. For example, at one moment, "fox A preys on rabbit B," and at the next moment, "rabbit C and rabbit D compete for grassland E," etc., are recorded.

[0112] Design a data transformation module to convert event information in a species interaction event stream into time-series structured data. Define a set of feature fields for each event, such as event type (predation, competition, symbiosis, etc.), species involved, time of occurrence, and location of occurrence. Organize and store each event according to these feature fields to form a time-series structured data sequence.

[0113] Determine the size of the sliding time window and the sliding step size. The sliding time window size is set based on the complexity of the game ecosystem and the forecasting needs, for example, set to 10 time units (e.g., 10 minutes). The sliding step size represents the time interval for each window movement, and can be set to 1 time unit (e.g., 1 minute).

[0114] According to the set sliding time window size and sliding step, data segments are sequentially extracted from the time-series structured data sequence, and each data segment is encapsulated into a sliding time window data packet. Each data packet contains structured data information of all events within the window, and records the start and end times of the window. For example, the first sliding time window data packet contains all event data from 2024-01-01 09:50:00 to 2024-01-01 10:00:00.

[0115] A dual-time-series prediction engine based on LSTM (Long Short-Term Memory) networks is constructed. This engine comprises two parallel LSTM network branches, each processing data at different time scales. One branch handles shorter timescales (e.g., minutes) to capture short-term dynamic changes in the ecosystem; the other branch handles longer timescales (e.g., hours) to consider long-term trends and periodic changes in the ecosystem. The outputs of the two branches are combined through a fusion layer to obtain the final prediction result.

[0116] The time-series structured data in the sliding time window data packets undergoes preprocessing, including data normalization and missing value imputation. The preprocessed sliding time window data packets are then input into the short-timescale branch of the LSTM dual-time-series prediction engine. The engine learns short-term change patterns in the ecosystem based on the input data and predicts potential species interaction events and ecological parameter changes within the next sliding time window. The prediction results are then output as short-term predictions.

[0117] Design a trigger mechanism to initiate long-term food chain effect predictions when certain key events or changes in ecological parameters exceed a preset threshold in short-term predictions. For example, if short-term predictions show a sharp decline in the population of a species within a short period, potentially triggering a chain reaction among species upstream and downstream in the food chain, then long-term food chain effect predictions would be triggered.

[0118] Once the long-term food chain effect prediction is triggered, the short-term prediction results and relevant historical data are input into the longer-term branch of the LSTM dual-time-series prediction engine. The engine comprehensively considers the long-term trends and short-term changes of the ecosystem to predict the evolution of the ecosystem over a longer period (such as several hours or days), including long-term changes in species abundance and ecosystem stability. The prediction results are then output as the long-term prediction outcome.

[0119] Short-term and long-term forecasts are combined and visualized to generate an ecological evolution map. This map graphically displays changes in the population size of individual species, interactions between species, and the overall evolutionary trend of the ecosystem. For example, line graphs show changes in species population size over time, and network diagrams illustrate interactions between species. Through this ecological evolution map, game developers and players can intuitively understand the dynamic changes in the ecosystem, providing a basis for formulating and adjusting game strategies.

[0120] Furthermore, the step of triggering long-term ecosystem effect prediction based on short-term prediction results, generating long-term prediction results, and merging the short-term and long-term prediction results to form an ecological evolution map specifically includes: Receive short-term ecological prediction results and determine the calculation path for long-term predictions based on the direction of change in the short-term prediction results; When the rate of change of a key species is identified as positive and exceeds a preset first rate of change threshold, the reproductive chain prediction process is performed through the calculation path to obtain the first long-term prediction result. When the rate of change of a key species is identified as negative and exceeds the preset second rate of change threshold, extinction chain prediction processing is performed through the calculation path to obtain the second long-term prediction result. The short-term forecast results and the first or second long-term forecast results are aligned along the time axis, and the data at the connection point are smoothed and calibrated using the species association strength coefficient to generate an ecological evolution map. In this ecological evolution map, the node size is used to represent the population size ratio of species, the node color is used to represent the species change trend, the line thickness of the ecological evolution map is used to represent the energy transfer intensity of species interaction, and the line style is used to represent the predation type or competition type of species interaction.

[0121] In this embodiment, the change direction information in the short-term prediction results is parsed out, and the change direction is divided into positive change (increase in species numbers) and negative change (decrease in species numbers). Based on the change direction, the system selects the corresponding long-term prediction calculation path from a preset calculation path library. The calculation path library stores the prediction algorithms and processes corresponding to different change directions. For example, the positive change direction corresponds to the breeding chain prediction path, and the negative change direction corresponds to the extinction chain prediction path.

[0122] In short-term predictions, the system identifies key species using pre-defined key species identification rules. Key species are typically those that significantly impact ecosystem stability, such as top predators and basic producers. After identifying key species, their rate of change is extracted.

[0123] When a key species' rate of change is identified as positive and exceeds a preset first threshold, the system triggers reproductive chain prediction processing. Following the selected reproductive chain prediction calculation path, the system first analyzes the key species' reproductive characteristics, including reproductive cycle, reproductive success rate, and offspring survival rate. Then, it considers the impact of factors such as food availability and environmental suitability on reproduction. By simulating the key species' reproductive process over a period of time, the system predicts its population growth trend and its impact on upstream and downstream species. For example, if the key species is a rabbit, and its positive rate of change exceeds the threshold, it predicts that its large-scale reproduction will lead to a reduction in grassland resources, thereby affecting the populations of other herbivores that feed on grasslands. The prediction results are used as the first long-term prediction outcome.

[0124] When a critical species' rate of change is identified as negative and exceeds a preset second rate of change threshold, the system performs extinction chain prediction processing. Based on the extinction chain prediction calculation path, the system analyzes the causes of the critical species' extinction, such as food shortages, excessive predators, and disease transmission. Simultaneously, it considers the cascading effects of its extinction on other species in the ecosystem. For example, if the critical species is the wolf, and its negative rate of change exceeds the threshold, its extinction is predicted to lead to a significant increase in the population of herbivores such as rabbits, resulting in excessive damage to grassland resources and disrupting the balance of the entire ecosystem. The prediction results are used as a second long-term prediction outcome.

[0125] The system aligns short-term forecasts with either the first or second long-term forecasts along a timeline. A unified time base is established, and data from corresponding time points in different forecasts are matched.

[0126] After timeline alignment, the data at the transition points are smoothed and calibrated. The system introduces a species association strength coefficient, which reflects the degree of mutual influence between different species. Based on this coefficient, the data at the transition points is adjusted to make the transition between short-term and long-term forecasts more natural and reasonable. For example, if a species' population shows a slight increase at a certain point in the short-term forecast, but a significant increase in the population in the long-term forecast, the data at the transition point is appropriately adjusted based on the species association strength coefficient to avoid abrupt data changes.

[0127] The system uses species as nodes, with node size representing the proportion of a species' population size. The system determines node size according to a certain proportional relationship based on the quantity information of each species in the prediction results. For example, species with larger populations have relatively larger nodes, while species with smaller populations have relatively smaller nodes. Node color represents the trend of species change; green indicates positive change (increase in quantity), and red indicates negative change (decrease in quantity). The intensity of the color reflects the magnitude of the change.

[0128] The lines connecting nodes represent species interactions. The thickness of the lines indicates the intensity of energy transfer in the interaction; the greater the energy transfer, the thicker the line. For example, the energy transfer between apex predators and prey is relatively strong, resulting in a thicker line; while the lines between species with weaker competitive relationships are thinner. The line style indicates the type of predation or competition in the interaction; solid lines represent predation, and dashed lines represent competition.

[0129] Based on the design rules for the nodes and connections mentioned above, visualization tools (such as the game's built-in graphics rendering engine) are used to plot the prediction results into an ecological evolution map. The ecological evolution map visually displays the changes in the number of species in the ecosystem, the interactions between species, and the overall evolutionary trend of the ecosystem in an intuitive graphical way.

[0130] In some embodiments, step S104 above, which involves generating a dynamic visualization interface based on an ecological evolution map and triggering a difficulty adjuster when an ecological imbalance index exceeds the current player difficulty threshold, specifically includes: Real-time monitoring of early warning signals in ecological evolution maps, extracting multi-dimensional elements including species extinction risk, number of food chain breaks and resource depletion rate; The real-time ecological imbalance index is calculated based on multi-dimensional factors and compared with the current player difficulty threshold; When the ecological imbalance index is detected to exceed the current player difficulty threshold, the corresponding difficulty adjustment strategy is selected based on the degree of excess, according to the difficulty regulator. Implement a difficulty adjustment strategy, generate new difficulty parameters, and synchronously feed these new difficulty parameters back to the game system.

[0131] In this embodiment, a graphical user interface (GUI) design tool is used in the game system to build the basic framework of a dynamic visualization interface. This framework includes a main display area for displaying the ecological evolution map, and auxiliary functional areas such as a status bar and operation buttons. The ecological evolution map is integrated into the main display area. Through data binding technology, the nodes and lines in the ecological evolution map can be dynamically updated based on real-time prediction data. For example, when the population of a species changes, the size and color of the corresponding node will be adjusted in real time; when the interaction relationships between species change, the style and thickness of the lines will also change accordingly. Warning signal markers are designed in the ecological evolution map. For species or areas facing extinction risks, food chain disruptions, or resource depletion, specific colors, shapes, or icons are used for marking. For example, a flashing red exclamation mark is used to mark species at risk of extinction, and a dashed box is used to mark areas where the food chain is disrupted.

[0132] The system periodically scans ecological evolution maps to identify species with early warning signals of extinction risk. It extracts relevant information about these species, including current population size, habitat suitability, and the number of natural enemies. Based on a pre-defined species extinction risk assessment model, it calculates the extinction risk value for each species. For example, if a species' population continues to decline, its habitat deteriorates, and the number of natural enemies increases, its extinction risk value is high.

[0133] Analyze the connections between species in an ecological evolution map and count the number of food chain breaks. A food chain break is defined as the disappearance or transformation of the connection between two species that were originally in a predatory or competitive relationship. Record the location of the food chain break and the information of the species involved.

[0134] For resource nodes in an ecosystem (such as grasslands and water sources), monitor changes in their resource quantities. Calculate the rate of resource depletion by measuring the decrease in resource quantity per unit time. For example, if the area of ​​a grassland decreases significantly in a short period, its rate of resource depletion is relatively fast.

[0135] Based on game design and ecological principles, different weights are assigned to three multidimensional factors: species extinction risk, number of food chain breaks, and resource depletion rate. For example, species extinction risk may be given a higher weight because it has a greater impact on ecosystem stability; the number of food chain breaks has a lower weight; and the resource depletion rate has a relatively lower weight. The value of each multidimensional factor is multiplied by its corresponding weight, and the results are summed to obtain the real-time ecological imbalance index. For example, if the species extinction risk value is A with a weight of a; the number of food chain breaks is B with a weight of b; and the resource depletion rate is C with a weight of c, then the real-time ecological imbalance index I = aA + bB + cC.

[0136] At the start of the game, based on the player's selected difficulty level (e.g., easy, medium, hard), the corresponding current player difficulty threshold is retrieved from a preset difficulty threshold database. This database stores the upper limits of the ecological imbalance index for different difficulty levels. The calculated real-time ecological imbalance index is compared to the current player difficulty threshold. If the real-time ecological imbalance index is less than or equal to the current player difficulty threshold, the ecosystem is considered to be in a relatively balanced state; if the real-time ecological imbalance index is greater than the current player difficulty threshold, the ecosystem is considered to be in an unbalanced state.

[0137] A difficulty adjustment strategy library is built within the difficulty adjuster. This library stores various difficulty adjustment strategies for different levels of exceeding the limit. For example, when the level of exceeding the limit is small, the strategy library may include strategies such as increasing game time limits and reducing resource generation speed; when the level of exceeding the limit is large, the strategy library may include strategies such as introducing new predator species and reducing player character abilities. The extent by which the real-time ecological imbalance index exceeds the current player difficulty threshold is calculated, i.e., the level of exceeding the limit S = I − T (where I is the real-time ecological imbalance index and T is the current player difficulty threshold).

[0138] Based on the calculated over-limit range, select the corresponding difficulty adjustment strategy from the difficulty adjustment strategy library. For example, if the over-limit range is between 0% and 10%, choose the strategy of increasing the game time limit; if the over-limit range is between 10% and 30%, choose the strategy of reducing the resource generation speed and introducing a small number of new natural enemies; if the over-limit range exceeds 30%, choose the strategy of reducing player character abilities and significantly adjusting the ecosystem.

[0139] Based on the selected difficulty adjustment strategy, relevant parameters in the game are adjusted. For example, if the strategy of increasing the game time limit is selected, the game's time counter parameter is modified; if the strategy of introducing a new natural enemy species is selected, a new natural enemy Agent instance is created in the ecosystem, and its attributes and behavioral rules are set. The adjusted parameters are then integrated to generate a new set of difficulty parameters. This new set of difficulty parameters includes parameters related to game time limits, resource generation speed, species attributes, player character abilities, and more. Through the game system's internal communication mechanism, the new set of difficulty parameters is transmitted to the relevant modules of the game system. For example, the new resource generation speed parameter is transmitted to the resource management module, the new species attribute parameter to the ecosystem simulation module, and the new player character ability parameter to the player control module. Each module of the game system updates according to the received new difficulty parameters, allowing the game difficulty to adjust in real time to adapt to imbalances in the ecosystem.

[0140] In some embodiments, step S105 above, which involves updating the ability evaluation parameters based on the topological integrity of the player's action chain data and optimizing the player ability evaluation model using a reinforcement learning algorithm, specifically includes: Real-time capture of player action chain data; analysis of nodes, paths, and loops in the topology of player action chain data. Based on nodes, paths, and loops in the topology, a multi-dimensional index including node coverage, path completeness, and loop health value is calculated. The player's ability parameters are dynamically adjusted based on multi-dimensional indicators to generate an updated ability vector. The updated ability vectors are input into a reinforcement learning network for processing, and the weight parameters of the player ability evaluation model are optimized based on the feedback results from the reinforcement learning network.

[0141] In this embodiment, a dedicated operation chain data capture module is set up in the game system, which is tightly integrated with the game's input system. When the player performs game operations (such as clicking, swiping, pressing buttons, etc.), the input system transmits the operation signal to the operation chain data capture module. The operation chain data capture module records each player operation in a time sequence, including operation type (such as movement, attack, use of items, etc.), operation object (such as game characters, items, scene elements, etc.), operation timestamp, and other information. This information is combined according to the chronological order of the operations to form the player's operation chain data.

[0142] Each action in the player's action chain data is considered a node in the topology. For example, in a role-playing game, the player's character, encountered monsters, and used items can all be considered nodes. By analyzing the action chain data, all distinct action objects are extracted and assigned a unique identifier as the node's identifier.

[0143] Determine the connection paths between nodes based on the order of operations. If a player operates on node A first, and then on node B, then there exists a path from node A to node B. Record all the connection relationships between nodes to form a path network of operation chains.

[0144] The algorithm detects loops in the path network. A loop is a path that starts from a node, performs a series of operations, and returns to that node. For example, a player attacks monster A, then uses an item to heal themselves, and then attacks monster A again, forming a simple loop. The algorithm traverses the path network using algorithms such as depth-first search to detect and record all loops.

[0145] Based on the game's design and rules, determine the set N of all possible operable nodes. This set contains all elements that players can potentially interact with in the game. Count the actual number of nodes n involved in the player's action chain data, and then calculate the node coverage C = n / |N|, where |N| represents the total number of nodes in the node set N. Node coverage reflects the extent to which players explore and utilize different operable objects in the game.

[0146] Based on the game's strategy and objectives, a theoretically optimal path set P is constructed for players to complete game tasks or achieve specific goals. This set can be obtained through the experience of game designers, expert analysis, or summarizing a large amount of data on the actions of excellent players. Paths in the player's action chain data are matched against the optimal path set P, and the number p of paths in the player's action chain that are similar to the ideal path, as well as the coverage of these similar paths in the ideal path set, are counted. Path completeness M is calculated based on the matching results. For example, it can be calculated using M = p / |P| with weighted calculations for coverage. The specific weights can be adjusted according to the characteristics and needs of the game. Path completeness reflects the strategic and rational nature of the player's actions.

[0147] Based on the impact of loops on game progress, loops are categorized into beneficial loops, neutral loops, and harmful loops. For example, a player attacking monsters through a reasonable skill rotation is a beneficial loop; a player repeatedly switching items unnecessarily is a neutral loop; a player getting stuck in an infinite attack loop due to operational errors and unable to inflict effective damage is a harmful loop. Different health value weights are assigned to different types of loops: beneficial loops have a positive weight, neutral loops have a weight of 0, and harmful loops have a negative weight. The number of each type of loop in the player's operation chain is counted, and the loop health value H is calculated based on the weights. For example, if the number of beneficial loops is h_1 with a weight of x; the number of neutral loops is h_2 with a weight of y; and the number of harmful loops is h_3 with a weight of z, then the loop health value H = x*h_1 + y*h_2 + z*h_3. The loop health value reflects the efficiency of the player's operations and their ability to avoid ineffective or erroneous operations.

[0148] Based on the game's characteristics and requirements, a set of player ability parameters A = {a_1, a_2, ..., a_m} is defined, where each parameter a_i represents a player's ability, such as operational proficiency, strategic planning ability, and reaction speed. A mapping relationship and adjustment rules between multi-dimensional indicators and ability parameters are established. For example, higher node coverage may indicate a player's strong exploration ability, allowing for an appropriate increase in operational proficiency and strategic planning ability parameters; higher path completeness indicates a reasonable player strategy, allowing for an increase in strategic planning ability and reaction speed parameters; higher loop health indicates high player operational efficiency, allowing for an increase in operational proficiency and reaction speed parameters. Adjustment ranges for ability parameters are set according to different indicator value ranges. For example, when node coverage is between 0 and 0.3, the operational proficiency parameter increases by 0.1; when node coverage is between 0.3 and 0.6, the operational proficiency parameter increases by 0.2, and so on. Based on the values ​​of the multidimensional indicators and the established adjustment rules, the player's ability parameters are adjusted to obtain the updated set of ability parameters A'={a_1',a_2',⋯,a_m'}. Design a reinforcement learning network, which can employ common reinforcement learning architectures such as Deep Q-Networks (DQNs) and Policy Gradient Networks. The network input is the updated player ability vector, and the output is the player's optimal action strategy or ability evaluation value in the current game state. Initialize the weight parameters W of the reinforcement learning network; these weight parameters will be continuously optimized through training. Input the updated ability vector into the reinforcement learning network, and the network calculates and outputs the prediction result based on the current weight parameters. Define a reward function based on the player's performance in the actual game and the game objective. The reward function can be designed based on factors such as the player's score, level completion status, and operational efficiency. For example, a higher positive reward is given for successful level completion, while a negative reward is given for mistakes or failures. Update the weight parameters of the reinforcement learning network using reinforcement learning algorithms (such as Q-learning, policy gradient algorithms, etc.) based on the reward value calculated by the reward function. By continuously inputting the ability vector, calculating rewards, and updating weight parameters, the reinforcement learning network gradually learns the relationship between player ability and optimal action strategy. After the reinforcement learning network is optimized, its weight parameters are applied to the player ability evaluation model. The player ability assessment model can more accurately evaluate a player's skill level based on the input player ability vector and optimized weight parameters. By periodically repeating the above steps, continuously updating the ability vector based on new player action chain data, and then re-inputting it into the reinforcement learning network for training and optimization, the player ability assessment model can adapt to changes in player actions and dynamic adjustments in the game environment in real time.

[0149] Furthermore, the updated ability vector is input into a reinforcement learning network for processing, and the weight parameters of the player ability evaluation model are optimized based on the feedback results of the reinforcement learning network. Specifically, this includes: The updated player ability vectors are input into the reinforcement learning policy network to generate action suggestions and their weight distributions. Based on the operational suggestions and their weight distribution, the actual utility value of the operational suggestions is calculated by analyzing the performance data of subsequent levels through reinforcement learning value networks. Feedback results are generated based on actual utility values. A weight adjustment strategy is selected based on the feedback results. The weight update is performed using a gradient backpropagation mechanism, and the weight parameters of the optimized player ability evaluation model are output.

[0150] In this embodiment, a deep neural network is designed as the reinforcement learning policy network, comprising an input layer, multiple hidden layers, and an output layer. The input layer receives the updated player ability vector, and its number of neurons matches the dimension of the ability vector. Hidden layers, using fully connected or convolutional layers, are used for feature extraction and transformation of the input data. The number of hidden layers and the number of neurons per layer can be adjusted according to the game's complexity and computational resources. The output layer consists of multiple neurons, each corresponding to a possible action suggestion in the game. For example, in a role-playing game, action suggestions could include attacking, defending, using skills, or moving. The output value of each neuron in the output layer represents the weight of the corresponding action suggestion; a larger weight value indicates a higher probability that the action suggestion will be recommended.

[0151] Once the updated player ability vector is obtained, it is input into the reinforcement learning policy network. The network performs a series of linear transformations and non-linear activation functions on the input vector through forward propagation, ultimately generating a weight distribution for each action suggestion at the output layer. For example, if the output layer has four neurons, corresponding to the four action suggestions of attack, defend, use a skill, and move, the output weight values ​​might be [0.3, 0.2, 0.4, 0.1]. This indicates that, given the player's current ability state, the action suggestion of using a skill has the highest weight and is most likely to be recommended.

[0152] Another deep neural network is constructed as the reinforcement learning value network. Its structure is similar to the policy network, also containing an input layer, hidden layers, and an output layer. The input layer receives game state information and player action suggestions as input. Game state information can include level difficulty, monster distribution, player health, etc. The hidden layers are also used for feature extraction and transformation. The output layer has only one neuron, and its output value represents the expected reward value for executing the corresponding action suggestion in the current game state, i.e., the actual utility value of the action suggestion.

[0153] Players select and execute actions based on action suggestions and their weight distributions generated by a reinforcement learning policy network. During gameplay, performance data for subsequent levels is collected, including scores, completion times, deaths, and resource usage. This collected performance data, along with game state information, is input into the reinforcement learning value network. The value network calculates the actual utility value of each action suggestion through forward propagation. For example, if a player executes an action suggestion that uses a skill and achieves a high score and completes the level quickly, the value network may output a high actual utility value; conversely, if the player performs poorly after executing the action, the value network will output a low actual utility value.

[0154] Feedback is generated based on the actual utility values ​​of the action suggestions calculated by the reinforcement learning value network. This feedback can be a comprehensive evaluation metric, such as a weighted average of the actual utility values ​​of all action suggestions to obtain an overall score reflecting the player's effectiveness in the current situation. Alternatively, feedback can be generated separately for each action suggestion, clearly indicating which suggestions were effective and which were ineffective, providing more detailed information for subsequent weight adjustments.

[0155] Develop a weight adjustment strategy based on feedback results. If the feedback indicates that the player's actions were effective (i.e., the overall score is high or the actual utility value of most action suggestions is high), it suggests that the current weight distribution of the player's ability vector and the policy network is reasonable. In this case, a small weight adjustment can be made to fine-tune the weights related to these effective action suggestions in the policy and value networks. Conversely, if the feedback indicates that the player's actions were ineffective (i.e., the overall score is low or the actual utility value of most action suggestions is low), it suggests that the current weight distribution may be problematic, requiring a larger weight adjustment to significantly modify the relevant weights in the policy and value networks.

[0156] A gradient backpropagation mechanism is employed to update the weight parameters of the reinforcement learning policy network and value network. The objective is to minimize the loss function, which can be defined based on the difference between the feedback result and the expected target; for example, a mean squared error loss function can be used. During backpropagation, the gradient of the loss function with respect to the output layer of the value network is first calculated, and then the gradients with respect to the hidden and input layers are calculated layer by layer according to the chain rule. Based on the calculated gradient values, the weight parameters of the value network are updated using gradient descent or other optimization algorithms (such as the Adam algorithm). For the reinforcement learning policy network, the gradient is also calculated and the weight parameters are updated through backpropagation based on the actual utility value fed back by the value network and the loss function. During the update process, the relationship between the weight distribution of the action suggestions and the actual utility value is considered, enabling the policy network to generate action suggestions that better match the player's ability and the game state.

[0157] After multiple weight updates and training iterations, when the performance of the reinforcement learning policy network and value network reaches a stable state—that is, when the loss function value no longer decreases significantly or the preset convergence condition is met—the relevant weight parameters in the policy network and value network are extracted. These weight parameters, after adjustment and optimization, can more accurately reflect the relationship between player ability and game operations. They are then applied to the player ability assessment model, outputting the optimized weight parameters for evaluating player ability and adjusting game difficulty.

[0158] Reference Figure 2An embodiment of the present invention provides an adaptive design system 2 for puzzle games based on ecological balance, wherein system 2 specifically includes: The ability assessment module 201 is used to receive the ecological scene data selected by the player, combine it with the player ability assessment model to assess the player's ability level, and generate assessment results; The behavior modeling module 202 is used to dynamically generate hierarchical ecological problems based on the evaluation results, and synchronously activate the corresponding level of Agent behavior tree loader to form an Agent modeling system. The ecological evolution module 203 is used to simulate biological interaction networks through the Agent modeling system, and input the real-time operational data of the biological interaction network into the LSTM dual time series prediction engine for prediction processing to generate an ecological evolution map. Difficulty adjustment module 204 is used to generate a dynamic visualization interface based on the ecological evolution map. When the ecological imbalance index is detected to exceed the current player difficulty threshold, the difficulty adjuster is triggered. The parameter optimization module 205 is used to update the ability evaluation parameters based on the topological integrity of the player operation chain data, and to optimize the player ability evaluation model by combining reinforcement learning algorithms.

[0159] It is understandable that, such as Figure 1 The content of the illustrated ecological balance-based adaptive design method embodiment for puzzle games is applicable to this ecological balance-based adaptive design system embodiment. The specific functions implemented by this ecological balance-based adaptive design system embodiment are the same as those shown below. Figure 1 The illustrated example of the adaptive design method for puzzle games based on ecological balance is the same, and the beneficial effects achieved are the same as those shown. Figure 1 The beneficial effects achieved by the illustrated example of the adaptive design method for puzzle games based on ecological balance are also the same.

[0160] It should be noted that the information interaction and execution process between the above systems are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0161] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0162] Reference Figure 3 The present invention also provides a computer device 3, including: a memory 302 and a processor 301, and a computer program 303 stored on the memory 302. When the computer program 303 is executed on the processor 301, it implements the ecological balance-based adaptive design method for puzzle games as described in any of the above methods.

[0163] The computer device 3 may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will understand that... Figure 3 The computer device 3 is merely an example and does not constitute a limitation on the computer device 3. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0164] The processor 301 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0165] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may be an external storage device of the computer device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 3. Furthermore, the memory 302 may include both internal and external storage units of the computer device 3. The memory 302 is used to store the operating system, applications, boot loader, data, and other programs, such as the program code of the computer program. The memory 302 can also be used to temporarily store data that has been output or will be output.

[0166] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the ecological balance-based adaptive design method for puzzle games as described in any of the above methods.

[0167] In this embodiment, if the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0168] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0169] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0170] In the embodiments disclosed in this application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0171] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

Claims

1. An adaptive design method for puzzle games based on ecological balance, characterized in that, The method specifically includes: Receive ecological scene data selected by the player, combine it with the player ability assessment model to evaluate the player's ability level, and generate assessment results; Based on the assessment results, hierarchical ecological problems are dynamically generated, and the corresponding Agent behavior tree loaders are activated simultaneously to form an Agent modeling system. The biological interaction network is simulated by an agent modeling system, and the real-time operational data of the biological interaction network is input into the LSTM dual time series prediction engine for prediction processing to generate an ecological evolution map. A dynamic visualization interface is generated based on the ecological evolution map. When the ecological imbalance index is detected to exceed the current player difficulty threshold, the difficulty adjuster is triggered. The ability assessment parameters are updated based on the topological integrity of the player's action chain data, and the player ability assessment model is optimized by combining reinforcement learning algorithms.

2. The method according to claim 1, characterized in that, The process of receiving the ecological scene data selected by the player, combining it with the player ability assessment model to evaluate the player's ability level, and generating an assessment result specifically includes: Receive ecological scene data selected by the player and collect historical operation sequences in real time through the operation chain capture device; Based on historical operation sequences, calculate strategy complexity, operation accuracy, and stability metrics. The strategy complexity metric, operational accuracy metric, and stability metric are input into the vector mapping engine to generate a normalized player ability vector. Player ability vectors are input into a pre-defined player ability assessment model for processing, resulting in an assessment result.

3. The method according to claim 2, characterized in that, The process of inputting player ability vectors into a preset player ability evaluation model for processing to form evaluation results specifically includes: It receives the input of the player's ability vector and deconstructs it into a strategy complexity feature dimension, an operation accuracy feature dimension, and a stability feature dimension according to a preset weight ratio; The strategy complexity feature dimension, operation accuracy feature dimension, and stability feature dimension are input into the pre-built capability classification model for processing, and basic capability classification results are generated. Based on the basic ability classification results, a dimensional balance analysis and potential ability prediction are performed using a pre-built player ability assessment model, generating balance analysis results and potential ability prediction results. A subject imbalance index is generated based on the results of the balance analysis. When the subject imbalance index exceeds the preset subject imbalance index threshold, the ability correction requirement is marked. The growth coefficient is determined based on the potential ability prediction results. The ability correction needs and the growth coefficient are combined to calculate and output the player's ability rating, and the evaluation result is obtained.

4. The method according to claim 2, characterized in that, The process of dynamically generating hierarchical ecological problems based on the evaluation results and simultaneously activating the corresponding agent behavior tree loader to form an agent modeling system specifically includes: The strategy complexity index in the evaluation results is analyzed, and a basic problem framework is selected from the ecological problem template library based on the strategy complexity index. The operational accuracy indicators in the evaluation results are analyzed, and dynamic perturbation parameters are injected into the basic problem framework based on the operational accuracy indicators to form a hierarchical ecological problem. Analyze the stability indicators in the evaluation results and determine the depth of the Agent behavior tree based on the stability indicators; Modular behavior units are configured based on the depth of the Agent behavior tree to generate a hierarchical behavior tree; By coupling hierarchical ecological problems with hierarchical behavior trees, an Agent modeling system is formed that includes behavior tree depth parameters, ecological tolerance thresholds, and early warning time windows.

5. The method according to claim 1, characterized in that, The process of simulating a biological interaction network using an agent-based modeling system and inputting the real-time operational data of the biological interaction network into an LSTM dual-time-series prediction engine for prediction processing to generate an ecological evolution map specifically includes: Simulate biological interaction networks using an agent-based modeling system to generate real-time streams of species interaction events. Convert species interaction event streams into time-series structured data and encapsulate them into sliding time window data packets; The sliding time window data packets are input into the LSTM dual time series prediction engine to perform short-term ecological change prediction and obtain short-term prediction results. Based on the short-term forecast results, long-term ecosystem effect forecasts are triggered, generating long-term forecast results. The short-term and long-term forecast results are then merged to form an ecological evolution map.

6. The method according to claim 5, characterized in that, The process of triggering long-term ecosystem chain effect prediction based on short-term prediction results, generating long-term prediction results, and merging the short-term and long-term prediction results to form an ecological evolution map specifically includes: Receive short-term ecological prediction results and determine the calculation path for long-term predictions based on the direction of change in the short-term prediction results; When the rate of change of a key species is identified as positive and exceeds a preset first rate of change threshold, the reproductive chain prediction process is performed through the calculation path to obtain the first long-term prediction result. When the rate of change of a key species is identified as negative and exceeds the preset second rate of change threshold, extinction chain prediction processing is performed through the calculation path to obtain the second long-term prediction result. The short-term forecast results and the first or second long-term forecast results are aligned along the time axis, and the data at the connection point are smoothed and calibrated using the species association strength coefficient to generate an ecological evolution map. In this ecological evolution map, the node size is used to represent the population size ratio of species, the node color is used to represent the species change trend, the line thickness of the ecological evolution map is used to represent the energy transfer intensity of species interaction, and the line style is used to represent the predation type or competition type of species interaction.

7. The method according to claim 1, characterized in that, The dynamic visualization interface generated based on the ecological evolution map, when the ecological imbalance index is detected to exceed the current player difficulty threshold, triggers a difficulty adjuster, specifically including: Real-time monitoring of early warning signals in ecological evolution maps, extracting multi-dimensional elements including species extinction risk, number of food chain breaks and resource depletion rate; The real-time ecological imbalance index is calculated based on multi-dimensional factors and compared with the current player difficulty threshold; When the ecological imbalance index is detected to exceed the current player difficulty threshold, the corresponding difficulty adjustment strategy is selected based on the degree of excess, according to the difficulty regulator. Implement a difficulty adjustment strategy, generate new difficulty parameters, and synchronously feed these new difficulty parameters back to the game system.

8. The method according to claim 1, characterized in that, The process of updating the ability assessment parameters based on the topological integrity of the player's action chain data, and optimizing the player ability assessment model using a reinforcement learning algorithm, specifically includes: Real-time capture of player action chain data; analysis of nodes, paths, and loops in the topology of player action chain data. Based on nodes, paths, and loops in the topology, a multi-dimensional index including node coverage, path completeness, and loop health value is calculated. The player's ability parameters are dynamically adjusted based on multi-dimensional indicators to generate an updated ability vector. The updated ability vectors are input into a reinforcement learning network for processing, and the weight parameters of the player ability evaluation model are optimized based on the feedback results from the reinforcement learning network.

9. The method according to claim 8, characterized in that, The process of inputting the updated ability vector into a reinforcement learning network for processing, and optimizing the weight parameters of the player ability evaluation model based on the feedback results of the reinforcement learning network, specifically includes: The updated player ability vectors are input into the reinforcement learning policy network to generate action suggestions and their weight distributions. Based on the operational suggestions and their weight distribution, the actual utility value of the operational suggestions is calculated by analyzing the performance data of subsequent levels through reinforcement learning value networks. Feedback results are generated based on actual utility values. A weight adjustment strategy is selected based on the feedback results. The weight update is performed using a gradient backpropagation mechanism, and the weight parameters of the optimized player ability evaluation model are output.

10. An adaptive design system for puzzle games based on ecological balance, characterized in that, The system specifically includes: The ability assessment module is used to receive ecological scenario data selected by players, combine it with the player ability assessment model to assess the player's ability level, and generate assessment results; The behavior modeling module is used to dynamically generate hierarchical ecological problems based on the evaluation results, and synchronously activate the corresponding level of Agent behavior tree loader to form an Agent modeling system. The ecological evolution module is used to simulate biological interaction networks through an agent modeling system, and input the real-time operational data of the biological interaction network into the LSTM dual time series prediction engine for prediction processing to generate an ecological evolution map. The difficulty adjustment module is used to generate a dynamic visualization interface based on the ecological evolution map. When the ecological imbalance index is detected to exceed the current player difficulty threshold, the difficulty adjuster is triggered. The parameter optimization module is used to update the ability evaluation parameters based on the topological integrity of the player's action chain data, and to optimize the player ability evaluation model by combining reinforcement learning algorithms.

Images